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);
547 ifa_free(&ia6->ia_ifa);
555 tcp_input(struct mbuf *m, int off0)
557 struct tcphdr *th = NULL;
558 struct ip *ip = NULL;
562 struct inpcb *inp = NULL;
563 struct tcpcb *tp = NULL;
564 struct socket *so = NULL;
573 int rstreason = 0; /* For badport_bandlim accounting purposes */
575 uint8_t sig_checked = 0;
578 #ifdef IPFIREWALL_FORWARD
579 struct m_tag *fwd_tag;
582 struct ip6_hdr *ip6 = NULL;
585 const void *ip6 = NULL;
587 struct tcpopt to; /* options in this segment */
588 char *s = NULL; /* address and port logging */
590 #define TI_UNLOCKED 1
595 * The size of tcp_saveipgen must be the size of the max ip header,
598 u_char tcp_saveipgen[IP6_HDR_LEN];
599 struct tcphdr tcp_savetcp;
604 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
608 TCPSTAT_INC(tcps_rcvtotal);
612 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */
613 ip6 = mtod(m, struct ip6_hdr *);
614 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
615 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) {
616 TCPSTAT_INC(tcps_rcvbadsum);
619 th = (struct tcphdr *)((caddr_t)ip6 + off0);
622 * Be proactive about unspecified IPv6 address in source.
623 * As we use all-zero to indicate unbounded/unconnected pcb,
624 * unspecified IPv6 address can be used to confuse us.
626 * Note that packets with unspecified IPv6 destination is
627 * already dropped in ip6_input.
629 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
635 #if defined(INET) && defined(INET6)
641 * Get IP and TCP header together in first mbuf.
642 * Note: IP leaves IP header in first mbuf.
644 if (off0 > sizeof (struct ip)) {
645 ip_stripoptions(m, (struct mbuf *)0);
646 off0 = sizeof(struct ip);
648 if (m->m_len < sizeof (struct tcpiphdr)) {
649 if ((m = m_pullup(m, sizeof (struct tcpiphdr)))
651 TCPSTAT_INC(tcps_rcvshort);
655 ip = mtod(m, struct ip *);
656 ipov = (struct ipovly *)ip;
657 th = (struct tcphdr *)((caddr_t)ip + off0);
660 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
661 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
662 th->th_sum = m->m_pkthdr.csum_data;
664 th->th_sum = in_pseudo(ip->ip_src.s_addr,
666 htonl(m->m_pkthdr.csum_data +
669 th->th_sum ^= 0xffff;
671 ipov->ih_len = (u_short)tlen;
672 ipov->ih_len = htons(ipov->ih_len);
676 * Checksum extended TCP header and data.
678 len = sizeof (struct ip) + tlen;
679 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
680 ipov->ih_len = (u_short)tlen;
681 ipov->ih_len = htons(ipov->ih_len);
682 th->th_sum = in_cksum(m, len);
685 TCPSTAT_INC(tcps_rcvbadsum);
688 /* Re-initialization for later version check */
689 ip->ip_v = IPVERSION;
695 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
697 #if defined(INET) && defined(INET6)
705 * Check that TCP offset makes sense,
706 * pull out TCP options and adjust length. XXX
708 off = th->th_off << 2;
709 if (off < sizeof (struct tcphdr) || off > tlen) {
710 TCPSTAT_INC(tcps_rcvbadoff);
713 tlen -= off; /* tlen is used instead of ti->ti_len */
714 if (off > sizeof (struct tcphdr)) {
717 IP6_EXTHDR_CHECK(m, off0, off, );
718 ip6 = mtod(m, struct ip6_hdr *);
719 th = (struct tcphdr *)((caddr_t)ip6 + off0);
722 #if defined(INET) && defined(INET6)
727 if (m->m_len < sizeof(struct ip) + off) {
728 if ((m = m_pullup(m, sizeof (struct ip) + off))
730 TCPSTAT_INC(tcps_rcvshort);
733 ip = mtod(m, struct ip *);
734 ipov = (struct ipovly *)ip;
735 th = (struct tcphdr *)((caddr_t)ip + off0);
739 optlen = off - sizeof (struct tcphdr);
740 optp = (u_char *)(th + 1);
742 thflags = th->th_flags;
745 * Convert TCP protocol specific fields to host format.
747 tcp_fields_to_host(th);
750 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options.
752 drop_hdrlen = off0 + off;
755 * Locate pcb for segment; if we're likely to add or remove a
756 * connection then first acquire pcbinfo lock. There are two cases
757 * where we might discover later we need a write lock despite the
758 * flags: ACKs moving a connection out of the syncache, and ACKs for
759 * a connection in TIMEWAIT.
761 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0) {
762 INP_INFO_WLOCK(&V_tcbinfo);
763 ti_locked = TI_WLOCKED;
765 ti_locked = TI_UNLOCKED;
769 if (ti_locked == TI_WLOCKED) {
770 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
772 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
776 #ifdef IPFIREWALL_FORWARD
778 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
780 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
781 #endif /* IPFIREWALL_FORWARD */
784 #ifdef IPFIREWALL_FORWARD
785 if (isipv6 && fwd_tag != NULL) {
786 struct sockaddr_in6 *next_hop6;
788 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1);
790 * Transparently forwarded. Pretend to be the destination.
791 * Already got one like this?
793 inp = in6_pcblookup_mbuf(&V_tcbinfo,
794 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport,
795 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m);
798 * It's new. Try to find the ambushing socket.
799 * Because we've rewritten the destination address,
800 * any hardware-generated hash is ignored.
802 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src,
803 th->th_sport, &next_hop6->sin6_addr,
804 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) :
805 th->th_dport, INPLOOKUP_WILDCARD |
806 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
808 /* Remove the tag from the packet. We don't need it anymore. */
809 m_tag_delete(m, fwd_tag);
811 #endif /* IPFIREWALL_FORWARD */
813 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src,
814 th->th_sport, &ip6->ip6_dst, th->th_dport,
815 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
816 m->m_pkthdr.rcvif, m);
819 #if defined(INET6) && defined(INET)
823 #ifdef IPFIREWALL_FORWARD
824 if (fwd_tag != NULL) {
825 struct sockaddr_in *next_hop;
827 next_hop = (struct sockaddr_in *)(fwd_tag+1);
829 * Transparently forwarded. Pretend to be the destination.
830 * already got one like this?
832 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport,
833 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB,
834 m->m_pkthdr.rcvif, m);
837 * It's new. Try to find the ambushing socket.
838 * Because we've rewritten the destination address,
839 * any hardware-generated hash is ignored.
841 inp = in_pcblookup(&V_tcbinfo, ip->ip_src,
842 th->th_sport, next_hop->sin_addr,
843 next_hop->sin_port ? ntohs(next_hop->sin_port) :
844 th->th_dport, INPLOOKUP_WILDCARD |
845 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
847 /* Remove the tag from the packet. We don't need it anymore. */
848 m_tag_delete(m, fwd_tag);
850 #endif /* IPFIREWALL_FORWARD */
851 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src,
852 th->th_sport, ip->ip_dst, th->th_dport,
853 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
854 m->m_pkthdr.rcvif, m);
858 * If the INPCB does not exist then all data in the incoming
859 * segment is discarded and an appropriate RST is sent back.
860 * XXX MRT Send RST using which routing table?
864 * Log communication attempts to ports that are not
867 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) ||
868 tcp_log_in_vain == 2) {
869 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6)))
870 log(LOG_INFO, "%s; %s: Connection attempt "
871 "to closed port\n", s, __func__);
874 * When blackholing do not respond with a RST but
875 * completely ignore the segment and drop it.
877 if ((V_blackhole == 1 && (thflags & TH_SYN)) ||
881 rstreason = BANDLIM_RST_CLOSEDPORT;
884 INP_WLOCK_ASSERT(inp);
885 if (!(inp->inp_flags & INP_HW_FLOWID)
886 && (m->m_flags & M_FLOWID)
887 && ((inp->inp_socket == NULL)
888 || !(inp->inp_socket->so_options & SO_ACCEPTCONN))) {
889 inp->inp_flags |= INP_HW_FLOWID;
890 inp->inp_flags &= ~INP_SW_FLOWID;
891 inp->inp_flowid = m->m_pkthdr.flowid;
895 if (isipv6 && ipsec6_in_reject(m, inp)) {
896 V_ipsec6stat.in_polvio++;
900 if (ipsec4_in_reject(m, inp) != 0) {
901 V_ipsec4stat.in_polvio++;
907 * Check the minimum TTL for socket.
909 if (inp->inp_ip_minttl != 0) {
911 if (isipv6 && inp->inp_ip_minttl > ip6->ip6_hlim)
915 if (inp->inp_ip_minttl > ip->ip_ttl)
920 * A previous connection in TIMEWAIT state is supposed to catch stray
921 * or duplicate segments arriving late. If this segment was a
922 * legitimate new connection attempt the old INPCB gets removed and
923 * we can try again to find a listening socket.
925 * At this point, due to earlier optimism, we may hold only an inpcb
926 * lock, and not the inpcbinfo write lock. If so, we need to try to
927 * acquire it, or if that fails, acquire a reference on the inpcb,
928 * drop all locks, acquire a global write lock, and then re-acquire
929 * the inpcb lock. We may at that point discover that another thread
930 * has tried to free the inpcb, in which case we need to loop back
931 * and try to find a new inpcb to deliver to.
933 * XXXRW: It may be time to rethink timewait locking.
