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>
67 #include <sys/signalvar.h>
68 #include <sys/socket.h>
69 #include <sys/socketvar.h>
70 #include <sys/sysctl.h>
71 #include <sys/syslog.h>
72 #include <sys/systm.h>
74 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
79 #include <net/if_var.h>
80 #include <net/route.h>
83 #define TCPSTATES /* for logging */
85 #include <netinet/cc.h>
86 #include <netinet/in.h>
87 #include <netinet/in_kdtrace.h>
88 #include <netinet/in_pcb.h>
89 #include <netinet/in_systm.h>
90 #include <netinet/in_var.h>
91 #include <netinet/ip.h>
92 #include <netinet/ip_icmp.h> /* required for icmp_var.h */
93 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */
94 #include <netinet/ip_var.h>
95 #include <netinet/ip_options.h>
96 #include <netinet/ip6.h>
97 #include <netinet/icmp6.h>
98 #include <netinet6/in6_pcb.h>
99 #include <netinet6/ip6_var.h>
100 #include <netinet6/nd6.h>
101 #include <netinet/tcp_fsm.h>
102 #include <netinet/tcp_seq.h>
103 #include <netinet/tcp_timer.h>
104 #include <netinet/tcp_var.h>
105 #include <netinet6/tcp6_var.h>
106 #include <netinet/tcpip.h>
107 #include <netinet/tcp_syncache.h>
109 #include <netinet/tcp_debug.h>
110 #endif /* TCPDEBUG */
112 #include <netinet/tcp_offload.h>
116 #include <netipsec/ipsec.h>
117 #include <netipsec/ipsec6.h>
120 #include <machine/in_cksum.h>
122 #include <security/mac/mac_framework.h>
124 const int tcprexmtthresh = 3;
126 int tcp_log_in_vain = 0;
127 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
129 "Log all incoming TCP segments to closed ports");
131 VNET_DEFINE(int, blackhole) = 0;
132 #define V_blackhole VNET(blackhole)
133 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW,
134 &VNET_NAME(blackhole), 0,
135 "Do not send RST on segments to closed ports");
137 VNET_DEFINE(int, tcp_delack_enabled) = 1;
138 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW,
139 &VNET_NAME(tcp_delack_enabled), 0,
140 "Delay ACK to try and piggyback it onto a data packet");
142 VNET_DEFINE(int, drop_synfin) = 0;
143 #define V_drop_synfin VNET(drop_synfin)
144 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW,
145 &VNET_NAME(drop_synfin), 0,
146 "Drop TCP packets with SYN+FIN set");
148 VNET_DEFINE(int, tcp_do_rfc3042) = 1;
149 #define V_tcp_do_rfc3042 VNET(tcp_do_rfc3042)
150 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_RW,
151 &VNET_NAME(tcp_do_rfc3042), 0,
152 "Enable RFC 3042 (Limited Transmit)");
154 VNET_DEFINE(int, tcp_do_rfc3390) = 1;
155 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW,
156 &VNET_NAME(tcp_do_rfc3390), 0,
157 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
159 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, experimental, CTLFLAG_RW, 0,
160 "Experimental TCP extensions");
162 VNET_DEFINE(int, tcp_do_initcwnd10) = 1;
163 SYSCTL_VNET_INT(_net_inet_tcp_experimental, OID_AUTO, initcwnd10, CTLFLAG_RW,
164 &VNET_NAME(tcp_do_initcwnd10), 0,
165 "Enable RFC 6928 (Increasing initial CWND to 10)");
167 VNET_DEFINE(int, tcp_do_rfc3465) = 1;
168 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_RW,
169 &VNET_NAME(tcp_do_rfc3465), 0,
170 "Enable RFC 3465 (Appropriate Byte Counting)");
172 VNET_DEFINE(int, tcp_abc_l_var) = 2;
173 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_RW,
174 &VNET_NAME(tcp_abc_l_var), 2,
175 "Cap the max cwnd increment during slow-start to this number of segments");
177 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, CTLFLAG_RW, 0, "TCP ECN");
179 VNET_DEFINE(int, tcp_do_ecn) = 0;
180 SYSCTL_VNET_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_RW,
181 &VNET_NAME(tcp_do_ecn), 0,
184 VNET_DEFINE(int, tcp_ecn_maxretries) = 1;
185 SYSCTL_VNET_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_RW,
186 &VNET_NAME(tcp_ecn_maxretries), 0,
187 "Max retries before giving up on ECN");
189 VNET_DEFINE(int, tcp_insecure_rst) = 0;
190 #define V_tcp_insecure_rst VNET(tcp_insecure_rst)
191 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_RW,
192 &VNET_NAME(tcp_insecure_rst), 0,
193 "Follow the old (insecure) criteria for accepting RST packets");
195 VNET_DEFINE(int, tcp_recvspace) = 1024*64;
196 #define V_tcp_recvspace VNET(tcp_recvspace)
197 SYSCTL_VNET_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
198 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size");
200 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1;
201 #define V_tcp_do_autorcvbuf VNET(tcp_do_autorcvbuf)
202 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_RW,
203 &VNET_NAME(tcp_do_autorcvbuf), 0,
204 "Enable automatic receive buffer sizing");
206 VNET_DEFINE(int, tcp_autorcvbuf_inc) = 16*1024;
207 #define V_tcp_autorcvbuf_inc VNET(tcp_autorcvbuf_inc)
208 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_RW,
209 &VNET_NAME(tcp_autorcvbuf_inc), 0,
210 "Incrementor step size of automatic receive buffer");
212 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024;
213 #define V_tcp_autorcvbuf_max VNET(tcp_autorcvbuf_max)
214 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_RW,
215 &VNET_NAME(tcp_autorcvbuf_max), 0,
216 "Max size of automatic receive buffer");
218 VNET_DEFINE(struct inpcbhead, tcb);
219 #define tcb6 tcb /* for KAME src sync over BSD*'s */
220 VNET_DEFINE(struct inpcbinfo, tcbinfo);
222 static void tcp_dooptions(struct tcpopt *, u_char *, int, int);
223 static void tcp_do_segment(struct mbuf *, struct tcphdr *,
224 struct socket *, struct tcpcb *, int, int, uint8_t,
226 static void tcp_dropwithreset(struct mbuf *, struct tcphdr *,
227 struct tcpcb *, int, int);
228 static void tcp_pulloutofband(struct socket *,
229 struct tcphdr *, struct mbuf *, int);
230 static void tcp_xmit_timer(struct tcpcb *, int);
231 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *);
232 static void inline cc_ack_received(struct tcpcb *tp, struct tcphdr *th,
234 static void inline cc_conn_init(struct tcpcb *tp);
235 static void inline cc_post_recovery(struct tcpcb *tp, struct tcphdr *th);
236 static void inline hhook_run_tcp_est_in(struct tcpcb *tp,
237 struct tcphdr *th, struct tcpopt *to);
240 * TCP statistics are stored in an "array" of counter(9)s.
242 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat);
243 VNET_PCPUSTAT_SYSINIT(tcpstat);
244 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat,
245 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
248 VNET_PCPUSTAT_SYSUNINIT(tcpstat);
251 * Kernel module interface for updating tcpstat. The argument is an index
252 * into tcpstat treated as an array.
255 kmod_tcpstat_inc(int statnum)
258 counter_u64_add(VNET(tcpstat)[statnum], 1);
262 * Wrapper for the TCP established input helper hook.
265 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to)
267 struct tcp_hhook_data hhook_data;
269 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) {
274 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data,
280 * CC wrapper hook functions
283 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t type)
285 INP_WLOCK_ASSERT(tp->t_inpcb);
287 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th);
288 if (tp->snd_cwnd <= tp->snd_wnd)
289 tp->ccv->flags |= CCF_CWND_LIMITED;
291 tp->ccv->flags &= ~CCF_CWND_LIMITED;
293 if (type == CC_ACK) {
294 if (tp->snd_cwnd > tp->snd_ssthresh) {
295 tp->t_bytes_acked += min(tp->ccv->bytes_this_ack,
296 V_tcp_abc_l_var * tp->t_maxseg);
297 if (tp->t_bytes_acked >= tp->snd_cwnd) {
298 tp->t_bytes_acked -= tp->snd_cwnd;
299 tp->ccv->flags |= CCF_ABC_SENTAWND;
302 tp->ccv->flags &= ~CCF_ABC_SENTAWND;
303 tp->t_bytes_acked = 0;
307 if (CC_ALGO(tp)->ack_received != NULL) {
308 /* XXXLAS: Find a way to live without this */
309 tp->ccv->curack = th->th_ack;
310 CC_ALGO(tp)->ack_received(tp->ccv, type);
315 cc_conn_init(struct tcpcb *tp)
317 struct hc_metrics_lite metrics;
318 struct inpcb *inp = tp->t_inpcb;
321 INP_WLOCK_ASSERT(tp->t_inpcb);
323 tcp_hc_get(&inp->inp_inc, &metrics);
325 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) {
327 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
328 TCPSTAT_INC(tcps_usedrtt);
329 if (metrics.rmx_rttvar) {
330 tp->t_rttvar = metrics.rmx_rttvar;
331 TCPSTAT_INC(tcps_usedrttvar);
333 /* default variation is +- 1 rtt */
335 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
337 TCPT_RANGESET(tp->t_rxtcur,
338 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
339 tp->t_rttmin, TCPTV_REXMTMAX);
341 if (metrics.rmx_ssthresh) {
343 * There's some sort of gateway or interface
344 * buffer limit on the path. Use this to set
345 * the slow start threshhold, but set the
346 * threshold to no less than 2*mss.
348 tp->snd_ssthresh = max(2 * tp->t_maxseg, metrics.rmx_ssthresh);
349 TCPSTAT_INC(tcps_usedssthresh);
353 * Set the initial slow-start flight size.
355 * RFC5681 Section 3.1 specifies the default conservative values.
356 * RFC3390 specifies slightly more aggressive values.
357 * RFC6928 increases it to ten segments.
359 * If a SYN or SYN/ACK was lost and retransmitted, we have to
360 * reduce the initial CWND to one segment as congestion is likely
361 * requiring us to be cautious.
363 if (tp->snd_cwnd == 1)
364 tp->snd_cwnd = tp->t_maxseg; /* SYN(-ACK) lost */
365 else if (V_tcp_do_initcwnd10)
366 tp->snd_cwnd = min(10 * tp->t_maxseg,
367 max(2 * tp->t_maxseg, 14600));
368 else if (V_tcp_do_rfc3390)
369 tp->snd_cwnd = min(4 * tp->t_maxseg,
370 max(2 * tp->t_maxseg, 4380));
372 /* Per RFC5681 Section 3.1 */
373 if (tp->t_maxseg > 2190)
374 tp->snd_cwnd = 2 * tp->t_maxseg;
375 else if (tp->t_maxseg > 1095)
376 tp->snd_cwnd = 3 * tp->t_maxseg;
378 tp->snd_cwnd = 4 * tp->t_maxseg;
381 if (CC_ALGO(tp)->conn_init != NULL)
382 CC_ALGO(tp)->conn_init(tp->ccv);
386 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type)
388 INP_WLOCK_ASSERT(tp->t_inpcb);
392 if (!IN_FASTRECOVERY(tp->t_flags)) {
393 tp->snd_recover = tp->snd_max;
394 if (tp->t_flags & TF_ECN_PERMIT)
395 tp->t_flags |= TF_ECN_SND_CWR;
399 if (!IN_CONGRECOVERY(tp->t_flags)) {
400 TCPSTAT_INC(tcps_ecn_rcwnd);
401 tp->snd_recover = tp->snd_max;
402 if (tp->t_flags & TF_ECN_PERMIT)
403 tp->t_flags |= TF_ECN_SND_CWR;
408 tp->t_bytes_acked = 0;
409 EXIT_RECOVERY(tp->t_flags);
410 tp->snd_ssthresh = max(2, min(tp->snd_wnd, tp->snd_cwnd) / 2 /
411 tp->t_maxseg) * tp->t_maxseg;
412 tp->snd_cwnd = tp->t_maxseg;
415 TCPSTAT_INC(tcps_sndrexmitbad);
416 /* RTO was unnecessary, so reset everything. */
417 tp->snd_cwnd = tp->snd_cwnd_prev;
418 tp->snd_ssthresh = tp->snd_ssthresh_prev;
419 tp->snd_recover = tp->snd_recover_prev;
420 if (tp->t_flags & TF_WASFRECOVERY)
421 ENTER_FASTRECOVERY(tp->t_flags);
422 if (tp->t_flags & TF_WASCRECOVERY)
423 ENTER_CONGRECOVERY(tp->t_flags);
424 tp->snd_nxt = tp->snd_max;
425 tp->t_flags &= ~TF_PREVVALID;
430 if (CC_ALGO(tp)->cong_signal != NULL) {
432 tp->ccv->curack = th->th_ack;
433 CC_ALGO(tp)->cong_signal(tp->ccv, type);
438 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th)
440 INP_WLOCK_ASSERT(tp->t_inpcb);
442 /* XXXLAS: KASSERT that we're in recovery? */
444 if (CC_ALGO(tp)->post_recovery != NULL) {
445 tp->ccv->curack = th->th_ack;
446 CC_ALGO(tp)->post_recovery(tp->ccv);
448 /* XXXLAS: EXIT_RECOVERY ? */
449 tp->t_bytes_acked = 0;
454 tcp_signature_verify_input(struct mbuf *m, int off0, int tlen, int optlen,
455 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag)
459 tcp_fields_to_net(th);
460 ret = tcp_signature_verify(m, off0, tlen, optlen, to, th, tcpbflag);
461 tcp_fields_to_host(th);
466 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
468 #define ND6_HINT(tp) \
470 if ((tp) && (tp)->t_inpcb && \
471 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \
472 nd6_nud_hint(NULL, NULL, 0); \
479 * Indicate whether this ack should be delayed. We can delay the ack if
480 * following conditions are met:
481 * - There is no delayed ack timer in progress.
