2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
3 * The Regents of the University of California. All rights reserved.
4 * Copyright (c) 2007-2008,2010
5 * Swinburne University of Technology, Melbourne, Australia.
6 * Copyright (c) 2009-2010 Lawrence Stewart <lstewart@freebsd.org>
7 * Copyright (c) 2010 The FreeBSD Foundation
8 * Copyright (c) 2010-2011 Juniper Networks, Inc.
11 * Portions of this software were developed at the Centre for Advanced Internet
12 * Architectures, Swinburne University of Technology, by Lawrence Stewart,
13 * James Healy and David Hayes, made possible in part by a grant from the Cisco
14 * University Research Program Fund at Community Foundation Silicon Valley.
16 * Portions of this software were developed at the Centre for Advanced
17 * Internet Architectures, Swinburne University of Technology, Melbourne,
18 * Australia by David Hayes under sponsorship from the FreeBSD Foundation.
20 * Portions of this software were developed by Robert N. M. Watson under
21 * contract to Juniper Networks, Inc.
23 * Redistribution and use in source and binary forms, with or without
24 * modification, are permitted provided that the following conditions
26 * 1. Redistributions of source code must retain the above copyright
27 * notice, this list of conditions and the following disclaimer.
28 * 2. Redistributions in binary form must reproduce the above copyright
29 * notice, this list of conditions and the following disclaimer in the
30 * documentation and/or other materials provided with the distribution.
31 * 4. Neither the name of the University nor the names of its contributors
32 * may be used to endorse or promote products derived from this software
33 * without specific prior written permission.
35 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
36 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
37 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
38 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
39 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
40 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
41 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
42 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
43 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
44 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
47 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95
50 #include <sys/cdefs.h>
51 __FBSDID("$FreeBSD$");
53 #include "opt_ipfw.h" /* for ipfw_fwd */
55 #include "opt_inet6.h"
56 #include "opt_ipsec.h"
57 #include "opt_tcpdebug.h"
59 #include <sys/param.h>
60 #include <sys/kernel.h>
61 #include <sys/hhook.h>
62 #include <sys/malloc.h>
64 #include <sys/proc.h> /* for proc0 declaration */
65 #include <sys/protosw.h>
66 #include <sys/signalvar.h>
67 #include <sys/socket.h>
68 #include <sys/socketvar.h>
69 #include <sys/sysctl.h>
70 #include <sys/syslog.h>
71 #include <sys/systm.h>
73 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
78 #include <net/route.h>
81 #define TCPSTATES /* for logging */
83 #include <netinet/cc.h>
84 #include <netinet/in.h>
85 #include <netinet/in_pcb.h>
86 #include <netinet/in_systm.h>
87 #include <netinet/in_var.h>
88 #include <netinet/ip.h>
89 #include <netinet/ip_icmp.h> /* required for icmp_var.h */
90 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */
91 #include <netinet/ip_var.h>
92 #include <netinet/ip_options.h>
93 #include <netinet/ip6.h>
94 #include <netinet/icmp6.h>
95 #include <netinet6/in6_pcb.h>
96 #include <netinet6/ip6_var.h>
97 #include <netinet6/nd6.h>
98 #include <netinet/tcp_fsm.h>
99 #include <netinet/tcp_seq.h>
100 #include <netinet/tcp_timer.h>
101 #include <netinet/tcp_var.h>
102 #include <netinet6/tcp6_var.h>
103 #include <netinet/tcpip.h>
104 #include <netinet/tcp_syncache.h>
106 #include <netinet/tcp_debug.h>
107 #endif /* TCPDEBUG */
110 #include <netipsec/ipsec.h>
111 #include <netipsec/ipsec6.h>
114 #include <machine/in_cksum.h>
116 #include <security/mac/mac_framework.h>
118 const int tcprexmtthresh = 3;
120 VNET_DEFINE(struct tcpstat, tcpstat);
121 SYSCTL_VNET_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW,
122 &VNET_NAME(tcpstat), tcpstat,
123 "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
125 int tcp_log_in_vain = 0;
126 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
128 "Log all incoming TCP segments to closed ports");
130 VNET_DEFINE(int, blackhole) = 0;
131 #define V_blackhole VNET(blackhole)
132 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW,
133 &VNET_NAME(blackhole), 0,
134 "Do not send RST on segments to closed ports");
136 VNET_DEFINE(int, tcp_delack_enabled) = 1;
137 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW,
138 &VNET_NAME(tcp_delack_enabled), 0,
139 "Delay ACK to try and piggyback it onto a data packet");
141 VNET_DEFINE(int, drop_synfin) = 0;
142 #define V_drop_synfin VNET(drop_synfin)
143 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW,
144 &VNET_NAME(drop_synfin), 0,
145 "Drop TCP packets with SYN+FIN set");
147 VNET_DEFINE(int, tcp_do_rfc3042) = 1;
148 #define V_tcp_do_rfc3042 VNET(tcp_do_rfc3042)
149 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_RW,
150 &VNET_NAME(tcp_do_rfc3042), 0,
151 "Enable RFC 3042 (Limited Transmit)");
153 VNET_DEFINE(int, tcp_do_rfc3390) = 1;
154 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW,
155 &VNET_NAME(tcp_do_rfc3390), 0,
156 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
158 VNET_DEFINE(int, tcp_do_rfc3465) = 1;
159 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_RW,
160 &VNET_NAME(tcp_do_rfc3465), 0,
161 "Enable RFC 3465 (Appropriate Byte Counting)");
163 VNET_DEFINE(int, tcp_abc_l_var) = 2;
164 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_RW,
165 &VNET_NAME(tcp_abc_l_var), 2,
166 "Cap the max cwnd increment during slow-start to this number of segments");
168 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, CTLFLAG_RW, 0, "TCP ECN");
170 VNET_DEFINE(int, tcp_do_ecn) = 0;
171 SYSCTL_VNET_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_RW,
172 &VNET_NAME(tcp_do_ecn), 0,
175 VNET_DEFINE(int, tcp_ecn_maxretries) = 1;
176 SYSCTL_VNET_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_RW,
177 &VNET_NAME(tcp_ecn_maxretries), 0,
178 "Max retries before giving up on ECN");
180 VNET_DEFINE(int, tcp_insecure_rst) = 0;
181 #define V_tcp_insecure_rst VNET(tcp_insecure_rst)
182 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_RW,
183 &VNET_NAME(tcp_insecure_rst), 0,
184 "Follow the old (insecure) criteria for accepting RST packets");
186 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1;
187 #define V_tcp_do_autorcvbuf VNET(tcp_do_autorcvbuf)
188 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_RW,
189 &VNET_NAME(tcp_do_autorcvbuf), 0,
190 "Enable automatic receive buffer sizing");
192 VNET_DEFINE(int, tcp_autorcvbuf_inc) = 16*1024;
193 #define V_tcp_autorcvbuf_inc VNET(tcp_autorcvbuf_inc)
194 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_RW,
195 &VNET_NAME(tcp_autorcvbuf_inc), 0,
196 "Incrementor step size of automatic receive buffer");
198 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024;
199 #define V_tcp_autorcvbuf_max VNET(tcp_autorcvbuf_max)
200 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_RW,
201 &VNET_NAME(tcp_autorcvbuf_max), 0,
202 "Max size of automatic receive buffer");
204 VNET_DEFINE(struct inpcbhead, tcb);
205 #define tcb6 tcb /* for KAME src sync over BSD*'s */
206 VNET_DEFINE(struct inpcbinfo, tcbinfo);
208 static void tcp_dooptions(struct tcpopt *, u_char *, int, int);
209 static void tcp_do_segment(struct mbuf *, struct tcphdr *,
210 struct socket *, struct tcpcb *, int, int, uint8_t,
212 static void tcp_dropwithreset(struct mbuf *, struct tcphdr *,
213 struct tcpcb *, int, int);
214 static void tcp_pulloutofband(struct socket *,
215 struct tcphdr *, struct mbuf *, int);
216 static void tcp_xmit_timer(struct tcpcb *, int);
217 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *);
218 static void inline tcp_fields_to_host(struct tcphdr *);
220 static void inline tcp_fields_to_net(struct tcphdr *);
221 static int inline tcp_signature_verify_input(struct mbuf *, int, int,
222 int, struct tcpopt *, struct tcphdr *, u_int);
224 static void inline cc_ack_received(struct tcpcb *tp, struct tcphdr *th,
226 static void inline cc_conn_init(struct tcpcb *tp);
227 static void inline cc_post_recovery(struct tcpcb *tp, struct tcphdr *th);
228 static void inline hhook_run_tcp_est_in(struct tcpcb *tp,
229 struct tcphdr *th, struct tcpopt *to);
232 * Kernel module interface for updating tcpstat. The argument is an index
233 * into tcpstat treated as an array of u_long. While this encodes the
234 * general layout of tcpstat into the caller, it doesn't encode its location,
235 * so that future changes to add, for example, per-CPU stats support won't
236 * cause binary compatibility problems for kernel modules.
239 kmod_tcpstat_inc(int statnum)
242 (*((u_long *)&V_tcpstat + statnum))++;
246 * Wrapper for the TCP established input helper hook.
249 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to)
251 struct tcp_hhook_data hhook_data;
253 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) {
258 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data,
264 * CC wrapper hook functions
267 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t type)
269 INP_WLOCK_ASSERT(tp->t_inpcb);
271 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th);
272 if (tp->snd_cwnd == min(tp->snd_cwnd, tp->snd_wnd))
273 tp->ccv->flags |= CCF_CWND_LIMITED;
275 tp->ccv->flags &= ~CCF_CWND_LIMITED;
277 if (type == CC_ACK) {
278 if (tp->snd_cwnd > tp->snd_ssthresh) {
279 tp->t_bytes_acked += min(tp->ccv->bytes_this_ack,
280 V_tcp_abc_l_var * tp->t_maxseg);
281 if (tp->t_bytes_acked >= tp->snd_cwnd) {
282 tp->t_bytes_acked -= tp->snd_cwnd;
283 tp->ccv->flags |= CCF_ABC_SENTAWND;
286 tp->ccv->flags &= ~CCF_ABC_SENTAWND;
287 tp->t_bytes_acked = 0;
291 if (CC_ALGO(tp)->ack_received != NULL) {
292 /* XXXLAS: Find a way to live without this */
293 tp->ccv->curack = th->th_ack;
294 CC_ALGO(tp)->ack_received(tp->ccv, type);
299 cc_conn_init(struct tcpcb *tp)
301 struct hc_metrics_lite metrics;
302 struct inpcb *inp = tp->t_inpcb;
305 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
308 INP_WLOCK_ASSERT(tp->t_inpcb);
310 tcp_hc_get(&inp->inp_inc, &metrics);
312 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) {
314 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
315 TCPSTAT_INC(tcps_usedrtt);
316 if (metrics.rmx_rttvar) {
317 tp->t_rttvar = metrics.rmx_rttvar;
318 TCPSTAT_INC(tcps_usedrttvar);
320 /* default variation is +- 1 rtt */
322 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
324 TCPT_RANGESET(tp->t_rxtcur,
325 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
326 tp->t_rttmin, TCPTV_REXMTMAX);
328 if (metrics.rmx_ssthresh) {
330 * There's some sort of gateway or interface
331 * buffer limit on the path. Use this to set
332 * the slow start threshhold, but set the
333 * threshold to no less than 2*mss.
335 tp->snd_ssthresh = max(2 * tp->t_maxseg, metrics.rmx_ssthresh);
336 TCPSTAT_INC(tcps_usedssthresh);
340 * Set the slow-start flight size depending on whether this
341 * is a local network or not.
343 * Extend this so we cache the cwnd too and retrieve it here.
344 * Make cwnd even bigger than RFC3390 suggests but only if we
345 * have previous experience with the remote host. Be careful
346 * not make cwnd bigger than remote receive window or our own
347 * send socket buffer. Maybe put some additional upper bound
348 * on the retrieved cwnd. Should do incremental updates to
349 * hostcache when cwnd collapses so next connection doesn't
350 * overloads the path again.
352 * XXXAO: Initializing the CWND from the hostcache is broken
353 * and in its current form not RFC conformant. It is disabled
354 * until fixed or removed entirely.
356 * RFC3390 says only do this if SYN or SYN/ACK didn't got lost.
357 * We currently check only in syncache_socket for that.
