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 two cases
752 * where we might discover later we need a write lock despite the
753 * flags: ACKs moving a connection out of the syncache, and ACKs for
754 * a connection in TIMEWAIT.
756 if ((thflags & (TH_SYN | 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_SYN | TH_FIN | TH_RST)) != 0)
986 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
988 if (tp->t_state != TCPS_ESTABLISHED) {
989 if (ti_locked == TI_UNLOCKED) {
990 if (INP_INFO_TRY_WLOCK(&V_tcbinfo) == 0) {
993 INP_INFO_WLOCK(&V_tcbinfo);
994 ti_locked = TI_WLOCKED;
996 if (in_pcbrele_wlocked(inp)) {
1002 ti_locked = TI_WLOCKED;
1004 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1008 INP_WLOCK_ASSERT(inp);
1009 if (mac_inpcb_check_deliver(inp, m))
1012 so = inp->inp_socket;
1013 KASSERT(so != NULL, ("%s: so == NULL", __func__));
1015 if (so->so_options & SO_DEBUG) {
1016 ostate = tp->t_state;
1019 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
1022 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
1025 #endif /* TCPDEBUG */
1027 * When the socket is accepting connections (the INPCB is in LISTEN
1028 * state) we look into the SYN cache if this is a new connection
1029 * attempt or the completion of a previous one. Because listen
1030 * sockets are never in TCPS_ESTABLISHED, the V_tcbinfo lock will be
1031 * held in this case.
1033 if (so->so_options & SO_ACCEPTCONN) {
1034 struct in_conninfo inc;
1036 KASSERT(tp->t_state == TCPS_LISTEN, ("%s: so accepting but "
1037 "tp not listening", __func__));
1038 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1040 bzero(&inc, sizeof(inc));
1043 inc.inc_flags |= INC_ISIPV6;
1044 inc.inc6_faddr = ip6->ip6_src;
1045 inc.inc6_laddr = ip6->ip6_dst;
1049 inc.inc_faddr = ip->ip_src;
1050 inc.inc_laddr = ip->ip_dst;
1052 inc.inc_fport = th->th_sport;
1053 inc.inc_lport = th->th_dport;
1054 inc.inc_fibnum = so->so_fibnum;
1057 * Check for an existing connection attempt in syncache if
1058 * the flag is only ACK. A successful lookup creates a new
1059 * socket appended to the listen queue in SYN_RECEIVED state.
1061 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
1063 * Parse the TCP options here because
1064 * syncookies need access to the reflected
1067 tcp_dooptions(&to, optp, optlen, 0);
1069 * NB: syncache_expand() doesn't unlock
1070 * inp and tcpinfo locks.
1072 if (!syncache_expand(&inc, &to, th, &so, m)) {
1074 * No syncache entry or ACK was not
1075 * for our SYN/ACK. Send a RST.
1076 * NB: syncache did its own logging
1077 * of the failure cause.
1079 rstreason = BANDLIM_RST_OPENPORT;
1084 * We completed the 3-way handshake
1085 * but could not allocate a socket
1086 * either due to memory shortage,
1087 * listen queue length limits or
1088 * global socket limits. Send RST
1089 * or wait and have the remote end
1090 * retransmit the ACK for another
1093 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1094 log(LOG_DEBUG, "%s; %s: Listen socket: "
1095 "Socket allocation failed due to "
1096 "limits or memory shortage, %s\n",
1098 V_tcp_sc_rst_sock_fail ?
1099 "sending RST" : "try again");
1100 if (V_tcp_sc_rst_sock_fail) {
1101 rstreason = BANDLIM_UNLIMITED;
1107 * Socket is created in state SYN_RECEIVED.
1108 * Unlock the listen socket, lock the newly
1109 * created socket and update the tp variable.
1111 INP_WUNLOCK(inp); /* listen socket */
1112 inp = sotoinpcb(so);
1113 INP_WLOCK(inp); /* new connection */
1114 tp = intotcpcb(inp);
1115 KASSERT(tp->t_state == TCPS_SYN_RECEIVED,
1116 ("%s: ", __func__));
1117 #ifdef TCP_SIGNATURE
1118 if (sig_checked == 0) {
1119 tcp_dooptions(&to, optp, optlen,
1120 (thflags & TH_SYN) ? TO_SYN : 0);
1121 if (!tcp_signature_verify_input(m, off0, tlen,
1122 optlen, &to, th, tp->t_flags)) {
1125 * In SYN_SENT state if it receives an
1126 * RST, it is allowed for further
1129 if ((thflags & TH_RST) == 0 ||
1130 (tp->t_state == TCPS_SYN_SENT) == 0)
1138 * Process the segment and the data it
1139 * contains. tcp_do_segment() consumes
1140 * the mbuf chain and unlocks the inpcb.
1142 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen,
1144 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1145 return (IPPROTO_DONE);
1148 * Segment flag validation for new connection attempts:
1150 * Our (SYN|ACK) response was rejected.
1151 * Check with syncache and remove entry to prevent
1154 * NB: syncache_chkrst does its own logging of failure
1157 if (thflags & TH_RST) {
1158 syncache_chkrst(&inc, th);
1162 * We can't do anything without SYN.
1164 if ((thflags & TH_SYN) == 0) {
1165 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1166 log(LOG_DEBUG, "%s; %s: Listen socket: "
1167 "SYN is missing, segment ignored\n",
1169 TCPSTAT_INC(tcps_badsyn);
1173 * (SYN|ACK) is bogus on a listen socket.
1175 if (thflags & TH_ACK) {
1176 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1177 log(LOG_DEBUG, "%s; %s: Listen socket: "
1178 "SYN|ACK invalid, segment rejected\n",
1180 syncache_badack(&inc); /* XXX: Not needed! */
1181 TCPSTAT_INC(tcps_badsyn);
1182 rstreason = BANDLIM_RST_OPENPORT;
1186 * If the drop_synfin option is enabled, drop all
1187 * segments with both the SYN and FIN bits set.
1188 * This prevents e.g. nmap from identifying the
1190 * XXX: Poor reasoning. nmap has other methods
1191 * and is constantly refining its stack detection
1193 * XXX: This is a violation of the TCP specification
1194 * and was used by RFC1644.
1196 if ((thflags & TH_FIN) && V_drop_synfin) {
1197 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1198 log(LOG_DEBUG, "%s; %s: Listen socket: "
1199 "SYN|FIN segment ignored (based on "
1200 "sysctl setting)\n", s, __func__);
1201 TCPSTAT_INC(tcps_badsyn);
1205 * Segment's flags are (SYN) or (SYN|FIN).
1207 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored
1208 * as they do not affect the state of the TCP FSM.
1209 * The data pointed to by TH_URG and th_urp is ignored.
1211 KASSERT((thflags & (TH_RST|TH_ACK)) == 0,
1212 ("%s: Listen socket: TH_RST or TH_ACK set", __func__));
1213 KASSERT(thflags & (TH_SYN),
1214 ("%s: Listen socket: TH_SYN not set", __func__));
1217 * If deprecated address is forbidden,
1218 * we do not accept SYN to deprecated interface
1219 * address to prevent any new inbound connection from
1220 * getting established.
1221 * When we do not accept SYN, we send a TCP RST,
1222 * with deprecated source address (instead of dropping
1223 * it). We compromise it as it is much better for peer
1224 * to send a RST, and RST will be the final packet
1227 * If we do not forbid deprecated addresses, we accept
1228 * the SYN packet. RFC2462 does not suggest dropping
1230 * If we decipher RFC2462 5.5.4, it says like this:
1231 * 1. use of deprecated addr with existing
1232 * communication is okay - "SHOULD continue to be
1234 * 2. use of it with new communication:
1235 * (2a) "SHOULD NOT be used if alternate address
1236 * with sufficient scope is available"
1237 * (2b) nothing mentioned otherwise.
1238 * Here we fall into (2b) case as we have no choice in
1239 * our source address selection - we must obey the peer.
1241 * The wording in RFC2462 is confusing, and there are
1242 * multiple description text for deprecated address
1243 * handling - worse, they are not exactly the same.
1244 * I believe 5.5.4 is the best one, so we follow 5.5.4.
1246 if (isipv6 && !V_ip6_use_deprecated) {
1247 struct in6_ifaddr *ia6;
1249 ia6 = ip6_getdstifaddr(m);
1251 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1252 ifa_free(&ia6->ia_ifa);
1253 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1254 log(LOG_DEBUG, "%s; %s: Listen socket: "
1255 "Connection attempt to deprecated "
1256 "IPv6 address rejected\n",
1258 rstreason = BANDLIM_RST_OPENPORT;
1262 ifa_free(&ia6->ia_ifa);
1266 * Basic sanity checks on incoming SYN requests:
1267 * Don't respond if the destination is a link layer
1268 * broadcast according to RFC1122 4.2.3.10, p. 104.
1269 * If it is from this socket it must be forged.
1270 * Don't respond if the source or destination is a
1271 * global or subnet broad- or multicast address.
1272 * Note that it is quite possible to receive unicast
1273 * link-layer packets with a broadcast IP address. Use
1274 * in_broadcast() to find them.