482 * - Our last ack wasn't a 0-sized window. We never want to delay
483 * the ack that opens up a 0-sized window.
484 * - LRO wasn't used for this segment. We make sure by checking that the
485 * segment size is not larger than the MSS.
486 * - Delayed acks are enabled or this is a half-synchronized T/TCP
489 #define DELAY_ACK(tp, tlen) \
490 ((!tcp_timer_active(tp, TT_DELACK) && \
491 (tp->t_flags & TF_RXWIN0SENT) == 0) && \
492 (tlen <= tp->t_maxopd) && \
493 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
496 * TCP input handling is split into multiple parts:
497 * tcp6_input is a thin wrapper around tcp_input for the extended
498 * ip6_protox[] call format in ip6_input
499 * tcp_input handles primary segment validation, inpcb lookup and
500 * SYN processing on listen sockets
501 * tcp_do_segment processes the ACK and text of the segment for
502 * establishing, established and closing connections
506 tcp6_input(struct mbuf **mp, int *offp, int proto)
508 struct mbuf *m = *mp;
509 struct in6_ifaddr *ia6;
511 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
514 * draft-itojun-ipv6-tcp-to-anycast
515 * better place to put this in?
517 ia6 = ip6_getdstifaddr(m);
518 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
521 ifa_free(&ia6->ia_ifa);
522 ip6 = mtod(m, struct ip6_hdr *);
523 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
524 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
525 return (IPPROTO_DONE);
528 ifa_free(&ia6->ia_ifa);
530 return (tcp_input(mp, offp, proto));
535 tcp_input(struct mbuf **mp, int *offp, int proto)
537 struct mbuf *m = *mp;
538 struct tcphdr *th = NULL;
539 struct ip *ip = NULL;
540 struct inpcb *inp = NULL;
541 struct tcpcb *tp = NULL;
542 struct socket *so = NULL;
552 int rstreason = 0; /* For badport_bandlim accounting purposes */
554 uint8_t sig_checked = 0;
557 struct m_tag *fwd_tag = NULL;
559 struct ip6_hdr *ip6 = NULL;
562 const void *ip6 = NULL;
564 struct tcpopt to; /* options in this segment */
565 char *s = NULL; /* address and port logging */
567 #define TI_UNLOCKED 1
572 * The size of tcp_saveipgen must be the size of the max ip header,
575 u_char tcp_saveipgen[IP6_HDR_LEN];
576 struct tcphdr tcp_savetcp;
581 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
588 TCPSTAT_INC(tcps_rcvtotal);
592 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */
594 if (m->m_len < (sizeof(*ip6) + sizeof(*th))) {
595 m = m_pullup(m, sizeof(*ip6) + sizeof(*th));
597 TCPSTAT_INC(tcps_rcvshort);
598 return (IPPROTO_DONE);
602 ip6 = mtod(m, struct ip6_hdr *);
603 th = (struct tcphdr *)((caddr_t)ip6 + off0);
604 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
605 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) {
606 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
607 th->th_sum = m->m_pkthdr.csum_data;
609 th->th_sum = in6_cksum_pseudo(ip6, tlen,
610 IPPROTO_TCP, m->m_pkthdr.csum_data);
611 th->th_sum ^= 0xffff;
613 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen);
615 TCPSTAT_INC(tcps_rcvbadsum);
620 * Be proactive about unspecified IPv6 address in source.
621 * As we use all-zero to indicate unbounded/unconnected pcb,
622 * unspecified IPv6 address can be used to confuse us.
624 * Note that packets with unspecified IPv6 destination is
625 * already dropped in ip6_input.
627 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
633 #if defined(INET) && defined(INET6)
639 * Get IP and TCP header together in first mbuf.
640 * Note: IP leaves IP header in first mbuf.
642 if (off0 > sizeof (struct ip)) {
644 off0 = sizeof(struct ip);
646 if (m->m_len < sizeof (struct tcpiphdr)) {
647 if ((m = m_pullup(m, sizeof (struct tcpiphdr)))
649 TCPSTAT_INC(tcps_rcvshort);
650 return (IPPROTO_DONE);
653 ip = mtod(m, struct ip *);
654 th = (struct tcphdr *)((caddr_t)ip + off0);
655 tlen = ntohs(ip->ip_len) - off0;
657 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
658 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
659 th->th_sum = m->m_pkthdr.csum_data;
661 th->th_sum = in_pseudo(ip->ip_src.s_addr,
663 htonl(m->m_pkthdr.csum_data + tlen +
665 th->th_sum ^= 0xffff;
667 struct ipovly *ipov = (struct ipovly *)ip;
670 * Checksum extended TCP header and data.
673 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
674 ipov->ih_len = htons(tlen);
675 th->th_sum = in_cksum(m, len);
676 /* Reset length for SDT probes. */
677 ip->ip_len = htons(tlen + off0);
681 TCPSTAT_INC(tcps_rcvbadsum);
684 /* Re-initialization for later version check */
685 ip->ip_v = IPVERSION;
691 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
693 #if defined(INET) && defined(INET6)
701 * Check that TCP offset makes sense,
702 * pull out TCP options and adjust length. XXX
704 off = th->th_off << 2;
705 if (off < sizeof (struct tcphdr) || off > tlen) {
706 TCPSTAT_INC(tcps_rcvbadoff);
709 tlen -= off; /* tlen is used instead of ti->ti_len */
710 if (off > sizeof (struct tcphdr)) {
713 IP6_EXTHDR_CHECK(m, off0, off, IPPROTO_DONE);
714 ip6 = mtod(m, struct ip6_hdr *);
715 th = (struct tcphdr *)((caddr_t)ip6 + off0);
718 #if defined(INET) && defined(INET6)
723 if (m->m_len < sizeof(struct ip) + off) {
724 if ((m = m_pullup(m, sizeof (struct ip) + off))
726 TCPSTAT_INC(tcps_rcvshort);
727 return (IPPROTO_DONE);
729 ip = mtod(m, struct ip *);
730 th = (struct tcphdr *)((caddr_t)ip + off0);
734 optlen = off - sizeof (struct tcphdr);
735 optp = (u_char *)(th + 1);
737 thflags = th->th_flags;
740 * Convert TCP protocol specific fields to host format.
742 tcp_fields_to_host(th);
745 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options.
747 drop_hdrlen = off0 + off;
750 * Locate pcb for segment; if we're likely to add or remove a
751 * connection then first acquire pcbinfo lock. There are three cases
752 * where we might discover later we need a write lock despite the
753 * flags: ACKs moving a connection out of the syncache, ACKs for a
754 * connection in TIMEWAIT and SYNs not targeting a listening socket.
756 if ((thflags & (TH_FIN | TH_RST)) != 0) {
757 INP_INFO_WLOCK(&V_tcbinfo);
758 ti_locked = TI_WLOCKED;
760 ti_locked = TI_UNLOCKED;
763 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
767 (isipv6 && (m->m_flags & M_IP6_NEXTHOP))
769 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP))
772 #if defined(INET) && !defined(INET6)
773 (m->m_flags & M_IP_NEXTHOP)
776 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
780 if (ti_locked == TI_WLOCKED) {
781 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
783 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
787 if (isipv6 && fwd_tag != NULL) {
788 struct sockaddr_in6 *next_hop6;
790 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1);
792 * Transparently forwarded. Pretend to be the destination.
793 * Already got one like this?
795 inp = in6_pcblookup_mbuf(&V_tcbinfo,
796 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport,
797 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m);
800 * It's new. Try to find the ambushing socket.
801 * Because we've rewritten the destination address,
802 * any hardware-generated hash is ignored.
804 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src,
805 th->th_sport, &next_hop6->sin6_addr,
806 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) :
807 th->th_dport, INPLOOKUP_WILDCARD |
808 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
811 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src,
812 th->th_sport, &ip6->ip6_dst, th->th_dport,
813 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
814 m->m_pkthdr.rcvif, m);
817 #if defined(INET6) && defined(INET)
821 if (fwd_tag != NULL) {
822 struct sockaddr_in *next_hop;
824 next_hop = (struct sockaddr_in *)(fwd_tag+1);
826 * Transparently forwarded. Pretend to be the destination.
827 * already got one like this?
829 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport,
830 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB,
831 m->m_pkthdr.rcvif, m);
834 * It's new. Try to find the ambushing socket.
835 * Because we've rewritten the destination address,
836 * any hardware-generated hash is ignored.
838 inp = in_pcblookup(&V_tcbinfo, ip->ip_src,
839 th->th_sport, next_hop->sin_addr,
840 next_hop->sin_port ? ntohs(next_hop->sin_port) :
841 th->th_dport, INPLOOKUP_WILDCARD |
842 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
845 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src,
846 th->th_sport, ip->ip_dst, th->th_dport,
847 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
848 m->m_pkthdr.rcvif, m);
852 * If the INPCB does not exist then all data in the incoming
853 * segment is discarded and an appropriate RST is sent back.
854 * XXX MRT Send RST using which routing table?
858 * Log communication attempts to ports that are not
861 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) ||
862 tcp_log_in_vain == 2) {
863 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6)))
864 log(LOG_INFO, "%s; %s: Connection attempt "
865 "to closed port\n", s, __func__);
868 * When blackholing do not respond with a RST but
869 * completely ignore the segment and drop it.
871 if ((V_blackhole == 1 && (thflags & TH_SYN)) ||
875 rstreason = BANDLIM_RST_CLOSEDPORT;
878 INP_WLOCK_ASSERT(inp);
879 if (!(inp->inp_flags & INP_HW_FLOWID)
880 && (m->m_flags & M_FLOWID)
881 && ((inp->inp_socket == NULL)
882 || !(inp->inp_socket->so_options & SO_ACCEPTCONN))) {
883 inp->inp_flags |= INP_HW_FLOWID;
884 inp->inp_flags &= ~INP_SW_FLOWID;
885 inp->inp_flowid = m->m_pkthdr.flowid;
886 inp->inp_flowtype = M_HASHTYPE_GET(m);
890 if (isipv6 && ipsec6_in_reject(m, inp)) {
891 IPSEC6STAT_INC(ips_in_polvio);
895 if (ipsec4_in_reject(m, inp) != 0) {
896 IPSECSTAT_INC(ips_in_polvio);
902 * Check the minimum TTL for socket.