359 /* #define TCP_METRICS_CWND */
360 #ifdef TCP_METRICS_CWND
361 if (metrics.rmx_cwnd)
362 tp->snd_cwnd = max(tp->t_maxseg, min(metrics.rmx_cwnd / 2,
363 min(tp->snd_wnd, so->so_snd.sb_hiwat)));
366 if (V_tcp_do_rfc3390)
367 tp->snd_cwnd = min(4 * tp->t_maxseg,
368 max(2 * tp->t_maxseg, 4380));
370 else if (isipv6 && in6_localaddr(&inp->in6p_faddr))
371 tp->snd_cwnd = tp->t_maxseg * V_ss_fltsz_local;
373 #if defined(INET) && defined(INET6)
374 else if (!isipv6 && in_localaddr(inp->inp_faddr))
375 tp->snd_cwnd = tp->t_maxseg * V_ss_fltsz_local;
378 else if (in_localaddr(inp->inp_faddr))
379 tp->snd_cwnd = tp->t_maxseg * V_ss_fltsz_local;
382 tp->snd_cwnd = tp->t_maxseg * V_ss_fltsz;
384 if (CC_ALGO(tp)->conn_init != NULL)
385 CC_ALGO(tp)->conn_init(tp->ccv);
389 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type)
391 INP_WLOCK_ASSERT(tp->t_inpcb);
395 if (!IN_FASTRECOVERY(tp->t_flags)) {
396 tp->snd_recover = tp->snd_max;
397 if (tp->t_flags & TF_ECN_PERMIT)
398 tp->t_flags |= TF_ECN_SND_CWR;
402 if (!IN_CONGRECOVERY(tp->t_flags)) {
403 TCPSTAT_INC(tcps_ecn_rcwnd);
404 tp->snd_recover = tp->snd_max;
405 if (tp->t_flags & TF_ECN_PERMIT)
406 tp->t_flags |= TF_ECN_SND_CWR;
411 tp->t_bytes_acked = 0;
412 EXIT_RECOVERY(tp->t_flags);
413 tp->snd_ssthresh = max(2, min(tp->snd_wnd, tp->snd_cwnd) / 2 /
414 tp->t_maxseg) * tp->t_maxseg;
415 tp->snd_cwnd = tp->t_maxseg;
418 TCPSTAT_INC(tcps_sndrexmitbad);
419 /* RTO was unnecessary, so reset everything. */
420 tp->snd_cwnd = tp->snd_cwnd_prev;
421 tp->snd_ssthresh = tp->snd_ssthresh_prev;
422 tp->snd_recover = tp->snd_recover_prev;
423 if (tp->t_flags & TF_WASFRECOVERY)
424 ENTER_FASTRECOVERY(tp->t_flags);
425 if (tp->t_flags & TF_WASCRECOVERY)
426 ENTER_CONGRECOVERY(tp->t_flags);
427 tp->snd_nxt = tp->snd_max;
428 tp->t_flags &= ~TF_PREVVALID;
433 if (CC_ALGO(tp)->cong_signal != NULL) {
435 tp->ccv->curack = th->th_ack;
436 CC_ALGO(tp)->cong_signal(tp->ccv, type);
441 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th)
443 INP_WLOCK_ASSERT(tp->t_inpcb);
445 /* XXXLAS: KASSERT that we're in recovery? */
447 if (CC_ALGO(tp)->post_recovery != NULL) {
448 tp->ccv->curack = th->th_ack;
449 CC_ALGO(tp)->post_recovery(tp->ccv);
451 /* XXXLAS: EXIT_RECOVERY ? */
452 tp->t_bytes_acked = 0;
456 tcp_fields_to_host(struct tcphdr *th)
459 th->th_seq = ntohl(th->th_seq);
460 th->th_ack = ntohl(th->th_ack);
461 th->th_win = ntohs(th->th_win);
462 th->th_urp = ntohs(th->th_urp);
467 tcp_fields_to_net(struct tcphdr *th)
470 th->th_seq = htonl(th->th_seq);
471 th->th_ack = htonl(th->th_ack);
472 th->th_win = htons(th->th_win);
473 th->th_urp = htons(th->th_urp);
477 tcp_signature_verify_input(struct mbuf *m, int off0, int tlen, int optlen,
478 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag)
482 tcp_fields_to_net(th);
483 ret = tcp_signature_verify(m, off0, tlen, optlen, to, th, tcpbflag);
484 tcp_fields_to_host(th);
489 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
491 #define ND6_HINT(tp) \
493 if ((tp) && (tp)->t_inpcb && \
494 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \
495 nd6_nud_hint(NULL, NULL, 0); \
502 * Indicate whether this ack should be delayed. We can delay the ack if
503 * - there is no delayed ack timer in progress and
504 * - our last ack wasn't a 0-sized window. We never want to delay
505 * the ack that opens up a 0-sized window and
506 * - delayed acks are enabled or
507 * - this is a half-synchronized T/TCP connection.
509 #define DELAY_ACK(tp) \
510 ((!tcp_timer_active(tp, TT_DELACK) && \
511 (tp->t_flags & TF_RXWIN0SENT) == 0) && \
512 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
515 * TCP input handling is split into multiple parts:
516 * tcp6_input is a thin wrapper around tcp_input for the extended
517 * ip6_protox[] call format in ip6_input
518 * tcp_input handles primary segment validation, inpcb lookup and
519 * SYN processing on listen sockets
520 * tcp_do_segment processes the ACK and text of the segment for
521 * establishing, established and closing connections
525 tcp6_input(struct mbuf **mp, int *offp, int proto)
527 struct mbuf *m = *mp;
528 struct in6_ifaddr *ia6;
530 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
533 * draft-itojun-ipv6-tcp-to-anycast
534 * better place to put this in?
536 ia6 = ip6_getdstifaddr(m);
537 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
540 ifa_free(&ia6->ia_ifa);
541 ip6 = mtod(m, struct ip6_hdr *);
542 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
543 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
553 tcp_input(struct mbuf *m, int off0)
555 struct tcphdr *th = NULL;
556 struct ip *ip = NULL;
560 struct inpcb *inp = NULL;
561 struct tcpcb *tp = NULL;
562 struct socket *so = NULL;
571 int rstreason = 0; /* For badport_bandlim accounting purposes */
573 uint8_t sig_checked = 0;
576 #ifdef IPFIREWALL_FORWARD
577 struct m_tag *fwd_tag;
580 struct ip6_hdr *ip6 = NULL;
583 const void *ip6 = NULL;
585 struct tcpopt to; /* options in this segment */
586 char *s = NULL; /* address and port logging */
588 #define TI_UNLOCKED 1
593 * The size of tcp_saveipgen must be the size of the max ip header,
596 u_char tcp_saveipgen[IP6_HDR_LEN];
597 struct tcphdr tcp_savetcp;
602 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
606 TCPSTAT_INC(tcps_rcvtotal);
610 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */
611 ip6 = mtod(m, struct ip6_hdr *);
612 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
613 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) {
614 TCPSTAT_INC(tcps_rcvbadsum);
617 th = (struct tcphdr *)((caddr_t)ip6 + off0);
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)) {
643 ip_stripoptions(m, (struct mbuf *)0);
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);
653 ip = mtod(m, struct ip *);
654 ipov = (struct ipovly *)ip;
655 th = (struct tcphdr *)((caddr_t)ip + off0);
658 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
659 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
660 th->th_sum = m->m_pkthdr.csum_data;
662 th->th_sum = in_pseudo(ip->ip_src.s_addr,
664 htonl(m->m_pkthdr.csum_data +
667 th->th_sum ^= 0xffff;
669 ipov->ih_len = (u_short)tlen;
670 ipov->ih_len = htons(ipov->ih_len);
674 * Checksum extended TCP header and data.
676 len = sizeof (struct ip) + tlen;
677 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
678 ipov->ih_len = (u_short)tlen;
679 ipov->ih_len = htons(ipov->ih_len);
680 th->th_sum = in_cksum(m, len);
683 TCPSTAT_INC(tcps_rcvbadsum);
686 /* Re-initialization for later version check */
687 ip->ip_v = IPVERSION;
693 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
695 #if defined(INET) && defined(INET6)
703 * Check that TCP offset makes sense,
704 * pull out TCP options and adjust length. XXX
706 off = th->th_off << 2;
707 if (off < sizeof (struct tcphdr) || off > tlen) {
708 TCPSTAT_INC(tcps_rcvbadoff);
711 tlen -= off; /* tlen is used instead of ti->ti_len */
712 if (off > sizeof (struct tcphdr)) {
715 IP6_EXTHDR_CHECK(m, off0, off, );
716 ip6 = mtod(m, struct ip6_hdr *);
717 th = (struct tcphdr *)((caddr_t)ip6 + off0);
720 #if defined(INET) && defined(INET6)
725 if (m->m_len < sizeof(struct ip) + off) {
726 if ((m = m_pullup(m, sizeof (struct ip) + off))
728 TCPSTAT_INC(tcps_rcvshort);
731 ip = mtod(m, struct ip *);
732 ipov = (struct ipovly *)ip;
733 th = (struct tcphdr *)((caddr_t)ip + off0);
737 optlen = off - sizeof (struct tcphdr);
738 optp = (u_char *)(th + 1);
740 thflags = th->th_flags;
743 * Convert TCP protocol specific fields to host format.
745 tcp_fields_to_host(th);
748 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options.
750 drop_hdrlen = off0 + off;
753 * Locate pcb for segment; if we're likely to add or remove a
754 * connection then first acquire pcbinfo lock. There are two cases
755 * where we might discover later we need a write lock despite the
756 * flags: ACKs moving a connection out of the syncache, and ACKs for
757 * a connection in TIMEWAIT.
759 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0) {
760 INP_INFO_WLOCK(&V_tcbinfo);
761 ti_locked = TI_WLOCKED;
763 ti_locked = TI_UNLOCKED;
767 if (ti_locked == TI_WLOCKED) {
768 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
770 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
774 #ifdef IPFIREWALL_FORWARD
776 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
778 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
779 #endif /* IPFIREWALL_FORWARD */
782 #ifdef IPFIREWALL_FORWARD
783 if (isipv6 && fwd_tag != NULL) {
784 struct sockaddr_in6 *next_hop6;
786 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1);
788 * Transparently forwarded. Pretend to be the destination.
789 * Already got one like this?
791 inp = in6_pcblookup_mbuf(&V_tcbinfo,
792 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport,
793 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m);
796 * It's new. Try to find the ambushing socket.
797 * Because we've rewritten the destination address,
798 * any hardware-generated hash is ignored.
800 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src,
801 th->th_sport, &next_hop6->sin6_addr,
802 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) :
803 th->th_dport, INPLOOKUP_WILDCARD |
804 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
806 /* Remove the tag from the packet. We don't need it anymore. */
807 m_tag_delete(m, fwd_tag);
809 #endif /* IPFIREWALL_FORWARD */
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 #ifdef IPFIREWALL_FORWARD
822 if (fwd_tag != NULL) {
823 struct sockaddr_in *next_hop;
825 next_hop = (struct sockaddr_in *)(fwd_tag+1);
827 * Transparently forwarded. Pretend to be the destination.
828 * already got one like this?
830 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport,
831 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB,
832 m->m_pkthdr.rcvif, m);
835 * It's new. Try to find the ambushing socket.
836 * Because we've rewritten the destination address,
837 * any hardware-generated hash is ignored.
839 inp = in_pcblookup(&V_tcbinfo, ip->ip_src,
840 th->th_sport, next_hop->sin_addr,
841 next_hop->sin_port ? ntohs(next_hop->sin_port) :
842 th->th_dport, INPLOOKUP_WILDCARD |
843 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
845 /* Remove the tag from the packet. We don't need it anymore. */
846 m_tag_delete(m, fwd_tag);
848 #endif /* IPFIREWALL_FORWARD */
849 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src,
850 th->th_sport, ip->ip_dst, th->th_dport,
851 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
852 m->m_pkthdr.rcvif, m);
856 * If the INPCB does not exist then all data in the incoming
857 * segment is discarded and an appropriate RST is sent back.
858 * XXX MRT Send RST using which routing table?
862 * Log communication attempts to ports that are not
865 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) ||
866 tcp_log_in_vain == 2) {
867 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6)))
868 log(LOG_INFO, "%s; %s: Connection attempt "
869 "to closed port\n", s, __func__);
872 * When blackholing do not respond with a RST but
873 * completely ignore the segment and drop it.
875 if ((V_blackhole == 1 && (thflags & TH_SYN)) ||
879 rstreason = BANDLIM_RST_CLOSEDPORT;
882 INP_WLOCK_ASSERT(inp);
883 if (!(inp->inp_flags & INP_HW_FLOWID)
884 && (m->m_flags & M_FLOWID)
885 && ((inp->inp_socket == NULL)
886 || !(inp->inp_socket->so_options & SO_ACCEPTCONN))) {
887 inp->inp_flags |= INP_HW_FLOWID;
888 inp->inp_flags &= ~INP_SW_FLOWID;
889 inp->inp_flowid = m->m_pkthdr.flowid;
893 if (isipv6 && ipsec6_in_reject(m, inp)) {
894 V_ipsec6stat.in_polvio++;
898 if (ipsec4_in_reject(m, inp) != 0) {
899 V_ipsec4stat.in_polvio++;
905 * Check the minimum TTL for socket.
907 if (inp->inp_ip_minttl != 0) {
909 if (isipv6 && inp->inp_ip_minttl > ip6->ip6_hlim)
913 if (inp->inp_ip_minttl > ip->ip_ttl)
918 * A previous connection in TIMEWAIT state is supposed to catch stray
919 * or duplicate segments arriving late. If this segment was a
920 * legitimate new connection attempt the old INPCB gets removed and
921 * we can try again to find a listening socket.
923 * At this point, due to earlier optimism, we may hold only an inpcb
924 * lock, and not the inpcbinfo write lock. If so, we need to try to
925 * acquire it, or if that fails, acquire a reference on the inpcb,
926 * drop all locks, acquire a global write lock, and then re-acquire
927 * the inpcb lock. We may at that point discover that another thread
928 * has tried to free the inpcb, in which case we need to loop back
929 * and try to find a new inpcb to deliver to.