1276 if (m->m_flags & (M_BCAST|M_MCAST)) {
1277 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1278 log(LOG_DEBUG, "%s; %s: Listen socket: "
1279 "Connection attempt from broad- or multicast "
1280 "link layer address ignored\n", s, __func__);
1285 if (th->th_dport == th->th_sport &&
1286 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) {
1287 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1288 log(LOG_DEBUG, "%s; %s: Listen socket: "
1289 "Connection attempt to/from self "
1290 "ignored\n", s, __func__);
1293 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1294 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) {
1295 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1296 log(LOG_DEBUG, "%s; %s: Listen socket: "
1297 "Connection attempt from/to multicast "
1298 "address ignored\n", s, __func__);
1303 #if defined(INET) && defined(INET6)
1308 if (th->th_dport == th->th_sport &&
1309 ip->ip_dst.s_addr == ip->ip_src.s_addr) {
1310 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1311 log(LOG_DEBUG, "%s; %s: Listen socket: "
1312 "Connection attempt from/to self "
1313 "ignored\n", s, __func__);
1316 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1317 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1318 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1319 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
1320 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1321 log(LOG_DEBUG, "%s; %s: Listen socket: "
1322 "Connection attempt from/to broad- "
1323 "or multicast address ignored\n",
1330 * SYN appears to be valid. Create compressed TCP state
1334 if (so->so_options & SO_DEBUG)
1335 tcp_trace(TA_INPUT, ostate, tp,
1336 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1338 tcp_dooptions(&to, optp, optlen, TO_SYN);
1339 syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL);
1341 * Entry added to syncache and mbuf consumed.
1342 * Everything already unlocked by syncache_add().
1344 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1345 return (IPPROTO_DONE);
1346 } else if (tp->t_state == TCPS_LISTEN) {
1348 * When a listen socket is torn down the SO_ACCEPTCONN
1349 * flag is removed first while connections are drained
1350 * from the accept queue in a unlock/lock cycle of the
1351 * ACCEPT_LOCK, opening a race condition allowing a SYN
1352 * attempt go through unhandled.
1357 #ifdef TCP_SIGNATURE
1358 if (sig_checked == 0) {
1359 tcp_dooptions(&to, optp, optlen,
1360 (thflags & TH_SYN) ? TO_SYN : 0);
1361 if (!tcp_signature_verify_input(m, off0, tlen, optlen, &to,
1365 * In SYN_SENT state if it receives an RST, it is
1366 * allowed for further processing.
1368 if ((thflags & TH_RST) == 0 ||
1369 (tp->t_state == TCPS_SYN_SENT) == 0)
1376 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th);
1379 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later
1380 * state. tcp_do_segment() always consumes the mbuf chain, unlocks
1381 * the inpcb, and unlocks pcbinfo.
1383 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos, ti_locked);
1384 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1385 return (IPPROTO_DONE);
1388 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th);
1390 if (ti_locked == TI_WLOCKED) {
1391 INP_INFO_WUNLOCK(&V_tcbinfo);
1392 ti_locked = TI_UNLOCKED;
1396 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset "
1397 "ti_locked: %d", __func__, ti_locked));
1398 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1403 tcp_dropwithreset(m, th, tp, tlen, rstreason);
1406 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
1407 m = NULL; /* mbuf chain got consumed. */
1412 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th);
1414 if (ti_locked == TI_WLOCKED) {
1415 INP_INFO_WUNLOCK(&V_tcbinfo);
1416 ti_locked = TI_UNLOCKED;
1420 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock "
1421 "ti_locked: %d", __func__, ti_locked));
1422 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1430 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1435 return (IPPROTO_DONE);
1439 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so,
1440 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos,
1443 int thflags, acked, ourfinisacked, needoutput = 0;
1444 int rstreason, todrop, win;
1447 struct in_conninfo *inc;
1453 * The size of tcp_saveipgen must be the size of the max ip header,
1456 u_char tcp_saveipgen[IP6_HDR_LEN];
1457 struct tcphdr tcp_savetcp;
1460 thflags = th->th_flags;
1461 inc = &tp->t_inpcb->inp_inc;
1462 tp->sackhint.last_sack_ack = 0;
1465 * If this is either a state-changing packet or current state isn't
1466 * established, we require a write lock on tcbinfo. Otherwise, we
1467 * allow the tcbinfo to be in either alocked or unlocked, as the
1468 * caller may have unnecessarily acquired a write lock due to a race.
1470 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 ||
1471 tp->t_state != TCPS_ESTABLISHED) {
1472 KASSERT(ti_locked == TI_WLOCKED, ("%s ti_locked %d for "
1473 "SYN/FIN/RST/!EST", __func__, ti_locked));
1474 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1477 if (ti_locked == TI_WLOCKED)
1478 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1480 KASSERT(ti_locked == TI_UNLOCKED, ("%s: EST "
1481 "ti_locked: %d", __func__, ti_locked));
1482 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1486 INP_WLOCK_ASSERT(tp->t_inpcb);
1487 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN",
1489 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT",
1493 * Segment received on connection.
1494 * Reset idle time and keep-alive timer.
1495 * XXX: This should be done after segment
1496 * validation to ignore broken/spoofed segs.
1498 tp->t_rcvtime = ticks;
1499 if (TCPS_HAVEESTABLISHED(tp->t_state))
1500 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
1503 * Unscale the window into a 32-bit value.
1504 * For the SYN_SENT state the scale is zero.
1506 tiwin = th->th_win << tp->snd_scale;
1509 * TCP ECN processing.
1511 if (tp->t_flags & TF_ECN_PERMIT) {
1512 if (thflags & TH_CWR)
1513 tp->t_flags &= ~TF_ECN_SND_ECE;
1514 switch (iptos & IPTOS_ECN_MASK) {
1516 tp->t_flags |= TF_ECN_SND_ECE;
1517 TCPSTAT_INC(tcps_ecn_ce);
1519 case IPTOS_ECN_ECT0:
1520 TCPSTAT_INC(tcps_ecn_ect0);
1522 case IPTOS_ECN_ECT1:
1523 TCPSTAT_INC(tcps_ecn_ect1);
1526 /* Congestion experienced. */
1527 if (thflags & TH_ECE) {
1528 cc_cong_signal(tp, th, CC_ECN);
1533 * Parse options on any incoming segment.
1535 tcp_dooptions(&to, (u_char *)(th + 1),
1536 (th->th_off << 2) - sizeof(struct tcphdr),
1537 (thflags & TH_SYN) ? TO_SYN : 0);
1540 * If echoed timestamp is later than the current time,
1541 * fall back to non RFC1323 RTT calculation. Normalize
1542 * timestamp if syncookies were used when this connection
1545 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1546 to.to_tsecr -= tp->ts_offset;
1547 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks()))
1551 * If timestamps were negotiated during SYN/ACK they should
1552 * appear on every segment during this session and vice versa.
1554 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) {
1555 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1556 log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1557 "no action\n", s, __func__);
1561 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) {
1562 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1563 log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
1564 "no action\n", s, __func__);
1570 * Process options only when we get SYN/ACK back. The SYN case
1571 * for incoming connections is handled in tcp_syncache.
1572 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
1573 * or <SYN,ACK>) segment itself is never scaled.
1574 * XXX this is traditional behavior, may need to be cleaned up.
1576 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1577 if ((to.to_flags & TOF_SCALE) &&
1578 (tp->t_flags & TF_REQ_SCALE)) {
1579 tp->t_flags |= TF_RCVD_SCALE;
1580 tp->snd_scale = to.to_wscale;
1583 * Initial send window. It will be updated with
1584 * the next incoming segment to the scaled value.
1586 tp->snd_wnd = th->th_win;
1587 if (to.to_flags & TOF_TS) {
1588 tp->t_flags |= TF_RCVD_TSTMP;
1589 tp->ts_recent = to.to_tsval;
1590 tp->ts_recent_age = tcp_ts_getticks();
1592 if (to.to_flags & TOF_MSS)
1593 tcp_mss(tp, to.to_mss);
1594 if ((tp->t_flags & TF_SACK_PERMIT) &&
1595 (to.to_flags & TOF_SACKPERM) == 0)
1596 tp->t_flags &= ~TF_SACK_PERMIT;
1600 * Header prediction: check for the two common cases
1601 * of a uni-directional data xfer. If the packet has
1602 * no control flags, is in-sequence, the window didn't
1603 * change and we're not retransmitting, it's a
1604 * candidate. If the length is zero and the ack moved
1605 * forward, we're the sender side of the xfer. Just
1606 * free the data acked & wake any higher level process
1607 * that was blocked waiting for space. If the length
1608 * is non-zero and the ack didn't move, we're the
1609 * receiver side. If we're getting packets in-order
1610 * (the reassembly queue is empty), add the data to
1611 * the socket buffer and note that we need a delayed ack.
1612 * Make sure that the hidden state-flags are also off.
1613 * Since we check for TCPS_ESTABLISHED first, it can only
1616 if (tp->t_state == TCPS_ESTABLISHED &&
1617 th->th_seq == tp->rcv_nxt &&
1618 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1619 tp->snd_nxt == tp->snd_max &&
1620 tiwin && tiwin == tp->snd_wnd &&
1621 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1622 tp->t_segq == NULL && ((to.to_flags & TOF_TS) == 0 ||
1623 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) {
1626 * If last ACK falls within this segment's sequence numbers,
1627 * record the timestamp.
1628 * NOTE that the test is modified according to the latest
1629 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1631 if ((to.to_flags & TOF_TS) != 0 &&
1632 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1633 tp->ts_recent_age = tcp_ts_getticks();
1634 tp->ts_recent = to.to_tsval;
1638 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1639 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1640 !IN_RECOVERY(tp->t_flags) &&
1641 (to.to_flags & TOF_SACK) == 0 &&
1642 TAILQ_EMPTY(&tp->snd_holes)) {
1644 * This is a pure ack for outstanding data.
1646 if (ti_locked == TI_WLOCKED)
1647 INP_INFO_WUNLOCK(&V_tcbinfo);
1648 ti_locked = TI_UNLOCKED;
1650 TCPSTAT_INC(tcps_predack);
1653 * "bad retransmit" recovery.
1655 if (tp->t_rxtshift == 1 &&
1656 tp->t_flags & TF_PREVVALID &&
1657 (int)(ticks - tp->t_badrxtwin) < 0) {
1658 cc_cong_signal(tp, th, CC_RTO_ERR);
1662 * Recalculate the transmit timer / rtt.