904 if (inp->inp_ip_minttl != 0) {
906 if (isipv6 && inp->inp_ip_minttl > ip6->ip6_hlim)
910 if (inp->inp_ip_minttl > ip->ip_ttl)
915 * A previous connection in TIMEWAIT state is supposed to catch stray
916 * or duplicate segments arriving late. If this segment was a
917 * legitimate new connection attempt, the old INPCB gets removed and
918 * we can try again to find a listening socket.
920 * At this point, due to earlier optimism, we may hold only an inpcb
921 * lock, and not the inpcbinfo write lock. If so, we need to try to
922 * acquire it, or if that fails, acquire a reference on the inpcb,
923 * drop all locks, acquire a global write lock, and then re-acquire
924 * the inpcb lock. We may at that point discover that another thread
925 * has tried to free the inpcb, in which case we need to loop back
926 * and try to find a new inpcb to deliver to.
928 * XXXRW: It may be time to rethink timewait locking.
931 if (inp->inp_flags & INP_TIMEWAIT) {
932 if (ti_locked == TI_UNLOCKED) {
933 if (INP_INFO_TRY_WLOCK(&V_tcbinfo) == 0) {
936 INP_INFO_WLOCK(&V_tcbinfo);
937 ti_locked = TI_WLOCKED;
939 if (in_pcbrele_wlocked(inp)) {
944 ti_locked = TI_WLOCKED;
946 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
948 if (thflags & TH_SYN)
949 tcp_dooptions(&to, optp, optlen, TO_SYN);
951 * NB: tcp_twcheck unlocks the INP and frees the mbuf.
953 if (tcp_twcheck(inp, &to, th, m, tlen))
955 INP_INFO_WUNLOCK(&V_tcbinfo);
956 return (IPPROTO_DONE);
959 * The TCPCB may no longer exist if the connection is winding
960 * down or it is in the CLOSED state. Either way we drop the
961 * segment and send an appropriate response.
964 if (tp == NULL || tp->t_state == TCPS_CLOSED) {
965 rstreason = BANDLIM_RST_CLOSEDPORT;
970 if (tp->t_flags & TF_TOE) {
971 tcp_offload_input(tp, m);
972 m = NULL; /* consumed by the TOE driver */
978 * We've identified a valid inpcb, but it could be that we need an
979 * inpcbinfo write lock but don't hold it. In this case, attempt to
980 * acquire using the same strategy as the TIMEWAIT case above. If we
981 * relock, we have to jump back to 'relocked' as the connection might
982 * now be in TIMEWAIT.
985 if ((thflags & (TH_FIN | TH_RST)) != 0)
986 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
988 if (!((tp->t_state == TCPS_ESTABLISHED && (thflags & TH_SYN) == 0) ||
989 (tp->t_state == TCPS_LISTEN && (thflags & TH_SYN)))) {
990 if (ti_locked == TI_UNLOCKED) {
991 if (INP_INFO_TRY_WLOCK(&V_tcbinfo) == 0) {
994 INP_INFO_WLOCK(&V_tcbinfo);
995 ti_locked = TI_WLOCKED;
997 if (in_pcbrele_wlocked(inp)) {
1003 ti_locked = TI_WLOCKED;
1005 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1009 INP_WLOCK_ASSERT(inp);
1010 if (mac_inpcb_check_deliver(inp, m))
1013 so = inp->inp_socket;
1014 KASSERT(so != NULL, ("%s: so == NULL", __func__));
1016 if (so->so_options & SO_DEBUG) {
1017 ostate = tp->t_state;
1020 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
1023 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
1026 #endif /* TCPDEBUG */
1028 * When the socket is accepting connections (the INPCB is in LISTEN
1029 * state) we look into the SYN cache if this is a new connection
1030 * attempt or the completion of a previous one.
1032 if (so->so_options & SO_ACCEPTCONN) {
1033 struct in_conninfo inc;
1035 KASSERT(tp->t_state == TCPS_LISTEN, ("%s: so accepting but "
1036 "tp not listening", __func__));
1037 bzero(&inc, sizeof(inc));
1040 inc.inc_flags |= INC_ISIPV6;
1041 inc.inc6_faddr = ip6->ip6_src;
1042 inc.inc6_laddr = ip6->ip6_dst;
1046 inc.inc_faddr = ip->ip_src;
1047 inc.inc_laddr = ip->ip_dst;
1049 inc.inc_fport = th->th_sport;
1050 inc.inc_lport = th->th_dport;
1051 inc.inc_fibnum = so->so_fibnum;
1054 * Check for an existing connection attempt in syncache if
1055 * the flag is only ACK. A successful lookup creates a new
1056 * socket appended to the listen queue in SYN_RECEIVED state.
1058 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
1060 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1062 * Parse the TCP options here because
1063 * syncookies need access to the reflected
1066 tcp_dooptions(&to, optp, optlen, 0);
1068 * NB: syncache_expand() doesn't unlock
1069 * inp and tcpinfo locks.
1071 if (!syncache_expand(&inc, &to, th, &so, m)) {
1073 * No syncache entry or ACK was not
1074 * for our SYN/ACK. Send a RST.
1075 * NB: syncache did its own logging
1076 * of the failure cause.
1078 rstreason = BANDLIM_RST_OPENPORT;
1083 * We completed the 3-way handshake
1084 * but could not allocate a socket
1085 * either due to memory shortage,
1086 * listen queue length limits or
1087 * global socket limits. Send RST
1088 * or wait and have the remote end
1089 * retransmit the ACK for another
1092 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1093 log(LOG_DEBUG, "%s; %s: Listen socket: "
1094 "Socket allocation failed due to "
1095 "limits or memory shortage, %s\n",
1097 V_tcp_sc_rst_sock_fail ?
1098 "sending RST" : "try again");
1099 if (V_tcp_sc_rst_sock_fail) {
1100 rstreason = BANDLIM_UNLIMITED;
1106 * Socket is created in state SYN_RECEIVED.
1107 * Unlock the listen socket, lock the newly
1108 * created socket and update the tp variable.
1110 INP_WUNLOCK(inp); /* listen socket */
1111 inp = sotoinpcb(so);
1112 INP_WLOCK(inp); /* new connection */
1113 tp = intotcpcb(inp);
1114 KASSERT(tp->t_state == TCPS_SYN_RECEIVED,
1115 ("%s: ", __func__));
1116 #ifdef TCP_SIGNATURE
1117 if (sig_checked == 0) {
1118 tcp_dooptions(&to, optp, optlen,
1119 (thflags & TH_SYN) ? TO_SYN : 0);
1120 if (!tcp_signature_verify_input(m, off0, tlen,
1121 optlen, &to, th, tp->t_flags)) {
1124 * In SYN_SENT state if it receives an
1125 * RST, it is allowed for further
1128 if ((thflags & TH_RST) == 0 ||
1129 (tp->t_state == TCPS_SYN_SENT) == 0)
1137 * Process the segment and the data it
1138 * contains. tcp_do_segment() consumes
1139 * the mbuf chain and unlocks the inpcb.
1141 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen,
1143 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1144 return (IPPROTO_DONE);
1147 * Segment flag validation for new connection attempts:
1149 * Our (SYN|ACK) response was rejected.
1150 * Check with syncache and remove entry to prevent
1153 * NB: syncache_chkrst does its own logging of failure
1156 if (thflags & TH_RST) {
1157 syncache_chkrst(&inc, th);
1161 * We can't do anything without SYN.
1163 if ((thflags & TH_SYN) == 0) {
1164 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1165 log(LOG_DEBUG, "%s; %s: Listen socket: "
1166 "SYN is missing, segment ignored\n",
1168 TCPSTAT_INC(tcps_badsyn);
1172 * (SYN|ACK) is bogus on a listen socket.
1174 if (thflags & TH_ACK) {
1175 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1176 log(LOG_DEBUG, "%s; %s: Listen socket: "
1177 "SYN|ACK invalid, segment rejected\n",
1179 syncache_badack(&inc); /* XXX: Not needed! */
1180 TCPSTAT_INC(tcps_badsyn);
1181 rstreason = BANDLIM_RST_OPENPORT;
1185 * If the drop_synfin option is enabled, drop all
1186 * segments with both the SYN and FIN bits set.
1187 * This prevents e.g. nmap from identifying the
1189 * XXX: Poor reasoning. nmap has other methods
1190 * and is constantly refining its stack detection
1192 * XXX: This is a violation of the TCP specification
1193 * and was used by RFC1644.
1195 if ((thflags & TH_FIN) && V_drop_synfin) {
1196 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1197 log(LOG_DEBUG, "%s; %s: Listen socket: "
1198 "SYN|FIN segment ignored (based on "
1199 "sysctl setting)\n", s, __func__);
1200 TCPSTAT_INC(tcps_badsyn);
1204 * Segment's flags are (SYN) or (SYN|FIN).
1206 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored
1207 * as they do not affect the state of the TCP FSM.
1208 * The data pointed to by TH_URG and th_urp is ignored.
1210 KASSERT((thflags & (TH_RST|TH_ACK)) == 0,
1211 ("%s: Listen socket: TH_RST or TH_ACK set", __func__));
1212 KASSERT(thflags & (TH_SYN),
1213 ("%s: Listen socket: TH_SYN not set", __func__));
1216 * If deprecated address is forbidden,
1217 * we do not accept SYN to deprecated interface
1218 * address to prevent any new inbound connection from
1219 * getting established.
1220 * When we do not accept SYN, we send a TCP RST,
1221 * with deprecated source address (instead of dropping
1222 * it). We compromise it as it is much better for peer
1223 * to send a RST, and RST will be the final packet
1226 * If we do not forbid deprecated addresses, we accept
1227 * the SYN packet. RFC2462 does not suggest dropping
1229 * If we decipher RFC2462 5.5.4, it says like this:
1230 * 1. use of deprecated addr with existing
1231 * communication is okay - "SHOULD continue to be
1233 * 2. use of it with new communication:
1234 * (2a) "SHOULD NOT be used if alternate address
1235 * with sufficient scope is available"
1236 * (2b) nothing mentioned otherwise.
1237 * Here we fall into (2b) case as we have no choice in
1238 * our source address selection - we must obey the peer.
1240 * The wording in RFC2462 is confusing, and there are
1241 * multiple description text for deprecated address
1242 * handling - worse, they are not exactly the same.
1243 * I believe 5.5.4 is the best one, so we follow 5.5.4.
1245 if (isipv6 && !V_ip6_use_deprecated) {
1246 struct in6_ifaddr *ia6;
1248 ia6 = ip6_getdstifaddr(m);
1250 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1251 ifa_free(&ia6->ia_ifa);
1252 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1253 log(LOG_DEBUG, "%s; %s: Listen socket: "
1254 "Connection attempt to deprecated "
1255 "IPv6 address rejected\n",
1257 rstreason = BANDLIM_RST_OPENPORT;
1261 ifa_free(&ia6->ia_ifa);
1265 * Basic sanity checks on incoming SYN requests:
1266 * Don't respond if the destination is a link layer
1267 * broadcast according to RFC1122 4.2.3.10, p. 104.
1268 * If it is from this socket it must be forged.
1269 * Don't respond if the source or destination is a
1270 * global or subnet broad- or multicast address.
1271 * Note that it is quite possible to receive unicast
1272 * link-layer packets with a broadcast IP address. Use
1273 * in_broadcast() to find them.
1275 if (m->m_flags & (M_BCAST|M_MCAST)) {
1276 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1277 log(LOG_DEBUG, "%s; %s: Listen socket: "
1278 "Connection attempt from broad- or multicast "
1279 "link layer address ignored\n", s, __func__);
1284 if (th->th_dport == th->th_sport &&
1285 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) {
1286 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1287 log(LOG_DEBUG, "%s; %s: Listen socket: "
1288 "Connection attempt to/from self "
1289 "ignored\n", s, __func__);
1292 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1293 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) {
1294 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1295 log(LOG_DEBUG, "%s; %s: Listen socket: "
1296 "Connection attempt from/to multicast "
1297 "address ignored\n", s, __func__);
1302 #if defined(INET) && defined(INET6)
1307 if (th->th_dport == th->th_sport &&
1308 ip->ip_dst.s_addr == ip->ip_src.s_addr) {
1309 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1310 log(LOG_DEBUG, "%s; %s: Listen socket: "
1311 "Connection attempt from/to self "
1312 "ignored\n", s, __func__);
1315 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1316 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1317 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1318 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
1319 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1320 log(LOG_DEBUG, "%s; %s: Listen socket: "
1321 "Connection attempt from/to broad- "
1322 "or multicast address ignored\n",
1329 * SYN appears to be valid. Create compressed TCP state
1333 if (so->so_options & SO_DEBUG)
1334 tcp_trace(TA_INPUT, ostate, tp,
1335 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1337 tcp_dooptions(&to, optp, optlen, TO_SYN);
1338 syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL);
1340 * Entry added to syncache and mbuf consumed.