931 * XXXRW: It may be time to rethink timewait locking.
934 if (inp->inp_flags & INP_TIMEWAIT) {
935 if (ti_locked == TI_UNLOCKED) {
936 if (INP_INFO_TRY_WLOCK(&V_tcbinfo) == 0) {
939 INP_INFO_WLOCK(&V_tcbinfo);
940 ti_locked = TI_WLOCKED;
942 if (in_pcbrele_wlocked(inp)) {
947 ti_locked = TI_WLOCKED;
949 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
952 tcp_dooptions(&to, optp, optlen,
953 (thflags & TH_SYN) ? TO_SYN : 0);
954 if (sig_checked == 0) {
956 if (tp == NULL || tp->t_state == TCPS_CLOSED) {
957 rstreason = BANDLIM_RST_CLOSEDPORT;
960 if (!tcp_signature_verify_input(m, off0, tlen, optlen,
961 &to, th, tp->t_flags))
966 if (thflags & TH_SYN)
967 tcp_dooptions(&to, optp, optlen, TO_SYN);
970 * NB: tcp_twcheck unlocks the INP and frees the mbuf.
972 if (tcp_twcheck(inp, &to, th, m, tlen))
974 INP_INFO_WUNLOCK(&V_tcbinfo);
978 * The TCPCB may no longer exist if the connection is winding
979 * down or it is in the CLOSED state. Either way we drop the
980 * segment and send an appropriate response.
983 if (tp == NULL || tp->t_state == TCPS_CLOSED) {
984 rstreason = BANDLIM_RST_CLOSEDPORT;
989 * We've identified a valid inpcb, but it could be that we need an
990 * inpcbinfo write lock but don't hold it. In this case, attempt to
991 * acquire using the same strategy as the TIMEWAIT case above. If we
992 * relock, we have to jump back to 'relocked' as the connection might
993 * now be in TIMEWAIT.
996 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0)
997 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
999 if (tp->t_state != TCPS_ESTABLISHED) {
1000 if (ti_locked == TI_UNLOCKED) {
1001 if (INP_INFO_TRY_WLOCK(&V_tcbinfo) == 0) {
1004 INP_INFO_WLOCK(&V_tcbinfo);
1005 ti_locked = TI_WLOCKED;
1007 if (in_pcbrele_wlocked(inp)) {
1013 ti_locked = TI_WLOCKED;
1015 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1019 INP_WLOCK_ASSERT(inp);
1020 if (mac_inpcb_check_deliver(inp, m))
1023 so = inp->inp_socket;
1024 KASSERT(so != NULL, ("%s: so == NULL", __func__));
1026 if (so->so_options & SO_DEBUG) {
1027 ostate = tp->t_state;
1030 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
1033 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
1036 #endif /* TCPDEBUG */
1038 * When the socket is accepting connections (the INPCB is in LISTEN
1039 * state) we look into the SYN cache if this is a new connection
1040 * attempt or the completion of a previous one. Because listen
1041 * sockets are never in TCPS_ESTABLISHED, the V_tcbinfo lock will be
1042 * held in this case.
1044 if (so->so_options & SO_ACCEPTCONN) {
1045 struct in_conninfo inc;
1047 KASSERT(tp->t_state == TCPS_LISTEN, ("%s: so accepting but "
1048 "tp not listening", __func__));
1049 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1051 bzero(&inc, sizeof(inc));
1054 inc.inc_flags |= INC_ISIPV6;
1055 inc.inc6_faddr = ip6->ip6_src;
1056 inc.inc6_laddr = ip6->ip6_dst;
1060 inc.inc_faddr = ip->ip_src;
1061 inc.inc_laddr = ip->ip_dst;
1063 inc.inc_fport = th->th_sport;
1064 inc.inc_lport = th->th_dport;
1065 inc.inc_fibnum = so->so_fibnum;
1068 * Check for an existing connection attempt in syncache if
1069 * the flag is only ACK. A successful lookup creates a new
1070 * socket appended to the listen queue in SYN_RECEIVED state.
1072 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
1074 * Parse the TCP options here because
1075 * syncookies need access to the reflected
1078 tcp_dooptions(&to, optp, optlen, 0);
1080 * NB: syncache_expand() doesn't unlock
1081 * inp and tcpinfo locks.
1083 if (!syncache_expand(&inc, &to, th, &so, m)) {
1085 * No syncache entry or ACK was not
1086 * for our SYN/ACK. Send a RST.
1087 * NB: syncache did its own logging
1088 * of the failure cause.
1090 rstreason = BANDLIM_RST_OPENPORT;
1095 * We completed the 3-way handshake
1096 * but could not allocate a socket
1097 * either due to memory shortage,
1098 * listen queue length limits or
1099 * global socket limits. Send RST
1100 * or wait and have the remote end
1101 * retransmit the ACK for another
1104 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1105 log(LOG_DEBUG, "%s; %s: Listen socket: "
1106 "Socket allocation failed due to "
1107 "limits or memory shortage, %s\n",
1109 V_tcp_sc_rst_sock_fail ?
1110 "sending RST" : "try again");
1111 if (V_tcp_sc_rst_sock_fail) {
1112 rstreason = BANDLIM_UNLIMITED;
1118 * Socket is created in state SYN_RECEIVED.
1119 * Unlock the listen socket, lock the newly
1120 * created socket and update the tp variable.
1122 INP_WUNLOCK(inp); /* listen socket */
1123 inp = sotoinpcb(so);
1124 INP_WLOCK(inp); /* new connection */
1125 tp = intotcpcb(inp);
1126 KASSERT(tp->t_state == TCPS_SYN_RECEIVED,
1127 ("%s: ", __func__));
1128 #ifdef TCP_SIGNATURE
1129 if (sig_checked == 0) {
1130 tcp_dooptions(&to, optp, optlen,
1131 (thflags & TH_SYN) ? TO_SYN : 0);
1132 if (!tcp_signature_verify_input(m, off0, tlen,
1133 optlen, &to, th, tp->t_flags)) {
1136 * In SYN_SENT state if it receives an
1137 * RST, it is allowed for further
1140 if ((thflags & TH_RST) == 0 ||
1141 (tp->t_state == TCPS_SYN_SENT) == 0)
1149 * Process the segment and the data it
1150 * contains. tcp_do_segment() consumes
1151 * the mbuf chain and unlocks the inpcb.
1153 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen,
1155 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1159 * Segment flag validation for new connection attempts:
1161 * Our (SYN|ACK) response was rejected.
1162 * Check with syncache and remove entry to prevent
1165 * NB: syncache_chkrst does its own logging of failure
1168 if (thflags & TH_RST) {
1169 syncache_chkrst(&inc, th);
1173 * We can't do anything without SYN.
1175 if ((thflags & TH_SYN) == 0) {
1176 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1177 log(LOG_DEBUG, "%s; %s: Listen socket: "
1178 "SYN is missing, segment ignored\n",
1180 TCPSTAT_INC(tcps_badsyn);
1184 * (SYN|ACK) is bogus on a listen socket.
1186 if (thflags & TH_ACK) {
1187 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1188 log(LOG_DEBUG, "%s; %s: Listen socket: "
1189 "SYN|ACK invalid, segment rejected\n",
1191 syncache_badack(&inc); /* XXX: Not needed! */
1192 TCPSTAT_INC(tcps_badsyn);
1193 rstreason = BANDLIM_RST_OPENPORT;
1197 * If the drop_synfin option is enabled, drop all
1198 * segments with both the SYN and FIN bits set.
1199 * This prevents e.g. nmap from identifying the
1201 * XXX: Poor reasoning. nmap has other methods
1202 * and is constantly refining its stack detection
1204 * XXX: This is a violation of the TCP specification
1205 * and was used by RFC1644.
1207 if ((thflags & TH_FIN) && V_drop_synfin) {
1208 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1209 log(LOG_DEBUG, "%s; %s: Listen socket: "
1210 "SYN|FIN segment ignored (based on "
1211 "sysctl setting)\n", s, __func__);
1212 TCPSTAT_INC(tcps_badsyn);
1216 * Segment's flags are (SYN) or (SYN|FIN).
1218 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored
1219 * as they do not affect the state of the TCP FSM.
1220 * The data pointed to by TH_URG and th_urp is ignored.
1222 KASSERT((thflags & (TH_RST|TH_ACK)) == 0,
1223 ("%s: Listen socket: TH_RST or TH_ACK set", __func__));
1224 KASSERT(thflags & (TH_SYN),
1225 ("%s: Listen socket: TH_SYN not set", __func__));
1228 * If deprecated address is forbidden,
1229 * we do not accept SYN to deprecated interface
1230 * address to prevent any new inbound connection from
1231 * getting established.
1232 * When we do not accept SYN, we send a TCP RST,
1233 * with deprecated source address (instead of dropping
1234 * it). We compromise it as it is much better for peer
1235 * to send a RST, and RST will be the final packet
1238 * If we do not forbid deprecated addresses, we accept
1239 * the SYN packet. RFC2462 does not suggest dropping
1241 * If we decipher RFC2462 5.5.4, it says like this:
1242 * 1. use of deprecated addr with existing
1243 * communication is okay - "SHOULD continue to be
1245 * 2. use of it with new communication:
1246 * (2a) "SHOULD NOT be used if alternate address
1247 * with sufficient scope is available"
1248 * (2b) nothing mentioned otherwise.
1249 * Here we fall into (2b) case as we have no choice in
1250 * our source address selection - we must obey the peer.
1252 * The wording in RFC2462 is confusing, and there are
1253 * multiple description text for deprecated address
1254 * handling - worse, they are not exactly the same.
1255 * I believe 5.5.4 is the best one, so we follow 5.5.4.
1257 if (isipv6 && !V_ip6_use_deprecated) {
1258 struct in6_ifaddr *ia6;
1260 ia6 = ip6_getdstifaddr(m);
1262 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1263 ifa_free(&ia6->ia_ifa);
1264 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1265 log(LOG_DEBUG, "%s; %s: Listen socket: "
1266 "Connection attempt to deprecated "
1267 "IPv6 address rejected\n",
1269 rstreason = BANDLIM_RST_OPENPORT;
1272 ifa_free(&ia6->ia_ifa);
1276 * Basic sanity checks on incoming SYN requests:
1277 * Don't respond if the destination is a link layer
1278 * broadcast according to RFC1122 4.2.3.10, p. 104.
1279 * If it is from this socket it must be forged.
1280 * Don't respond if the source or destination is a
1281 * global or subnet broad- or multicast address.
1282 * Note that it is quite possible to receive unicast
1283 * link-layer packets with a broadcast IP address. Use
1284 * in_broadcast() to find them.
1286 if (m->m_flags & (M_BCAST|M_MCAST)) {
1287 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1288 log(LOG_DEBUG, "%s; %s: Listen socket: "
1289 "Connection attempt from broad- or multicast "
1290 "link layer address ignored\n", s, __func__);
1295 if (th->th_dport == th->th_sport &&
1296 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) {
1297 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1298 log(LOG_DEBUG, "%s; %s: Listen socket: "
1299 "Connection attempt to/from self "
1300 "ignored\n", s, __func__);
1303 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1304 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) {
1305 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1306 log(LOG_DEBUG, "%s; %s: Listen socket: "
1307 "Connection attempt from/to multicast "
1308 "address ignored\n", s, __func__);
1313 #if defined(INET) && defined(INET6)
1318 if (th->th_dport == th->th_sport &&
1319 ip->ip_dst.s_addr == ip->ip_src.s_addr) {
1320 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1321 log(LOG_DEBUG, "%s; %s: Listen socket: "
1322 "Connection attempt from/to self "
1323 "ignored\n", s, __func__);
1326 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1327 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1328 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1329 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
1330 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1331 log(LOG_DEBUG, "%s; %s: Listen socket: "
1332 "Connection attempt from/to broad- "
1333 "or multicast address ignored\n",
1340 * SYN appears to be valid. Create compressed TCP state
1344 if (so->so_options & SO_DEBUG)
1345 tcp_trace(TA_INPUT, ostate, tp,
1346 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1348 tcp_dooptions(&to, optp, optlen, TO_SYN);
1349 syncache_add(&inc, &to, th, inp, &so, m);
1351 * Entry added to syncache and mbuf consumed.
1352 * Everything already unlocked by syncache_add().
1354 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1358 #ifdef TCP_SIGNATURE
1359 if (sig_checked == 0) {
1360 tcp_dooptions(&to, optp, optlen,
1361 (thflags & TH_SYN) ? TO_SYN : 0);
1362 if (!tcp_signature_verify_input(m, off0, tlen, optlen, &to,
1366 * In SYN_SENT state if it receives an RST, it is
1367 * allowed for further processing.
1369 if ((thflags & TH_RST) == 0 ||
1370 (tp->t_state == TCPS_SYN_SENT) == 0)
1378 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later
1379 * state. tcp_do_segment() always consumes the mbuf chain, unlocks
1380 * the inpcb, and unlocks pcbinfo.