936 if (inp->inp_flags & INP_TIMEWAIT) {
937 if (ti_locked == TI_UNLOCKED) {
938 if (INP_INFO_TRY_WLOCK(&V_tcbinfo) == 0) {
941 INP_INFO_WLOCK(&V_tcbinfo);
942 ti_locked = TI_WLOCKED;
944 if (in_pcbrele_wlocked(inp)) {
949 ti_locked = TI_WLOCKED;
951 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
953 if (thflags & TH_SYN)
954 tcp_dooptions(&to, optp, optlen, TO_SYN);
956 * NB: tcp_twcheck unlocks the INP and frees the mbuf.
958 if (tcp_twcheck(inp, &to, th, m, tlen))
960 INP_INFO_WUNLOCK(&V_tcbinfo);
964 * The TCPCB may no longer exist if the connection is winding
965 * down or it is in the CLOSED state. Either way we drop the
966 * segment and send an appropriate response.
969 if (tp == NULL || tp->t_state == TCPS_CLOSED) {
970 rstreason = BANDLIM_RST_CLOSEDPORT;
975 * We've identified a valid inpcb, but it could be that we need an
976 * inpcbinfo write lock but don't hold it. In this case, attempt to
977 * acquire using the same strategy as the TIMEWAIT case above. If we
978 * relock, we have to jump back to 'relocked' as the connection might
979 * now be in TIMEWAIT.
982 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0)
983 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
985 if (tp->t_state != TCPS_ESTABLISHED) {
986 if (ti_locked == TI_UNLOCKED) {
987 if (INP_INFO_TRY_WLOCK(&V_tcbinfo) == 0) {
990 INP_INFO_WLOCK(&V_tcbinfo);
991 ti_locked = TI_WLOCKED;
993 if (in_pcbrele_wlocked(inp)) {
999 ti_locked = TI_WLOCKED;
1001 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1005 INP_WLOCK_ASSERT(inp);
1006 if (mac_inpcb_check_deliver(inp, m))
1009 so = inp->inp_socket;
1010 KASSERT(so != NULL, ("%s: so == NULL", __func__));
1012 if (so->so_options & SO_DEBUG) {
1013 ostate = tp->t_state;
1016 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
1019 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
1022 #endif /* TCPDEBUG */
1024 * When the socket is accepting connections (the INPCB is in LISTEN
1025 * state) we look into the SYN cache if this is a new connection
1026 * attempt or the completion of a previous one. Because listen
1027 * sockets are never in TCPS_ESTABLISHED, the V_tcbinfo lock will be
1028 * held in this case.
1030 if (so->so_options & SO_ACCEPTCONN) {
1031 struct in_conninfo inc;
1033 KASSERT(tp->t_state == TCPS_LISTEN, ("%s: so accepting but "
1034 "tp not listening", __func__));
1035 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1037 bzero(&inc, sizeof(inc));
1040 inc.inc_flags |= INC_ISIPV6;
1041 inc.inc6_faddr = ip6->ip6_src;
1042 inc.inc6_laddr = ip6->ip6_dst;
1046 inc.inc_faddr = ip->ip_src;
1047 inc.inc_laddr = ip->ip_dst;
1049 inc.inc_fport = th->th_sport;
1050 inc.inc_lport = th->th_dport;
1051 inc.inc_fibnum = so->so_fibnum;
1054 * Check for an existing connection attempt in syncache if
1055 * the flag is only ACK. A successful lookup creates a new
1056 * socket appended to the listen queue in SYN_RECEIVED state.
1058 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
1060 * Parse the TCP options here because
1061 * syncookies need access to the reflected
1064 tcp_dooptions(&to, optp, optlen, 0);
1066 * NB: syncache_expand() doesn't unlock
1067 * inp and tcpinfo locks.
1069 if (!syncache_expand(&inc, &to, th, &so, m)) {
1071 * No syncache entry or ACK was not
1072 * for our SYN/ACK. Send a RST.
1073 * NB: syncache did its own logging
1074 * of the failure cause.
1076 rstreason = BANDLIM_RST_OPENPORT;
1081 * We completed the 3-way handshake
1082 * but could not allocate a socket
1083 * either due to memory shortage,
1084 * listen queue length limits or
1085 * global socket limits. Send RST
1086 * or wait and have the remote end
1087 * retransmit the ACK for another
1090 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1091 log(LOG_DEBUG, "%s; %s: Listen socket: "
1092 "Socket allocation failed due to "
1093 "limits or memory shortage, %s\n",
1095 V_tcp_sc_rst_sock_fail ?
1096 "sending RST" : "try again");
1097 if (V_tcp_sc_rst_sock_fail) {
1098 rstreason = BANDLIM_UNLIMITED;
1104 * Socket is created in state SYN_RECEIVED.
1105 * Unlock the listen socket, lock the newly
1106 * created socket and update the tp variable.
1108 INP_WUNLOCK(inp); /* listen socket */
1109 inp = sotoinpcb(so);
1110 INP_WLOCK(inp); /* new connection */
1111 tp = intotcpcb(inp);
1112 KASSERT(tp->t_state == TCPS_SYN_RECEIVED,
1113 ("%s: ", __func__));
1114 #ifdef TCP_SIGNATURE
1115 if (sig_checked == 0) {
1116 tcp_dooptions(&to, optp, optlen,
1117 (thflags & TH_SYN) ? TO_SYN : 0);
1118 if (!tcp_signature_verify_input(m, off0, tlen,
1119 optlen, &to, th, tp->t_flags)) {
1122 * In SYN_SENT state if it receives an
1123 * RST, it is allowed for further
1126 if ((thflags & TH_RST) == 0 ||
1127 (tp->t_state == TCPS_SYN_SENT) == 0)
1135 * Process the segment and the data it
1136 * contains. tcp_do_segment() consumes
1137 * the mbuf chain and unlocks the inpcb.
1139 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen,
1141 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1145 * Segment flag validation for new connection attempts:
1147 * Our (SYN|ACK) response was rejected.
1148 * Check with syncache and remove entry to prevent
1151 * NB: syncache_chkrst does its own logging of failure
1154 if (thflags & TH_RST) {
1155 syncache_chkrst(&inc, th);
1159 * We can't do anything without SYN.
1161 if ((thflags & TH_SYN) == 0) {
1162 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1163 log(LOG_DEBUG, "%s; %s: Listen socket: "
1164 "SYN is missing, segment ignored\n",
1166 TCPSTAT_INC(tcps_badsyn);
1170 * (SYN|ACK) is bogus on a listen socket.
1172 if (thflags & TH_ACK) {
1173 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1174 log(LOG_DEBUG, "%s; %s: Listen socket: "
1175 "SYN|ACK invalid, segment rejected\n",
1177 syncache_badack(&inc); /* XXX: Not needed! */
1178 TCPSTAT_INC(tcps_badsyn);
1179 rstreason = BANDLIM_RST_OPENPORT;
1183 * If the drop_synfin option is enabled, drop all
1184 * segments with both the SYN and FIN bits set.
1185 * This prevents e.g. nmap from identifying the
1187 * XXX: Poor reasoning. nmap has other methods
1188 * and is constantly refining its stack detection
1190 * XXX: This is a violation of the TCP specification
1191 * and was used by RFC1644.
1193 if ((thflags & TH_FIN) && V_drop_synfin) {
1194 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1195 log(LOG_DEBUG, "%s; %s: Listen socket: "
1196 "SYN|FIN segment ignored (based on "
1197 "sysctl setting)\n", s, __func__);
1198 TCPSTAT_INC(tcps_badsyn);
1202 * Segment's flags are (SYN) or (SYN|FIN).
1204 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored
1205 * as they do not affect the state of the TCP FSM.
1206 * The data pointed to by TH_URG and th_urp is ignored.
1208 KASSERT((thflags & (TH_RST|TH_ACK)) == 0,
1209 ("%s: Listen socket: TH_RST or TH_ACK set", __func__));
1210 KASSERT(thflags & (TH_SYN),
1211 ("%s: Listen socket: TH_SYN not set", __func__));
1214 * If deprecated address is forbidden,
1215 * we do not accept SYN to deprecated interface
1216 * address to prevent any new inbound connection from
1217 * getting established.
1218 * When we do not accept SYN, we send a TCP RST,
1219 * with deprecated source address (instead of dropping
1220 * it). We compromise it as it is much better for peer
1221 * to send a RST, and RST will be the final packet
1224 * If we do not forbid deprecated addresses, we accept
1225 * the SYN packet. RFC2462 does not suggest dropping
1227 * If we decipher RFC2462 5.5.4, it says like this:
1228 * 1. use of deprecated addr with existing
1229 * communication is okay - "SHOULD continue to be
1231 * 2. use of it with new communication:
1232 * (2a) "SHOULD NOT be used if alternate address
1233 * with sufficient scope is available"
1234 * (2b) nothing mentioned otherwise.
1235 * Here we fall into (2b) case as we have no choice in
1236 * our source address selection - we must obey the peer.
1238 * The wording in RFC2462 is confusing, and there are
1239 * multiple description text for deprecated address
1240 * handling - worse, they are not exactly the same.
1241 * I believe 5.5.4 is the best one, so we follow 5.5.4.
1243 if (isipv6 && !V_ip6_use_deprecated) {
1244 struct in6_ifaddr *ia6;
1246 ia6 = ip6_getdstifaddr(m);
1248 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1249 ifa_free(&ia6->ia_ifa);
1250 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1251 log(LOG_DEBUG, "%s; %s: Listen socket: "
1252 "Connection attempt to deprecated "
1253 "IPv6 address rejected\n",
1255 rstreason = BANDLIM_RST_OPENPORT;
1259 ifa_free(&ia6->ia_ifa);
1263 * Basic sanity checks on incoming SYN requests:
1264 * Don't respond if the destination is a link layer
1265 * broadcast according to RFC1122 4.2.3.10, p. 104.