1341 * Only the listen socket is unlocked by syncache_add().
1343 if (ti_locked == TI_WLOCKED) {
1344 INP_INFO_WUNLOCK(&V_tcbinfo);
1345 ti_locked = TI_UNLOCKED;
1347 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1348 return (IPPROTO_DONE);
1349 } else if (tp->t_state == TCPS_LISTEN) {
1351 * When a listen socket is torn down the SO_ACCEPTCONN
1352 * flag is removed first while connections are drained
1353 * from the accept queue in a unlock/lock cycle of the
1354 * ACCEPT_LOCK, opening a race condition allowing a SYN
1355 * attempt go through unhandled.
1360 #ifdef TCP_SIGNATURE
1361 if (sig_checked == 0) {
1362 tcp_dooptions(&to, optp, optlen,
1363 (thflags & TH_SYN) ? TO_SYN : 0);
1364 if (!tcp_signature_verify_input(m, off0, tlen, optlen, &to,
1368 * In SYN_SENT state if it receives an RST, it is
1369 * allowed for further processing.
1371 if ((thflags & TH_RST) == 0 ||
1372 (tp->t_state == TCPS_SYN_SENT) == 0)
1379 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th);
1382 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later
1383 * state. tcp_do_segment() always consumes the mbuf chain, unlocks
1384 * the inpcb, and unlocks pcbinfo.
1386 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos, ti_locked);
1387 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1388 return (IPPROTO_DONE);
1391 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th);
1393 if (ti_locked == TI_WLOCKED) {
1394 INP_INFO_WUNLOCK(&V_tcbinfo);
1395 ti_locked = TI_UNLOCKED;
1399 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset "
1400 "ti_locked: %d", __func__, ti_locked));
1401 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1406 tcp_dropwithreset(m, th, tp, tlen, rstreason);
1409 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
1410 m = NULL; /* mbuf chain got consumed. */
1415 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th);
1417 if (ti_locked == TI_WLOCKED) {
1418 INP_INFO_WUNLOCK(&V_tcbinfo);
1419 ti_locked = TI_UNLOCKED;
1423 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock "
1424 "ti_locked: %d", __func__, ti_locked));
1425 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1433 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1438 return (IPPROTO_DONE);
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;
1456 * The size of tcp_saveipgen must be the size of the max ip header,
1459 u_char tcp_saveipgen[IP6_HDR_LEN];
1460 struct tcphdr tcp_savetcp;
1463 thflags = th->th_flags;
1464 inc = &tp->t_inpcb->inp_inc;
1465 tp->sackhint.last_sack_ack = 0;
1468 * If this is either a state-changing packet or current state isn't
1469 * established, we require a write lock on tcbinfo. Otherwise, we
1470 * allow the tcbinfo to be in either alocked or unlocked, as the
1471 * caller may have unnecessarily acquired a write lock due to a race.
1473 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 ||
1474 tp->t_state != TCPS_ESTABLISHED) {
1475 KASSERT(ti_locked == TI_WLOCKED, ("%s ti_locked %d for "
1476 "SYN/FIN/RST/!EST", __func__, ti_locked));
1477 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1480 if (ti_locked == TI_WLOCKED)
1481 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1483 KASSERT(ti_locked == TI_UNLOCKED, ("%s: EST "
1484 "ti_locked: %d", __func__, ti_locked));
1485 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1489 INP_WLOCK_ASSERT(tp->t_inpcb);
1490 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN",
1492 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT",
1496 * Segment received on connection.
1497 * Reset idle time and keep-alive timer.
1498 * XXX: This should be done after segment
1499 * validation to ignore broken/spoofed segs.
1501 tp->t_rcvtime = ticks;
1502 if (TCPS_HAVEESTABLISHED(tp->t_state))
1503 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
1506 * Unscale the window into a 32-bit value.
1507 * For the SYN_SENT state the scale is zero.
1509 tiwin = th->th_win << tp->snd_scale;
1512 * TCP ECN processing.
1514 if (tp->t_flags & TF_ECN_PERMIT) {
1515 if (thflags & TH_CWR)
1516 tp->t_flags &= ~TF_ECN_SND_ECE;
1517 switch (iptos & IPTOS_ECN_MASK) {
1519 tp->t_flags |= TF_ECN_SND_ECE;
1520 TCPSTAT_INC(tcps_ecn_ce);
1522 case IPTOS_ECN_ECT0:
1523 TCPSTAT_INC(tcps_ecn_ect0);
1525 case IPTOS_ECN_ECT1:
1526 TCPSTAT_INC(tcps_ecn_ect1);
1529 /* Congestion experienced. */
1530 if (thflags & TH_ECE) {
1531 cc_cong_signal(tp, th, CC_ECN);
1536 * Parse options on any incoming segment.
1538 tcp_dooptions(&to, (u_char *)(th + 1),
1539 (th->th_off << 2) - sizeof(struct tcphdr),
1540 (thflags & TH_SYN) ? TO_SYN : 0);
1543 * If echoed timestamp is later than the current time,
1544 * fall back to non RFC1323 RTT calculation. Normalize
1545 * timestamp if syncookies were used when this connection
1548 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1549 to.to_tsecr -= tp->ts_offset;
1550 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks()))
1554 * If timestamps were negotiated during SYN/ACK they should
1555 * appear on every segment during this session and vice versa.
1557 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) {
1558 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1559 log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1560 "no action\n", s, __func__);
1564 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) {
1565 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1566 log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
1567 "no action\n", s, __func__);
1573 * Process options only when we get SYN/ACK back. The SYN case
1574 * for incoming connections is handled in tcp_syncache.
1575 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
1576 * or <SYN,ACK>) segment itself is never scaled.
1577 * XXX this is traditional behavior, may need to be cleaned up.
1579 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1580 if ((to.to_flags & TOF_SCALE) &&
1581 (tp->t_flags & TF_REQ_SCALE)) {
1582 tp->t_flags |= TF_RCVD_SCALE;
1583 tp->snd_scale = to.to_wscale;
1586 * Initial send window. It will be updated with
1587 * the next incoming segment to the scaled value.
1589 tp->snd_wnd = th->th_win;
1590 if (to.to_flags & TOF_TS) {
1591 tp->t_flags |= TF_RCVD_TSTMP;
1592 tp->ts_recent = to.to_tsval;
1593 tp->ts_recent_age = tcp_ts_getticks();
1595 if (to.to_flags & TOF_MSS)
1596 tcp_mss(tp, to.to_mss);
1597 if ((tp->t_flags & TF_SACK_PERMIT) &&
1598 (to.to_flags & TOF_SACKPERM) == 0)
1599 tp->t_flags &= ~TF_SACK_PERMIT;
1603 * Header prediction: check for the two common cases
1604 * of a uni-directional data xfer. If the packet has
1605 * no control flags, is in-sequence, the window didn't
1606 * change and we're not retransmitting, it's a
1607 * candidate. If the length is zero and the ack moved
1608 * forward, we're the sender side of the xfer. Just
1609 * free the data acked & wake any higher level process
1610 * that was blocked waiting for space. If the length
1611 * is non-zero and the ack didn't move, we're the
1612 * receiver side. If we're getting packets in-order
1613 * (the reassembly queue is empty), add the data to
1614 * the socket buffer and note that we need a delayed ack.
1615 * Make sure that the hidden state-flags are also off.
1616 * Since we check for TCPS_ESTABLISHED first, it can only
1619 if (tp->t_state == TCPS_ESTABLISHED &&
1620 th->th_seq == tp->rcv_nxt &&
1621 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1622 tp->snd_nxt == tp->snd_max &&
1623 tiwin && tiwin == tp->snd_wnd &&
1624 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1625 tp->t_segq == NULL && ((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. Application has not set receive buffer size with
1800 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE.
1801 * 2. the number of bytes received during the time it takes
1802 * one timestamp to be reflected back to us (the RTT);
1803 * 3. received bytes per RTT is within seven eighth of the
1804 * current socket buffer size;
1805 * 4. receive buffer size has not hit maximal automatic size;
1807 * This algorithm does one step per RTT at most and only if
1808 * we receive a bulk stream w/o packet losses or reorderings.
1809 * Shrinking the buffer during idle times is not necessary as
1810 * it doesn't consume any memory when idle.
1812 * TODO: Only step up if the application is actually serving
1813 * the buffer to better manage the socket buffer resources.
1815 if (V_tcp_do_autorcvbuf &&
1817 (so->so_rcv.sb_flags & SB_AUTOSIZE)) {
1818 if (TSTMP_GT(to.to_tsecr, tp->rfbuf_ts) &&
1819 to.to_tsecr - tp->rfbuf_ts < hz) {
1821 (so->so_rcv.sb_hiwat / 8 * 7) &&
1822 so->so_rcv.sb_hiwat <
1823 V_tcp_autorcvbuf_max) {
1825 min(so->so_rcv.sb_hiwat +
1826 V_tcp_autorcvbuf_inc,
1827 V_tcp_autorcvbuf_max);
1829 /* Start over with next RTT. */
1833 tp->rfbuf_cnt += tlen; /* add up */
1836 /* Add data to socket buffer. */
1837 SOCKBUF_LOCK(&so->so_rcv);
1838 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1842 * Set new socket buffer size.
1843 * Give up when limit is reached.
1846 if (!sbreserve_locked(&so->so_rcv,
1848 so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1849 m_adj(m, drop_hdrlen); /* delayed header drop */
1850 sbappendstream_locked(&so->so_rcv, m);
1852 /* NB: sorwakeup_locked() does an implicit unlock. */
1853 sorwakeup_locked(so);
1854 if (DELAY_ACK(tp, tlen)) {
1855 tp->t_flags |= TF_DELACK;
1857 tp->t_flags |= TF_ACKNOW;
1865 * Calculate amount of space in receive window,
1866 * and then do TCP input processing.
1867 * Receive window is amount of space in rcv queue,
1868 * but not less than advertised window.
1870 win = sbspace(&so->so_rcv);
1873 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1875 /* Reset receive buffer auto scaling when not in bulk receive mode. */
1879 switch (tp->t_state) {
1882 * If the state is SYN_RECEIVED:
1883 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1885 case TCPS_SYN_RECEIVED:
1886 if ((thflags & TH_ACK) &&
1887 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1888 SEQ_GT(th->th_ack, tp->snd_max))) {
1889 rstreason = BANDLIM_RST_OPENPORT;
1895 * If the state is SYN_SENT:
1896 * if seg contains an ACK, but not for our SYN, drop the input.
1897 * if seg contains a RST, then drop the connection.
1898 * if seg does not contain SYN, then drop it.
1899 * Otherwise this is an acceptable SYN segment
1900 * initialize tp->rcv_nxt and tp->irs
1901 * if seg contains ack then advance tp->snd_una
1902 * if seg contains an ECE and ECN support is enabled, the stream
1904 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1905 * arrange for segment to be acked (eventually)
1906 * continue processing rest of data/controls, beginning with URG
1909 if ((thflags & TH_ACK) &&
1910 (SEQ_LEQ(th->th_ack, tp->iss) ||
1911 SEQ_GT(th->th_ack, tp->snd_max))) {
1912 rstreason = BANDLIM_UNLIMITED;
1915 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) {
1916 TCP_PROBE5(connect__refused, NULL, tp,
1917 mtod(m, const char *), tp, th);
1918 tp = tcp_drop(tp, ECONNREFUSED);
1920 if (thflags & TH_RST)
1922 if (!(thflags & TH_SYN))
1925 tp->irs = th->th_seq;
1927 if (thflags & TH_ACK) {
1928 TCPSTAT_INC(tcps_connects);
1931 mac_socketpeer_set_from_mbuf(m, so);
1933 /* Do window scaling on this connection? */
1934 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1935 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1936 tp->rcv_scale = tp->request_r_scale;
1938 tp->rcv_adv += imin(tp->rcv_wnd,
1939 TCP_MAXWIN << tp->rcv_scale);
1940 tp->snd_una++; /* SYN is acked */
1942 * If there's data, delay ACK; if there's also a FIN
1943 * ACKNOW will be turned on later.