1382 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos, ti_locked);
1383 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1387 if (ti_locked == TI_WLOCKED) {
1388 INP_INFO_WUNLOCK(&V_tcbinfo);
1389 ti_locked = TI_UNLOCKED;
1393 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset "
1394 "ti_locked: %d", __func__, ti_locked));
1395 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1400 tcp_dropwithreset(m, th, tp, tlen, rstreason);
1403 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
1404 m = NULL; /* mbuf chain got consumed. */
1408 if (ti_locked == TI_WLOCKED) {
1409 INP_INFO_WUNLOCK(&V_tcbinfo);
1410 ti_locked = TI_UNLOCKED;
1414 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock "
1415 "ti_locked: %d", __func__, ti_locked));
1416 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1424 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1432 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so,
1433 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos,
1436 int thflags, acked, ourfinisacked, needoutput = 0;
1437 int rstreason, todrop, win;
1443 * The size of tcp_saveipgen must be the size of the max ip header,
1446 u_char tcp_saveipgen[IP6_HDR_LEN];
1447 struct tcphdr tcp_savetcp;
1450 thflags = th->th_flags;
1451 tp->sackhint.last_sack_ack = 0;
1454 * If this is either a state-changing packet or current state isn't
1455 * established, we require a write lock on tcbinfo. Otherwise, we
1456 * allow either a read lock or a write lock, as we may have acquired
1457 * a write lock due to a race.
1459 * Require a global write lock for SYN/FIN/RST segments or
1460 * non-established connections; otherwise accept either a read or
1461 * write lock, as we may have conservatively acquired a write lock in
1462 * certain cases in tcp_input() (is this still true?). Currently we
1463 * will never enter with no lock, so we try to drop it quickly in the
1464 * common pure ack/pure data cases.
1466 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 ||
1467 tp->t_state != TCPS_ESTABLISHED) {
1468 KASSERT(ti_locked == TI_WLOCKED, ("%s ti_locked %d for "
1469 "SYN/FIN/RST/!EST", __func__, ti_locked));
1470 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1473 if (ti_locked == TI_WLOCKED)
1474 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1476 KASSERT(ti_locked == TI_UNLOCKED, ("%s: EST "
1477 "ti_locked: %d", __func__, ti_locked));
1478 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1482 INP_WLOCK_ASSERT(tp->t_inpcb);
1483 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN",
1485 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT",
1489 * Segment received on connection.
1490 * Reset idle time and keep-alive timer.
1491 * XXX: This should be done after segment
1492 * validation to ignore broken/spoofed segs.
1494 tp->t_rcvtime = ticks;
1495 if (TCPS_HAVEESTABLISHED(tp->t_state))
1496 tcp_timer_activate(tp, TT_KEEP, tcp_keepidle);
1499 * Unscale the window into a 32-bit value.
1500 * For the SYN_SENT state the scale is zero.
1502 tiwin = th->th_win << tp->snd_scale;
1505 * TCP ECN processing.
1507 if (tp->t_flags & TF_ECN_PERMIT) {
1508 if (thflags & TH_CWR)
1509 tp->t_flags &= ~TF_ECN_SND_ECE;
1510 switch (iptos & IPTOS_ECN_MASK) {
1512 tp->t_flags |= TF_ECN_SND_ECE;
1513 TCPSTAT_INC(tcps_ecn_ce);
1515 case IPTOS_ECN_ECT0:
1516 TCPSTAT_INC(tcps_ecn_ect0);
1518 case IPTOS_ECN_ECT1:
1519 TCPSTAT_INC(tcps_ecn_ect1);
1522 /* Congestion experienced. */
1523 if (thflags & TH_ECE) {
1524 cc_cong_signal(tp, th, CC_ECN);
1529 * Parse options on any incoming segment.
1531 tcp_dooptions(&to, (u_char *)(th + 1),
1532 (th->th_off << 2) - sizeof(struct tcphdr),
1533 (thflags & TH_SYN) ? TO_SYN : 0);
1536 * If echoed timestamp is later than the current time,
1537 * fall back to non RFC1323 RTT calculation. Normalize
1538 * timestamp if syncookies were used when this connection
1541 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1542 to.to_tsecr -= tp->ts_offset;
1543 if (TSTMP_GT(to.to_tsecr, ticks))
1548 * Process options only when we get SYN/ACK back. The SYN case
1549 * for incoming connections is handled in tcp_syncache.
1550 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
1551 * or <SYN,ACK>) segment itself is never scaled.
1552 * XXX this is traditional behavior, may need to be cleaned up.
1554 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1555 if ((to.to_flags & TOF_SCALE) &&
1556 (tp->t_flags & TF_REQ_SCALE)) {
1557 tp->t_flags |= TF_RCVD_SCALE;
1558 tp->snd_scale = to.to_wscale;
1561 * Initial send window. It will be updated with
1562 * the next incoming segment to the scaled value.
1564 tp->snd_wnd = th->th_win;
1565 if (to.to_flags & TOF_TS) {
1566 tp->t_flags |= TF_RCVD_TSTMP;
1567 tp->ts_recent = to.to_tsval;
1568 tp->ts_recent_age = ticks;
1570 if (to.to_flags & TOF_MSS)
1571 tcp_mss(tp, to.to_mss);
1572 if ((tp->t_flags & TF_SACK_PERMIT) &&
1573 (to.to_flags & TOF_SACKPERM) == 0)
1574 tp->t_flags &= ~TF_SACK_PERMIT;
1578 * Header prediction: check for the two common cases
1579 * of a uni-directional data xfer. If the packet has
1580 * no control flags, is in-sequence, the window didn't
1581 * change and we're not retransmitting, it's a
1582 * candidate. If the length is zero and the ack moved
1583 * forward, we're the sender side of the xfer. Just
1584 * free the data acked & wake any higher level process
1585 * that was blocked waiting for space. If the length
1586 * is non-zero and the ack didn't move, we're the
1587 * receiver side. If we're getting packets in-order
1588 * (the reassembly queue is empty), add the data to
1589 * the socket buffer and note that we need a delayed ack.
1590 * Make sure that the hidden state-flags are also off.
1591 * Since we check for TCPS_ESTABLISHED first, it can only
1594 if (tp->t_state == TCPS_ESTABLISHED &&
1595 th->th_seq == tp->rcv_nxt &&
1596 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1597 tp->snd_nxt == tp->snd_max &&
1598 tiwin && tiwin == tp->snd_wnd &&
1599 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1600 LIST_EMPTY(&tp->t_segq) &&
1601 ((to.to_flags & TOF_TS) == 0 ||
1602 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) {
1605 * If last ACK falls within this segment's sequence numbers,
1606 * record the timestamp.
1607 * NOTE that the test is modified according to the latest
1608 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1610 if ((to.to_flags & TOF_TS) != 0 &&
1611 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1612 tp->ts_recent_age = ticks;
1613 tp->ts_recent = to.to_tsval;
1617 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1618 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1619 !IN_RECOVERY(tp->t_flags) &&
1620 (to.to_flags & TOF_SACK) == 0 &&
1621 TAILQ_EMPTY(&tp->snd_holes)) {
1623 * This is a pure ack for outstanding data.
1625 if (ti_locked == TI_WLOCKED)
1626 INP_INFO_WUNLOCK(&V_tcbinfo);
1627 ti_locked = TI_UNLOCKED;
1629 TCPSTAT_INC(tcps_predack);
1632 * "bad retransmit" recovery.
1634 if (tp->t_rxtshift == 1 &&
1635 tp->t_flags & TF_PREVVALID &&
1636 (int)(ticks - tp->t_badrxtwin) < 0) {
1637 cc_cong_signal(tp, th, CC_RTO_ERR);
1641 * Recalculate the transmit timer / rtt.
1643 * Some boxes send broken timestamp replies
1644 * during the SYN+ACK phase, ignore
1645 * timestamps of 0 or we could calculate a
1646 * huge RTT and blow up the retransmit timer.
1648 if ((to.to_flags & TOF_TS) != 0 &&
1650 if (!tp->t_rttlow ||
1651 tp->t_rttlow > ticks - to.to_tsecr)
1652 tp->t_rttlow = ticks - to.to_tsecr;
1654 ticks - to.to_tsecr + 1);
1655 } else if (tp->t_rtttime &&
1656 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1657 if (!tp->t_rttlow ||
1658 tp->t_rttlow > ticks - tp->t_rtttime)
1659 tp->t_rttlow = ticks - tp->t_rtttime;
1661 ticks - tp->t_rtttime);
1663 acked = BYTES_THIS_ACK(tp, th);
1665 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
1666 hhook_run_tcp_est_in(tp, th, &to);
1668 TCPSTAT_INC(tcps_rcvackpack);
1669 TCPSTAT_ADD(tcps_rcvackbyte, acked);
1670 sbdrop(&so->so_snd, acked);
1671 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1672 SEQ_LEQ(th->th_ack, tp->snd_recover))
1673 tp->snd_recover = th->th_ack - 1;
1676 * Let the congestion control algorithm update
1677 * congestion control related information. This
1678 * typically means increasing the congestion
1681 cc_ack_received(tp, th, CC_ACK);
1683 tp->snd_una = th->th_ack;
1685 * Pull snd_wl2 up to prevent seq wrap relative
1688 tp->snd_wl2 = th->th_ack;
1691 ND6_HINT(tp); /* Some progress has been made. */
1694 * If all outstanding data are acked, stop
1695 * retransmit timer, otherwise restart timer
1696 * using current (possibly backed-off) value.
1697 * If process is waiting for space,
1698 * wakeup/selwakeup/signal. If data
1699 * are ready to send, let tcp_output
1700 * decide between more output or persist.
1703 if (so->so_options & SO_DEBUG)
1704 tcp_trace(TA_INPUT, ostate, tp,
1705 (void *)tcp_saveipgen,
1708 if (tp->snd_una == tp->snd_max)
1709 tcp_timer_activate(tp, TT_REXMT, 0);
1710 else if (!tcp_timer_active(tp, TT_PERSIST))
1711 tcp_timer_activate(tp, TT_REXMT,
1714 if (so->so_snd.sb_cc)
1715 (void) tcp_output(tp);
1718 } else if (th->th_ack == tp->snd_una &&
1719 tlen <= sbspace(&so->so_rcv)) {
1720 int newsize = 0; /* automatic sockbuf scaling */
1723 * This is a pure, in-sequence data packet with
1724 * nothing on the reassembly queue and we have enough
1725 * buffer space to take it.
1727 if (ti_locked == TI_WLOCKED)
1728 INP_INFO_WUNLOCK(&V_tcbinfo);
1729 ti_locked = TI_UNLOCKED;
1731 /* Clean receiver SACK report if present */
1732 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks)
1733 tcp_clean_sackreport(tp);
1734 TCPSTAT_INC(tcps_preddat);
1735 tp->rcv_nxt += tlen;
1737 * Pull snd_wl1 up to prevent seq wrap relative to
1740 tp->snd_wl1 = th->th_seq;
1742 * Pull rcv_up up to prevent seq wrap relative to
1745 tp->rcv_up = tp->rcv_nxt;
1746 TCPSTAT_INC(tcps_rcvpack);
1747 TCPSTAT_ADD(tcps_rcvbyte, tlen);
1748 ND6_HINT(tp); /* Some progress has been made */
1750 if (so->so_options & SO_DEBUG)
1751 tcp_trace(TA_INPUT, ostate, tp,
1752 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1755 * Automatic sizing of receive socket buffer. Often the send
1756 * buffer size is not optimally adjusted to the actual network
1757 * conditions at hand (delay bandwidth product). Setting the
1758 * buffer size too small limits throughput on links with high
1759 * bandwidth and high delay (eg. trans-continental/oceanic links).
1761 * On the receive side the socket buffer memory is only rarely
1762 * used to any significant extent. This allows us to be much
1763 * more aggressive in scaling the receive socket buffer. For
1764 * the case that the buffer space is actually used to a large
1765 * extent and we run out of kernel memory we can simply drop
1766 * the new segments; TCP on the sender will just retransmit it
1767 * later. Setting the buffer size too big may only consume too
1768 * much kernel memory if the application doesn't read() from
1769 * the socket or packet loss or reordering makes use of the
1772 * The criteria to step up the receive buffer one notch are:
1773 * 1. the number of bytes received during the time it takes
1774 * one timestamp to be reflected back to us (the RTT);
1775 * 2. received bytes per RTT is within seven eighth of the
1776 * current socket buffer size;
1777 * 3. receive buffer size has not hit maximal automatic size;
1779 * This algorithm does one step per RTT at most and only if
1780 * we receive a bulk stream w/o packet losses or reorderings.
1781 * Shrinking the buffer during idle times is not necessary as
1782 * it doesn't consume any memory when idle.
1784 * TODO: Only step up if the application is actually serving
1785 * the buffer to better manage the socket buffer resources.