1266 * If it is from this socket it must be forged.
1267 * Don't respond if the source or destination is a
1268 * global or subnet broad- or multicast address.
1269 * Note that it is quite possible to receive unicast
1270 * link-layer packets with a broadcast IP address. Use
1271 * in_broadcast() to find them.
1273 if (m->m_flags & (M_BCAST|M_MCAST)) {
1274 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1275 log(LOG_DEBUG, "%s; %s: Listen socket: "
1276 "Connection attempt from broad- or multicast "
1277 "link layer address ignored\n", s, __func__);
1282 if (th->th_dport == th->th_sport &&
1283 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) {
1284 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1285 log(LOG_DEBUG, "%s; %s: Listen socket: "
1286 "Connection attempt to/from self "
1287 "ignored\n", s, __func__);
1290 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1291 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) {
1292 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1293 log(LOG_DEBUG, "%s; %s: Listen socket: "
1294 "Connection attempt from/to multicast "
1295 "address ignored\n", s, __func__);
1300 #if defined(INET) && defined(INET6)
1305 if (th->th_dport == th->th_sport &&
1306 ip->ip_dst.s_addr == ip->ip_src.s_addr) {
1307 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1308 log(LOG_DEBUG, "%s; %s: Listen socket: "
1309 "Connection attempt from/to self "
1310 "ignored\n", s, __func__);
1313 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1314 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1315 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1316 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
1317 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1318 log(LOG_DEBUG, "%s; %s: Listen socket: "
1319 "Connection attempt from/to broad- "
1320 "or multicast address ignored\n",
1327 * SYN appears to be valid. Create compressed TCP state
1331 if (so->so_options & SO_DEBUG)
1332 tcp_trace(TA_INPUT, ostate, tp,
1333 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1335 tcp_dooptions(&to, optp, optlen, TO_SYN);
1336 syncache_add(&inc, &to, th, inp, &so, m);
1338 * Entry added to syncache and mbuf consumed.
1339 * Everything already unlocked by syncache_add().
1341 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1345 #ifdef TCP_SIGNATURE
1346 if (sig_checked == 0) {
1347 tcp_dooptions(&to, optp, optlen,
1348 (thflags & TH_SYN) ? TO_SYN : 0);
1349 if (!tcp_signature_verify_input(m, off0, tlen, optlen, &to,
1353 * In SYN_SENT state if it receives an RST, it is
1354 * allowed for further processing.
1356 if ((thflags & TH_RST) == 0 ||
1357 (tp->t_state == TCPS_SYN_SENT) == 0)
1365 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later
1366 * state. tcp_do_segment() always consumes the mbuf chain, unlocks
1367 * the inpcb, and unlocks pcbinfo.
1369 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos, ti_locked);
1370 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1374 if (ti_locked == TI_WLOCKED) {
1375 INP_INFO_WUNLOCK(&V_tcbinfo);
1376 ti_locked = TI_UNLOCKED;
1380 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset "
1381 "ti_locked: %d", __func__, ti_locked));
1382 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1387 tcp_dropwithreset(m, th, tp, tlen, rstreason);
1390 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
1391 m = NULL; /* mbuf chain got consumed. */
1395 if (ti_locked == TI_WLOCKED) {
1396 INP_INFO_WUNLOCK(&V_tcbinfo);
1397 ti_locked = TI_UNLOCKED;
1401 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock "
1402 "ti_locked: %d", __func__, ti_locked));
1403 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1411 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1419 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so,
1420 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos,
1423 int thflags, acked, ourfinisacked, needoutput = 0;
1424 int rstreason, todrop, win;
1430 * The size of tcp_saveipgen must be the size of the max ip header,
1433 u_char tcp_saveipgen[IP6_HDR_LEN];
1434 struct tcphdr tcp_savetcp;
1437 thflags = th->th_flags;
1438 tp->sackhint.last_sack_ack = 0;
1441 * If this is either a state-changing packet or current state isn't
1442 * established, we require a write lock on tcbinfo. Otherwise, we
1443 * allow either a read lock or a write lock, as we may have acquired
1444 * a write lock due to a race.
1446 * Require a global write lock for SYN/FIN/RST segments or
1447 * non-established connections; otherwise accept either a read or
1448 * write lock, as we may have conservatively acquired a write lock in
1449 * certain cases in tcp_input() (is this still true?). Currently we
1450 * will never enter with no lock, so we try to drop it quickly in the
1451 * common pure ack/pure data cases.
1453 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 ||
1454 tp->t_state != TCPS_ESTABLISHED) {
1455 KASSERT(ti_locked == TI_WLOCKED, ("%s ti_locked %d for "
1456 "SYN/FIN/RST/!EST", __func__, ti_locked));
1457 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1460 if (ti_locked == TI_WLOCKED)
1461 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1463 KASSERT(ti_locked == TI_UNLOCKED, ("%s: EST "
1464 "ti_locked: %d", __func__, ti_locked));
1465 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1469 INP_WLOCK_ASSERT(tp->t_inpcb);
1470 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN",
1472 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT",
1476 * Segment received on connection.
1477 * Reset idle time and keep-alive timer.
1478 * XXX: This should be done after segment
1479 * validation to ignore broken/spoofed segs.
1481 tp->t_rcvtime = ticks;
1482 if (TCPS_HAVEESTABLISHED(tp->t_state))
1483 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
1486 * Unscale the window into a 32-bit value.
1487 * For the SYN_SENT state the scale is zero.
1489 tiwin = th->th_win << tp->snd_scale;
1492 * TCP ECN processing.
1494 if (tp->t_flags & TF_ECN_PERMIT) {
1495 if (thflags & TH_CWR)
1496 tp->t_flags &= ~TF_ECN_SND_ECE;
1497 switch (iptos & IPTOS_ECN_MASK) {
1499 tp->t_flags |= TF_ECN_SND_ECE;
1500 TCPSTAT_INC(tcps_ecn_ce);
1502 case IPTOS_ECN_ECT0:
1503 TCPSTAT_INC(tcps_ecn_ect0);
1505 case IPTOS_ECN_ECT1:
1506 TCPSTAT_INC(tcps_ecn_ect1);
1509 /* Congestion experienced. */
1510 if (thflags & TH_ECE) {
1511 cc_cong_signal(tp, th, CC_ECN);
1516 * Parse options on any incoming segment.
1518 tcp_dooptions(&to, (u_char *)(th + 1),
1519 (th->th_off << 2) - sizeof(struct tcphdr),
1520 (thflags & TH_SYN) ? TO_SYN : 0);
1523 * If echoed timestamp is later than the current time,
1524 * fall back to non RFC1323 RTT calculation. Normalize
1525 * timestamp if syncookies were used when this connection
1528 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1529 to.to_tsecr -= tp->ts_offset;
1530 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks()))
1535 * Process options only when we get SYN/ACK back. The SYN case
1536 * for incoming connections is handled in tcp_syncache.
1537 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
1538 * or <SYN,ACK>) segment itself is never scaled.
1539 * XXX this is traditional behavior, may need to be cleaned up.
1541 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1542 if ((to.to_flags & TOF_SCALE) &&
1543 (tp->t_flags & TF_REQ_SCALE)) {
1544 tp->t_flags |= TF_RCVD_SCALE;
1545 tp->snd_scale = to.to_wscale;
1548 * Initial send window. It will be updated with
1549 * the next incoming segment to the scaled value.
1551 tp->snd_wnd = th->th_win;
1552 if (to.to_flags & TOF_TS) {
1553 tp->t_flags |= TF_RCVD_TSTMP;
1554 tp->ts_recent = to.to_tsval;
1555 tp->ts_recent_age = tcp_ts_getticks();
1557 if (to.to_flags & TOF_MSS)
1558 tcp_mss(tp, to.to_mss);
1559 if ((tp->t_flags & TF_SACK_PERMIT) &&
1560 (to.to_flags & TOF_SACKPERM) == 0)
1561 tp->t_flags &= ~TF_SACK_PERMIT;
1565 * Header prediction: check for the two common cases
1566 * of a uni-directional data xfer. If the packet has
1567 * no control flags, is in-sequence, the window didn't
1568 * change and we're not retransmitting, it's a
1569 * candidate. If the length is zero and the ack moved
1570 * forward, we're the sender side of the xfer. Just
1571 * free the data acked & wake any higher level process
1572 * that was blocked waiting for space. If the length
1573 * is non-zero and the ack didn't move, we're the
1574 * receiver side. If we're getting packets in-order
1575 * (the reassembly queue is empty), add the data to
1576 * the socket buffer and note that we need a delayed ack.
1577 * Make sure that the hidden state-flags are also off.
1578 * Since we check for TCPS_ESTABLISHED first, it can only
1581 if (tp->t_state == TCPS_ESTABLISHED &&
1582 th->th_seq == tp->rcv_nxt &&
1583 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1584 tp->snd_nxt == tp->snd_max &&
1585 tiwin && tiwin == tp->snd_wnd &&
1586 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1587 LIST_EMPTY(&tp->t_segq) &&
1588 ((to.to_flags & TOF_TS) == 0 ||
1589 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) {
1592 * If last ACK falls within this segment's sequence numbers,
1593 * record the timestamp.