1945 if (DELAY_ACK(tp, tlen) && tlen != 0)
1946 tcp_timer_activate(tp, TT_DELACK,
1949 tp->t_flags |= TF_ACKNOW;
1951 if ((thflags & TH_ECE) && V_tcp_do_ecn) {
1952 tp->t_flags |= TF_ECN_PERMIT;
1953 TCPSTAT_INC(tcps_ecn_shs);
1957 * Received <SYN,ACK> in SYN_SENT[*] state.
1959 * SYN_SENT --> ESTABLISHED
1960 * SYN_SENT* --> FIN_WAIT_1
1962 tp->t_starttime = ticks;
1963 if (tp->t_flags & TF_NEEDFIN) {
1964 tcp_state_change(tp, TCPS_FIN_WAIT_1);
1965 tp->t_flags &= ~TF_NEEDFIN;
1968 tcp_state_change(tp, TCPS_ESTABLISHED);
1969 TCP_PROBE5(connect__established, NULL, tp,
1970 mtod(m, const char *), tp, th);
1972 tcp_timer_activate(tp, TT_KEEP,
1977 * Received initial SYN in SYN-SENT[*] state =>
1978 * simultaneous open.
1979 * If it succeeds, connection is * half-synchronized.
1980 * Otherwise, do 3-way handshake:
1981 * SYN-SENT -> SYN-RECEIVED
1982 * SYN-SENT* -> SYN-RECEIVED*
1984 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
1985 tcp_timer_activate(tp, TT_REXMT, 0);
1986 tcp_state_change(tp, TCPS_SYN_RECEIVED);
1989 KASSERT(ti_locked == TI_WLOCKED, ("%s: trimthenstep6: "
1990 "ti_locked %d", __func__, ti_locked));
1991 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1992 INP_WLOCK_ASSERT(tp->t_inpcb);
1995 * Advance th->th_seq to correspond to first data byte.
1996 * If data, trim to stay within window,
1997 * dropping FIN if necessary.
2000 if (tlen > tp->rcv_wnd) {
2001 todrop = tlen - tp->rcv_wnd;
2005 TCPSTAT_INC(tcps_rcvpackafterwin);
2006 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2008 tp->snd_wl1 = th->th_seq - 1;
2009 tp->rcv_up = th->th_seq;
2011 * Client side of transaction: already sent SYN and data.
2012 * If the remote host used T/TCP to validate the SYN,
2013 * our data will be ACK'd; if so, enter normal data segment
2014 * processing in the middle of step 5, ack processing.
2015 * Otherwise, goto step 6.
2017 if (thflags & TH_ACK)
2023 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
2024 * do normal processing.
2026 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later.
2030 break; /* continue normal processing */
2034 * States other than LISTEN or SYN_SENT.
2035 * First check the RST flag and sequence number since reset segments
2036 * are exempt from the timestamp and connection count tests. This
2037 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
2038 * below which allowed reset segments in half the sequence space
2039 * to fall though and be processed (which gives forged reset
2040 * segments with a random sequence number a 50 percent chance of
2041 * killing a connection).
2042 * Then check timestamp, if present.
2043 * Then check the connection count, if present.
2044 * Then check that at least some bytes of segment are within
2045 * receive window. If segment begins before rcv_nxt,
2046 * drop leading data (and SYN); if nothing left, just ack.
2049 * If the RST bit is set, check the sequence number to see
2050 * if this is a valid reset segment.
2052 * In all states except SYN-SENT, all reset (RST) segments
2053 * are validated by checking their SEQ-fields. A reset is
2054 * valid if its sequence number is in the window.
2055 * Note: this does not take into account delayed ACKs, so
2056 * we should test against last_ack_sent instead of rcv_nxt.
2057 * The sequence number in the reset segment is normally an
2058 * echo of our outgoing acknowlegement numbers, but some hosts
2059 * send a reset with the sequence number at the rightmost edge
2060 * of our receive window, and we have to handle this case.
2061 * Note 2: Paul Watson's paper "Slipping in the Window" has shown
2062 * that brute force RST attacks are possible. To combat this,
2063 * we use a much stricter check while in the ESTABLISHED state,
2064 * only accepting RSTs where the sequence number is equal to
2065 * last_ack_sent. In all other states (the states in which a
2066 * RST is more likely), the more permissive check is used.
2067 * If we have multiple segments in flight, the initial reset
2068 * segment sequence numbers will be to the left of last_ack_sent,
2069 * but they will eventually catch up.
2070 * In any case, it never made sense to trim reset segments to
2071 * fit the receive window since RFC 1122 says:
2072 * 4.2.2.12 RST Segment: RFC-793 Section 3.4
2074 * A TCP SHOULD allow a received RST segment to include data.
2077 * It has been suggested that a RST segment could contain
2078 * ASCII text that encoded and explained the cause of the
2079 * RST. No standard has yet been established for such
2082 * If the reset segment passes the sequence number test examine
2084 * SYN_RECEIVED STATE:
2085 * If passive open, return to LISTEN state.
2086 * If active open, inform user that connection was refused.
2087 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
2088 * Inform user that connection was reset, and close tcb.
2089 * CLOSING, LAST_ACK STATES:
2092 * Drop the segment - see Stevens, vol. 2, p. 964 and
2095 if (thflags & TH_RST) {
2096 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) &&
2097 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
2098 switch (tp->t_state) {
2100 case TCPS_SYN_RECEIVED:
2101 so->so_error = ECONNREFUSED;
2104 case TCPS_ESTABLISHED:
2105 if (V_tcp_insecure_rst == 0 &&
2106 !(SEQ_GEQ(th->th_seq, tp->rcv_nxt - 1) &&
2107 SEQ_LEQ(th->th_seq, tp->rcv_nxt + 1)) &&
2108 !(SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) &&
2109 SEQ_LEQ(th->th_seq, tp->last_ack_sent + 1))) {
2110 TCPSTAT_INC(tcps_badrst);
2114 case TCPS_FIN_WAIT_1:
2115 case TCPS_FIN_WAIT_2:
2116 case TCPS_CLOSE_WAIT:
2117 so->so_error = ECONNRESET;
2119 KASSERT(ti_locked == TI_WLOCKED,
2120 ("tcp_do_segment: TH_RST 1 ti_locked %d",
2122 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2124 tcp_state_change(tp, TCPS_CLOSED);
2125 TCPSTAT_INC(tcps_drops);
2131 KASSERT(ti_locked == TI_WLOCKED,
2132 ("tcp_do_segment: TH_RST 2 ti_locked %d",
2134 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2144 * RFC 1323 PAWS: If we have a timestamp reply on this segment
2145 * and it's less than ts_recent, drop it.
2147 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
2148 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
2150 /* Check to see if ts_recent is over 24 days old. */
2151 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) {
2153 * Invalidate ts_recent. If this segment updates
2154 * ts_recent, the age will be reset later and ts_recent
2155 * will get a valid value. If it does not, setting
2156 * ts_recent to zero will at least satisfy the
2157 * requirement that zero be placed in the timestamp
2158 * echo reply when ts_recent isn't valid. The
2159 * age isn't reset until we get a valid ts_recent
2160 * because we don't want out-of-order segments to be
2161 * dropped when ts_recent is old.
2165 TCPSTAT_INC(tcps_rcvduppack);
2166 TCPSTAT_ADD(tcps_rcvdupbyte, tlen);
2167 TCPSTAT_INC(tcps_pawsdrop);
2175 * In the SYN-RECEIVED state, validate that the packet belongs to
2176 * this connection before trimming the data to fit the receive
2177 * window. Check the sequence number versus IRS since we know
2178 * the sequence numbers haven't wrapped. This is a partial fix
2179 * for the "LAND" DoS attack.
2181 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
2182 rstreason = BANDLIM_RST_OPENPORT;
2186 todrop = tp->rcv_nxt - th->th_seq;
2189 * If this is a duplicate SYN for our current connection,
2190 * advance over it and pretend and it's not a SYN.
2192 if (thflags & TH_SYN && th->th_seq == tp->irs) {
2202 * Following if statement from Stevens, vol. 2, p. 960.
2205 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
2207 * Any valid FIN must be to the left of the window.
2208 * At this point the FIN must be a duplicate or out
2209 * of sequence; drop it.
2214 * Send an ACK to resynchronize and drop any data.
2215 * But keep on processing for RST or ACK.
2217 tp->t_flags |= TF_ACKNOW;
2219 TCPSTAT_INC(tcps_rcvduppack);
2220 TCPSTAT_ADD(tcps_rcvdupbyte, todrop);
2222 TCPSTAT_INC(tcps_rcvpartduppack);
2223 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop);
2225 drop_hdrlen += todrop; /* drop from the top afterwards */
2226 th->th_seq += todrop;
2228 if (th->th_urp > todrop)
2229 th->th_urp -= todrop;
2237 * If new data are received on a connection after the
2238 * user processes are gone, then RST the other end.
2240 if ((so->so_state & SS_NOFDREF) &&
2241 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
2242 KASSERT(ti_locked == TI_WLOCKED, ("%s: SS_NOFDEREF && "
2243 "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked));
2244 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2246 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
2247 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data "
2248 "after socket was closed, "
2249 "sending RST and removing tcpcb\n",
2250 s, __func__, tcpstates[tp->t_state], tlen);
2254 TCPSTAT_INC(tcps_rcvafterclose);
2255 rstreason = BANDLIM_UNLIMITED;
2260 * If segment ends after window, drop trailing data
2261 * (and PUSH and FIN); if nothing left, just ACK.
2263 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
2265 TCPSTAT_INC(tcps_rcvpackafterwin);
2266 if (todrop >= tlen) {
2267 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen);
2269 * If window is closed can only take segments at
2270 * window edge, and have to drop data and PUSH from
2271 * incoming segments. Continue processing, but
2272 * remember to ack. Otherwise, drop segment
2275 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2276 tp->t_flags |= TF_ACKNOW;
2277 TCPSTAT_INC(tcps_rcvwinprobe);
2281 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2284 thflags &= ~(TH_PUSH|TH_FIN);
2288 * If last ACK falls within this segment's sequence numbers,
2289 * record its timestamp.
2291 * 1) That the test incorporates suggestions from the latest
2292 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
2293 * 2) That updating only on newer timestamps interferes with
2294 * our earlier PAWS tests, so this check should be solely
2295 * predicated on the sequence space of this segment.
2296 * 3) That we modify the segment boundary check to be
2297 * Last.ACK.Sent <= SEG.SEQ + SEG.Len
2298 * instead of RFC1323's
2299 * Last.ACK.Sent < SEG.SEQ + SEG.Len,
2300 * This modified check allows us to overcome RFC1323's
2301 * limitations as described in Stevens TCP/IP Illustrated
2302 * Vol. 2 p.869. In such cases, we can still calculate the
2303 * RTT correctly when RCV.NXT == Last.ACK.Sent.
2305 if ((to.to_flags & TOF_TS) != 0 &&
2306 SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2307 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
2308 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
2309 tp->ts_recent_age = tcp_ts_getticks();
2310 tp->ts_recent = to.to_tsval;
2314 * If a SYN is in the window, then this is an
2315 * error and we send an RST and drop the connection.
2317 if (thflags & TH_SYN) {
2318 KASSERT(ti_locked == TI_WLOCKED,
2319 ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked));
2320 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2322 tp = tcp_drop(tp, ECONNRESET);
2323 rstreason = BANDLIM_UNLIMITED;
2328 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
2329 * flag is on (half-synchronized state), then queue data for
2330 * later processing; else drop segment and return.
2332 if ((thflags & TH_ACK) == 0) {
2333 if (tp->t_state == TCPS_SYN_RECEIVED ||
2334 (tp->t_flags & TF_NEEDSYN))
2336 else if (tp->t_flags & TF_ACKNOW)
2345 switch (tp->t_state) {
2348 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
2349 * ESTABLISHED state and continue processing.