1787 if (V_tcp_do_autorcvbuf &&
1789 (so->so_rcv.sb_flags & SB_AUTOSIZE)) {
1790 if (TSTMP_GT(to.to_tsecr, tp->rfbuf_ts) &&
1791 to.to_tsecr - tp->rfbuf_ts < hz) {
1793 (so->so_rcv.sb_hiwat / 8 * 7) &&
1794 so->so_rcv.sb_hiwat <
1795 V_tcp_autorcvbuf_max) {
1797 min(so->so_rcv.sb_hiwat +
1798 V_tcp_autorcvbuf_inc,
1799 V_tcp_autorcvbuf_max);
1801 /* Start over with next RTT. */
1805 tp->rfbuf_cnt += tlen; /* add up */
1808 /* Add data to socket buffer. */
1809 SOCKBUF_LOCK(&so->so_rcv);
1810 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1814 * Set new socket buffer size.
1815 * Give up when limit is reached.
1818 if (!sbreserve_locked(&so->so_rcv,
1820 so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1821 m_adj(m, drop_hdrlen); /* delayed header drop */
1822 sbappendstream_locked(&so->so_rcv, m);
1824 /* NB: sorwakeup_locked() does an implicit unlock. */
1825 sorwakeup_locked(so);
1826 if (DELAY_ACK(tp)) {
1827 tp->t_flags |= TF_DELACK;
1829 tp->t_flags |= TF_ACKNOW;
1837 * Calculate amount of space in receive window,
1838 * and then do TCP input processing.
1839 * Receive window is amount of space in rcv queue,
1840 * but not less than advertised window.
1842 win = sbspace(&so->so_rcv);
1845 KASSERT(SEQ_GEQ(tp->rcv_adv, tp->rcv_nxt),
1846 ("tcp_input negative window: tp %p rcv_nxt %u rcv_adv %u", tp,
1847 tp->rcv_nxt, tp->rcv_adv));
1848 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1850 /* Reset receive buffer auto scaling when not in bulk receive mode. */
1854 switch (tp->t_state) {
1857 * If the state is SYN_RECEIVED:
1858 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1860 case TCPS_SYN_RECEIVED:
1861 if ((thflags & TH_ACK) &&
1862 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1863 SEQ_GT(th->th_ack, tp->snd_max))) {
1864 rstreason = BANDLIM_RST_OPENPORT;
1870 * If the state is SYN_SENT:
1871 * if seg contains an ACK, but not for our SYN, drop the input.
1872 * if seg contains a RST, then drop the connection.
1873 * if seg does not contain SYN, then drop it.
1874 * Otherwise this is an acceptable SYN segment
1875 * initialize tp->rcv_nxt and tp->irs
1876 * if seg contains ack then advance tp->snd_una
1877 * if seg contains an ECE and ECN support is enabled, the stream
1879 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1880 * arrange for segment to be acked (eventually)
1881 * continue processing rest of data/controls, beginning with URG
1884 if ((thflags & TH_ACK) &&
1885 (SEQ_LEQ(th->th_ack, tp->iss) ||
1886 SEQ_GT(th->th_ack, tp->snd_max))) {
1887 rstreason = BANDLIM_UNLIMITED;
1890 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST))
1891 tp = tcp_drop(tp, ECONNREFUSED);
1892 if (thflags & TH_RST)
1894 if (!(thflags & TH_SYN))
1897 tp->irs = th->th_seq;
1899 if (thflags & TH_ACK) {
1900 TCPSTAT_INC(tcps_connects);
1903 mac_socketpeer_set_from_mbuf(m, so);
1905 /* Do window scaling on this connection? */
1906 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1907 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1908 tp->rcv_scale = tp->request_r_scale;
1910 tp->rcv_adv += imin(tp->rcv_wnd,
1911 TCP_MAXWIN << tp->rcv_scale);
1912 tp->snd_una++; /* SYN is acked */
1914 * If there's data, delay ACK; if there's also a FIN
1915 * ACKNOW will be turned on later.
1917 if (DELAY_ACK(tp) && tlen != 0)
1918 tcp_timer_activate(tp, TT_DELACK,
1921 tp->t_flags |= TF_ACKNOW;
1923 if ((thflags & TH_ECE) && V_tcp_do_ecn) {
1924 tp->t_flags |= TF_ECN_PERMIT;
1925 TCPSTAT_INC(tcps_ecn_shs);
1929 * Received <SYN,ACK> in SYN_SENT[*] state.
1931 * SYN_SENT --> ESTABLISHED
1932 * SYN_SENT* --> FIN_WAIT_1
1934 tp->t_starttime = ticks;
1935 if (tp->t_flags & TF_NEEDFIN) {
1936 tp->t_state = TCPS_FIN_WAIT_1;
1937 tp->t_flags &= ~TF_NEEDFIN;
1940 tp->t_state = TCPS_ESTABLISHED;
1942 tcp_timer_activate(tp, TT_KEEP, tcp_keepidle);
1946 * Received initial SYN in SYN-SENT[*] state =>
1947 * simultaneous open. If segment contains CC option
1948 * and there is a cached CC, apply TAO test.
1949 * If it succeeds, connection is * half-synchronized.
1950 * Otherwise, do 3-way handshake:
1951 * SYN-SENT -> SYN-RECEIVED
1952 * SYN-SENT* -> SYN-RECEIVED*
1953 * If there was no CC option, clear cached CC value.
1955 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
1956 tcp_timer_activate(tp, TT_REXMT, 0);
1957 tp->t_state = TCPS_SYN_RECEIVED;
1960 KASSERT(ti_locked == TI_WLOCKED, ("%s: trimthenstep6: "
1961 "ti_locked %d", __func__, ti_locked));
1962 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1963 INP_WLOCK_ASSERT(tp->t_inpcb);
1966 * Advance th->th_seq to correspond to first data byte.
1967 * If data, trim to stay within window,
1968 * dropping FIN if necessary.
1971 if (tlen > tp->rcv_wnd) {
1972 todrop = tlen - tp->rcv_wnd;
1976 TCPSTAT_INC(tcps_rcvpackafterwin);
1977 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
1979 tp->snd_wl1 = th->th_seq - 1;
1980 tp->rcv_up = th->th_seq;
1982 * Client side of transaction: already sent SYN and data.
1983 * If the remote host used T/TCP to validate the SYN,
1984 * our data will be ACK'd; if so, enter normal data segment
1985 * processing in the middle of step 5, ack processing.
1986 * Otherwise, goto step 6.
1988 if (thflags & TH_ACK)
1994 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
1995 * do normal processing.
1997 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later.
2001 break; /* continue normal processing */
2005 * States other than LISTEN or SYN_SENT.
2006 * First check the RST flag and sequence number since reset segments
2007 * are exempt from the timestamp and connection count tests. This
2008 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
2009 * below which allowed reset segments in half the sequence space
2010 * to fall though and be processed (which gives forged reset
2011 * segments with a random sequence number a 50 percent chance of
2012 * killing a connection).
2013 * Then check timestamp, if present.
2014 * Then check the connection count, if present.
2015 * Then check that at least some bytes of segment are within
2016 * receive window. If segment begins before rcv_nxt,
2017 * drop leading data (and SYN); if nothing left, just ack.
2020 * If the RST bit is set, check the sequence number to see
2021 * if this is a valid reset segment.
2023 * In all states except SYN-SENT, all reset (RST) segments
2024 * are validated by checking their SEQ-fields. A reset is
2025 * valid if its sequence number is in the window.
2026 * Note: this does not take into account delayed ACKs, so
2027 * we should test against last_ack_sent instead of rcv_nxt.
2028 * The sequence number in the reset segment is normally an
2029 * echo of our outgoing acknowlegement numbers, but some hosts
2030 * send a reset with the sequence number at the rightmost edge
2031 * of our receive window, and we have to handle this case.
2032 * Note 2: Paul Watson's paper "Slipping in the Window" has shown
2033 * that brute force RST attacks are possible. To combat this,
2034 * we use a much stricter check while in the ESTABLISHED state,
2035 * only accepting RSTs where the sequence number is equal to
2036 * last_ack_sent. In all other states (the states in which a
2037 * RST is more likely), the more permissive check is used.
2038 * If we have multiple segments in flight, the initial reset
2039 * segment sequence numbers will be to the left of last_ack_sent,
2040 * but they will eventually catch up.
2041 * In any case, it never made sense to trim reset segments to
2042 * fit the receive window since RFC 1122 says:
2043 * 4.2.2.12 RST Segment: RFC-793 Section 3.4
2045 * A TCP SHOULD allow a received RST segment to include data.
2048 * It has been suggested that a RST segment could contain
2049 * ASCII text that encoded and explained the cause of the
2050 * RST. No standard has yet been established for such
2053 * If the reset segment passes the sequence number test examine
2055 * SYN_RECEIVED STATE:
2056 * If passive open, return to LISTEN state.
2057 * If active open, inform user that connection was refused.
2058 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
2059 * Inform user that connection was reset, and close tcb.
2060 * CLOSING, LAST_ACK STATES:
2063 * Drop the segment - see Stevens, vol. 2, p. 964 and
2066 if (thflags & TH_RST) {
2067 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) &&
2068 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
2069 switch (tp->t_state) {
2071 case TCPS_SYN_RECEIVED:
2072 so->so_error = ECONNREFUSED;
2075 case TCPS_ESTABLISHED:
2076 if (V_tcp_insecure_rst == 0 &&
2077 !(SEQ_GEQ(th->th_seq, tp->rcv_nxt - 1) &&
2078 SEQ_LEQ(th->th_seq, tp->rcv_nxt + 1)) &&
2079 !(SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) &&
2080 SEQ_LEQ(th->th_seq, tp->last_ack_sent + 1))) {
2081 TCPSTAT_INC(tcps_badrst);
2085 case TCPS_FIN_WAIT_1:
2086 case TCPS_FIN_WAIT_2:
2087 case TCPS_CLOSE_WAIT:
2088 so->so_error = ECONNRESET;
2090 KASSERT(ti_locked == TI_WLOCKED,
2091 ("tcp_do_segment: TH_RST 1 ti_locked %d",
2093 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2095 tp->t_state = TCPS_CLOSED;
2096 TCPSTAT_INC(tcps_drops);
2102 KASSERT(ti_locked == TI_WLOCKED,
2103 ("tcp_do_segment: TH_RST 2 ti_locked %d",
2105 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2115 * RFC 1323 PAWS: If we have a timestamp reply on this segment
2116 * and it's less than ts_recent, drop it.
2118 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
2119 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
2121 /* Check to see if ts_recent is over 24 days old. */
2122 if (ticks - tp->ts_recent_age > TCP_PAWS_IDLE) {
2124 * Invalidate ts_recent. If this segment updates
2125 * ts_recent, the age will be reset later and ts_recent
2126 * will get a valid value. If it does not, setting
2127 * ts_recent to zero will at least satisfy the
2128 * requirement that zero be placed in the timestamp
2129 * echo reply when ts_recent isn't valid. The
2130 * age isn't reset until we get a valid ts_recent
2131 * because we don't want out-of-order segments to be
2132 * dropped when ts_recent is old.
2136 TCPSTAT_INC(tcps_rcvduppack);
2137 TCPSTAT_ADD(tcps_rcvdupbyte, tlen);
2138 TCPSTAT_INC(tcps_pawsdrop);
2146 * In the SYN-RECEIVED state, validate that the packet belongs to
2147 * this connection before trimming the data to fit the receive
2148 * window. Check the sequence number versus IRS since we know
2149 * the sequence numbers haven't wrapped. This is a partial fix
2150 * for the "LAND" DoS attack.
2152 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
2153 rstreason = BANDLIM_RST_OPENPORT;
2157 todrop = tp->rcv_nxt - th->th_seq;
2160 * If this is a duplicate SYN for our current connection,
2161 * advance over it and pretend and it's not a SYN.
2163 if (thflags & TH_SYN && th->th_seq == tp->irs) {
2173 * Following if statement from Stevens, vol. 2, p. 960.
2176 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
2178 * Any valid FIN must be to the left of the window.
2179 * At this point the FIN must be a duplicate or out
2180 * of sequence; drop it.
2185 * Send an ACK to resynchronize and drop any data.
2186 * But keep on processing for RST or ACK.
2188 tp->t_flags |= TF_ACKNOW;
2190 TCPSTAT_INC(tcps_rcvduppack);
2191 TCPSTAT_ADD(tcps_rcvdupbyte, todrop);
2193 TCPSTAT_INC(tcps_rcvpartduppack);
2194 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop);
2196 drop_hdrlen += todrop; /* drop from the top afterwards */
2197 th->th_seq += todrop;
2199 if (th->th_urp > todrop)
2200 th->th_urp -= todrop;
2208 * If new data are received on a connection after the
2209 * user processes are gone, then RST the other end.
2211 if ((so->so_state & SS_NOFDREF) &&
2212 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
2215 KASSERT(ti_locked == TI_WLOCKED, ("%s: SS_NOFDEREF && "
2216 "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked));
2217 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2219 if ((s = tcp_log_addrs(&tp->t_inpcb->inp_inc, th, NULL, NULL))) {
2220 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data after socket "
2221 "was closed, sending RST and removing tcpcb\n",
2222 s, __func__, tcpstates[tp->t_state], tlen);
2226 TCPSTAT_INC(tcps_rcvafterclose);
2227 rstreason = BANDLIM_UNLIMITED;
2232 * If segment ends after window, drop trailing data
2233 * (and PUSH and FIN); if nothing left, just ACK.