1594 * NOTE that the test is modified according to the latest
1595 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1597 if ((to.to_flags & TOF_TS) != 0 &&
1598 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1599 tp->ts_recent_age = tcp_ts_getticks();
1600 tp->ts_recent = to.to_tsval;
1604 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1605 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1606 !IN_RECOVERY(tp->t_flags) &&
1607 (to.to_flags & TOF_SACK) == 0 &&
1608 TAILQ_EMPTY(&tp->snd_holes)) {
1610 * This is a pure ack for outstanding data.
1612 if (ti_locked == TI_WLOCKED)
1613 INP_INFO_WUNLOCK(&V_tcbinfo);
1614 ti_locked = TI_UNLOCKED;
1616 TCPSTAT_INC(tcps_predack);
1619 * "bad retransmit" recovery.
1621 if (tp->t_rxtshift == 1 &&
1622 tp->t_flags & TF_PREVVALID &&
1623 (int)(ticks - tp->t_badrxtwin) < 0) {
1624 cc_cong_signal(tp, th, CC_RTO_ERR);
1628 * Recalculate the transmit timer / rtt.
1630 * Some boxes send broken timestamp replies
1631 * during the SYN+ACK phase, ignore
1632 * timestamps of 0 or we could calculate a
1633 * huge RTT and blow up the retransmit timer.
1635 if ((to.to_flags & TOF_TS) != 0 &&
1639 t = tcp_ts_getticks() - to.to_tsecr;
1640 if (!tp->t_rttlow || tp->t_rttlow > t)
1643 TCP_TS_TO_TICKS(t) + 1);
1644 } else if (tp->t_rtttime &&
1645 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1646 if (!tp->t_rttlow ||
1647 tp->t_rttlow > ticks - tp->t_rtttime)
1648 tp->t_rttlow = ticks - tp->t_rtttime;
1650 ticks - tp->t_rtttime);
1652 acked = BYTES_THIS_ACK(tp, th);
1654 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
1655 hhook_run_tcp_est_in(tp, th, &to);
1657 TCPSTAT_INC(tcps_rcvackpack);
1658 TCPSTAT_ADD(tcps_rcvackbyte, acked);
1659 sbdrop(&so->so_snd, acked);
1660 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1661 SEQ_LEQ(th->th_ack, tp->snd_recover))
1662 tp->snd_recover = th->th_ack - 1;
1665 * Let the congestion control algorithm update
1666 * congestion control related information. This
1667 * typically means increasing the congestion
1670 cc_ack_received(tp, th, CC_ACK);
1672 tp->snd_una = th->th_ack;
1674 * Pull snd_wl2 up to prevent seq wrap relative
1677 tp->snd_wl2 = th->th_ack;
1680 ND6_HINT(tp); /* Some progress has been made. */
1683 * If all outstanding data are acked, stop
1684 * retransmit timer, otherwise restart timer
1685 * using current (possibly backed-off) value.
1686 * If process is waiting for space,
1687 * wakeup/selwakeup/signal. If data
1688 * are ready to send, let tcp_output
1689 * decide between more output or persist.
1692 if (so->so_options & SO_DEBUG)
1693 tcp_trace(TA_INPUT, ostate, tp,
1694 (void *)tcp_saveipgen,
1697 if (tp->snd_una == tp->snd_max)
1698 tcp_timer_activate(tp, TT_REXMT, 0);
1699 else if (!tcp_timer_active(tp, TT_PERSIST))
1700 tcp_timer_activate(tp, TT_REXMT,
1703 if (so->so_snd.sb_cc)
1704 (void) tcp_output(tp);
1707 } else if (th->th_ack == tp->snd_una &&
1708 tlen <= sbspace(&so->so_rcv)) {
1709 int newsize = 0; /* automatic sockbuf scaling */
1712 * This is a pure, in-sequence data packet with
1713 * nothing on the reassembly queue and we have enough
1714 * buffer space to take it.
1716 if (ti_locked == TI_WLOCKED)
1717 INP_INFO_WUNLOCK(&V_tcbinfo);
1718 ti_locked = TI_UNLOCKED;
1720 /* Clean receiver SACK report if present */
1721 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks)
1722 tcp_clean_sackreport(tp);
1723 TCPSTAT_INC(tcps_preddat);
1724 tp->rcv_nxt += tlen;
1726 * Pull snd_wl1 up to prevent seq wrap relative to
1729 tp->snd_wl1 = th->th_seq;
1731 * Pull rcv_up up to prevent seq wrap relative to
1734 tp->rcv_up = tp->rcv_nxt;
1735 TCPSTAT_INC(tcps_rcvpack);
1736 TCPSTAT_ADD(tcps_rcvbyte, tlen);
1737 ND6_HINT(tp); /* Some progress has been made */
1739 if (so->so_options & SO_DEBUG)
1740 tcp_trace(TA_INPUT, ostate, tp,
1741 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1744 * Automatic sizing of receive socket buffer. Often the send
1745 * buffer size is not optimally adjusted to the actual network
1746 * conditions at hand (delay bandwidth product). Setting the
1747 * buffer size too small limits throughput on links with high
1748 * bandwidth and high delay (eg. trans-continental/oceanic links).
1750 * On the receive side the socket buffer memory is only rarely
1751 * used to any significant extent. This allows us to be much
1752 * more aggressive in scaling the receive socket buffer. For
1753 * the case that the buffer space is actually used to a large
1754 * extent and we run out of kernel memory we can simply drop
1755 * the new segments; TCP on the sender will just retransmit it
1756 * later. Setting the buffer size too big may only consume too
1757 * much kernel memory if the application doesn't read() from
1758 * the socket or packet loss or reordering makes use of the
1761 * The criteria to step up the receive buffer one notch are:
1762 * 1. the number of bytes received during the time it takes
1763 * one timestamp to be reflected back to us (the RTT);
1764 * 2. received bytes per RTT is within seven eighth of the
1765 * current socket buffer size;
1766 * 3. receive buffer size has not hit maximal automatic size;
1768 * This algorithm does one step per RTT at most and only if
1769 * we receive a bulk stream w/o packet losses or reorderings.
1770 * Shrinking the buffer during idle times is not necessary as
1771 * it doesn't consume any memory when idle.
1773 * TODO: Only step up if the application is actually serving
1774 * the buffer to better manage the socket buffer resources.
1776 if (V_tcp_do_autorcvbuf &&
1778 (so->so_rcv.sb_flags & SB_AUTOSIZE)) {
1779 if (TSTMP_GT(to.to_tsecr, tp->rfbuf_ts) &&
1780 to.to_tsecr - tp->rfbuf_ts < hz) {
1782 (so->so_rcv.sb_hiwat / 8 * 7) &&
1783 so->so_rcv.sb_hiwat <
1784 V_tcp_autorcvbuf_max) {
1786 min(so->so_rcv.sb_hiwat +
1787 V_tcp_autorcvbuf_inc,
1788 V_tcp_autorcvbuf_max);
1790 /* Start over with next RTT. */
1794 tp->rfbuf_cnt += tlen; /* add up */
1797 /* Add data to socket buffer. */
1798 SOCKBUF_LOCK(&so->so_rcv);
1799 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1803 * Set new socket buffer size.
1804 * Give up when limit is reached.
1807 if (!sbreserve_locked(&so->so_rcv,
1809 so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1810 m_adj(m, drop_hdrlen); /* delayed header drop */
1811 sbappendstream_locked(&so->so_rcv, m);
1813 /* NB: sorwakeup_locked() does an implicit unlock. */
1814 sorwakeup_locked(so);
1815 if (DELAY_ACK(tp)) {
1816 tp->t_flags |= TF_DELACK;
1818 tp->t_flags |= TF_ACKNOW;
1826 * Calculate amount of space in receive window,
1827 * and then do TCP input processing.
1828 * Receive window is amount of space in rcv queue,
1829 * but not less than advertised window.
1831 win = sbspace(&so->so_rcv);
1834 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1836 /* Reset receive buffer auto scaling when not in bulk receive mode. */
1840 switch (tp->t_state) {
1843 * If the state is SYN_RECEIVED:
1844 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1846 case TCPS_SYN_RECEIVED:
1847 if ((thflags & TH_ACK) &&
1848 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1849 SEQ_GT(th->th_ack, tp->snd_max))) {
1850 rstreason = BANDLIM_RST_OPENPORT;
1856 * If the state is SYN_SENT:
1857 * if seg contains an ACK, but not for our SYN, drop the input.
1858 * if seg contains a RST, then drop the connection.
1859 * if seg does not contain SYN, then drop it.
1860 * Otherwise this is an acceptable SYN segment
1861 * initialize tp->rcv_nxt and tp->irs
1862 * if seg contains ack then advance tp->snd_una
1863 * if seg contains an ECE and ECN support is enabled, the stream
1865 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1866 * arrange for segment to be acked (eventually)
1867 * continue processing rest of data/controls, beginning with URG
1870 if ((thflags & TH_ACK) &&
1871 (SEQ_LEQ(th->th_ack, tp->iss) ||
1872 SEQ_GT(th->th_ack, tp->snd_max))) {
1873 rstreason = BANDLIM_UNLIMITED;
1876 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST))
1877 tp = tcp_drop(tp, ECONNREFUSED);
1878 if (thflags & TH_RST)
1880 if (!(thflags & TH_SYN))
1883 tp->irs = th->th_seq;
1885 if (thflags & TH_ACK) {
1886 TCPSTAT_INC(tcps_connects);
1889 mac_socketpeer_set_from_mbuf(m, so);
1891 /* Do window scaling on this connection? */
1892 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1893 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1894 tp->rcv_scale = tp->request_r_scale;
1896 tp->rcv_adv += imin(tp->rcv_wnd,
1897 TCP_MAXWIN << tp->rcv_scale);
1898 tp->snd_una++; /* SYN is acked */
1900 * If there's data, delay ACK; if there's also a FIN
1901 * ACKNOW will be turned on later.