2350 * The ACK was checked above.
2352 case TCPS_SYN_RECEIVED:
2354 TCPSTAT_INC(tcps_connects);
2356 /* Do window scaling? */
2357 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2358 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2359 tp->rcv_scale = tp->request_r_scale;
2360 tp->snd_wnd = tiwin;
2364 * SYN-RECEIVED -> ESTABLISHED
2365 * SYN-RECEIVED* -> FIN-WAIT-1
2367 tp->t_starttime = ticks;
2368 if (tp->t_flags & TF_NEEDFIN) {
2369 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2370 tp->t_flags &= ~TF_NEEDFIN;
2372 tcp_state_change(tp, TCPS_ESTABLISHED);
2373 TCP_PROBE5(accept__established, NULL, tp,
2374 mtod(m, const char *), tp, th);
2376 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
2379 * If segment contains data or ACK, will call tcp_reass()
2380 * later; if not, do so now to pass queued data to user.
2382 if (tlen == 0 && (thflags & TH_FIN) == 0)
2383 (void) tcp_reass(tp, (struct tcphdr *)0, 0,
2385 tp->snd_wl1 = th->th_seq - 1;
2389 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2390 * ACKs. If the ack is in the range
2391 * tp->snd_una < th->th_ack <= tp->snd_max
2392 * then advance tp->snd_una to th->th_ack and drop
2393 * data from the retransmission queue. If this ACK reflects
2394 * more up to date window information we update our window information.
2396 case TCPS_ESTABLISHED:
2397 case TCPS_FIN_WAIT_1:
2398 case TCPS_FIN_WAIT_2:
2399 case TCPS_CLOSE_WAIT:
2402 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2403 TCPSTAT_INC(tcps_rcvacktoomuch);
2406 if ((tp->t_flags & TF_SACK_PERMIT) &&
2407 ((to.to_flags & TOF_SACK) ||
2408 !TAILQ_EMPTY(&tp->snd_holes)))
2409 tcp_sack_doack(tp, &to, th->th_ack);
2411 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
2412 hhook_run_tcp_est_in(tp, th, &to);
2414 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2415 if (tlen == 0 && tiwin == tp->snd_wnd) {
2417 * If this is the first time we've seen a
2418 * FIN from the remote, this is not a
2419 * duplicate and it needs to be processed
2420 * normally. This happens during a
2421 * simultaneous close.
2423 if ((thflags & TH_FIN) &&
2424 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2428 TCPSTAT_INC(tcps_rcvdupack);
2430 * If we have outstanding data (other than
2431 * a window probe), this is a completely
2432 * duplicate ack (ie, window info didn't
2433 * change and FIN isn't set),
2434 * the ack is the biggest we've
2435 * seen and we've seen exactly our rexmt
2436 * threshhold of them, assume a packet
2437 * has been dropped and retransmit it.
2438 * Kludge snd_nxt & the congestion
2439 * window so we send only this one
2442 * We know we're losing at the current
2443 * window size so do congestion avoidance
2444 * (set ssthresh to half the current window
2445 * and pull our congestion window back to
2446 * the new ssthresh).
2448 * Dup acks mean that packets have left the
2449 * network (they're now cached at the receiver)
2450 * so bump cwnd by the amount in the receiver
2451 * to keep a constant cwnd packets in the
2454 * When using TCP ECN, notify the peer that
2455 * we reduced the cwnd.
2457 if (!tcp_timer_active(tp, TT_REXMT) ||
2458 th->th_ack != tp->snd_una)
2460 else if (++tp->t_dupacks > tcprexmtthresh ||
2461 IN_FASTRECOVERY(tp->t_flags)) {
2462 cc_ack_received(tp, th, CC_DUPACK);
2463 if ((tp->t_flags & TF_SACK_PERMIT) &&
2464 IN_FASTRECOVERY(tp->t_flags)) {
2468 * Compute the amount of data in flight first.
2469 * We can inject new data into the pipe iff
2470 * we have less than 1/2 the original window's
2471 * worth of data in flight.
2473 awnd = (tp->snd_nxt - tp->snd_fack) +
2474 tp->sackhint.sack_bytes_rexmit;
2475 if (awnd < tp->snd_ssthresh) {
2476 tp->snd_cwnd += tp->t_maxseg;
2477 if (tp->snd_cwnd > tp->snd_ssthresh)
2478 tp->snd_cwnd = tp->snd_ssthresh;
2481 tp->snd_cwnd += tp->t_maxseg;
2482 (void) tcp_output(tp);
2484 } else if (tp->t_dupacks == tcprexmtthresh) {
2485 tcp_seq onxt = tp->snd_nxt;
2488 * If we're doing sack, check to
2489 * see if we're already in sack
2490 * recovery. If we're not doing sack,
2491 * check to see if we're in newreno
2494 if (tp->t_flags & TF_SACK_PERMIT) {
2495 if (IN_FASTRECOVERY(tp->t_flags)) {
2500 if (SEQ_LEQ(th->th_ack,
2506 /* Congestion signal before ack. */
2507 cc_cong_signal(tp, th, CC_NDUPACK);
2508 cc_ack_received(tp, th, CC_DUPACK);
2509 tcp_timer_activate(tp, TT_REXMT, 0);
2511 if (tp->t_flags & TF_SACK_PERMIT) {
2513 tcps_sack_recovery_episode);
2514 tp->sack_newdata = tp->snd_nxt;
2515 tp->snd_cwnd = tp->t_maxseg;
2516 (void) tcp_output(tp);
2519 tp->snd_nxt = th->th_ack;
2520 tp->snd_cwnd = tp->t_maxseg;
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) *
2549 * Only call tcp_output when there
2550 * is new data available to be sent.
2551 * Otherwise we would send pure ACKs.
2553 SOCKBUF_LOCK(&so->so_snd);
2554 avail = so->so_snd.sb_cc -
2555 (tp->snd_nxt - tp->snd_una);
2556 SOCKBUF_UNLOCK(&so->so_snd);
2558 (void) tcp_output(tp);
2559 sent = tp->snd_max - oldsndmax;
2560 if (sent > tp->t_maxseg) {
2561 KASSERT((tp->t_dupacks == 2 &&
2562 tp->snd_limited == 0) ||
2563 (sent == tp->t_maxseg + 1 &&
2564 tp->t_flags & TF_SENTFIN),
2565 ("%s: sent too much",
2567 tp->snd_limited = 2;
2568 } else if (sent > 0)
2570 tp->snd_cwnd = oldcwnd;
2578 KASSERT(SEQ_GT(th->th_ack, tp->snd_una),
2579 ("%s: th_ack <= snd_una", __func__));
2582 * If the congestion window was inflated to account
2583 * for the other side's cached packets, retract it.
2585 if (IN_FASTRECOVERY(tp->t_flags)) {
2586 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2587 if (tp->t_flags & TF_SACK_PERMIT)
2588 tcp_sack_partialack(tp, th);
2590 tcp_newreno_partial_ack(tp, th);
2592 cc_post_recovery(tp, th);
2596 * If we reach this point, ACK is not a duplicate,
2597 * i.e., it ACKs something we sent.
2599 if (tp->t_flags & TF_NEEDSYN) {
2601 * T/TCP: Connection was half-synchronized, and our
2602 * SYN has been ACK'd (so connection is now fully
2603 * synchronized). Go to non-starred state,
2604 * increment snd_una for ACK of SYN, and check if
2605 * we can do window scaling.
2607 tp->t_flags &= ~TF_NEEDSYN;
2609 /* Do window scaling? */
2610 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2611 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2612 tp->rcv_scale = tp->request_r_scale;
2613 /* Send window already scaled. */
2618 INP_WLOCK_ASSERT(tp->t_inpcb);
2620 acked = BYTES_THIS_ACK(tp, th);
2621 TCPSTAT_INC(tcps_rcvackpack);
2622 TCPSTAT_ADD(tcps_rcvackbyte, acked);
2625 * If we just performed our first retransmit, and the ACK
2626 * arrives within our recovery window, then it was a mistake
2627 * to do the retransmit in the first place. Recover our
2628 * original cwnd and ssthresh, and proceed to transmit where
2631 if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID &&
2632 (int)(ticks - tp->t_badrxtwin) < 0)
2633 cc_cong_signal(tp, th, CC_RTO_ERR);
2636 * If we have a timestamp reply, update smoothed
2637 * round trip time. If no timestamp is present but
2638 * transmit timer is running and timed sequence
2639 * number was acked, update smoothed round trip time.
2640 * Since we now have an rtt measurement, cancel the
2641 * timer backoff (cf., Phil Karn's retransmit alg.).
2642 * Recompute the initial retransmit timer.
2644 * Some boxes send broken timestamp replies
2645 * during the SYN+ACK phase, ignore
2646 * timestamps of 0 or we could calculate a
2647 * huge RTT and blow up the retransmit timer.
2649 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) {
2652 t = tcp_ts_getticks() - to.to_tsecr;
2653 if (!tp->t_rttlow || tp->t_rttlow > t)
2655 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1);
2656 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2657 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2658 tp->t_rttlow = ticks - tp->t_rtttime;
2659 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2663 * If all outstanding data is acked, stop retransmit
2664 * timer and remember to restart (more output or persist).
2665 * If there is more data to be acked, restart retransmit
2666 * timer, using current (possibly backed-off) value.
2668 if (th->th_ack == tp->snd_max) {
2669 tcp_timer_activate(tp, TT_REXMT, 0);
2671 } else if (!tcp_timer_active(tp, TT_PERSIST))
2672 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
2675 * If no data (only SYN) was ACK'd,
2676 * skip rest of ACK processing.
2682 * Let the congestion control algorithm update congestion
2683 * control related information. This typically means increasing
2684 * the congestion window.
2686 cc_ack_received(tp, th, CC_ACK);
2688 SOCKBUF_LOCK(&so->so_snd);
2689 if (acked > so->so_snd.sb_cc) {
2690 tp->snd_wnd -= so->so_snd.sb_cc;
2691 mfree = sbcut_locked(&so->so_snd,
2692 (int)so->so_snd.sb_cc);
2695 mfree = sbcut_locked(&so->so_snd, acked);
2696 tp->snd_wnd -= acked;
2699 /* NB: sowwakeup_locked() does an implicit unlock. */
2700 sowwakeup_locked(so);
2702 /* Detect una wraparound. */
2703 if (!IN_RECOVERY(tp->t_flags) &&
2704 SEQ_GT(tp->snd_una, tp->snd_recover) &&
2705 SEQ_LEQ(th->th_ack, tp->snd_recover))
2706 tp->snd_recover = th->th_ack - 1;
2707 /* XXXLAS: Can this be moved up into cc_post_recovery? */
2708 if (IN_RECOVERY(tp->t_flags) &&
2709 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2710 EXIT_RECOVERY(tp->t_flags);
2712 tp->snd_una = th->th_ack;
2713 if (tp->t_flags & TF_SACK_PERMIT) {
2714 if (SEQ_GT(tp->snd_una, tp->snd_recover))
2715 tp->snd_recover = tp->snd_una;
2717 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2718 tp->snd_nxt = tp->snd_una;
2720 switch (tp->t_state) {
2723 * In FIN_WAIT_1 STATE in addition to the processing
2724 * for the ESTABLISHED state if our FIN is now acknowledged
2725 * then enter FIN_WAIT_2.
2727 case TCPS_FIN_WAIT_1:
2728 if (ourfinisacked) {
2730 * If we can't receive any more
2731 * data, then closing user can proceed.
2732 * Starting the timer is contrary to the
2733 * specification, but if we don't get a FIN
2734 * we'll hang forever.
2737 * we should release the tp also, and use a
2740 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2741 soisdisconnected(so);
2742 tcp_timer_activate(tp, TT_2MSL,
2743 (tcp_fast_finwait2_recycle ?