2235 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
2237 TCPSTAT_INC(tcps_rcvpackafterwin);
2238 if (todrop >= tlen) {
2239 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen);
2241 * If window is closed can only take segments at
2242 * window edge, and have to drop data and PUSH from
2243 * incoming segments. Continue processing, but
2244 * remember to ack. Otherwise, drop segment
2247 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2248 tp->t_flags |= TF_ACKNOW;
2249 TCPSTAT_INC(tcps_rcvwinprobe);
2253 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2256 thflags &= ~(TH_PUSH|TH_FIN);
2260 * If last ACK falls within this segment's sequence numbers,
2261 * record its timestamp.
2263 * 1) That the test incorporates suggestions from the latest
2264 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
2265 * 2) That updating only on newer timestamps interferes with
2266 * our earlier PAWS tests, so this check should be solely
2267 * predicated on the sequence space of this segment.
2268 * 3) That we modify the segment boundary check to be
2269 * Last.ACK.Sent <= SEG.SEQ + SEG.Len
2270 * instead of RFC1323's
2271 * Last.ACK.Sent < SEG.SEQ + SEG.Len,
2272 * This modified check allows us to overcome RFC1323's
2273 * limitations as described in Stevens TCP/IP Illustrated
2274 * Vol. 2 p.869. In such cases, we can still calculate the
2275 * RTT correctly when RCV.NXT == Last.ACK.Sent.
2277 if ((to.to_flags & TOF_TS) != 0 &&
2278 SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2279 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
2280 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
2281 tp->ts_recent_age = ticks;
2282 tp->ts_recent = to.to_tsval;
2286 * If a SYN is in the window, then this is an
2287 * error and we send an RST and drop the connection.
2289 if (thflags & TH_SYN) {
2290 KASSERT(ti_locked == TI_WLOCKED,
2291 ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked));
2292 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2294 tp = tcp_drop(tp, ECONNRESET);
2295 rstreason = BANDLIM_UNLIMITED;
2300 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
2301 * flag is on (half-synchronized state), then queue data for
2302 * later processing; else drop segment and return.
2304 if ((thflags & TH_ACK) == 0) {
2305 if (tp->t_state == TCPS_SYN_RECEIVED ||
2306 (tp->t_flags & TF_NEEDSYN))
2308 else if (tp->t_flags & TF_ACKNOW)
2317 switch (tp->t_state) {
2320 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
2321 * ESTABLISHED state and continue processing.
2322 * The ACK was checked above.
2324 case TCPS_SYN_RECEIVED:
2326 TCPSTAT_INC(tcps_connects);
2328 /* Do window scaling? */
2329 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2330 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2331 tp->rcv_scale = tp->request_r_scale;
2332 tp->snd_wnd = tiwin;
2336 * SYN-RECEIVED -> ESTABLISHED
2337 * SYN-RECEIVED* -> FIN-WAIT-1
2339 tp->t_starttime = ticks;
2340 if (tp->t_flags & TF_NEEDFIN) {
2341 tp->t_state = TCPS_FIN_WAIT_1;
2342 tp->t_flags &= ~TF_NEEDFIN;
2344 tp->t_state = TCPS_ESTABLISHED;
2346 tcp_timer_activate(tp, TT_KEEP, tcp_keepidle);
2349 * If segment contains data or ACK, will call tcp_reass()
2350 * later; if not, do so now to pass queued data to user.
2352 if (tlen == 0 && (thflags & TH_FIN) == 0)
2353 (void) tcp_reass(tp, (struct tcphdr *)0, 0,
2355 tp->snd_wl1 = th->th_seq - 1;
2359 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2360 * ACKs. If the ack is in the range
2361 * tp->snd_una < th->th_ack <= tp->snd_max
2362 * then advance tp->snd_una to th->th_ack and drop
2363 * data from the retransmission queue. If this ACK reflects
2364 * more up to date window information we update our window information.
2366 case TCPS_ESTABLISHED:
2367 case TCPS_FIN_WAIT_1:
2368 case TCPS_FIN_WAIT_2:
2369 case TCPS_CLOSE_WAIT:
2372 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2373 TCPSTAT_INC(tcps_rcvacktoomuch);
2376 if ((tp->t_flags & TF_SACK_PERMIT) &&
2377 ((to.to_flags & TOF_SACK) ||
2378 !TAILQ_EMPTY(&tp->snd_holes)))
2379 tcp_sack_doack(tp, &to, th->th_ack);
2381 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
2382 hhook_run_tcp_est_in(tp, th, &to);
2384 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2385 if (tlen == 0 && tiwin == tp->snd_wnd) {
2386 TCPSTAT_INC(tcps_rcvdupack);
2388 * If we have outstanding data (other than
2389 * a window probe), this is a completely
2390 * duplicate ack (ie, window info didn't
2391 * change), the ack is the biggest we've
2392 * seen and we've seen exactly our rexmt
2393 * threshhold of them, assume a packet
2394 * has been dropped and retransmit it.
2395 * Kludge snd_nxt & the congestion
2396 * window so we send only this one
2399 * We know we're losing at the current
2400 * window size so do congestion avoidance
2401 * (set ssthresh to half the current window
2402 * and pull our congestion window back to
2403 * the new ssthresh).
2405 * Dup acks mean that packets have left the
2406 * network (they're now cached at the receiver)
2407 * so bump cwnd by the amount in the receiver
2408 * to keep a constant cwnd packets in the
2411 * When using TCP ECN, notify the peer that
2412 * we reduced the cwnd.
2414 if (!tcp_timer_active(tp, TT_REXMT) ||
2415 th->th_ack != tp->snd_una)
2417 else if (++tp->t_dupacks > tcprexmtthresh ||
2418 IN_FASTRECOVERY(tp->t_flags)) {
2419 cc_ack_received(tp, th, CC_DUPACK);
2420 if ((tp->t_flags & TF_SACK_PERMIT) &&
2421 IN_FASTRECOVERY(tp->t_flags)) {
2425 * Compute the amount of data in flight first.
2426 * We can inject new data into the pipe iff
2427 * we have less than 1/2 the original window's
2428 * worth of data in flight.
2430 awnd = (tp->snd_nxt - tp->snd_fack) +
2431 tp->sackhint.sack_bytes_rexmit;
2432 if (awnd < tp->snd_ssthresh) {
2433 tp->snd_cwnd += tp->t_maxseg;
2434 if (tp->snd_cwnd > tp->snd_ssthresh)
2435 tp->snd_cwnd = tp->snd_ssthresh;
2438 tp->snd_cwnd += tp->t_maxseg;
2439 (void) tcp_output(tp);
2441 } else if (tp->t_dupacks == tcprexmtthresh) {
2442 tcp_seq onxt = tp->snd_nxt;
2445 * If we're doing sack, check to
2446 * see if we're already in sack
2447 * recovery. If we're not doing sack,
2448 * check to see if we're in newreno
2451 if (tp->t_flags & TF_SACK_PERMIT) {
2452 if (IN_FASTRECOVERY(tp->t_flags)) {
2457 if (SEQ_LEQ(th->th_ack,
2463 /* Congestion signal before ack. */
2464 cc_cong_signal(tp, th, CC_NDUPACK);
2465 cc_ack_received(tp, th, CC_DUPACK);
2466 tcp_timer_activate(tp, TT_REXMT, 0);
2468 if (tp->t_flags & TF_SACK_PERMIT) {
2470 tcps_sack_recovery_episode);
2471 tp->sack_newdata = tp->snd_nxt;
2472 tp->snd_cwnd = tp->t_maxseg;
2473 (void) tcp_output(tp);
2476 tp->snd_nxt = th->th_ack;
2477 tp->snd_cwnd = tp->t_maxseg;
2478 (void) tcp_output(tp);
2479 KASSERT(tp->snd_limited <= 2,
2480 ("%s: tp->snd_limited too big",
2482 tp->snd_cwnd = tp->snd_ssthresh +
2484 (tp->t_dupacks - tp->snd_limited);
2485 if (SEQ_GT(onxt, tp->snd_nxt))
2488 } else if (V_tcp_do_rfc3042) {
2489 cc_ack_received(tp, th, CC_DUPACK);
2490 u_long oldcwnd = tp->snd_cwnd;
2491 tcp_seq oldsndmax = tp->snd_max;
2494 KASSERT(tp->t_dupacks == 1 ||
2496 ("%s: dupacks not 1 or 2",
2498 if (tp->t_dupacks == 1)
2499 tp->snd_limited = 0;
2501 (tp->snd_nxt - tp->snd_una) +
2502 (tp->t_dupacks - tp->snd_limited) *
2504 (void) tcp_output(tp);
2505 sent = tp->snd_max - oldsndmax;
2506 if (sent > tp->t_maxseg) {
2507 KASSERT((tp->t_dupacks == 2 &&
2508 tp->snd_limited == 0) ||
2509 (sent == tp->t_maxseg + 1 &&
2510 tp->t_flags & TF_SENTFIN),
2511 ("%s: sent too much",
2513 tp->snd_limited = 2;
2514 } else if (sent > 0)
2516 tp->snd_cwnd = oldcwnd;
2524 KASSERT(SEQ_GT(th->th_ack, tp->snd_una),
2525 ("%s: th_ack <= snd_una", __func__));
2528 * If the congestion window was inflated to account
2529 * for the other side's cached packets, retract it.
2531 if (IN_FASTRECOVERY(tp->t_flags)) {
2532 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2533 if (tp->t_flags & TF_SACK_PERMIT)
2534 tcp_sack_partialack(tp, th);
2536 tcp_newreno_partial_ack(tp, th);
2538 cc_post_recovery(tp, th);
2542 * If we reach this point, ACK is not a duplicate,
2543 * i.e., it ACKs something we sent.
2545 if (tp->t_flags & TF_NEEDSYN) {
2547 * T/TCP: Connection was half-synchronized, and our
2548 * SYN has been ACK'd (so connection is now fully
2549 * synchronized). Go to non-starred state,
2550 * increment snd_una for ACK of SYN, and check if
2551 * we can do window scaling.
2553 tp->t_flags &= ~TF_NEEDSYN;
2555 /* Do window scaling? */
2556 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2557 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2558 tp->rcv_scale = tp->request_r_scale;
2559 /* Send window already scaled. */
2564 INP_WLOCK_ASSERT(tp->t_inpcb);
2566 acked = BYTES_THIS_ACK(tp, th);
2567 TCPSTAT_INC(tcps_rcvackpack);
2568 TCPSTAT_ADD(tcps_rcvackbyte, acked);
2571 * If we just performed our first retransmit, and the ACK
2572 * arrives within our recovery window, then it was a mistake
2573 * to do the retransmit in the first place. Recover our
2574 * original cwnd and ssthresh, and proceed to transmit where
2577 if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID &&
2578 (int)(ticks - tp->t_badrxtwin) < 0)
2579 cc_cong_signal(tp, th, CC_RTO_ERR);
2582 * If we have a timestamp reply, update smoothed
2583 * round trip time. If no timestamp is present but
2584 * transmit timer is running and timed sequence
2585 * number was acked, update smoothed round trip time.
2586 * Since we now have an rtt measurement, cancel the
2587 * timer backoff (cf., Phil Karn's retransmit alg.).
2588 * Recompute the initial retransmit timer.
2590 * Some boxes send broken timestamp replies
2591 * during the SYN+ACK phase, ignore
2592 * timestamps of 0 or we could calculate a
2593 * huge RTT and blow up the retransmit timer.
2595 if ((to.to_flags & TOF_TS) != 0 &&
2597 if (!tp->t_rttlow || tp->t_rttlow > ticks - to.to_tsecr)
2598 tp->t_rttlow = ticks - to.to_tsecr;
2599 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1);
2600 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2601 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2602 tp->t_rttlow = ticks - tp->t_rtttime;
2603 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2607 * If all outstanding data is acked, stop retransmit
2608 * timer and remember to restart (more output or persist).
2609 * If there is more data to be acked, restart retransmit
2610 * timer, using current (possibly backed-off) value.
2612 if (th->th_ack == tp->snd_max) {
2613 tcp_timer_activate(tp, TT_REXMT, 0);
2615 } else if (!tcp_timer_active(tp, TT_PERSIST))
2616 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
2619 * If no data (only SYN) was ACK'd,
2620 * skip rest of ACK processing.
2626 * Let the congestion control algorithm update congestion
2627 * control related information. This typically means increasing
2628 * the congestion window.