1903 if (DELAY_ACK(tp) && tlen != 0)
1904 tcp_timer_activate(tp, TT_DELACK,
1907 tp->t_flags |= TF_ACKNOW;
1909 if ((thflags & TH_ECE) && V_tcp_do_ecn) {
1910 tp->t_flags |= TF_ECN_PERMIT;
1911 TCPSTAT_INC(tcps_ecn_shs);
1915 * Received <SYN,ACK> in SYN_SENT[*] state.
1917 * SYN_SENT --> ESTABLISHED
1918 * SYN_SENT* --> FIN_WAIT_1
1920 tp->t_starttime = ticks;
1921 if (tp->t_flags & TF_NEEDFIN) {
1922 tp->t_state = TCPS_FIN_WAIT_1;
1923 tp->t_flags &= ~TF_NEEDFIN;
1926 tp->t_state = TCPS_ESTABLISHED;
1928 tcp_timer_activate(tp, TT_KEEP,
1933 * Received initial SYN in SYN-SENT[*] state =>
1934 * simultaneous open. If segment contains CC option
1935 * and there is a cached CC, apply TAO test.
1936 * If it succeeds, connection is * half-synchronized.
1937 * Otherwise, do 3-way handshake:
1938 * SYN-SENT -> SYN-RECEIVED
1939 * SYN-SENT* -> SYN-RECEIVED*
1940 * If there was no CC option, clear cached CC value.
1942 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
1943 tcp_timer_activate(tp, TT_REXMT, 0);
1944 tp->t_state = TCPS_SYN_RECEIVED;
1947 KASSERT(ti_locked == TI_WLOCKED, ("%s: trimthenstep6: "
1948 "ti_locked %d", __func__, ti_locked));
1949 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1950 INP_WLOCK_ASSERT(tp->t_inpcb);
1953 * Advance th->th_seq to correspond to first data byte.
1954 * If data, trim to stay within window,
1955 * dropping FIN if necessary.
1958 if (tlen > tp->rcv_wnd) {
1959 todrop = tlen - tp->rcv_wnd;
1963 TCPSTAT_INC(tcps_rcvpackafterwin);
1964 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
1966 tp->snd_wl1 = th->th_seq - 1;
1967 tp->rcv_up = th->th_seq;
1969 * Client side of transaction: already sent SYN and data.
1970 * If the remote host used T/TCP to validate the SYN,
1971 * our data will be ACK'd; if so, enter normal data segment
1972 * processing in the middle of step 5, ack processing.
1973 * Otherwise, goto step 6.
1975 if (thflags & TH_ACK)
1981 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
1982 * do normal processing.
1984 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later.
1988 break; /* continue normal processing */
1992 * States other than LISTEN or SYN_SENT.
1993 * First check the RST flag and sequence number since reset segments
1994 * are exempt from the timestamp and connection count tests. This
1995 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
1996 * below which allowed reset segments in half the sequence space
1997 * to fall though and be processed (which gives forged reset
1998 * segments with a random sequence number a 50 percent chance of
1999 * killing a connection).
2000 * Then check timestamp, if present.
2001 * Then check the connection count, if present.
2002 * Then check that at least some bytes of segment are within
2003 * receive window. If segment begins before rcv_nxt,
2004 * drop leading data (and SYN); if nothing left, just ack.
2007 * If the RST bit is set, check the sequence number to see
2008 * if this is a valid reset segment.
2010 * In all states except SYN-SENT, all reset (RST) segments
2011 * are validated by checking their SEQ-fields. A reset is
2012 * valid if its sequence number is in the window.
2013 * Note: this does not take into account delayed ACKs, so
2014 * we should test against last_ack_sent instead of rcv_nxt.
2015 * The sequence number in the reset segment is normally an
2016 * echo of our outgoing acknowlegement numbers, but some hosts
2017 * send a reset with the sequence number at the rightmost edge
2018 * of our receive window, and we have to handle this case.
2019 * Note 2: Paul Watson's paper "Slipping in the Window" has shown
2020 * that brute force RST attacks are possible. To combat this,
2021 * we use a much stricter check while in the ESTABLISHED state,
2022 * only accepting RSTs where the sequence number is equal to
2023 * last_ack_sent. In all other states (the states in which a
2024 * RST is more likely), the more permissive check is used.
2025 * If we have multiple segments in flight, the initial reset
2026 * segment sequence numbers will be to the left of last_ack_sent,
2027 * but they will eventually catch up.
2028 * In any case, it never made sense to trim reset segments to
2029 * fit the receive window since RFC 1122 says:
2030 * 4.2.2.12 RST Segment: RFC-793 Section 3.4
2032 * A TCP SHOULD allow a received RST segment to include data.
2035 * It has been suggested that a RST segment could contain
2036 * ASCII text that encoded and explained the cause of the
2037 * RST. No standard has yet been established for such
2040 * If the reset segment passes the sequence number test examine
2042 * SYN_RECEIVED STATE:
2043 * If passive open, return to LISTEN state.
2044 * If active open, inform user that connection was refused.
2045 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
2046 * Inform user that connection was reset, and close tcb.
2047 * CLOSING, LAST_ACK STATES:
2050 * Drop the segment - see Stevens, vol. 2, p. 964 and
2053 if (thflags & TH_RST) {
2054 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) &&
2055 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
2056 switch (tp->t_state) {
2058 case TCPS_SYN_RECEIVED:
2059 so->so_error = ECONNREFUSED;
2062 case TCPS_ESTABLISHED:
2063 if (V_tcp_insecure_rst == 0 &&
2064 !(SEQ_GEQ(th->th_seq, tp->rcv_nxt - 1) &&
2065 SEQ_LEQ(th->th_seq, tp->rcv_nxt + 1)) &&
2066 !(SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) &&
2067 SEQ_LEQ(th->th_seq, tp->last_ack_sent + 1))) {
2068 TCPSTAT_INC(tcps_badrst);
2072 case TCPS_FIN_WAIT_1:
2073 case TCPS_FIN_WAIT_2:
2074 case TCPS_CLOSE_WAIT:
2075 so->so_error = ECONNRESET;
2077 KASSERT(ti_locked == TI_WLOCKED,
2078 ("tcp_do_segment: TH_RST 1 ti_locked %d",
2080 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2082 tp->t_state = TCPS_CLOSED;
2083 TCPSTAT_INC(tcps_drops);
2089 KASSERT(ti_locked == TI_WLOCKED,
2090 ("tcp_do_segment: TH_RST 2 ti_locked %d",
2092 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2102 * RFC 1323 PAWS: If we have a timestamp reply on this segment
2103 * and it's less than ts_recent, drop it.
2105 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
2106 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
2108 /* Check to see if ts_recent is over 24 days old. */
2109 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) {
2111 * Invalidate ts_recent. If this segment updates
2112 * ts_recent, the age will be reset later and ts_recent
2113 * will get a valid value. If it does not, setting
2114 * ts_recent to zero will at least satisfy the
2115 * requirement that zero be placed in the timestamp
2116 * echo reply when ts_recent isn't valid. The
2117 * age isn't reset until we get a valid ts_recent
2118 * because we don't want out-of-order segments to be
2119 * dropped when ts_recent is old.
2123 TCPSTAT_INC(tcps_rcvduppack);
2124 TCPSTAT_ADD(tcps_rcvdupbyte, tlen);
2125 TCPSTAT_INC(tcps_pawsdrop);
2133 * In the SYN-RECEIVED state, validate that the packet belongs to
2134 * this connection before trimming the data to fit the receive
2135 * window. Check the sequence number versus IRS since we know
2136 * the sequence numbers haven't wrapped. This is a partial fix
2137 * for the "LAND" DoS attack.
2139 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
2140 rstreason = BANDLIM_RST_OPENPORT;
2144 todrop = tp->rcv_nxt - th->th_seq;
2147 * If this is a duplicate SYN for our current connection,
2148 * advance over it and pretend and it's not a SYN.
2150 if (thflags & TH_SYN && th->th_seq == tp->irs) {
2160 * Following if statement from Stevens, vol. 2, p. 960.
2163 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
2165 * Any valid FIN must be to the left of the window.
2166 * At this point the FIN must be a duplicate or out
2167 * of sequence; drop it.
2172 * Send an ACK to resynchronize and drop any data.
2173 * But keep on processing for RST or ACK.
2175 tp->t_flags |= TF_ACKNOW;
2177 TCPSTAT_INC(tcps_rcvduppack);
2178 TCPSTAT_ADD(tcps_rcvdupbyte, todrop);
2180 TCPSTAT_INC(tcps_rcvpartduppack);
2181 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop);
2183 drop_hdrlen += todrop; /* drop from the top afterwards */
2184 th->th_seq += todrop;
2186 if (th->th_urp > todrop)
2187 th->th_urp -= todrop;
2195 * If new data are received on a connection after the
2196 * user processes are gone, then RST the other end.