1664 * Some boxes send broken timestamp replies
1665 * during the SYN+ACK phase, ignore
1666 * timestamps of 0 or we could calculate a
1667 * huge RTT and blow up the retransmit timer.
1669 if ((to.to_flags & TOF_TS) != 0 &&
1673 t = tcp_ts_getticks() - to.to_tsecr;
1674 if (!tp->t_rttlow || tp->t_rttlow > t)
1677 TCP_TS_TO_TICKS(t) + 1);
1678 } else if (tp->t_rtttime &&
1679 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1680 if (!tp->t_rttlow ||
1681 tp->t_rttlow > ticks - tp->t_rtttime)
1682 tp->t_rttlow = ticks - tp->t_rtttime;
1684 ticks - tp->t_rtttime);
1686 acked = BYTES_THIS_ACK(tp, th);
1688 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
1689 hhook_run_tcp_est_in(tp, th, &to);
1691 TCPSTAT_INC(tcps_rcvackpack);
1692 TCPSTAT_ADD(tcps_rcvackbyte, acked);
1693 sbdrop(&so->so_snd, acked);
1694 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1695 SEQ_LEQ(th->th_ack, tp->snd_recover))
1696 tp->snd_recover = th->th_ack - 1;
1699 * Let the congestion control algorithm update
1700 * congestion control related information. This
1701 * typically means increasing the congestion
1704 cc_ack_received(tp, th, CC_ACK);
1706 tp->snd_una = th->th_ack;
1708 * Pull snd_wl2 up to prevent seq wrap relative
1711 tp->snd_wl2 = th->th_ack;
1714 ND6_HINT(tp); /* Some progress has been made. */
1717 * If all outstanding data are acked, stop
1718 * retransmit timer, otherwise restart timer
1719 * using current (possibly backed-off) value.
1720 * If process is waiting for space,
1721 * wakeup/selwakeup/signal. If data
1722 * are ready to send, let tcp_output
1723 * decide between more output or persist.
1726 if (so->so_options & SO_DEBUG)
1727 tcp_trace(TA_INPUT, ostate, tp,
1728 (void *)tcp_saveipgen,
1731 if (tp->snd_una == tp->snd_max)
1732 tcp_timer_activate(tp, TT_REXMT, 0);
1733 else if (!tcp_timer_active(tp, TT_PERSIST))
1734 tcp_timer_activate(tp, TT_REXMT,
1737 if (so->so_snd.sb_cc)
1738 (void) tcp_output(tp);
1741 } else if (th->th_ack == tp->snd_una &&
1742 tlen <= sbspace(&so->so_rcv)) {
1743 int newsize = 0; /* automatic sockbuf scaling */
1746 * This is a pure, in-sequence data packet with
1747 * nothing on the reassembly queue and we have enough
1748 * buffer space to take it.
1750 if (ti_locked == TI_WLOCKED)
1751 INP_INFO_WUNLOCK(&V_tcbinfo);
1752 ti_locked = TI_UNLOCKED;
1754 /* Clean receiver SACK report if present */
1755 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks)
1756 tcp_clean_sackreport(tp);
1757 TCPSTAT_INC(tcps_preddat);
1758 tp->rcv_nxt += tlen;
1760 * Pull snd_wl1 up to prevent seq wrap relative to
1763 tp->snd_wl1 = th->th_seq;
1765 * Pull rcv_up up to prevent seq wrap relative to
1768 tp->rcv_up = tp->rcv_nxt;
1769 TCPSTAT_INC(tcps_rcvpack);
1770 TCPSTAT_ADD(tcps_rcvbyte, tlen);
1771 ND6_HINT(tp); /* Some progress has been made */
1773 if (so->so_options & SO_DEBUG)
1774 tcp_trace(TA_INPUT, ostate, tp,
1775 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1778 * Automatic sizing of receive socket buffer. Often the send
1779 * buffer size is not optimally adjusted to the actual network
1780 * conditions at hand (delay bandwidth product). Setting the
1781 * buffer size too small limits throughput on links with high
1782 * bandwidth and high delay (eg. trans-continental/oceanic links).
1784 * On the receive side the socket buffer memory is only rarely
1785 * used to any significant extent. This allows us to be much
1786 * more aggressive in scaling the receive socket buffer. For
1787 * the case that the buffer space is actually used to a large
1788 * extent and we run out of kernel memory we can simply drop
1789 * the new segments; TCP on the sender will just retransmit it
1790 * later. Setting the buffer size too big may only consume too
1791 * much kernel memory if the application doesn't read() from
1792 * the socket or packet loss or reordering makes use of the
1795 * The criteria to step up the receive buffer one notch are:
1796 * 1. Application has not set receive buffer size with
1797 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE.
1798 * 2. the number of bytes received during the time it takes
1799 * one timestamp to be reflected back to us (the RTT);
1800 * 3. received bytes per RTT is within seven eighth of the
1801 * current socket buffer size;
1802 * 4. receive buffer size has not hit maximal automatic size;
1804 * This algorithm does one step per RTT at most and only if
1805 * we receive a bulk stream w/o packet losses or reorderings.
1806 * Shrinking the buffer during idle times is not necessary as
1807 * it doesn't consume any memory when idle.
1809 * TODO: Only step up if the application is actually serving
1810 * the buffer to better manage the socket buffer resources.
1812 if (V_tcp_do_autorcvbuf &&
1814 (so->so_rcv.sb_flags & SB_AUTOSIZE)) {
1815 if (TSTMP_GT(to.to_tsecr, tp->rfbuf_ts) &&
1816 to.to_tsecr - tp->rfbuf_ts < hz) {
1818 (so->so_rcv.sb_hiwat / 8 * 7) &&
1819 so->so_rcv.sb_hiwat <
1820 V_tcp_autorcvbuf_max) {
1822 min(so->so_rcv.sb_hiwat +
1823 V_tcp_autorcvbuf_inc,
1824 V_tcp_autorcvbuf_max);
1826 /* Start over with next RTT. */
1830 tp->rfbuf_cnt += tlen; /* add up */
1833 /* Add data to socket buffer. */
1834 SOCKBUF_LOCK(&so->so_rcv);
1835 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1839 * Set new socket buffer size.
1840 * Give up when limit is reached.
1843 if (!sbreserve_locked(&so->so_rcv,
1845 so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1846 m_adj(m, drop_hdrlen); /* delayed header drop */
1847 sbappendstream_locked(&so->so_rcv, m);
1849 /* NB: sorwakeup_locked() does an implicit unlock. */
1850 sorwakeup_locked(so);
1851 if (DELAY_ACK(tp, tlen)) {
1852 tp->t_flags |= TF_DELACK;
1854 tp->t_flags |= TF_ACKNOW;
1862 * Calculate amount of space in receive window,
1863 * and then do TCP input processing.
1864 * Receive window is amount of space in rcv queue,
1865 * but not less than advertised window.
1867 win = sbspace(&so->so_rcv);
1870 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1872 /* Reset receive buffer auto scaling when not in bulk receive mode. */
1876 switch (tp->t_state) {
1879 * If the state is SYN_RECEIVED:
1880 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1882 case TCPS_SYN_RECEIVED:
1883 if ((thflags & TH_ACK) &&
1884 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1885 SEQ_GT(th->th_ack, tp->snd_max))) {
1886 rstreason = BANDLIM_RST_OPENPORT;
1892 * If the state is SYN_SENT:
1893 * if seg contains an ACK, but not for our SYN, drop the input.
1894 * if seg contains a RST, then drop the connection.
1895 * if seg does not contain SYN, then drop it.
1896 * Otherwise this is an acceptable SYN segment
1897 * initialize tp->rcv_nxt and tp->irs
1898 * if seg contains ack then advance tp->snd_una
1899 * if seg contains an ECE and ECN support is enabled, the stream
1901 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1902 * arrange for segment to be acked (eventually)
1903 * continue processing rest of data/controls, beginning with URG
1906 if ((thflags & TH_ACK) &&
1907 (SEQ_LEQ(th->th_ack, tp->iss) ||
1908 SEQ_GT(th->th_ack, tp->snd_max))) {
1909 rstreason = BANDLIM_UNLIMITED;
1912 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) {
1913 TCP_PROBE5(connect__refused, NULL, tp,
1914 mtod(m, const char *), tp, th);
1915 tp = tcp_drop(tp, ECONNREFUSED);
1917 if (thflags & TH_RST)
1919 if (!(thflags & TH_SYN))
1922 tp->irs = th->th_seq;
1924 if (thflags & TH_ACK) {
1925 TCPSTAT_INC(tcps_connects);
1928 mac_socketpeer_set_from_mbuf(m, so);
1930 /* Do window scaling on this connection? */
1931 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1932 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1933 tp->rcv_scale = tp->request_r_scale;
1935 tp->rcv_adv += imin(tp->rcv_wnd,
1936 TCP_MAXWIN << tp->rcv_scale);
1937 tp->snd_una++; /* SYN is acked */
1939 * If there's data, delay ACK; if there's also a FIN
1940 * ACKNOW will be turned on later.
1942 if (DELAY_ACK(tp, tlen) && tlen != 0)
1943 tcp_timer_activate(tp, TT_DELACK,
1946 tp->t_flags |= TF_ACKNOW;
1948 if ((thflags & TH_ECE) && V_tcp_do_ecn) {
1949 tp->t_flags |= TF_ECN_PERMIT;
1950 TCPSTAT_INC(tcps_ecn_shs);
1954 * Received <SYN,ACK> in SYN_SENT[*] state.