2630 cc_ack_received(tp, th, CC_ACK);
2632 SOCKBUF_LOCK(&so->so_snd);
2633 if (acked > so->so_snd.sb_cc) {
2634 tp->snd_wnd -= so->so_snd.sb_cc;
2635 sbdrop_locked(&so->so_snd, (int)so->so_snd.sb_cc);
2638 sbdrop_locked(&so->so_snd, acked);
2639 tp->snd_wnd -= acked;
2642 /* NB: sowwakeup_locked() does an implicit unlock. */
2643 sowwakeup_locked(so);
2644 /* Detect una wraparound. */
2645 if (!IN_RECOVERY(tp->t_flags) &&
2646 SEQ_GT(tp->snd_una, tp->snd_recover) &&
2647 SEQ_LEQ(th->th_ack, tp->snd_recover))
2648 tp->snd_recover = th->th_ack - 1;
2649 /* XXXLAS: Can this be moved up into cc_post_recovery? */
2650 if (IN_RECOVERY(tp->t_flags) &&
2651 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2652 EXIT_RECOVERY(tp->t_flags);
2654 tp->snd_una = th->th_ack;
2655 if (tp->t_flags & TF_SACK_PERMIT) {
2656 if (SEQ_GT(tp->snd_una, tp->snd_recover))
2657 tp->snd_recover = tp->snd_una;
2659 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2660 tp->snd_nxt = tp->snd_una;
2662 switch (tp->t_state) {
2665 * In FIN_WAIT_1 STATE in addition to the processing
2666 * for the ESTABLISHED state if our FIN is now acknowledged
2667 * then enter FIN_WAIT_2.
2669 case TCPS_FIN_WAIT_1:
2670 if (ourfinisacked) {
2672 * If we can't receive any more
2673 * data, then closing user can proceed.
2674 * Starting the timer is contrary to the
2675 * specification, but if we don't get a FIN
2676 * we'll hang forever.
2679 * we should release the tp also, and use a
2682 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2685 soisdisconnected(so);
2686 timeout = (tcp_fast_finwait2_recycle) ?
2687 tcp_finwait2_timeout : tcp_maxidle;
2688 tcp_timer_activate(tp, TT_2MSL, timeout);
2690 tp->t_state = TCPS_FIN_WAIT_2;
2695 * In CLOSING STATE in addition to the processing for
2696 * the ESTABLISHED state if the ACK acknowledges our FIN
2697 * then enter the TIME-WAIT state, otherwise ignore
2701 if (ourfinisacked) {
2702 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2704 INP_INFO_WUNLOCK(&V_tcbinfo);
2711 * In LAST_ACK, we may still be waiting for data to drain
2712 * and/or to be acked, as well as for the ack of our FIN.
2713 * If our FIN is now acknowledged, delete the TCB,
2714 * enter the closed state and return.
2717 if (ourfinisacked) {
2718 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2727 INP_WLOCK_ASSERT(tp->t_inpcb);
2730 * Update window information.
2731 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2733 if ((thflags & TH_ACK) &&
2734 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2735 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2736 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2737 /* keep track of pure window updates */
2739 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2740 TCPSTAT_INC(tcps_rcvwinupd);
2741 tp->snd_wnd = tiwin;
2742 tp->snd_wl1 = th->th_seq;
2743 tp->snd_wl2 = th->th_ack;
2744 if (tp->snd_wnd > tp->max_sndwnd)
2745 tp->max_sndwnd = tp->snd_wnd;
2750 * Process segments with URG.
2752 if ((thflags & TH_URG) && th->th_urp &&
2753 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2755 * This is a kludge, but if we receive and accept
2756 * random urgent pointers, we'll crash in
2757 * soreceive. It's hard to imagine someone
2758 * actually wanting to send this much urgent data.
2760 SOCKBUF_LOCK(&so->so_rcv);
2761 if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2762 th->th_urp = 0; /* XXX */
2763 thflags &= ~TH_URG; /* XXX */
2764 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */
2765 goto dodata; /* XXX */
2768 * If this segment advances the known urgent pointer,
2769 * then mark the data stream. This should not happen
2770 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2771 * a FIN has been received from the remote side.
2772 * In these states we ignore the URG.
2774 * According to RFC961 (Assigned Protocols),
2775 * the urgent pointer points to the last octet
2776 * of urgent data. We continue, however,
2777 * to consider it to indicate the first octet
2778 * of data past the urgent section as the original
2779 * spec states (in one of two places).
2781 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2782 tp->rcv_up = th->th_seq + th->th_urp;
2783 so->so_oobmark = so->so_rcv.sb_cc +
2784 (tp->rcv_up - tp->rcv_nxt) - 1;
2785 if (so->so_oobmark == 0)
2786 so->so_rcv.sb_state |= SBS_RCVATMARK;
2788 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2790 SOCKBUF_UNLOCK(&so->so_rcv);
2792 * Remove out of band data so doesn't get presented to user.
2793 * This can happen independent of advancing the URG pointer,
2794 * but if two URG's are pending at once, some out-of-band
2795 * data may creep in... ick.
2797 if (th->th_urp <= (u_long)tlen &&
2798 !(so->so_options & SO_OOBINLINE)) {
2799 /* hdr drop is delayed */
2800 tcp_pulloutofband(so, th, m, drop_hdrlen);
2804 * If no out of band data is expected,
2805 * pull receive urgent pointer along
2806 * with the receive window.
2808 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2809 tp->rcv_up = tp->rcv_nxt;
2812 INP_WLOCK_ASSERT(tp->t_inpcb);
2815 * Process the segment text, merging it into the TCP sequencing queue,
2816 * and arranging for acknowledgment of receipt if necessary.
2817 * This process logically involves adjusting tp->rcv_wnd as data
2818 * is presented to the user (this happens in tcp_usrreq.c,
2819 * case PRU_RCVD). If a FIN has already been received on this
2820 * connection then we just ignore the text.
2822 if ((tlen || (thflags & TH_FIN)) &&
2823 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2824 tcp_seq save_start = th->th_seq;
2825 m_adj(m, drop_hdrlen); /* delayed header drop */
2827 * Insert segment which includes th into TCP reassembly queue
2828 * with control block tp. Set thflags to whether reassembly now
2829 * includes a segment with FIN. This handles the common case
2830 * inline (segment is the next to be received on an established
2831 * connection, and the queue is empty), avoiding linkage into
2832 * and removal from the queue and repetition of various
2834 * Set DELACK for segments received in order, but ack
2835 * immediately when segments are out of order (so
2836 * fast retransmit can work).
2838 if (th->th_seq == tp->rcv_nxt &&
2839 LIST_EMPTY(&tp->t_segq) &&
2840 TCPS_HAVEESTABLISHED(tp->t_state)) {
2842 tp->t_flags |= TF_DELACK;
2844 tp->t_flags |= TF_ACKNOW;
2845 tp->rcv_nxt += tlen;
2846 thflags = th->th_flags & TH_FIN;
2847 TCPSTAT_INC(tcps_rcvpack);
2848 TCPSTAT_ADD(tcps_rcvbyte, tlen);
2850 SOCKBUF_LOCK(&so->so_rcv);
2851 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
2854 sbappendstream_locked(&so->so_rcv, m);
2855 /* NB: sorwakeup_locked() does an implicit unlock. */
2856 sorwakeup_locked(so);
2859 * XXX: Due to the header drop above "th" is
2860 * theoretically invalid by now. Fortunately
2861 * m_adj() doesn't actually frees any mbufs
2862 * when trimming from the head.
2864 thflags = tcp_reass(tp, th, &tlen, m);
2865 tp->t_flags |= TF_ACKNOW;
2867 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT))
2868 tcp_update_sack_list(tp, save_start, save_start + tlen);
2871 * Note the amount of data that peer has sent into
2872 * our window, in order to estimate the sender's
2876 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
2877 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2879 len = so->so_rcv.sb_hiwat;
2887 * If FIN is received ACK the FIN and let the user know
2888 * that the connection is closing.
2890 if (thflags & TH_FIN) {
2891 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2894 * If connection is half-synchronized
2895 * (ie NEEDSYN flag on) then delay ACK,
2896 * so it may be piggybacked when SYN is sent.
2897 * Otherwise, since we received a FIN then no
2898 * more input can be expected, send ACK now.
2900 if (tp->t_flags & TF_NEEDSYN)
2901 tp->t_flags |= TF_DELACK;
2903 tp->t_flags |= TF_ACKNOW;
2906 switch (tp->t_state) {
2909 * In SYN_RECEIVED and ESTABLISHED STATES
2910 * enter the CLOSE_WAIT state.
2912 case TCPS_SYN_RECEIVED:
2913 tp->t_starttime = ticks;
2915 case TCPS_ESTABLISHED:
2916 tp->t_state = TCPS_CLOSE_WAIT;
2920 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2921 * enter the CLOSING state.
2923 case TCPS_FIN_WAIT_1:
2924 tp->t_state = TCPS_CLOSING;
2928 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2929 * starting the time-wait timer, turning off the other
2932 case TCPS_FIN_WAIT_2:
2933 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2934 KASSERT(ti_locked == TI_WLOCKED, ("%s: dodata "
2935 "TCP_FIN_WAIT_2 ti_locked: %d", __func__,
2939 INP_INFO_WUNLOCK(&V_tcbinfo);
2943 if (ti_locked == TI_WLOCKED)
2944 INP_INFO_WUNLOCK(&V_tcbinfo);
2945 ti_locked = TI_UNLOCKED;
2948 if (so->so_options & SO_DEBUG)
2949 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
2954 * Return any desired output.
2956 if (needoutput || (tp->t_flags & TF_ACKNOW))
2957 (void) tcp_output(tp);
2960 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d",
2961 __func__, ti_locked));
2962 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
2963 INP_WLOCK_ASSERT(tp->t_inpcb);
2965 if (tp->t_flags & TF_DELACK) {
2966 tp->t_flags &= ~TF_DELACK;
2967 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
2969 INP_WUNLOCK(tp->t_inpcb);
2974 * Generate an ACK dropping incoming segment if it occupies
2975 * sequence space, where the ACK reflects our state.
2977 * We can now skip the test for the RST flag since all
2978 * paths to this code happen after packets containing
2979 * RST have been dropped.
2981 * In the SYN-RECEIVED state, don't send an ACK unless the
2982 * segment we received passes the SYN-RECEIVED ACK test.
2983 * If it fails send a RST. This breaks the loop in the
2984 * "LAND" DoS attack, and also prevents an ACK storm
2985 * between two listening ports that have been sent forged
2986 * SYN segments, each with the source address of the other.
2988 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
2989 (SEQ_GT(tp->snd_una, th->th_ack) ||
2990 SEQ_GT(th->th_ack, tp->snd_max)) ) {
2991 rstreason = BANDLIM_RST_OPENPORT;
2995 if (so->so_options & SO_DEBUG)
2996 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2999 if (ti_locked == TI_WLOCKED)
3000 INP_INFO_WUNLOCK(&V_tcbinfo);
3001 ti_locked = TI_UNLOCKED;
3003 tp->t_flags |= TF_ACKNOW;
3004 (void) tcp_output(tp);
3005 INP_WUNLOCK(tp->t_inpcb);
3010 if (ti_locked == TI_WLOCKED)
3011 INP_INFO_WUNLOCK(&V_tcbinfo);
3012 ti_locked = TI_UNLOCKED;
3015 tcp_dropwithreset(m, th, tp, tlen, rstreason);
3016 INP_WUNLOCK(tp->t_inpcb);
3018 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
3022 if (ti_locked == TI_WLOCKED) {
3023 INP_INFO_WUNLOCK(&V_tcbinfo);
3024 ti_locked = TI_UNLOCKED;
3028 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
3032 * Drop space held by incoming segment and return.
3035 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
3036 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3040 INP_WUNLOCK(tp->t_inpcb);
3045 * Issue RST and make ACK acceptable to originator of segment.
3046 * The mbuf must still include the original packet header.
3050 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
3051 int tlen, int rstreason)
3057 struct ip6_hdr *ip6;
3061 INP_WLOCK_ASSERT(tp->t_inpcb);
3064 /* Don't bother if destination was broadcast/multicast. */
3065 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
3068 if (mtod(m, struct ip *)->ip_v == 6) {
3069 ip6 = mtod(m, struct ip6_hdr *);
3070 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3071 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3073 /* IPv6 anycast check is done at tcp6_input() */
3076 #if defined(INET) && defined(INET6)
3081 ip = mtod(m, struct ip *);
3082 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3083 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3084 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3085 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3090 /* Perform bandwidth limiting. */
3091 if (badport_bandlim(rstreason) < 0)
3094 /* tcp_respond consumes the mbuf chain. */
3095 if (th->th_flags & TH_ACK) {
3096 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0,
3097 th->th_ack, TH_RST);
3099 if (th->th_flags & TH_SYN)
3101 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
3102 (tcp_seq)0, TH_RST|TH_ACK);
3110 * Parse TCP options and place in tcpopt.
3113 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags)
3118 for (; cnt > 0; cnt -= optlen, cp += optlen) {
3120 if (opt == TCPOPT_EOL)
3122 if (opt == TCPOPT_NOP)
3128 if (optlen < 2 || optlen > cnt)
3133 if (optlen != TCPOLEN_MAXSEG)
3135 if (!(flags & TO_SYN))
3137 to->to_flags |= TOF_MSS;
3138 bcopy((char *)cp + 2,
3139 (char *)&to->to_mss, sizeof(to->to_mss));
3140 to->to_mss = ntohs(to->to_mss);
3143 if (optlen != TCPOLEN_WINDOW)
3145 if (!(flags & TO_SYN))
3147 to->to_flags |= TOF_SCALE;
3148 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT);
3150 case TCPOPT_TIMESTAMP:
3151 if (optlen != TCPOLEN_TIMESTAMP)
3153 to->to_flags |= TOF_TS;
3154 bcopy((char *)cp + 2,
3155 (char *)&to->to_tsval, sizeof(to->to_tsval));
3156 to->to_tsval = ntohl(to->to_tsval);
3157 bcopy((char *)cp + 6,
3158 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
3159 to->to_tsecr = ntohl(to->to_tsecr);
3161 #ifdef TCP_SIGNATURE
3163 * XXX In order to reply to a host which has set the
3164 * TCP_SIGNATURE option in its initial SYN, we have to
3165 * record the fact that the option was observed here
3166 * for the syncache code to perform the correct response.