2744 tcp_finwait2_timeout :
2747 tcp_state_change(tp, TCPS_FIN_WAIT_2);
2752 * In CLOSING STATE in addition to the processing for
2753 * the ESTABLISHED state if the ACK acknowledges our FIN
2754 * then enter the TIME-WAIT state, otherwise ignore
2758 if (ourfinisacked) {
2759 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2761 INP_INFO_WUNLOCK(&V_tcbinfo);
2768 * In LAST_ACK, we may still be waiting for data to drain
2769 * and/or to be acked, as well as for the ack of our FIN.
2770 * If our FIN is now acknowledged, delete the TCB,
2771 * enter the closed state and return.
2774 if (ourfinisacked) {
2775 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2784 INP_WLOCK_ASSERT(tp->t_inpcb);
2787 * Update window information.
2788 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2790 if ((thflags & TH_ACK) &&
2791 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2792 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2793 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2794 /* keep track of pure window updates */
2796 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2797 TCPSTAT_INC(tcps_rcvwinupd);
2798 tp->snd_wnd = tiwin;
2799 tp->snd_wl1 = th->th_seq;
2800 tp->snd_wl2 = th->th_ack;
2801 if (tp->snd_wnd > tp->max_sndwnd)
2802 tp->max_sndwnd = tp->snd_wnd;
2807 * Process segments with URG.
2809 if ((thflags & TH_URG) && th->th_urp &&
2810 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2812 * This is a kludge, but if we receive and accept
2813 * random urgent pointers, we'll crash in
2814 * soreceive. It's hard to imagine someone
2815 * actually wanting to send this much urgent data.
2817 SOCKBUF_LOCK(&so->so_rcv);
2818 if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2819 th->th_urp = 0; /* XXX */
2820 thflags &= ~TH_URG; /* XXX */
2821 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */
2822 goto dodata; /* XXX */
2825 * If this segment advances the known urgent pointer,
2826 * then mark the data stream. This should not happen
2827 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2828 * a FIN has been received from the remote side.
2829 * In these states we ignore the URG.
2831 * According to RFC961 (Assigned Protocols),
2832 * the urgent pointer points to the last octet
2833 * of urgent data. We continue, however,
2834 * to consider it to indicate the first octet
2835 * of data past the urgent section as the original
2836 * spec states (in one of two places).
2838 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2839 tp->rcv_up = th->th_seq + th->th_urp;
2840 so->so_oobmark = so->so_rcv.sb_cc +
2841 (tp->rcv_up - tp->rcv_nxt) - 1;
2842 if (so->so_oobmark == 0)
2843 so->so_rcv.sb_state |= SBS_RCVATMARK;
2845 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2847 SOCKBUF_UNLOCK(&so->so_rcv);
2849 * Remove out of band data so doesn't get presented to user.
2850 * This can happen independent of advancing the URG pointer,
2851 * but if two URG's are pending at once, some out-of-band
2852 * data may creep in... ick.
2854 if (th->th_urp <= (u_long)tlen &&
2855 !(so->so_options & SO_OOBINLINE)) {
2856 /* hdr drop is delayed */
2857 tcp_pulloutofband(so, th, m, drop_hdrlen);
2861 * If no out of band data is expected,
2862 * pull receive urgent pointer along
2863 * with the receive window.
2865 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2866 tp->rcv_up = tp->rcv_nxt;
2869 INP_WLOCK_ASSERT(tp->t_inpcb);
2872 * Process the segment text, merging it into the TCP sequencing queue,
2873 * and arranging for acknowledgment of receipt if necessary.
2874 * This process logically involves adjusting tp->rcv_wnd as data
2875 * is presented to the user (this happens in tcp_usrreq.c,
2876 * case PRU_RCVD). If a FIN has already been received on this
2877 * connection then we just ignore the text.
2879 if ((tlen || (thflags & TH_FIN)) &&
2880 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2881 tcp_seq save_start = th->th_seq;
2882 m_adj(m, drop_hdrlen); /* delayed header drop */
2884 * Insert segment which includes th into TCP reassembly queue
2885 * with control block tp. Set thflags to whether reassembly now
2886 * includes a segment with FIN. This handles the common case
2887 * inline (segment is the next to be received on an established
2888 * connection, and the queue is empty), avoiding linkage into
2889 * and removal from the queue and repetition of various
2891 * Set DELACK for segments received in order, but ack
2892 * immediately when segments are out of order (so
2893 * fast retransmit can work).
2895 if (th->th_seq == tp->rcv_nxt && tp->t_segq == NULL &&
2896 TCPS_HAVEESTABLISHED(tp->t_state)) {
2897 if (DELAY_ACK(tp, tlen))
2898 tp->t_flags |= TF_DELACK;
2900 tp->t_flags |= TF_ACKNOW;
2901 tp->rcv_nxt += tlen;
2902 thflags = th->th_flags & TH_FIN;
2903 TCPSTAT_INC(tcps_rcvpack);
2904 TCPSTAT_ADD(tcps_rcvbyte, tlen);
2906 SOCKBUF_LOCK(&so->so_rcv);
2907 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
2910 sbappendstream_locked(&so->so_rcv, m);
2911 /* NB: sorwakeup_locked() does an implicit unlock. */
2912 sorwakeup_locked(so);
2915 * XXX: Due to the header drop above "th" is
2916 * theoretically invalid by now. Fortunately
2917 * m_adj() doesn't actually frees any mbufs
2918 * when trimming from the head.
2920 thflags = tcp_reass(tp, th, &tlen, m);
2921 tp->t_flags |= TF_ACKNOW;
2923 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT))
2924 tcp_update_sack_list(tp, save_start, save_start + tlen);
2927 * Note the amount of data that peer has sent into
2928 * our window, in order to estimate the sender's
2932 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
2933 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2935 len = so->so_rcv.sb_hiwat;
2943 * If FIN is received ACK the FIN and let the user know
2944 * that the connection is closing.
2946 if (thflags & TH_FIN) {
2947 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2950 * If connection is half-synchronized
2951 * (ie NEEDSYN flag on) then delay ACK,
2952 * so it may be piggybacked when SYN is sent.
2953 * Otherwise, since we received a FIN then no
2954 * more input can be expected, send ACK now.
2956 if (tp->t_flags & TF_NEEDSYN)
2957 tp->t_flags |= TF_DELACK;
2959 tp->t_flags |= TF_ACKNOW;
2962 switch (tp->t_state) {
2965 * In SYN_RECEIVED and ESTABLISHED STATES
2966 * enter the CLOSE_WAIT state.
2968 case TCPS_SYN_RECEIVED:
2969 tp->t_starttime = ticks;
2971 case TCPS_ESTABLISHED:
2972 tcp_state_change(tp, TCPS_CLOSE_WAIT);
2976 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2977 * enter the CLOSING state.
2979 case TCPS_FIN_WAIT_1:
2980 tcp_state_change(tp, TCPS_CLOSING);
2984 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2985 * starting the time-wait timer, turning off the other
2988 case TCPS_FIN_WAIT_2:
2989 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2990 KASSERT(ti_locked == TI_WLOCKED, ("%s: dodata "
2991 "TCP_FIN_WAIT_2 ti_locked: %d", __func__,
2995 INP_INFO_WUNLOCK(&V_tcbinfo);
2999 if (ti_locked == TI_WLOCKED)
3000 INP_INFO_WUNLOCK(&V_tcbinfo);
3001 ti_locked = TI_UNLOCKED;
3004 if (so->so_options & SO_DEBUG)
3005 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
3010 * Return any desired output.
3012 if (needoutput || (tp->t_flags & TF_ACKNOW))
3013 (void) tcp_output(tp);
3016 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d",
3017 __func__, ti_locked));
3018 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
3019 INP_WLOCK_ASSERT(tp->t_inpcb);
3021 if (tp->t_flags & TF_DELACK) {
3022 tp->t_flags &= ~TF_DELACK;
3023 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
3025 INP_WUNLOCK(tp->t_inpcb);
3030 * Generate an ACK dropping incoming segment if it occupies
3031 * sequence space, where the ACK reflects our state.
3033 * We can now skip the test for the RST flag since all
3034 * paths to this code happen after packets containing
3035 * RST have been dropped.
3037 * In the SYN-RECEIVED state, don't send an ACK unless the
3038 * segment we received passes the SYN-RECEIVED ACK test.
3039 * If it fails send a RST. This breaks the loop in the
3040 * "LAND" DoS attack, and also prevents an ACK storm
3041 * between two listening ports that have been sent forged
3042 * SYN segments, each with the source address of the other.
3044 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
3045 (SEQ_GT(tp->snd_una, th->th_ack) ||
3046 SEQ_GT(th->th_ack, tp->snd_max)) ) {
3047 rstreason = BANDLIM_RST_OPENPORT;
3051 if (so->so_options & SO_DEBUG)
3052 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3055 if (ti_locked == TI_WLOCKED)
3056 INP_INFO_WUNLOCK(&V_tcbinfo);
3057 ti_locked = TI_UNLOCKED;
3059 tp->t_flags |= TF_ACKNOW;
3060 (void) tcp_output(tp);
3061 INP_WUNLOCK(tp->t_inpcb);
3066 if (ti_locked == TI_WLOCKED)
3067 INP_INFO_WUNLOCK(&V_tcbinfo);
3068 ti_locked = TI_UNLOCKED;
3071 tcp_dropwithreset(m, th, tp, tlen, rstreason);
3072 INP_WUNLOCK(tp->t_inpcb);
3074 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
3078 if (ti_locked == TI_WLOCKED) {
3079 INP_INFO_WUNLOCK(&V_tcbinfo);
3080 ti_locked = TI_UNLOCKED;
3084 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
3088 * Drop space held by incoming segment and return.
3091 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
3092 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3096 INP_WUNLOCK(tp->t_inpcb);
3101 * Issue RST and make ACK acceptable to originator of segment.
3102 * The mbuf must still include the original packet header.
3106 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
3107 int tlen, int rstreason)
3113 struct ip6_hdr *ip6;
3117 INP_WLOCK_ASSERT(tp->t_inpcb);
3120 /* Don't bother if destination was broadcast/multicast. */
3121 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
3124 if (mtod(m, struct ip *)->ip_v == 6) {
3125 ip6 = mtod(m, struct ip6_hdr *);
3126 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3127 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3129 /* IPv6 anycast check is done at tcp6_input() */
3132 #if defined(INET) && defined(INET6)
3137 ip = mtod(m, struct ip *);
3138 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3139 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3140 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3141 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3146 /* Perform bandwidth limiting. */
3147 if (badport_bandlim(rstreason) < 0)
3150 /* tcp_respond consumes the mbuf chain. */
3151 if (th->th_flags & TH_ACK) {
3152 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0,
3153 th->th_ack, TH_RST);
3155 if (th->th_flags & TH_SYN)
3157 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
3158 (tcp_seq)0, TH_RST|TH_ACK);
3166 * Parse TCP options and place in tcpopt.
3169 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags)
3174 for (; cnt > 0; cnt -= optlen, cp += optlen) {
3176 if (opt == TCPOPT_EOL)
3178 if (opt == TCPOPT_NOP)
3184 if (optlen < 2 || optlen > cnt)
3189 if (optlen != TCPOLEN_MAXSEG)
3191 if (!(flags & TO_SYN))
3193 to->to_flags |= TOF_MSS;
3194 bcopy((char *)cp + 2,
3195 (char *)&to->to_mss, sizeof(to->to_mss));
3196 to->to_mss = ntohs(to->to_mss);
3199 if (optlen != TCPOLEN_WINDOW)
3201 if (!(flags & TO_SYN))
3203 to->to_flags |= TOF_SCALE;
3204 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT);
3206 case TCPOPT_TIMESTAMP:
3207 if (optlen != TCPOLEN_TIMESTAMP)
3209 to->to_flags |= TOF_TS;
3210 bcopy((char *)cp + 2,
3211 (char *)&to->to_tsval, sizeof(to->to_tsval));
3212 to->to_tsval = ntohl(to->to_tsval);
3213 bcopy((char *)cp + 6,
3214 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
3215 to->to_tsecr = ntohl(to->to_tsecr);
3217 #ifdef TCP_SIGNATURE
3219 * XXX In order to reply to a host which has set the
3220 * TCP_SIGNATURE option in its initial SYN, we have to
3221 * record the fact that the option was observed here
3222 * for the syncache code to perform the correct response.