1956 * SYN_SENT --> ESTABLISHED
1957 * SYN_SENT* --> FIN_WAIT_1
1959 tp->t_starttime = ticks;
1960 if (tp->t_flags & TF_NEEDFIN) {
1961 tcp_state_change(tp, TCPS_FIN_WAIT_1);
1962 tp->t_flags &= ~TF_NEEDFIN;
1965 tcp_state_change(tp, TCPS_ESTABLISHED);
1966 TCP_PROBE5(connect__established, NULL, tp,
1967 mtod(m, const char *), tp, th);
1969 tcp_timer_activate(tp, TT_KEEP,
1974 * Received initial SYN in SYN-SENT[*] state =>
1975 * simultaneous open.
1976 * If it succeeds, connection is * half-synchronized.
1977 * Otherwise, do 3-way handshake:
1978 * SYN-SENT -> SYN-RECEIVED
1979 * SYN-SENT* -> SYN-RECEIVED*
1981 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
1982 tcp_timer_activate(tp, TT_REXMT, 0);
1983 tcp_state_change(tp, TCPS_SYN_RECEIVED);
1986 KASSERT(ti_locked == TI_WLOCKED, ("%s: trimthenstep6: "
1987 "ti_locked %d", __func__, ti_locked));
1988 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1989 INP_WLOCK_ASSERT(tp->t_inpcb);
1992 * Advance th->th_seq to correspond to first data byte.
1993 * If data, trim to stay within window,
1994 * dropping FIN if necessary.
1997 if (tlen > tp->rcv_wnd) {
1998 todrop = tlen - tp->rcv_wnd;
2002 TCPSTAT_INC(tcps_rcvpackafterwin);
2003 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2005 tp->snd_wl1 = th->th_seq - 1;
2006 tp->rcv_up = th->th_seq;
2008 * Client side of transaction: already sent SYN and data.
2009 * If the remote host used T/TCP to validate the SYN,
2010 * our data will be ACK'd; if so, enter normal data segment
2011 * processing in the middle of step 5, ack processing.
2012 * Otherwise, goto step 6.
2014 if (thflags & TH_ACK)
2020 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
2021 * do normal processing.
2023 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later.
2027 break; /* continue normal processing */
2031 * States other than LISTEN or SYN_SENT.
2032 * First check the RST flag and sequence number since reset segments
2033 * are exempt from the timestamp and connection count tests. This
2034 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
2035 * below which allowed reset segments in half the sequence space
2036 * to fall though and be processed (which gives forged reset
2037 * segments with a random sequence number a 50 percent chance of
2038 * killing a connection).
2039 * Then check timestamp, if present.
2040 * Then check the connection count, if present.
2041 * Then check that at least some bytes of segment are within
2042 * receive window. If segment begins before rcv_nxt,
2043 * drop leading data (and SYN); if nothing left, just ack.
2046 * If the RST bit is set, check the sequence number to see
2047 * if this is a valid reset segment.
2049 * In all states except SYN-SENT, all reset (RST) segments
2050 * are validated by checking their SEQ-fields. A reset is
2051 * valid if its sequence number is in the window.
2052 * Note: this does not take into account delayed ACKs, so
2053 * we should test against last_ack_sent instead of rcv_nxt.
2054 * The sequence number in the reset segment is normally an
2055 * echo of our outgoing acknowlegement numbers, but some hosts
2056 * send a reset with the sequence number at the rightmost edge
2057 * of our receive window, and we have to handle this case.
2058 * Note 2: Paul Watson's paper "Slipping in the Window" has shown
2059 * that brute force RST attacks are possible. To combat this,
2060 * we use a much stricter check while in the ESTABLISHED state,
2061 * only accepting RSTs where the sequence number is equal to
2062 * last_ack_sent. In all other states (the states in which a
2063 * RST is more likely), the more permissive check is used.
2064 * If we have multiple segments in flight, the initial reset
2065 * segment sequence numbers will be to the left of last_ack_sent,
2066 * but they will eventually catch up.
2067 * In any case, it never made sense to trim reset segments to
2068 * fit the receive window since RFC 1122 says:
2069 * 4.2.2.12 RST Segment: RFC-793 Section 3.4
2071 * A TCP SHOULD allow a received RST segment to include data.
2074 * It has been suggested that a RST segment could contain
2075 * ASCII text that encoded and explained the cause of the
2076 * RST. No standard has yet been established for such
2079 * If the reset segment passes the sequence number test examine
2081 * SYN_RECEIVED STATE:
2082 * If passive open, return to LISTEN state.
2083 * If active open, inform user that connection was refused.
2084 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
2085 * Inform user that connection was reset, and close tcb.
2086 * CLOSING, LAST_ACK STATES:
2089 * Drop the segment - see Stevens, vol. 2, p. 964 and
2092 if (thflags & TH_RST) {
2093 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) &&
2094 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
2095 switch (tp->t_state) {
2097 case TCPS_SYN_RECEIVED:
2098 so->so_error = ECONNREFUSED;
2101 case TCPS_ESTABLISHED:
2102 if (V_tcp_insecure_rst == 0 &&
2103 !(SEQ_GEQ(th->th_seq, tp->rcv_nxt - 1) &&
2104 SEQ_LEQ(th->th_seq, tp->rcv_nxt + 1)) &&
2105 !(SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) &&
2106 SEQ_LEQ(th->th_seq, tp->last_ack_sent + 1))) {
2107 TCPSTAT_INC(tcps_badrst);
2111 case TCPS_FIN_WAIT_1:
2112 case TCPS_FIN_WAIT_2:
2113 case TCPS_CLOSE_WAIT:
2114 so->so_error = ECONNRESET;
2116 KASSERT(ti_locked == TI_WLOCKED,
2117 ("tcp_do_segment: TH_RST 1 ti_locked %d",
2119 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2121 tcp_state_change(tp, TCPS_CLOSED);
2122 TCPSTAT_INC(tcps_drops);
2128 KASSERT(ti_locked == TI_WLOCKED,
2129 ("tcp_do_segment: TH_RST 2 ti_locked %d",
2131 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2141 * RFC 1323 PAWS: If we have a timestamp reply on this segment
2142 * and it's less than ts_recent, drop it.
2144 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
2145 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
2147 /* Check to see if ts_recent is over 24 days old. */
2148 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) {
2150 * Invalidate ts_recent. If this segment updates
2151 * ts_recent, the age will be reset later and ts_recent
2152 * will get a valid value. If it does not, setting
2153 * ts_recent to zero will at least satisfy the
2154 * requirement that zero be placed in the timestamp
2155 * echo reply when ts_recent isn't valid. The
2156 * age isn't reset until we get a valid ts_recent
2157 * because we don't want out-of-order segments to be
2158 * dropped when ts_recent is old.
2162 TCPSTAT_INC(tcps_rcvduppack);
2163 TCPSTAT_ADD(tcps_rcvdupbyte, tlen);
2164 TCPSTAT_INC(tcps_pawsdrop);
2172 * In the SYN-RECEIVED state, validate that the packet belongs to
2173 * this connection before trimming the data to fit the receive
2174 * window. Check the sequence number versus IRS since we know
2175 * the sequence numbers haven't wrapped. This is a partial fix
2176 * for the "LAND" DoS attack.
2178 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
2179 rstreason = BANDLIM_RST_OPENPORT;
2183 todrop = tp->rcv_nxt - th->th_seq;
2186 * If this is a duplicate SYN for our current connection,
2187 * advance over it and pretend and it's not a SYN.
2189 if (thflags & TH_SYN && th->th_seq == tp->irs) {
2199 * Following if statement from Stevens, vol. 2, p. 960.
2202 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
2204 * Any valid FIN must be to the left of the window.
2205 * At this point the FIN must be a duplicate or out
2206 * of sequence; drop it.
2211 * Send an ACK to resynchronize and drop any data.
2212 * But keep on processing for RST or ACK.
2214 tp->t_flags |= TF_ACKNOW;
2216 TCPSTAT_INC(tcps_rcvduppack);
2217 TCPSTAT_ADD(tcps_rcvdupbyte, todrop);
2219 TCPSTAT_INC(tcps_rcvpartduppack);
2220 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop);
2222 drop_hdrlen += todrop; /* drop from the top afterwards */
2223 th->th_seq += todrop;
2225 if (th->th_urp > todrop)
2226 th->th_urp -= todrop;
2234 * If new data are received on a connection after the
2235 * user processes are gone, then RST the other end.
2237 if ((so->so_state & SS_NOFDREF) &&
2238 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
2239 KASSERT(ti_locked == TI_WLOCKED, ("%s: SS_NOFDEREF && "
2240 "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked));
2241 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2243 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
2244 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data "
2245 "after socket was closed, "
2246 "sending RST and removing tcpcb\n",
2247 s, __func__, tcpstates[tp->t_state], tlen);
2251 TCPSTAT_INC(tcps_rcvafterclose);
2252 rstreason = BANDLIM_UNLIMITED;
2257 * If segment ends after window, drop trailing data
2258 * (and PUSH and FIN); if nothing left, just ACK.
2260 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
2262 TCPSTAT_INC(tcps_rcvpackafterwin);
2263 if (todrop >= tlen) {
2264 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen);
2266 * If window is closed can only take segments at
2267 * window edge, and have to drop data and PUSH from
2268 * incoming segments. Continue processing, but
2269 * remember to ack. Otherwise, drop segment
2272 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2273 tp->t_flags |= TF_ACKNOW;
2274 TCPSTAT_INC(tcps_rcvwinprobe);
2278 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2281 thflags &= ~(TH_PUSH|TH_FIN);
2285 * If last ACK falls within this segment's sequence numbers,
2286 * record its timestamp.
2288 * 1) That the test incorporates suggestions from the latest
2289 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
2290 * 2) That updating only on newer timestamps interferes with
2291 * our earlier PAWS tests, so this check should be solely
2292 * predicated on the sequence space of this segment.