2198 if ((so->so_state & SS_NOFDREF) &&
2199 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
2202 KASSERT(ti_locked == TI_WLOCKED, ("%s: SS_NOFDEREF && "
2203 "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked));
2204 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2206 if ((s = tcp_log_addrs(&tp->t_inpcb->inp_inc, th, NULL, NULL))) {
2207 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data after socket "
2208 "was closed, sending RST and removing tcpcb\n",
2209 s, __func__, tcpstates[tp->t_state], tlen);
2213 TCPSTAT_INC(tcps_rcvafterclose);
2214 rstreason = BANDLIM_UNLIMITED;
2219 * If segment ends after window, drop trailing data
2220 * (and PUSH and FIN); if nothing left, just ACK.
2222 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
2224 TCPSTAT_INC(tcps_rcvpackafterwin);
2225 if (todrop >= tlen) {
2226 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen);
2228 * If window is closed can only take segments at
2229 * window edge, and have to drop data and PUSH from
2230 * incoming segments. Continue processing, but
2231 * remember to ack. Otherwise, drop segment
2234 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2235 tp->t_flags |= TF_ACKNOW;
2236 TCPSTAT_INC(tcps_rcvwinprobe);
2240 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2243 thflags &= ~(TH_PUSH|TH_FIN);
2247 * If last ACK falls within this segment's sequence numbers,
2248 * record its timestamp.
2250 * 1) That the test incorporates suggestions from the latest
2251 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
2252 * 2) That updating only on newer timestamps interferes with
2253 * our earlier PAWS tests, so this check should be solely
2254 * predicated on the sequence space of this segment.
2255 * 3) That we modify the segment boundary check to be
2256 * Last.ACK.Sent <= SEG.SEQ + SEG.Len
2257 * instead of RFC1323's
2258 * Last.ACK.Sent < SEG.SEQ + SEG.Len,
2259 * This modified check allows us to overcome RFC1323's
2260 * limitations as described in Stevens TCP/IP Illustrated
2261 * Vol. 2 p.869. In such cases, we can still calculate the
2262 * RTT correctly when RCV.NXT == Last.ACK.Sent.
2264 if ((to.to_flags & TOF_TS) != 0 &&
2265 SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2266 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
2267 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
2268 tp->ts_recent_age = tcp_ts_getticks();
2269 tp->ts_recent = to.to_tsval;
2273 * If a SYN is in the window, then this is an
2274 * error and we send an RST and drop the connection.
2276 if (thflags & TH_SYN) {
2277 KASSERT(ti_locked == TI_WLOCKED,
2278 ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked));
2279 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2281 tp = tcp_drop(tp, ECONNRESET);
2282 rstreason = BANDLIM_UNLIMITED;
2287 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
2288 * flag is on (half-synchronized state), then queue data for
2289 * later processing; else drop segment and return.
2291 if ((thflags & TH_ACK) == 0) {
2292 if (tp->t_state == TCPS_SYN_RECEIVED ||
2293 (tp->t_flags & TF_NEEDSYN))
2295 else if (tp->t_flags & TF_ACKNOW)
2304 switch (tp->t_state) {
2307 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
2308 * ESTABLISHED state and continue processing.
2309 * The ACK was checked above.
2311 case TCPS_SYN_RECEIVED:
2313 TCPSTAT_INC(tcps_connects);
2315 /* Do window scaling? */
2316 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2317 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2318 tp->rcv_scale = tp->request_r_scale;
2319 tp->snd_wnd = tiwin;
2323 * SYN-RECEIVED -> ESTABLISHED
2324 * SYN-RECEIVED* -> FIN-WAIT-1
2326 tp->t_starttime = ticks;
2327 if (tp->t_flags & TF_NEEDFIN) {
2328 tp->t_state = TCPS_FIN_WAIT_1;
2329 tp->t_flags &= ~TF_NEEDFIN;
2331 tp->t_state = TCPS_ESTABLISHED;
2333 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
2336 * If segment contains data or ACK, will call tcp_reass()
2337 * later; if not, do so now to pass queued data to user.
2339 if (tlen == 0 && (thflags & TH_FIN) == 0)
2340 (void) tcp_reass(tp, (struct tcphdr *)0, 0,
2342 tp->snd_wl1 = th->th_seq - 1;
2346 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2347 * ACKs. If the ack is in the range
2348 * tp->snd_una < th->th_ack <= tp->snd_max
2349 * then advance tp->snd_una to th->th_ack and drop
2350 * data from the retransmission queue. If this ACK reflects
2351 * more up to date window information we update our window information.
2353 case TCPS_ESTABLISHED:
2354 case TCPS_FIN_WAIT_1:
2355 case TCPS_FIN_WAIT_2:
2356 case TCPS_CLOSE_WAIT:
2359 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2360 TCPSTAT_INC(tcps_rcvacktoomuch);
2363 if ((tp->t_flags & TF_SACK_PERMIT) &&
2364 ((to.to_flags & TOF_SACK) ||
2365 !TAILQ_EMPTY(&tp->snd_holes)))
2366 tcp_sack_doack(tp, &to, th->th_ack);
2368 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
2369 hhook_run_tcp_est_in(tp, th, &to);
2371 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2372 if (tlen == 0 && tiwin == tp->snd_wnd) {
2373 TCPSTAT_INC(tcps_rcvdupack);
2375 * If we have outstanding data (other than
2376 * a window probe), this is a completely
2377 * duplicate ack (ie, window info didn't
2378 * change), the ack is the biggest we've
2379 * seen and we've seen exactly our rexmt
2380 * threshhold of them, assume a packet
2381 * has been dropped and retransmit it.
2382 * Kludge snd_nxt & the congestion
2383 * window so we send only this one
2386 * We know we're losing at the current
2387 * window size so do congestion avoidance
2388 * (set ssthresh to half the current window
2389 * and pull our congestion window back to
2390 * the new ssthresh).
2392 * Dup acks mean that packets have left the
2393 * network (they're now cached at the receiver)
2394 * so bump cwnd by the amount in the receiver
2395 * to keep a constant cwnd packets in the
2398 * When using TCP ECN, notify the peer that
2399 * we reduced the cwnd.
2401 if (!tcp_timer_active(tp, TT_REXMT) ||
2402 th->th_ack != tp->snd_una)
2404 else if (++tp->t_dupacks > tcprexmtthresh ||
2405 IN_FASTRECOVERY(tp->t_flags)) {
2406 cc_ack_received(tp, th, CC_DUPACK);
2407 if ((tp->t_flags & TF_SACK_PERMIT) &&
2408 IN_FASTRECOVERY(tp->t_flags)) {
2412 * Compute the amount of data in flight first.
2413 * We can inject new data into the pipe iff
2414 * we have less than 1/2 the original window's
2415 * worth of data in flight.
2417 awnd = (tp->snd_nxt - tp->snd_fack) +
2418 tp->sackhint.sack_bytes_rexmit;
2419 if (awnd < tp->snd_ssthresh) {
2420 tp->snd_cwnd += tp->t_maxseg;
2421 if (tp->snd_cwnd > tp->snd_ssthresh)
2422 tp->snd_cwnd = tp->snd_ssthresh;
2425 tp->snd_cwnd += tp->t_maxseg;
2426 (void) tcp_output(tp);
2428 } else if (tp->t_dupacks == tcprexmtthresh) {
2429 tcp_seq onxt = tp->snd_nxt;
2432 * If we're doing sack, check to
2433 * see if we're already in sack
2434 * recovery. If we're not doing sack,
2435 * check to see if we're in newreno
2438 if (tp->t_flags & TF_SACK_PERMIT) {
2439 if (IN_FASTRECOVERY(tp->t_flags)) {
2444 if (SEQ_LEQ(th->th_ack,
2450 /* Congestion signal before ack. */
2451 cc_cong_signal(tp, th, CC_NDUPACK);
2452 cc_ack_received(tp, th, CC_DUPACK);
2453 tcp_timer_activate(tp, TT_REXMT, 0);
2455 if (tp->t_flags & TF_SACK_PERMIT) {
2457 tcps_sack_recovery_episode);
2458 tp->sack_newdata = tp->snd_nxt;
2459 tp->snd_cwnd = tp->t_maxseg;
2460 (void) tcp_output(tp);
2463 tp->snd_nxt = th->th_ack;
2464 tp->snd_cwnd = tp->t_maxseg;
2465 (void) tcp_output(tp);
2466 KASSERT(tp->snd_limited <= 2,
2467 ("%s: tp->snd_limited too big",
2469 tp->snd_cwnd = tp->snd_ssthresh +
2471 (tp->t_dupacks - tp->snd_limited);
2472 if (SEQ_GT(onxt, tp->snd_nxt))
2475 } else if (V_tcp_do_rfc3042) {
2476 cc_ack_received(tp, th, CC_DUPACK);
2477 u_long oldcwnd = tp->snd_cwnd;
2478 tcp_seq oldsndmax = tp->snd_max;
2481 KASSERT(tp->t_dupacks == 1 ||
2483 ("%s: dupacks not 1 or 2",
2485 if (tp->t_dupacks == 1)
2486 tp->snd_limited = 0;
2488 (tp->snd_nxt - tp->snd_una) +
2489 (tp->t_dupacks - tp->snd_limited) *
2491 (void) tcp_output(tp);
2492 sent = tp->snd_max - oldsndmax;
2493 if (sent > tp->t_maxseg) {
2494 KASSERT((tp->t_dupacks == 2 &&
2495 tp->snd_limited == 0) ||
2496 (sent == tp->t_maxseg + 1 &&
2497 tp->t_flags & TF_SENTFIN),
2498 ("%s: sent too much",
2500 tp->snd_limited = 2;
2501 } else if (sent > 0)
2503 tp->snd_cwnd = oldcwnd;
2511 KASSERT(SEQ_GT(th->th_ack, tp->snd_una),
2512 ("%s: th_ack <= snd_una", __func__));
2515 * If the congestion window was inflated to account
2516 * for the other side's cached packets, retract it.