3224 case TCPOPT_SIGNATURE:
3225 if (optlen != TCPOLEN_SIGNATURE)
3227 to->to_flags |= TOF_SIGNATURE;
3228 to->to_signature = cp + 2;
3231 case TCPOPT_SACK_PERMITTED:
3232 if (optlen != TCPOLEN_SACK_PERMITTED)
3234 if (!(flags & TO_SYN))
3238 to->to_flags |= TOF_SACKPERM;
3241 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
3245 to->to_flags |= TOF_SACK;
3246 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
3247 to->to_sacks = cp + 2;
3248 TCPSTAT_INC(tcps_sack_rcv_blocks);
3257 * Pull out of band byte out of a segment so
3258 * it doesn't appear in the user's data queue.
3259 * It is still reflected in the segment length for
3260 * sequencing purposes.
3263 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m,
3266 int cnt = off + th->th_urp - 1;
3269 if (m->m_len > cnt) {
3270 char *cp = mtod(m, caddr_t) + cnt;
3271 struct tcpcb *tp = sototcpcb(so);
3273 INP_WLOCK_ASSERT(tp->t_inpcb);
3276 tp->t_oobflags |= TCPOOB_HAVEDATA;
3277 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3279 if (m->m_flags & M_PKTHDR)
3288 panic("tcp_pulloutofband");
3292 * Collect new round-trip time estimate
3293 * and update averages and current timeout.
3296 tcp_xmit_timer(struct tcpcb *tp, int rtt)
3300 INP_WLOCK_ASSERT(tp->t_inpcb);
3302 TCPSTAT_INC(tcps_rttupdated);
3304 if (tp->t_srtt != 0) {
3306 * srtt is stored as fixed point with 5 bits after the
3307 * binary point (i.e., scaled by 8). The following magic
3308 * is equivalent to the smoothing algorithm in rfc793 with
3309 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3310 * point). Adjust rtt to origin 0.
3312 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
3313 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
3315 if ((tp->t_srtt += delta) <= 0)
3319 * We accumulate a smoothed rtt variance (actually, a
3320 * smoothed mean difference), then set the retransmit
3321 * timer to smoothed rtt + 4 times the smoothed variance.
3322 * rttvar is stored as fixed point with 4 bits after the
3323 * binary point (scaled by 16). The following is
3324 * equivalent to rfc793 smoothing with an alpha of .75
3325 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
3326 * rfc793's wired-in beta.
3330 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
3331 if ((tp->t_rttvar += delta) <= 0)
3333 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
3334 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3337 * No rtt measurement yet - use the unsmoothed rtt.
3338 * Set the variance to half the rtt (so our first
3339 * retransmit happens at 3*rtt).
3341 tp->t_srtt = rtt << TCP_RTT_SHIFT;
3342 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
3343 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3349 * the retransmit should happen at rtt + 4 * rttvar.
3350 * Because of the way we do the smoothing, srtt and rttvar
3351 * will each average +1/2 tick of bias. When we compute
3352 * the retransmit timer, we want 1/2 tick of rounding and
3353 * 1 extra tick because of +-1/2 tick uncertainty in the
3354 * firing of the timer. The bias will give us exactly the
3355 * 1.5 tick we need. But, because the bias is
3356 * statistical, we have to test that we don't drop below
3357 * the minimum feasible timer (which is 2 ticks).
3359 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3360 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3363 * We received an ack for a packet that wasn't retransmitted;
3364 * it is probably safe to discard any error indications we've
3365 * received recently. This isn't quite right, but close enough
3366 * for now (a route might have failed after we sent a segment,
3367 * and the return path might not be symmetrical).
3369 tp->t_softerror = 0;
3373 * Determine a reasonable value for maxseg size.
3374 * If the route is known, check route for mtu.
3375 * If none, use an mss that can be handled on the outgoing interface
3376 * without forcing IP to fragment. If no route is found, route has no mtu,
3377 * or the destination isn't local, use a default, hopefully conservative
3378 * size (usually 512 or the default IP max size, but no more than the mtu
3379 * of the interface), as we can't discover anything about intervening
3380 * gateways or networks. We also initialize the congestion/slow start
3381 * window to be a single segment if the destination isn't local.
3382 * While looking at the routing entry, we also initialize other path-dependent
3383 * parameters from pre-set or cached values in the routing entry.
3385 * Also take into account the space needed for options that we
3386 * send regularly. Make maxseg shorter by that amount to assure
3387 * that we can send maxseg amount of data even when the options
3388 * are present. Store the upper limit of the length of options plus
3391 * NOTE that this routine is only called when we process an incoming
3392 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS
3393 * settings are handled in tcp_mssopt().
3396 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer,
3397 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap)
3401 struct inpcb *inp = tp->t_inpcb;
3402 struct hc_metrics_lite metrics;
3405 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
3406 size_t min_protoh = isipv6 ?
3407 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
3408 sizeof (struct tcpiphdr);
3410 const size_t min_protoh = sizeof(struct tcpiphdr);
3413 INP_WLOCK_ASSERT(tp->t_inpcb);
3415 if (mtuoffer != -1) {
3416 KASSERT(offer == -1, ("%s: conflict", __func__));
3417 offer = mtuoffer - min_protoh;
3424 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap);
3425 tp->t_maxopd = tp->t_maxseg = V_tcp_v6mssdflt;
3428 #if defined(INET) && defined(INET6)
3433 maxmtu = tcp_maxmtu(&inp->inp_inc, cap);
3434 tp->t_maxopd = tp->t_maxseg = V_tcp_mssdflt;
3439 * No route to sender, stay with default mss and return.
3443 * In case we return early we need to initialize metrics
3444 * to a defined state as tcp_hc_get() would do for us
3445 * if there was no cache hit.
3447 if (metricptr != NULL)
3448 bzero(metricptr, sizeof(struct hc_metrics_lite));
3452 /* What have we got? */
3456 * Offer == 0 means that there was no MSS on the SYN
3457 * segment, in this case we use tcp_mssdflt as
3458 * already assigned to t_maxopd above.
3460 offer = tp->t_maxopd;
3465 * Offer == -1 means that we didn't receive SYN yet.
3471 * Prevent DoS attack with too small MSS. Round up
3472 * to at least minmss.
3474 offer = max(offer, V_tcp_minmss);
3478 * rmx information is now retrieved from tcp_hostcache.
3480 tcp_hc_get(&inp->inp_inc, &metrics);
3481 if (metricptr != NULL)
3482 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite));
3485 * If there's a discovered mtu in tcp hostcache, use it.
3486 * Else, use the link mtu.
3488 if (metrics.rmx_mtu)
3489 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
3493 mss = maxmtu - min_protoh;
3494 if (!V_path_mtu_discovery &&
3495 !in6_localaddr(&inp->in6p_faddr))
3496 mss = min(mss, V_tcp_v6mssdflt);
3499 #if defined(INET) && defined(INET6)
3504 mss = maxmtu - min_protoh;
3505 if (!V_path_mtu_discovery &&
3506 !in_localaddr(inp->inp_faddr))
3507 mss = min(mss, V_tcp_mssdflt);
3511 * XXX - The above conditional (mss = maxmtu - min_protoh)
3512 * probably violates the TCP spec.
3513 * The problem is that, since we don't know the
3514 * other end's MSS, we are supposed to use a conservative
3515 * default. But, if we do that, then MTU discovery will
3516 * never actually take place, because the conservative
3517 * default is much less than the MTUs typically seen
3518 * on the Internet today. For the moment, we'll sweep
3519 * this under the carpet.
3521 * The conservative default might not actually be a problem
3522 * if the only case this occurs is when sending an initial
3523 * SYN with options and data to a host we've never talked
3524 * to before. Then, they will reply with an MSS value which
3525 * will get recorded and the new parameters should get
3526 * recomputed. For Further Study.
3529 mss = min(mss, offer);
3532 * Sanity check: make sure that maxopd will be large
3533 * enough to allow some data on segments even if the
3534 * all the option space is used (40bytes). Otherwise
3535 * funny things may happen in tcp_output.
3540 * maxopd stores the maximum length of data AND options
3541 * in a segment; maxseg is the amount of data in a normal
3542 * segment. We need to store this value (maxopd) apart
3543 * from maxseg, because now every segment carries options
3544 * and thus we normally have somewhat less data in segments.
3549 * origoffer==-1 indicates that no segments were received yet.
3550 * In this case we just guess.
3552 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
3554 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
3555 mss -= TCPOLEN_TSTAMP_APPA;
3561 tcp_mss(struct tcpcb *tp, int offer)
3567 struct hc_metrics_lite metrics;
3568 struct tcp_ifcap cap;
3570 KASSERT(tp != NULL, ("%s: tp == NULL", __func__));
3572 bzero(&cap, sizeof(cap));
3573 tcp_mss_update(tp, offer, -1, &metrics, &cap);
3579 * If there's a pipesize, change the socket buffer to that size,
3580 * don't change if sb_hiwat is different than default (then it
3581 * has been changed on purpose with setsockopt).
3582 * Make the socket buffers an integral number of mss units;
3583 * if the mss is larger than the socket buffer, decrease the mss.
3585 so = inp->inp_socket;
3586 SOCKBUF_LOCK(&so->so_snd);
3587 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe)
3588 bufsize = metrics.rmx_sendpipe;
3590 bufsize = so->so_snd.sb_hiwat;
3594 bufsize = roundup(bufsize, mss);
3595 if (bufsize > sb_max)
3597 if (bufsize > so->so_snd.sb_hiwat)
3598 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
3600 SOCKBUF_UNLOCK(&so->so_snd);
3603 SOCKBUF_LOCK(&so->so_rcv);
3604 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe)
3605 bufsize = metrics.rmx_recvpipe;
3607 bufsize = so->so_rcv.sb_hiwat;
3608 if (bufsize > mss) {
3609 bufsize = roundup(bufsize, mss);
3610 if (bufsize > sb_max)
3612 if (bufsize > so->so_rcv.sb_hiwat)
3613 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
3615 SOCKBUF_UNLOCK(&so->so_rcv);
3617 /* Check the interface for TSO capabilities. */
3618 if (cap.ifcap & CSUM_TSO) {
3619 tp->t_flags |= TF_TSO;
3620 tp->t_tsomax = cap.tsomax;
3625 * Determine the MSS option to send on an outgoing SYN.
3628 tcp_mssopt(struct in_conninfo *inc)
3635 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3638 if (inc->inc_flags & INC_ISIPV6) {
3639 mss = V_tcp_v6mssdflt;
3640 maxmtu = tcp_maxmtu6(inc, NULL);
3641 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3644 #if defined(INET) && defined(INET6)
3649 mss = V_tcp_mssdflt;
3650 maxmtu = tcp_maxmtu(inc, NULL);
3651 min_protoh = sizeof(struct tcpiphdr);
3654 #if defined(INET6) || defined(INET)
3655 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3658 if (maxmtu && thcmtu)
3659 mss = min(maxmtu, thcmtu) - min_protoh;
3660 else if (maxmtu || thcmtu)
3661 mss = max(maxmtu, thcmtu) - min_protoh;
3668 * On a partial ack arrives, force the retransmission of the
3669 * next unacknowledged segment. Do not clear tp->t_dupacks.
3670 * By setting snd_nxt to ti_ack, this forces retransmission timer to
3674 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th)
3676 tcp_seq onxt = tp->snd_nxt;
3677 u_long ocwnd = tp->snd_cwnd;
3679 INP_WLOCK_ASSERT(tp->t_inpcb);
3681 tcp_timer_activate(tp, TT_REXMT, 0);
3683 tp->snd_nxt = th->th_ack;
3685 * Set snd_cwnd to one segment beyond acknowledged offset.
3686 * (tp->snd_una has not yet been updated when this function is called.)
3688 tp->snd_cwnd = tp->t_maxseg + BYTES_THIS_ACK(tp, th);
3689 tp->t_flags |= TF_ACKNOW;
3690 (void) tcp_output(tp);
3691 tp->snd_cwnd = ocwnd;
3692 if (SEQ_GT(onxt, tp->snd_nxt))
3695 * Partial window deflation. Relies on fact that tp->snd_una
3698 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th))
3699 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th);
3702 tp->snd_cwnd += tp->t_maxseg;