2293 * 3) That we modify the segment boundary check to be
2294 * Last.ACK.Sent <= SEG.SEQ + SEG.Len
2295 * instead of RFC1323's
2296 * Last.ACK.Sent < SEG.SEQ + SEG.Len,
2297 * This modified check allows us to overcome RFC1323's
2298 * limitations as described in Stevens TCP/IP Illustrated
2299 * Vol. 2 p.869. In such cases, we can still calculate the
2300 * RTT correctly when RCV.NXT == Last.ACK.Sent.
2302 if ((to.to_flags & TOF_TS) != 0 &&
2303 SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2304 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
2305 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
2306 tp->ts_recent_age = tcp_ts_getticks();
2307 tp->ts_recent = to.to_tsval;
2311 * If a SYN is in the window, then this is an
2312 * error and we send an RST and drop the connection.
2314 if (thflags & TH_SYN) {
2315 KASSERT(ti_locked == TI_WLOCKED,
2316 ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked));
2317 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2319 tp = tcp_drop(tp, ECONNRESET);
2320 rstreason = BANDLIM_UNLIMITED;
2325 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
2326 * flag is on (half-synchronized state), then queue data for
2327 * later processing; else drop segment and return.
2329 if ((thflags & TH_ACK) == 0) {
2330 if (tp->t_state == TCPS_SYN_RECEIVED ||
2331 (tp->t_flags & TF_NEEDSYN))
2333 else if (tp->t_flags & TF_ACKNOW)
2342 switch (tp->t_state) {
2345 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
2346 * ESTABLISHED state and continue processing.
2347 * The ACK was checked above.
2349 case TCPS_SYN_RECEIVED:
2351 TCPSTAT_INC(tcps_connects);
2353 /* Do window scaling? */
2354 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2355 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2356 tp->rcv_scale = tp->request_r_scale;
2357 tp->snd_wnd = tiwin;
2361 * SYN-RECEIVED -> ESTABLISHED
2362 * SYN-RECEIVED* -> FIN-WAIT-1
2364 tp->t_starttime = ticks;
2365 if (tp->t_flags & TF_NEEDFIN) {
2366 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2367 tp->t_flags &= ~TF_NEEDFIN;
2369 tcp_state_change(tp, TCPS_ESTABLISHED);
2370 TCP_PROBE5(accept__established, NULL, tp,
2371 mtod(m, const char *), tp, th);
2373 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
2376 * If segment contains data or ACK, will call tcp_reass()
2377 * later; if not, do so now to pass queued data to user.
2379 if (tlen == 0 && (thflags & TH_FIN) == 0)
2380 (void) tcp_reass(tp, (struct tcphdr *)0, 0,
2382 tp->snd_wl1 = th->th_seq - 1;
2386 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2387 * ACKs. If the ack is in the range
2388 * tp->snd_una < th->th_ack <= tp->snd_max
2389 * then advance tp->snd_una to th->th_ack and drop
2390 * data from the retransmission queue. If this ACK reflects
2391 * more up to date window information we update our window information.
2393 case TCPS_ESTABLISHED:
2394 case TCPS_FIN_WAIT_1:
2395 case TCPS_FIN_WAIT_2:
2396 case TCPS_CLOSE_WAIT:
2399 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2400 TCPSTAT_INC(tcps_rcvacktoomuch);
2403 if ((tp->t_flags & TF_SACK_PERMIT) &&
2404 ((to.to_flags & TOF_SACK) ||
2405 !TAILQ_EMPTY(&tp->snd_holes)))
2406 tcp_sack_doack(tp, &to, th->th_ack);
2408 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
2409 hhook_run_tcp_est_in(tp, th, &to);
2411 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2412 if (tlen == 0 && tiwin == tp->snd_wnd) {
2414 * If this is the first time we've seen a
2415 * FIN from the remote, this is not a
2416 * duplicate and it needs to be processed
2417 * normally. This happens during a
2418 * simultaneous close.
2420 if ((thflags & TH_FIN) &&
2421 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2425 TCPSTAT_INC(tcps_rcvdupack);
2427 * If we have outstanding data (other than
2428 * a window probe), this is a completely
2429 * duplicate ack (ie, window info didn't
2430 * change and FIN isn't set),
2431 * the ack is the biggest we've
2432 * seen and we've seen exactly our rexmt
2433 * threshhold of them, assume a packet
2434 * has been dropped and retransmit it.
2435 * Kludge snd_nxt & the congestion
2436 * window so we send only this one
2439 * We know we're losing at the current
2440 * window size so do congestion avoidance
2441 * (set ssthresh to half the current window
2442 * and pull our congestion window back to
2443 * the new ssthresh).
2445 * Dup acks mean that packets have left the
2446 * network (they're now cached at the receiver)
2447 * so bump cwnd by the amount in the receiver
2448 * to keep a constant cwnd packets in the
2451 * When using TCP ECN, notify the peer that
2452 * we reduced the cwnd.
2454 if (!tcp_timer_active(tp, TT_REXMT) ||
2455 th->th_ack != tp->snd_una)
2457 else if (++tp->t_dupacks > tcprexmtthresh ||
2458 IN_FASTRECOVERY(tp->t_flags)) {
2459 cc_ack_received(tp, th, CC_DUPACK);
2460 if ((tp->t_flags & TF_SACK_PERMIT) &&
2461 IN_FASTRECOVERY(tp->t_flags)) {
2465 * Compute the amount of data in flight first.
2466 * We can inject new data into the pipe iff
2467 * we have less than 1/2 the original window's
2468 * worth of data in flight.
2470 awnd = (tp->snd_nxt - tp->snd_fack) +
2471 tp->sackhint.sack_bytes_rexmit;
2472 if (awnd < tp->snd_ssthresh) {
2473 tp->snd_cwnd += tp->t_maxseg;
2474 if (tp->snd_cwnd > tp->snd_ssthresh)
2475 tp->snd_cwnd = tp->snd_ssthresh;
2478 tp->snd_cwnd += tp->t_maxseg;
2479 (void) tcp_output(tp);
2481 } else if (tp->t_dupacks == tcprexmtthresh) {
2482 tcp_seq onxt = tp->snd_nxt;
2485 * If we're doing sack, check to
2486 * see if we're already in sack
2487 * recovery. If we're not doing sack,
2488 * check to see if we're in newreno
2491 if (tp->t_flags & TF_SACK_PERMIT) {
2492 if (IN_FASTRECOVERY(tp->t_flags)) {
2497 if (SEQ_LEQ(th->th_ack,
2503 /* Congestion signal before ack. */
2504 cc_cong_signal(tp, th, CC_NDUPACK);
2505 cc_ack_received(tp, th, CC_DUPACK);
2506 tcp_timer_activate(tp, TT_REXMT, 0);
2508 if (tp->t_flags & TF_SACK_PERMIT) {
2510 tcps_sack_recovery_episode);
2511 tp->sack_newdata = tp->snd_nxt;
2512 tp->snd_cwnd = tp->t_maxseg;
2513 (void) tcp_output(tp);
2516 tp->snd_nxt = th->th_ack;
2517 tp->snd_cwnd = tp->t_maxseg;
2518 (void) tcp_output(tp);
2519 KASSERT(tp->snd_limited <= 2,
2520 ("%s: tp->snd_limited too big",
2522 tp->snd_cwnd = tp->snd_ssthresh +
2524 (tp->t_dupacks - tp->snd_limited);
2525 if (SEQ_GT(onxt, tp->snd_nxt))
2528 } else if (V_tcp_do_rfc3042) {
2529 cc_ack_received(tp, th, CC_DUPACK);
2530 u_long oldcwnd = tp->snd_cwnd;
2531 tcp_seq oldsndmax = tp->snd_max;
2535 KASSERT(tp->t_dupacks == 1 ||
2537 ("%s: dupacks not 1 or 2",
2539 if (tp->t_dupacks == 1)
2540 tp->snd_limited = 0;
2542 (tp->snd_nxt - tp->snd_una) +
2543 (tp->t_dupacks - tp->snd_limited) *
2546 * Only call tcp_output when there
2547 * is new data available to be sent.
2548 * Otherwise we would send pure ACKs.
2550 SOCKBUF_LOCK(&so->so_snd);
2551 avail = so->so_snd.sb_cc -
2552 (tp->snd_nxt - tp->snd_una);
2553 SOCKBUF_UNLOCK(&so->so_snd);
2555 (void) tcp_output(tp);
2556 sent = tp->snd_max - oldsndmax;
2557 if (sent > tp->t_maxseg) {
2558 KASSERT((tp->t_dupacks == 2 &&
2559 tp->snd_limited == 0) ||
2560 (sent == tp->t_maxseg + 1 &&
2561 tp->t_flags & TF_SENTFIN),
2562 ("%s: sent too much",
2564 tp->snd_limited = 2;
2565 } else if (sent > 0)
2567 tp->snd_cwnd = oldcwnd;
2575 KASSERT(SEQ_GT(th->th_ack, tp->snd_una),
2576 ("%s: th_ack <= snd_una", __func__));
2579 * If the congestion window was inflated to account
2580 * for the other side's cached packets, retract it.
2582 if (IN_FASTRECOVERY(tp->t_flags)) {
2583 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2584 if (tp->t_flags & TF_SACK_PERMIT)
2585 tcp_sack_partialack(tp, th);
2587 tcp_newreno_partial_ack(tp, th);
2589 cc_post_recovery(tp, th);
2593 * If we reach this point, ACK is not a duplicate,
2594 * i.e., it ACKs something we sent.
2596 if (tp->t_flags & TF_NEEDSYN) {
2598 * T/TCP: Connection was half-synchronized, and our
2599 * SYN has been ACK'd (so connection is now fully
2600 * synchronized). Go to non-starred state,
2601 * increment snd_una for ACK of SYN, and check if
2602 * we can do window scaling.