2518 if (IN_FASTRECOVERY(tp->t_flags)) {
2519 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2520 if (tp->t_flags & TF_SACK_PERMIT)
2521 tcp_sack_partialack(tp, th);
2523 tcp_newreno_partial_ack(tp, th);
2525 cc_post_recovery(tp, th);
2529 * If we reach this point, ACK is not a duplicate,
2530 * i.e., it ACKs something we sent.
2532 if (tp->t_flags & TF_NEEDSYN) {
2534 * T/TCP: Connection was half-synchronized, and our
2535 * SYN has been ACK'd (so connection is now fully
2536 * synchronized). Go to non-starred state,
2537 * increment snd_una for ACK of SYN, and check if
2538 * we can do window scaling.
2540 tp->t_flags &= ~TF_NEEDSYN;
2542 /* Do window scaling? */
2543 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2544 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2545 tp->rcv_scale = tp->request_r_scale;
2546 /* Send window already scaled. */
2551 INP_WLOCK_ASSERT(tp->t_inpcb);
2553 acked = BYTES_THIS_ACK(tp, th);
2554 TCPSTAT_INC(tcps_rcvackpack);
2555 TCPSTAT_ADD(tcps_rcvackbyte, acked);
2558 * If we just performed our first retransmit, and the ACK
2559 * arrives within our recovery window, then it was a mistake
2560 * to do the retransmit in the first place. Recover our
2561 * original cwnd and ssthresh, and proceed to transmit where
2564 if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID &&
2565 (int)(ticks - tp->t_badrxtwin) < 0)
2566 cc_cong_signal(tp, th, CC_RTO_ERR);
2569 * If we have a timestamp reply, update smoothed
2570 * round trip time. If no timestamp is present but
2571 * transmit timer is running and timed sequence
2572 * number was acked, update smoothed round trip time.
2573 * Since we now have an rtt measurement, cancel the
2574 * timer backoff (cf., Phil Karn's retransmit alg.).
2575 * Recompute the initial retransmit timer.
2577 * Some boxes send broken timestamp replies
2578 * during the SYN+ACK phase, ignore
2579 * timestamps of 0 or we could calculate a
2580 * huge RTT and blow up the retransmit timer.
2582 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) {
2585 t = tcp_ts_getticks() - to.to_tsecr;
2586 if (!tp->t_rttlow || tp->t_rttlow > t)
2588 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1);
2589 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2590 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2591 tp->t_rttlow = ticks - tp->t_rtttime;
2592 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2596 * If all outstanding data is acked, stop retransmit
2597 * timer and remember to restart (more output or persist).
2598 * If there is more data to be acked, restart retransmit
2599 * timer, using current (possibly backed-off) value.
2601 if (th->th_ack == tp->snd_max) {
2602 tcp_timer_activate(tp, TT_REXMT, 0);
2604 } else if (!tcp_timer_active(tp, TT_PERSIST))
2605 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
2608 * If no data (only SYN) was ACK'd,
2609 * skip rest of ACK processing.
2615 * Let the congestion control algorithm update congestion
2616 * control related information. This typically means increasing
2617 * the congestion window.
2619 cc_ack_received(tp, th, CC_ACK);
2621 SOCKBUF_LOCK(&so->so_snd);
2622 if (acked > so->so_snd.sb_cc) {
2623 tp->snd_wnd -= so->so_snd.sb_cc;
2624 sbdrop_locked(&so->so_snd, (int)so->so_snd.sb_cc);
2627 sbdrop_locked(&so->so_snd, acked);
2628 tp->snd_wnd -= acked;
2631 /* NB: sowwakeup_locked() does an implicit unlock. */
2632 sowwakeup_locked(so);
2633 /* Detect una wraparound. */
2634 if (!IN_RECOVERY(tp->t_flags) &&
2635 SEQ_GT(tp->snd_una, tp->snd_recover) &&
2636 SEQ_LEQ(th->th_ack, tp->snd_recover))
2637 tp->snd_recover = th->th_ack - 1;
2638 /* XXXLAS: Can this be moved up into cc_post_recovery? */
2639 if (IN_RECOVERY(tp->t_flags) &&
2640 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2641 EXIT_RECOVERY(tp->t_flags);
2643 tp->snd_una = th->th_ack;
2644 if (tp->t_flags & TF_SACK_PERMIT) {
2645 if (SEQ_GT(tp->snd_una, tp->snd_recover))
2646 tp->snd_recover = tp->snd_una;
2648 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2649 tp->snd_nxt = tp->snd_una;
2651 switch (tp->t_state) {
2654 * In FIN_WAIT_1 STATE in addition to the processing
2655 * for the ESTABLISHED state if our FIN is now acknowledged
2656 * then enter FIN_WAIT_2.
2658 case TCPS_FIN_WAIT_1:
2659 if (ourfinisacked) {
2661 * If we can't receive any more
2662 * data, then closing user can proceed.
2663 * Starting the timer is contrary to the
2664 * specification, but if we don't get a FIN
2665 * we'll hang forever.
2668 * we should release the tp also, and use a
2671 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2672 soisdisconnected(so);
2673 tcp_timer_activate(tp, TT_2MSL,
2674 (tcp_fast_finwait2_recycle ?
2675 tcp_finwait2_timeout :
2678 tp->t_state = TCPS_FIN_WAIT_2;
2683 * In CLOSING STATE in addition to the processing for
2684 * the ESTABLISHED state if the ACK acknowledges our FIN
2685 * then enter the TIME-WAIT state, otherwise ignore
2689 if (ourfinisacked) {
2690 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2692 INP_INFO_WUNLOCK(&V_tcbinfo);
2699 * In LAST_ACK, we may still be waiting for data to drain
2700 * and/or to be acked, as well as for the ack of our FIN.
2701 * If our FIN is now acknowledged, delete the TCB,
2702 * enter the closed state and return.
2705 if (ourfinisacked) {
2706 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2715 INP_WLOCK_ASSERT(tp->t_inpcb);
2718 * Update window information.
2719 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2721 if ((thflags & TH_ACK) &&
2722 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2723 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2724 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2725 /* keep track of pure window updates */
2727 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2728 TCPSTAT_INC(tcps_rcvwinupd);
2729 tp->snd_wnd = tiwin;
2730 tp->snd_wl1 = th->th_seq;
2731 tp->snd_wl2 = th->th_ack;
2732 if (tp->snd_wnd > tp->max_sndwnd)
2733 tp->max_sndwnd = tp->snd_wnd;
2738 * Process segments with URG.
2740 if ((thflags & TH_URG) && th->th_urp &&
2741 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2743 * This is a kludge, but if we receive and accept
2744 * random urgent pointers, we'll crash in
2745 * soreceive. It's hard to imagine someone
2746 * actually wanting to send this much urgent data.
2748 SOCKBUF_LOCK(&so->so_rcv);
2749 if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2750 th->th_urp = 0; /* XXX */
2751 thflags &= ~TH_URG; /* XXX */
2752 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */
2753 goto dodata; /* XXX */
2756 * If this segment advances the known urgent pointer,
2757 * then mark the data stream. This should not happen
2758 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2759 * a FIN has been received from the remote side.
2760 * In these states we ignore the URG.
2762 * According to RFC961 (Assigned Protocols),
2763 * the urgent pointer points to the last octet
2764 * of urgent data. We continue, however,
2765 * to consider it to indicate the first octet
2766 * of data past the urgent section as the original
2767 * spec states (in one of two places).
2769 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2770 tp->rcv_up = th->th_seq + th->th_urp;
2771 so->so_oobmark = so->so_rcv.sb_cc +
2772 (tp->rcv_up - tp->rcv_nxt) - 1;
2773 if (so->so_oobmark == 0)
2774 so->so_rcv.sb_state |= SBS_RCVATMARK;
2776 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2778 SOCKBUF_UNLOCK(&so->so_rcv);
2780 * Remove out of band data so doesn't get presented to user.
2781 * This can happen independent of advancing the URG pointer,
2782 * but if two URG's are pending at once, some out-of-band
2783 * data may creep in... ick.
2785 if (th->th_urp <= (u_long)tlen &&
2786 !(so->so_options & SO_OOBINLINE)) {
2787 /* hdr drop is delayed */
2788 tcp_pulloutofband(so, th, m, drop_hdrlen);
2792 * If no out of band data is expected,
2793 * pull receive urgent pointer along
2794 * with the receive window.
2796 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2797 tp->rcv_up = tp->rcv_nxt;
2800 INP_WLOCK_ASSERT(tp->t_inpcb);
2803 * Process the segment text, merging it into the TCP sequencing queue,
2804 * and arranging for acknowledgment of receipt if necessary.
2805 * This process logically involves adjusting tp->rcv_wnd as data
2806 * is presented to the user (this happens in tcp_usrreq.c,
2807 * case PRU_RCVD). If a FIN has already been received on this
2808 * connection then we just ignore the text.
2810 if ((tlen || (thflags & TH_FIN)) &&
2811 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2812 tcp_seq save_start = th->th_seq;
2813 m_adj(m, drop_hdrlen); /* delayed header drop */
2815 * Insert segment which includes th into TCP reassembly queue
2816 * with control block tp. Set thflags to whether reassembly now
2817 * includes a segment with FIN. This handles the common case
2818 * inline (segment is the next to be received on an established
2819 * connection, and the queue is empty), avoiding linkage into
2820 * and removal from the queue and repetition of various
2822 * Set DELACK for segments received in order, but ack
2823 * immediately when segments are out of order (so
2824 * fast retransmit can work).