3168 case TCPOPT_SIGNATURE:
3169 if (optlen != TCPOLEN_SIGNATURE)
3171 to->to_flags |= TOF_SIGNATURE;
3172 to->to_signature = cp + 2;
3175 case TCPOPT_SACK_PERMITTED:
3176 if (optlen != TCPOLEN_SACK_PERMITTED)
3178 if (!(flags & TO_SYN))
3182 to->to_flags |= TOF_SACKPERM;
3185 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
3189 to->to_flags |= TOF_SACK;
3190 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
3191 to->to_sacks = cp + 2;
3192 TCPSTAT_INC(tcps_sack_rcv_blocks);
3201 * Pull out of band byte out of a segment so
3202 * it doesn't appear in the user's data queue.
3203 * It is still reflected in the segment length for
3204 * sequencing purposes.
3207 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m,
3210 int cnt = off + th->th_urp - 1;
3213 if (m->m_len > cnt) {
3214 char *cp = mtod(m, caddr_t) + cnt;
3215 struct tcpcb *tp = sototcpcb(so);
3217 INP_WLOCK_ASSERT(tp->t_inpcb);
3220 tp->t_oobflags |= TCPOOB_HAVEDATA;
3221 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3223 if (m->m_flags & M_PKTHDR)
3232 panic("tcp_pulloutofband");
3236 * Collect new round-trip time estimate
3237 * and update averages and current timeout.
3240 tcp_xmit_timer(struct tcpcb *tp, int rtt)
3244 INP_WLOCK_ASSERT(tp->t_inpcb);
3246 TCPSTAT_INC(tcps_rttupdated);
3248 if (tp->t_srtt != 0) {
3250 * srtt is stored as fixed point with 5 bits after the
3251 * binary point (i.e., scaled by 8). The following magic
3252 * is equivalent to the smoothing algorithm in rfc793 with
3253 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3254 * point). Adjust rtt to origin 0.
3256 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
3257 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
3259 if ((tp->t_srtt += delta) <= 0)
3263 * We accumulate a smoothed rtt variance (actually, a
3264 * smoothed mean difference), then set the retransmit
3265 * timer to smoothed rtt + 4 times the smoothed variance.
3266 * rttvar is stored as fixed point with 4 bits after the
3267 * binary point (scaled by 16). The following is
3268 * equivalent to rfc793 smoothing with an alpha of .75
3269 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
3270 * rfc793's wired-in beta.
3274 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
3275 if ((tp->t_rttvar += delta) <= 0)
3277 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
3278 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3281 * No rtt measurement yet - use the unsmoothed rtt.
3282 * Set the variance to half the rtt (so our first
3283 * retransmit happens at 3*rtt).
3285 tp->t_srtt = rtt << TCP_RTT_SHIFT;
3286 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
3287 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3293 * the retransmit should happen at rtt + 4 * rttvar.
3294 * Because of the way we do the smoothing, srtt and rttvar
3295 * will each average +1/2 tick of bias. When we compute
3296 * the retransmit timer, we want 1/2 tick of rounding and
3297 * 1 extra tick because of +-1/2 tick uncertainty in the
3298 * firing of the timer. The bias will give us exactly the
3299 * 1.5 tick we need. But, because the bias is
3300 * statistical, we have to test that we don't drop below
3301 * the minimum feasible timer (which is 2 ticks).
3303 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3304 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3307 * We received an ack for a packet that wasn't retransmitted;
3308 * it is probably safe to discard any error indications we've
3309 * received recently. This isn't quite right, but close enough
3310 * for now (a route might have failed after we sent a segment,
3311 * and the return path might not be symmetrical).
3313 tp->t_softerror = 0;
3317 * Determine a reasonable value for maxseg size.
3318 * If the route is known, check route for mtu.
3319 * If none, use an mss that can be handled on the outgoing
3320 * interface without forcing IP to fragment; if bigger than
3321 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
3322 * to utilize large mbufs. If no route is found, route has no mtu,
3323 * or the destination isn't local, use a default, hopefully conservative
3324 * size (usually 512 or the default IP max size, but no more than the mtu
3325 * of the interface), as we can't discover anything about intervening
3326 * gateways or networks. We also initialize the congestion/slow start
3327 * window to be a single segment if the destination isn't local.
3328 * While looking at the routing entry, we also initialize other path-dependent
3329 * parameters from pre-set or cached values in the routing entry.
3331 * Also take into account the space needed for options that we
3332 * send regularly. Make maxseg shorter by that amount to assure
3333 * that we can send maxseg amount of data even when the options
3334 * are present. Store the upper limit of the length of options plus
3337 * In case of T/TCP, we call this routine during implicit connection
3338 * setup as well (offer = -1), to initialize maxseg from the cached
3341 * NOTE that this routine is only called when we process an incoming
3342 * segment. Outgoing SYN/ACK MSS settings are handled in tcp_mssopt().
3345 tcp_mss_update(struct tcpcb *tp, int offer,
3346 struct hc_metrics_lite *metricptr, int *mtuflags)
3350 struct inpcb *inp = tp->t_inpcb;
3351 struct hc_metrics_lite metrics;
3352 int origoffer = offer;
3354 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
3355 size_t min_protoh = isipv6 ?
3356 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
3357 sizeof (struct tcpiphdr);
3359 const size_t min_protoh = sizeof(struct tcpiphdr);
3362 INP_WLOCK_ASSERT(tp->t_inpcb);
3367 maxmtu = tcp_maxmtu6(&inp->inp_inc, mtuflags);
3368 tp->t_maxopd = tp->t_maxseg = V_tcp_v6mssdflt;
3371 #if defined(INET) && defined(INET6)
3376 maxmtu = tcp_maxmtu(&inp->inp_inc, mtuflags);
3377 tp->t_maxopd = tp->t_maxseg = V_tcp_mssdflt;
3382 * No route to sender, stay with default mss and return.
3386 * In case we return early we need to initialize metrics
3387 * to a defined state as tcp_hc_get() would do for us
3388 * if there was no cache hit.
3390 if (metricptr != NULL)
3391 bzero(metricptr, sizeof(struct hc_metrics_lite));
3395 /* What have we got? */
3399 * Offer == 0 means that there was no MSS on the SYN
3400 * segment, in this case we use tcp_mssdflt as
3401 * already assigned to t_maxopd above.
3403 offer = tp->t_maxopd;
3408 * Offer == -1 means that we didn't receive SYN yet.
3414 * Prevent DoS attack with too small MSS. Round up
3415 * to at least minmss.
3417 offer = max(offer, V_tcp_minmss);
3421 * rmx information is now retrieved from tcp_hostcache.
3423 tcp_hc_get(&inp->inp_inc, &metrics);
3424 if (metricptr != NULL)
3425 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite));
3428 * If there's a discovered mtu int tcp hostcache, use it
3429 * else, use the link mtu.
3431 if (metrics.rmx_mtu)
3432 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
3436 mss = maxmtu - min_protoh;
3437 if (!V_path_mtu_discovery &&
3438 !in6_localaddr(&inp->in6p_faddr))
3439 mss = min(mss, V_tcp_v6mssdflt);
3442 #if defined(INET) && defined(INET6)
3447 mss = maxmtu - min_protoh;
3448 if (!V_path_mtu_discovery &&
3449 !in_localaddr(inp->inp_faddr))
3450 mss = min(mss, V_tcp_mssdflt);
3454 * XXX - The above conditional (mss = maxmtu - min_protoh)
3455 * probably violates the TCP spec.
3456 * The problem is that, since we don't know the
3457 * other end's MSS, we are supposed to use a conservative
3458 * default. But, if we do that, then MTU discovery will
3459 * never actually take place, because the conservative
3460 * default is much less than the MTUs typically seen
3461 * on the Internet today. For the moment, we'll sweep
3462 * this under the carpet.
3464 * The conservative default might not actually be a problem
3465 * if the only case this occurs is when sending an initial
3466 * SYN with options and data to a host we've never talked
3467 * to before. Then, they will reply with an MSS value which
3468 * will get recorded and the new parameters should get
3469 * recomputed. For Further Study.
3472 mss = min(mss, offer);
3475 * Sanity check: make sure that maxopd will be large
3476 * enough to allow some data on segments even if the
3477 * all the option space is used (40bytes). Otherwise
3478 * funny things may happen in tcp_output.
3483 * maxopd stores the maximum length of data AND options
3484 * in a segment; maxseg is the amount of data in a normal
3485 * segment. We need to store this value (maxopd) apart
3486 * from maxseg, because now every segment carries options
3487 * and thus we normally have somewhat less data in segments.
3492 * origoffer==-1 indicates that no segments were received yet.
3493 * In this case we just guess.
3495 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
3497 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
3498 mss -= TCPOLEN_TSTAMP_APPA;
3500 #if (MCLBYTES & (MCLBYTES - 1)) == 0
3502 mss &= ~(MCLBYTES-1);
3505 mss = mss / MCLBYTES * MCLBYTES;
3511 tcp_mss(struct tcpcb *tp, int offer)
3517 struct hc_metrics_lite metrics;
3520 KASSERT(tp != NULL, ("%s: tp == NULL", __func__));
3522 tcp_mss_update(tp, offer, &metrics, &mtuflags);
3528 * If there's a pipesize, change the socket buffer to that size,
3529 * don't change if sb_hiwat is different than default (then it
3530 * has been changed on purpose with setsockopt).
3531 * Make the socket buffers an integral number of mss units;
3532 * if the mss is larger than the socket buffer, decrease the mss.
3534 so = inp->inp_socket;
3535 SOCKBUF_LOCK(&so->so_snd);
3536 if ((so->so_snd.sb_hiwat == tcp_sendspace) && metrics.rmx_sendpipe)
3537 bufsize = metrics.rmx_sendpipe;
3539 bufsize = so->so_snd.sb_hiwat;
3543 bufsize = roundup(bufsize, mss);
3544 if (bufsize > sb_max)
3546 if (bufsize > so->so_snd.sb_hiwat)
3547 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
3549 SOCKBUF_UNLOCK(&so->so_snd);
3552 SOCKBUF_LOCK(&so->so_rcv);
3553 if ((so->so_rcv.sb_hiwat == tcp_recvspace) && metrics.rmx_recvpipe)
3554 bufsize = metrics.rmx_recvpipe;
3556 bufsize = so->so_rcv.sb_hiwat;
3557 if (bufsize > mss) {
3558 bufsize = roundup(bufsize, mss);
3559 if (bufsize > sb_max)
3561 if (bufsize > so->so_rcv.sb_hiwat)
3562 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
3564 SOCKBUF_UNLOCK(&so->so_rcv);
3566 /* Check the interface for TSO capabilities. */
3567 if (mtuflags & CSUM_TSO)
3568 tp->t_flags |= TF_TSO;
3572 * Determine the MSS option to send on an outgoing SYN.
3575 tcp_mssopt(struct in_conninfo *inc)
3582 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3585 if (inc->inc_flags & INC_ISIPV6) {
3586 mss = V_tcp_v6mssdflt;
3587 maxmtu = tcp_maxmtu6(inc, NULL);
3588 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3589 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3592 #if defined(INET) && defined(INET6)
3597 mss = V_tcp_mssdflt;
3598 maxmtu = tcp_maxmtu(inc, NULL);
3599 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3600 min_protoh = sizeof(struct tcpiphdr);
3603 if (maxmtu && thcmtu)
3604 mss = min(maxmtu, thcmtu) - min_protoh;
3605 else if (maxmtu || thcmtu)
3606 mss = max(maxmtu, thcmtu) - min_protoh;
3613 * On a partial ack arrives, force the retransmission of the
3614 * next unacknowledged segment. Do not clear tp->t_dupacks.
3615 * By setting snd_nxt to ti_ack, this forces retransmission timer to
3619 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th)
3621 tcp_seq onxt = tp->snd_nxt;
3622 u_long ocwnd = tp->snd_cwnd;
3624 INP_WLOCK_ASSERT(tp->t_inpcb);
3626 tcp_timer_activate(tp, TT_REXMT, 0);
3628 tp->snd_nxt = th->th_ack;
3630 * Set snd_cwnd to one segment beyond acknowledged offset.
3631 * (tp->snd_una has not yet been updated when this function is called.)
3633 tp->snd_cwnd = tp->t_maxseg + BYTES_THIS_ACK(tp, th);
3634 tp->t_flags |= TF_ACKNOW;
3635 (void) tcp_output(tp);
3636 tp->snd_cwnd = ocwnd;
3637 if (SEQ_GT(onxt, tp->snd_nxt))
3640 * Partial window deflation. Relies on fact that tp->snd_una
3643 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th))
3644 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th);
3647 tp->snd_cwnd += tp->t_maxseg;