2604 tp->t_flags &= ~TF_NEEDSYN;
2606 /* Do window scaling? */
2607 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2608 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2609 tp->rcv_scale = tp->request_r_scale;
2610 /* Send window already scaled. */
2615 INP_WLOCK_ASSERT(tp->t_inpcb);
2617 acked = BYTES_THIS_ACK(tp, th);
2618 TCPSTAT_INC(tcps_rcvackpack);
2619 TCPSTAT_ADD(tcps_rcvackbyte, acked);
2622 * If we just performed our first retransmit, and the ACK
2623 * arrives within our recovery window, then it was a mistake
2624 * to do the retransmit in the first place. Recover our
2625 * original cwnd and ssthresh, and proceed to transmit where
2628 if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID &&
2629 (int)(ticks - tp->t_badrxtwin) < 0)
2630 cc_cong_signal(tp, th, CC_RTO_ERR);
2633 * If we have a timestamp reply, update smoothed
2634 * round trip time. If no timestamp is present but
2635 * transmit timer is running and timed sequence
2636 * number was acked, update smoothed round trip time.
2637 * Since we now have an rtt measurement, cancel the
2638 * timer backoff (cf., Phil Karn's retransmit alg.).
2639 * Recompute the initial retransmit timer.
2641 * Some boxes send broken timestamp replies
2642 * during the SYN+ACK phase, ignore
2643 * timestamps of 0 or we could calculate a
2644 * huge RTT and blow up the retransmit timer.
2646 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) {
2649 t = tcp_ts_getticks() - to.to_tsecr;
2650 if (!tp->t_rttlow || tp->t_rttlow > t)
2652 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1);
2653 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2654 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2655 tp->t_rttlow = ticks - tp->t_rtttime;
2656 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2660 * If all outstanding data is acked, stop retransmit
2661 * timer and remember to restart (more output or persist).
2662 * If there is more data to be acked, restart retransmit
2663 * timer, using current (possibly backed-off) value.
2665 if (th->th_ack == tp->snd_max) {
2666 tcp_timer_activate(tp, TT_REXMT, 0);
2668 } else if (!tcp_timer_active(tp, TT_PERSIST))
2669 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
2672 * If no data (only SYN) was ACK'd,
2673 * skip rest of ACK processing.
2679 * Let the congestion control algorithm update congestion
2680 * control related information. This typically means increasing
2681 * the congestion window.
2683 cc_ack_received(tp, th, CC_ACK);
2685 SOCKBUF_LOCK(&so->so_snd);
2686 if (acked > so->so_snd.sb_cc) {
2687 tp->snd_wnd -= so->so_snd.sb_cc;
2688 mfree = sbcut_locked(&so->so_snd,
2689 (int)so->so_snd.sb_cc);
2692 mfree = sbcut_locked(&so->so_snd, acked);
2693 tp->snd_wnd -= acked;
2696 /* NB: sowwakeup_locked() does an implicit unlock. */
2697 sowwakeup_locked(so);
2699 /* Detect una wraparound. */
2700 if (!IN_RECOVERY(tp->t_flags) &&
2701 SEQ_GT(tp->snd_una, tp->snd_recover) &&
2702 SEQ_LEQ(th->th_ack, tp->snd_recover))
2703 tp->snd_recover = th->th_ack - 1;
2704 /* XXXLAS: Can this be moved up into cc_post_recovery? */
2705 if (IN_RECOVERY(tp->t_flags) &&
2706 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2707 EXIT_RECOVERY(tp->t_flags);
2709 tp->snd_una = th->th_ack;
2710 if (tp->t_flags & TF_SACK_PERMIT) {
2711 if (SEQ_GT(tp->snd_una, tp->snd_recover))
2712 tp->snd_recover = tp->snd_una;
2714 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2715 tp->snd_nxt = tp->snd_una;
2717 switch (tp->t_state) {
2720 * In FIN_WAIT_1 STATE in addition to the processing
2721 * for the ESTABLISHED state if our FIN is now acknowledged
2722 * then enter FIN_WAIT_2.
2724 case TCPS_FIN_WAIT_1:
2725 if (ourfinisacked) {
2727 * If we can't receive any more
2728 * data, then closing user can proceed.
2729 * Starting the timer is contrary to the
2730 * specification, but if we don't get a FIN
2731 * we'll hang forever.
2734 * we should release the tp also, and use a
2737 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2738 soisdisconnected(so);
2739 tcp_timer_activate(tp, TT_2MSL,
2740 (tcp_fast_finwait2_recycle ?
2741 tcp_finwait2_timeout :
2744 tcp_state_change(tp, TCPS_FIN_WAIT_2);
2749 * In CLOSING STATE in addition to the processing for
2750 * the ESTABLISHED state if the ACK acknowledges our FIN
2751 * then enter the TIME-WAIT state, otherwise ignore
2755 if (ourfinisacked) {
2756 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2758 INP_INFO_WUNLOCK(&V_tcbinfo);
2765 * In LAST_ACK, we may still be waiting for data to drain
2766 * and/or to be acked, as well as for the ack of our FIN.
2767 * If our FIN is now acknowledged, delete the TCB,
2768 * enter the closed state and return.
2771 if (ourfinisacked) {
2772 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2781 INP_WLOCK_ASSERT(tp->t_inpcb);
2784 * Update window information.
2785 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2787 if ((thflags & TH_ACK) &&
2788 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2789 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2790 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2791 /* keep track of pure window updates */
2793 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2794 TCPSTAT_INC(tcps_rcvwinupd);
2795 tp->snd_wnd = tiwin;
2796 tp->snd_wl1 = th->th_seq;
2797 tp->snd_wl2 = th->th_ack;
2798 if (tp->snd_wnd > tp->max_sndwnd)
2799 tp->max_sndwnd = tp->snd_wnd;
2804 * Process segments with URG.
2806 if ((thflags & TH_URG) && th->th_urp &&
2807 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2809 * This is a kludge, but if we receive and accept
2810 * random urgent pointers, we'll crash in
2811 * soreceive. It's hard to imagine someone
2812 * actually wanting to send this much urgent data.
2814 SOCKBUF_LOCK(&so->so_rcv);
2815 if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2816 th->th_urp = 0; /* XXX */
2817 thflags &= ~TH_URG; /* XXX */
2818 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */
2819 goto dodata; /* XXX */
2822 * If this segment advances the known urgent pointer,
2823 * then mark the data stream. This should not happen
2824 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2825 * a FIN has been received from the remote side.
2826 * In these states we ignore the URG.
2828 * According to RFC961 (Assigned Protocols),
2829 * the urgent pointer points to the last octet
2830 * of urgent data. We continue, however,
2831 * to consider it to indicate the first octet
2832 * of data past the urgent section as the original
2833 * spec states (in one of two places).
2835 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2836 tp->rcv_up = th->th_seq + th->th_urp;
2837 so->so_oobmark = so->so_rcv.sb_cc +
2838 (tp->rcv_up - tp->rcv_nxt) - 1;
2839 if (so->so_oobmark == 0)
2840 so->so_rcv.sb_state |= SBS_RCVATMARK;
2842 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2844 SOCKBUF_UNLOCK(&so->so_rcv);
2846 * Remove out of band data so doesn't get presented to user.
2847 * This can happen independent of advancing the URG pointer,
2848 * but if two URG's are pending at once, some out-of-band
2849 * data may creep in... ick.
2851 if (th->th_urp <= (u_long)tlen &&
2852 !(so->so_options & SO_OOBINLINE)) {
2853 /* hdr drop is delayed */
2854 tcp_pulloutofband(so, th, m, drop_hdrlen);
2858 * If no out of band data is expected,
2859 * pull receive urgent pointer along
2860 * with the receive window.
2862 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2863 tp->rcv_up = tp->rcv_nxt;
2866 INP_WLOCK_ASSERT(tp->t_inpcb);
2869 * Process the segment text, merging it into the TCP sequencing queue,
2870 * and arranging for acknowledgment of receipt if necessary.
2871 * This process logically involves adjusting tp->rcv_wnd as data
2872 * is presented to the user (this happens in tcp_usrreq.c,
2873 * case PRU_RCVD). If a FIN has already been received on this
2874 * connection then we just ignore the text.
2876 if ((tlen || (thflags & TH_FIN)) &&
2877 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2878 tcp_seq save_start = th->th_seq;
2879 m_adj(m, drop_hdrlen); /* delayed header drop */
2881 * Insert segment which includes th into TCP reassembly queue
2882 * with control block tp. Set thflags to whether reassembly now
2883 * includes a segment with FIN. This handles the common case
2884 * inline (segment is the next to be received on an established
2885 * connection, and the queue is empty), avoiding linkage into
2886 * and removal from the queue and repetition of various
2888 * Set DELACK for segments received in order, but ack
2889 * immediately when segments are out of order (so
2890 * fast retransmit can work).
2892 if (th->th_seq == tp->rcv_nxt && tp->t_segq == NULL &&
2893 TCPS_HAVEESTABLISHED(tp->t_state)) {
2894 if (DELAY_ACK(tp, tlen))
2895 tp->t_flags |= TF_DELACK;
2897 tp->t_flags |= TF_ACKNOW;
2898 tp->rcv_nxt += tlen;
2899 thflags = th->th_flags & TH_FIN;
2900 TCPSTAT_INC(tcps_rcvpack);
2901 TCPSTAT_ADD(tcps_rcvbyte, tlen);
2903 SOCKBUF_LOCK(&so->so_rcv);
2904 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
2907 sbappendstream_locked(&so->so_rcv, m);
2908 /* NB: sorwakeup_locked() does an implicit unlock. */
2909 sorwakeup_locked(so);
2912 * XXX: Due to the header drop above "th" is
2913 * theoretically invalid by now. Fortunately
2914 * m_adj() doesn't actually frees any mbufs
2915 * when trimming from the head.
2917 thflags = tcp_reass(tp, th, &tlen, m);
2918 tp->t_flags |= TF_ACKNOW;
2920 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT))
2921 tcp_update_sack_list(tp, save_start, save_start + tlen);
2924 * Note the amount of data that peer has sent into
2925 * our window, in order to estimate the sender's
2929 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
2930 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2932 len = so->so_rcv.sb_hiwat;
2940 * If FIN is received ACK the FIN and let the user know
2941 * that the connection is closing.
2943 if (thflags & TH_FIN) {
2944 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2947 * If connection is half-synchronized
2948 * (ie NEEDSYN flag on) then delay ACK,
2949 * so it may be piggybacked when SYN is sent.
2950 * Otherwise, since we received a FIN then no
2951 * more input can be expected, send ACK now.
2953 if (tp->t_flags & TF_NEEDSYN)
2954 tp->t_flags |= TF_DELACK;
2956 tp->t_flags |= TF_ACKNOW;
2959 switch (tp->t_state) {
2962 * In SYN_RECEIVED and ESTABLISHED STATES
2963 * enter the CLOSE_WAIT state.
2965 case TCPS_SYN_RECEIVED:
2966 tp->t_starttime = ticks;
2968 case TCPS_ESTABLISHED:
2969 tcp_state_change(tp, TCPS_CLOSE_WAIT);
2973 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2974 * enter the CLOSING state.
2976 case TCPS_FIN_WAIT_1:
2977 tcp_state_change(tp, TCPS_CLOSING);
2981 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2982 * starting the time-wait timer, turning off the other
2985 case TCPS_FIN_WAIT_2:
2986 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2987 KASSERT(ti_locked == TI_WLOCKED, ("%s: dodata "
2988 "TCP_FIN_WAIT_2 ti_locked: %d", __func__,
2992 INP_INFO_WUNLOCK(&V_tcbinfo);
2996 if (ti_locked == TI_WLOCKED)
2997 INP_INFO_WUNLOCK(&V_tcbinfo);
2998 ti_locked = TI_UNLOCKED;
3001 if (so->so_options & SO_DEBUG)
3002 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
3007 * Return any desired output.
3009 if (needoutput || (tp->t_flags & TF_ACKNOW))
3010 (void) tcp_output(tp);
3013 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d",
3014 __func__, ti_locked));
3015 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
3016 INP_WLOCK_ASSERT(tp->t_inpcb);
3018 if (tp->t_flags & TF_DELACK) {
3019 tp->t_flags &= ~TF_DELACK;
3020 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
3022 INP_WUNLOCK(tp->t_inpcb);
3027 * Generate an ACK dropping incoming segment if it occupies
3028 * sequence space, where the ACK reflects our state.
3030 * We can now skip the test for the RST flag since all
3031 * paths to this code happen after packets containing
3032 * RST have been dropped.
3034 * In the SYN-RECEIVED state, don't send an ACK unless the
3035 * segment we received passes the SYN-RECEIVED ACK test.
3036 * If it fails send a RST. This breaks the loop in the
3037 * "LAND" DoS attack, and also prevents an ACK storm
3038 * between two listening ports that have been sent forged
3039 * SYN segments, each with the source address of the other.
3041 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
3042 (SEQ_GT(tp->snd_una, th->th_ack) ||
3043 SEQ_GT(th->th_ack, tp->snd_max)) ) {
3044 rstreason = BANDLIM_RST_OPENPORT;
3048 if (so->so_options & SO_DEBUG)
3049 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3052 if (ti_locked == TI_WLOCKED)
3053 INP_INFO_WUNLOCK(&V_tcbinfo);
3054 ti_locked = TI_UNLOCKED;
3056 tp->t_flags |= TF_ACKNOW;
3057 (void) tcp_output(tp);
3058 INP_WUNLOCK(tp->t_inpcb);
3063 if (ti_locked == TI_WLOCKED)
3064 INP_INFO_WUNLOCK(&V_tcbinfo);
3065 ti_locked = TI_UNLOCKED;
3068 tcp_dropwithreset(m, th, tp, tlen, rstreason);
3069 INP_WUNLOCK(tp->t_inpcb);
3071 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
3075 if (ti_locked == TI_WLOCKED) {
3076 INP_INFO_WUNLOCK(&V_tcbinfo);
3077 ti_locked = TI_UNLOCKED;
3081 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
3085 * Drop space held by incoming segment and return.
3088 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
3089 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3093 INP_WUNLOCK(tp->t_inpcb);
3098 * Issue RST and make ACK acceptable to originator of segment.
3099 * The mbuf must still include the original packet header.
3103 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
3104 int tlen, int rstreason)
3110 struct ip6_hdr *ip6;
3114 INP_WLOCK_ASSERT(tp->t_inpcb);
3117 /* Don't bother if destination was broadcast/multicast. */
3118 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
3121 if (mtod(m, struct ip *)->ip_v == 6) {
3122 ip6 = mtod(m, struct ip6_hdr *);
3123 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3124 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3126 /* IPv6 anycast check is done at tcp6_input() */
3129 #if defined(INET) && defined(INET6)
3134 ip = mtod(m, struct ip *);
3135 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3136 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3137 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3138 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3143 /* Perform bandwidth limiting. */
3144 if (badport_bandlim(rstreason) < 0)
3147 /* tcp_respond consumes the mbuf chain. */
3148 if (th->th_flags & TH_ACK) {
3149 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0,
3150 th->th_ack, TH_RST);
3152 if (th->th_flags & TH_SYN)
3154 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
3155 (tcp_seq)0, TH_RST|TH_ACK);
3163 * Parse TCP options and place in tcpopt.
3166 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags)
3171 for (; cnt > 0; cnt -= optlen, cp += optlen) {
3173 if (opt == TCPOPT_EOL)
3175 if (opt == TCPOPT_NOP)
3181 if (optlen < 2 || optlen > cnt)
3186 if (optlen != TCPOLEN_MAXSEG)
3188 if (!(flags & TO_SYN))
3190 to->to_flags |= TOF_MSS;
3191 bcopy((char *)cp + 2,
3192 (char *)&to->to_mss, sizeof(to->to_mss));
3193 to->to_mss = ntohs(to->to_mss);
3196 if (optlen != TCPOLEN_WINDOW)
3198 if (!(flags & TO_SYN))
3200 to->to_flags |= TOF_SCALE;
3201 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT);
3203 case TCPOPT_TIMESTAMP:
3204 if (optlen != TCPOLEN_TIMESTAMP)
3206 to->to_flags |= TOF_TS;
3207 bcopy((char *)cp + 2,
3208 (char *)&to->to_tsval, sizeof(to->to_tsval));
3209 to->to_tsval = ntohl(to->to_tsval);
3210 bcopy((char *)cp + 6,
3211 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
3212 to->to_tsecr = ntohl(to->to_tsecr);
3214 #ifdef TCP_SIGNATURE
3216 * XXX In order to reply to a host which has set the
3217 * TCP_SIGNATURE option in its initial SYN, we have to
3218 * record the fact that the option was observed here
3219 * for the syncache code to perform the correct response.
3221 case TCPOPT_SIGNATURE:
3222 if (optlen != TCPOLEN_SIGNATURE)
3224 to->to_flags |= TOF_SIGNATURE;
3225 to->to_signature = cp + 2;
3228 case TCPOPT_SACK_PERMITTED:
3229 if (optlen != TCPOLEN_SACK_PERMITTED)
3231 if (!(flags & TO_SYN))
3235 to->to_flags |= TOF_SACKPERM;
3238 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
3242 to->to_flags |= TOF_SACK;
3243 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
3244 to->to_sacks = cp + 2;
3245 TCPSTAT_INC(tcps_sack_rcv_blocks);
3254 * Pull out of band byte out of a segment so
3255 * it doesn't appear in the user's data queue.
3256 * It is still reflected in the segment length for
3257 * sequencing purposes.
3260 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m,
3263 int cnt = off + th->th_urp - 1;
3266 if (m->m_len > cnt) {
3267 char *cp = mtod(m, caddr_t) + cnt;
3268 struct tcpcb *tp = sototcpcb(so);
3270 INP_WLOCK_ASSERT(tp->t_inpcb);
3273 tp->t_oobflags |= TCPOOB_HAVEDATA;
3274 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3276 if (m->m_flags & M_PKTHDR)
3285 panic("tcp_pulloutofband");
3289 * Collect new round-trip time estimate
3290 * and update averages and current timeout.
3293 tcp_xmit_timer(struct tcpcb *tp, int rtt)
3297 INP_WLOCK_ASSERT(tp->t_inpcb);
3299 TCPSTAT_INC(tcps_rttupdated);
3301 if (tp->t_srtt != 0) {
3303 * srtt is stored as fixed point with 5 bits after the
3304 * binary point (i.e., scaled by 8). The following magic
3305 * is equivalent to the smoothing algorithm in rfc793 with
3306 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3307 * point). Adjust rtt to origin 0.
3309 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
3310 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
3312 if ((tp->t_srtt += delta) <= 0)
3316 * We accumulate a smoothed rtt variance (actually, a
3317 * smoothed mean difference), then set the retransmit
3318 * timer to smoothed rtt + 4 times the smoothed variance.
3319 * rttvar is stored as fixed point with 4 bits after the
3320 * binary point (scaled by 16). The following is
3321 * equivalent to rfc793 smoothing with an alpha of .75
3322 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
3323 * rfc793's wired-in beta.
3327 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
3328 if ((tp->t_rttvar += delta) <= 0)
3330 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
3331 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3334 * No rtt measurement yet - use the unsmoothed rtt.
3335 * Set the variance to half the rtt (so our first
3336 * retransmit happens at 3*rtt).
3338 tp->t_srtt = rtt << TCP_RTT_SHIFT;
3339 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
3340 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3346 * the retransmit should happen at rtt + 4 * rttvar.
3347 * Because of the way we do the smoothing, srtt and rttvar
3348 * will each average +1/2 tick of bias. When we compute
3349 * the retransmit timer, we want 1/2 tick of rounding and
3350 * 1 extra tick because of +-1/2 tick uncertainty in the
3351 * firing of the timer. The bias will give us exactly the
3352 * 1.5 tick we need. But, because the bias is
3353 * statistical, we have to test that we don't drop below
3354 * the minimum feasible timer (which is 2 ticks).
3356 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3357 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3360 * We received an ack for a packet that wasn't retransmitted;
3361 * it is probably safe to discard any error indications we've
3362 * received recently. This isn't quite right, but close enough
3363 * for now (a route might have failed after we sent a segment,
3364 * and the return path might not be symmetrical).
3366 tp->t_softerror = 0;
3370 * Determine a reasonable value for maxseg size.
3371 * If the route is known, check route for mtu.
3372 * If none, use an mss that can be handled on the outgoing interface
3373 * without forcing IP to fragment. If no route is found, route has no mtu,
3374 * or the destination isn't local, use a default, hopefully conservative
3375 * size (usually 512 or the default IP max size, but no more than the mtu
3376 * of the interface), as we can't discover anything about intervening
3377 * gateways or networks. We also initialize the congestion/slow start
3378 * window to be a single segment if the destination isn't local.
3379 * While looking at the routing entry, we also initialize other path-dependent
3380 * parameters from pre-set or cached values in the routing entry.
3382 * Also take into account the space needed for options that we
3383 * send regularly. Make maxseg shorter by that amount to assure
3384 * that we can send maxseg amount of data even when the options
3385 * are present. Store the upper limit of the length of options plus
3388 * NOTE that this routine is only called when we process an incoming
3389 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS
3390 * settings are handled in tcp_mssopt().
3393 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer,
3394 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap)
3398 struct inpcb *inp = tp->t_inpcb;
3399 struct hc_metrics_lite metrics;
3402 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
3403 size_t min_protoh = isipv6 ?
3404 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
3405 sizeof (struct tcpiphdr);
3407 const size_t min_protoh = sizeof(struct tcpiphdr);
3410 INP_WLOCK_ASSERT(tp->t_inpcb);
3412 if (mtuoffer != -1) {
3413 KASSERT(offer == -1, ("%s: conflict", __func__));
3414 offer = mtuoffer - min_protoh;
3421 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap);
3422 tp->t_maxopd = tp->t_maxseg = V_tcp_v6mssdflt;
3425 #if defined(INET) && defined(INET6)
3430 maxmtu = tcp_maxmtu(&inp->inp_inc, cap);
3431 tp->t_maxopd = tp->t_maxseg = V_tcp_mssdflt;
3436 * No route to sender, stay with default mss and return.
3440 * In case we return early we need to initialize metrics
3441 * to a defined state as tcp_hc_get() would do for us
3442 * if there was no cache hit.
3444 if (metricptr != NULL)
3445 bzero(metricptr, sizeof(struct hc_metrics_lite));
3449 /* What have we got? */
3453 * Offer == 0 means that there was no MSS on the SYN
3454 * segment, in this case we use tcp_mssdflt as
3455 * already assigned to t_maxopd above.
3457 offer = tp->t_maxopd;
3462 * Offer == -1 means that we didn't receive SYN yet.
3468 * Prevent DoS attack with too small MSS. Round up
3469 * to at least minmss.
3471 offer = max(offer, V_tcp_minmss);
3475 * rmx information is now retrieved from tcp_hostcache.
3477 tcp_hc_get(&inp->inp_inc, &metrics);
3478 if (metricptr != NULL)
3479 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite));
3482 * If there's a discovered mtu in tcp hostcache, use it.
3483 * Else, use the link mtu.
3485 if (metrics.rmx_mtu)
3486 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
3490 mss = maxmtu - min_protoh;
3491 if (!V_path_mtu_discovery &&
3492 !in6_localaddr(&inp->in6p_faddr))
3493 mss = min(mss, V_tcp_v6mssdflt);
3496 #if defined(INET) && defined(INET6)
3501 mss = maxmtu - min_protoh;
3502 if (!V_path_mtu_discovery &&
3503 !in_localaddr(inp->inp_faddr))
3504 mss = min(mss, V_tcp_mssdflt);
3508 * XXX - The above conditional (mss = maxmtu - min_protoh)
3509 * probably violates the TCP spec.
3510 * The problem is that, since we don't know the
3511 * other end's MSS, we are supposed to use a conservative
3512 * default. But, if we do that, then MTU discovery will
3513 * never actually take place, because the conservative
3514 * default is much less than the MTUs typically seen
3515 * on the Internet today. For the moment, we'll sweep
3516 * this under the carpet.
3518 * The conservative default might not actually be a problem
3519 * if the only case this occurs is when sending an initial
3520 * SYN with options and data to a host we've never talked
3521 * to before. Then, they will reply with an MSS value which
3522 * will get recorded and the new parameters should get
3523 * recomputed. For Further Study.
3526 mss = min(mss, offer);
3529 * Sanity check: make sure that maxopd will be large
3530 * enough to allow some data on segments even if the
3531 * all the option space is used (40bytes). Otherwise
3532 * funny things may happen in tcp_output.
3537 * maxopd stores the maximum length of data AND options
3538 * in a segment; maxseg is the amount of data in a normal
3539 * segment. We need to store this value (maxopd) apart
3540 * from maxseg, because now every segment carries options
3541 * and thus we normally have somewhat less data in segments.
3546 * origoffer==-1 indicates that no segments were received yet.
3547 * In this case we just guess.
3549 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
3551 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
3552 mss -= TCPOLEN_TSTAMP_APPA;
3558 tcp_mss(struct tcpcb *tp, int offer)
3564 struct hc_metrics_lite metrics;
3565 struct tcp_ifcap cap;
3567 KASSERT(tp != NULL, ("%s: tp == NULL", __func__));
3569 bzero(&cap, sizeof(cap));
3570 tcp_mss_update(tp, offer, -1, &metrics, &cap);
3576 * If there's a pipesize, change the socket buffer to that size,
3577 * don't change if sb_hiwat is different than default (then it
3578 * has been changed on purpose with setsockopt).
3579 * Make the socket buffers an integral number of mss units;
3580 * if the mss is larger than the socket buffer, decrease the mss.
3582 so = inp->inp_socket;
3583 SOCKBUF_LOCK(&so->so_snd);
3584 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe)
3585 bufsize = metrics.rmx_sendpipe;
3587 bufsize = so->so_snd.sb_hiwat;
3591 bufsize = roundup(bufsize, mss);
3592 if (bufsize > sb_max)
3594 if (bufsize > so->so_snd.sb_hiwat)
3595 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
3597 SOCKBUF_UNLOCK(&so->so_snd);
3600 SOCKBUF_LOCK(&so->so_rcv);
3601 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe)
3602 bufsize = metrics.rmx_recvpipe;
3604 bufsize = so->so_rcv.sb_hiwat;
3605 if (bufsize > mss) {
3606 bufsize = roundup(bufsize, mss);
3607 if (bufsize > sb_max)
3609 if (bufsize > so->so_rcv.sb_hiwat)
3610 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
3612 SOCKBUF_UNLOCK(&so->so_rcv);
3614 /* Check the interface for TSO capabilities. */
3615 if (cap.ifcap & CSUM_TSO) {
3616 tp->t_flags |= TF_TSO;
3617 tp->t_tsomax = cap.tsomax;
3622 * Determine the MSS option to send on an outgoing SYN.
3625 tcp_mssopt(struct in_conninfo *inc)
3632 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3635 if (inc->inc_flags & INC_ISIPV6) {
3636 mss = V_tcp_v6mssdflt;
3637 maxmtu = tcp_maxmtu6(inc, NULL);
3638 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3641 #if defined(INET) && defined(INET6)
3646 mss = V_tcp_mssdflt;
3647 maxmtu = tcp_maxmtu(inc, NULL);
3648 min_protoh = sizeof(struct tcpiphdr);
3651 #if defined(INET6) || defined(INET)
3652 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3655 if (maxmtu && thcmtu)
3656 mss = min(maxmtu, thcmtu) - min_protoh;
3657 else if (maxmtu || thcmtu)
3658 mss = max(maxmtu, thcmtu) - min_protoh;
3665 * On a partial ack arrives, force the retransmission of the
3666 * next unacknowledged segment. Do not clear tp->t_dupacks.
3667 * By setting snd_nxt to ti_ack, this forces retransmission timer to
3671 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th)
3673 tcp_seq onxt = tp->snd_nxt;
3674 u_long ocwnd = tp->snd_cwnd;
3676 INP_WLOCK_ASSERT(tp->t_inpcb);
3678 tcp_timer_activate(tp, TT_REXMT, 0);
3680 tp->snd_nxt = th->th_ack;
3682 * Set snd_cwnd to one segment beyond acknowledged offset.
3683 * (tp->snd_una has not yet been updated when this function is called.)
3685 tp->snd_cwnd = tp->t_maxseg + BYTES_THIS_ACK(tp, th);
3686 tp->t_flags |= TF_ACKNOW;
3687 (void) tcp_output(tp);
3688 tp->snd_cwnd = ocwnd;
3689 if (SEQ_GT(onxt, tp->snd_nxt))
3692 * Partial window deflation. Relies on fact that tp->snd_una
3695 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th))
3696 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th);
3699 tp->snd_cwnd += tp->t_maxseg;