2826 if (th->th_seq == tp->rcv_nxt &&
2827 LIST_EMPTY(&tp->t_segq) &&
2828 TCPS_HAVEESTABLISHED(tp->t_state)) {
2830 tp->t_flags |= TF_DELACK;
2832 tp->t_flags |= TF_ACKNOW;
2833 tp->rcv_nxt += tlen;
2834 thflags = th->th_flags & TH_FIN;
2835 TCPSTAT_INC(tcps_rcvpack);
2836 TCPSTAT_ADD(tcps_rcvbyte, tlen);
2838 SOCKBUF_LOCK(&so->so_rcv);
2839 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
2842 sbappendstream_locked(&so->so_rcv, m);
2843 /* NB: sorwakeup_locked() does an implicit unlock. */
2844 sorwakeup_locked(so);
2847 * XXX: Due to the header drop above "th" is
2848 * theoretically invalid by now. Fortunately
2849 * m_adj() doesn't actually frees any mbufs
2850 * when trimming from the head.
2852 thflags = tcp_reass(tp, th, &tlen, m);
2853 tp->t_flags |= TF_ACKNOW;
2855 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT))
2856 tcp_update_sack_list(tp, save_start, save_start + tlen);
2859 * Note the amount of data that peer has sent into
2860 * our window, in order to estimate the sender's
2864 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
2865 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2867 len = so->so_rcv.sb_hiwat;
2875 * If FIN is received ACK the FIN and let the user know
2876 * that the connection is closing.
2878 if (thflags & TH_FIN) {
2879 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2882 * If connection is half-synchronized
2883 * (ie NEEDSYN flag on) then delay ACK,
2884 * so it may be piggybacked when SYN is sent.
2885 * Otherwise, since we received a FIN then no
2886 * more input can be expected, send ACK now.
2888 if (tp->t_flags & TF_NEEDSYN)
2889 tp->t_flags |= TF_DELACK;
2891 tp->t_flags |= TF_ACKNOW;
2894 switch (tp->t_state) {
2897 * In SYN_RECEIVED and ESTABLISHED STATES
2898 * enter the CLOSE_WAIT state.
2900 case TCPS_SYN_RECEIVED:
2901 tp->t_starttime = ticks;
2903 case TCPS_ESTABLISHED:
2904 tp->t_state = TCPS_CLOSE_WAIT;
2908 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2909 * enter the CLOSING state.
2911 case TCPS_FIN_WAIT_1:
2912 tp->t_state = TCPS_CLOSING;
2916 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2917 * starting the time-wait timer, turning off the other
2920 case TCPS_FIN_WAIT_2:
2921 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2922 KASSERT(ti_locked == TI_WLOCKED, ("%s: dodata "
2923 "TCP_FIN_WAIT_2 ti_locked: %d", __func__,
2927 INP_INFO_WUNLOCK(&V_tcbinfo);
2931 if (ti_locked == TI_WLOCKED)
2932 INP_INFO_WUNLOCK(&V_tcbinfo);
2933 ti_locked = TI_UNLOCKED;
2936 if (so->so_options & SO_DEBUG)
2937 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
2942 * Return any desired output.
2944 if (needoutput || (tp->t_flags & TF_ACKNOW))
2945 (void) tcp_output(tp);
2948 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d",
2949 __func__, ti_locked));
2950 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
2951 INP_WLOCK_ASSERT(tp->t_inpcb);
2953 if (tp->t_flags & TF_DELACK) {
2954 tp->t_flags &= ~TF_DELACK;
2955 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
2957 INP_WUNLOCK(tp->t_inpcb);
2962 * Generate an ACK dropping incoming segment if it occupies
2963 * sequence space, where the ACK reflects our state.
2965 * We can now skip the test for the RST flag since all
2966 * paths to this code happen after packets containing
2967 * RST have been dropped.
2969 * In the SYN-RECEIVED state, don't send an ACK unless the
2970 * segment we received passes the SYN-RECEIVED ACK test.
2971 * If it fails send a RST. This breaks the loop in the
2972 * "LAND" DoS attack, and also prevents an ACK storm
2973 * between two listening ports that have been sent forged
2974 * SYN segments, each with the source address of the other.
2976 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
2977 (SEQ_GT(tp->snd_una, th->th_ack) ||
2978 SEQ_GT(th->th_ack, tp->snd_max)) ) {
2979 rstreason = BANDLIM_RST_OPENPORT;
2983 if (so->so_options & SO_DEBUG)
2984 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2987 if (ti_locked == TI_WLOCKED)
2988 INP_INFO_WUNLOCK(&V_tcbinfo);
2989 ti_locked = TI_UNLOCKED;
2991 tp->t_flags |= TF_ACKNOW;
2992 (void) tcp_output(tp);
2993 INP_WUNLOCK(tp->t_inpcb);
2998 if (ti_locked == TI_WLOCKED)
2999 INP_INFO_WUNLOCK(&V_tcbinfo);
3000 ti_locked = TI_UNLOCKED;
3003 tcp_dropwithreset(m, th, tp, tlen, rstreason);
3004 INP_WUNLOCK(tp->t_inpcb);
3006 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
3010 if (ti_locked == TI_WLOCKED) {
3011 INP_INFO_WUNLOCK(&V_tcbinfo);
3012 ti_locked = TI_UNLOCKED;
3016 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
3020 * Drop space held by incoming segment and return.
3023 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
3024 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3028 INP_WUNLOCK(tp->t_inpcb);
3033 * Issue RST and make ACK acceptable to originator of segment.
3034 * The mbuf must still include the original packet header.
3038 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
3039 int tlen, int rstreason)
3045 struct ip6_hdr *ip6;
3049 INP_WLOCK_ASSERT(tp->t_inpcb);
3052 /* Don't bother if destination was broadcast/multicast. */
3053 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
3056 if (mtod(m, struct ip *)->ip_v == 6) {
3057 ip6 = mtod(m, struct ip6_hdr *);
3058 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3059 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3061 /* IPv6 anycast check is done at tcp6_input() */
3064 #if defined(INET) && defined(INET6)
3069 ip = mtod(m, struct ip *);
3070 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3071 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3072 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3073 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3078 /* Perform bandwidth limiting. */
3079 if (badport_bandlim(rstreason) < 0)
3082 /* tcp_respond consumes the mbuf chain. */
3083 if (th->th_flags & TH_ACK) {
3084 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0,
3085 th->th_ack, TH_RST);
3087 if (th->th_flags & TH_SYN)
3089 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
3090 (tcp_seq)0, TH_RST|TH_ACK);
3098 * Parse TCP options and place in tcpopt.
3101 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags)
3106 for (; cnt > 0; cnt -= optlen, cp += optlen) {
3108 if (opt == TCPOPT_EOL)
3110 if (opt == TCPOPT_NOP)
3116 if (optlen < 2 || optlen > cnt)
3121 if (optlen != TCPOLEN_MAXSEG)
3123 if (!(flags & TO_SYN))
3125 to->to_flags |= TOF_MSS;
3126 bcopy((char *)cp + 2,
3127 (char *)&to->to_mss, sizeof(to->to_mss));
3128 to->to_mss = ntohs(to->to_mss);
3131 if (optlen != TCPOLEN_WINDOW)
3133 if (!(flags & TO_SYN))
3135 to->to_flags |= TOF_SCALE;
3136 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT);
3138 case TCPOPT_TIMESTAMP:
3139 if (optlen != TCPOLEN_TIMESTAMP)
3141 to->to_flags |= TOF_TS;
3142 bcopy((char *)cp + 2,
3143 (char *)&to->to_tsval, sizeof(to->to_tsval));
3144 to->to_tsval = ntohl(to->to_tsval);
3145 bcopy((char *)cp + 6,
3146 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
3147 to->to_tsecr = ntohl(to->to_tsecr);
3149 #ifdef TCP_SIGNATURE
3151 * XXX In order to reply to a host which has set the
3152 * TCP_SIGNATURE option in its initial SYN, we have to
3153 * record the fact that the option was observed here
3154 * for the syncache code to perform the correct response.
3156 case TCPOPT_SIGNATURE:
3157 if (optlen != TCPOLEN_SIGNATURE)
3159 to->to_flags |= TOF_SIGNATURE;
3160 to->to_signature = cp + 2;
3163 case TCPOPT_SACK_PERMITTED:
3164 if (optlen != TCPOLEN_SACK_PERMITTED)
3166 if (!(flags & TO_SYN))
3170 to->to_flags |= TOF_SACKPERM;
3173 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
3177 to->to_flags |= TOF_SACK;
3178 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
3179 to->to_sacks = cp + 2;
3180 TCPSTAT_INC(tcps_sack_rcv_blocks);
3189 * Pull out of band byte out of a segment so
3190 * it doesn't appear in the user's data queue.
3191 * It is still reflected in the segment length for
3192 * sequencing purposes.
3195 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m,
3198 int cnt = off + th->th_urp - 1;
3201 if (m->m_len > cnt) {
3202 char *cp = mtod(m, caddr_t) + cnt;
3203 struct tcpcb *tp = sototcpcb(so);
3205 INP_WLOCK_ASSERT(tp->t_inpcb);
3208 tp->t_oobflags |= TCPOOB_HAVEDATA;
3209 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3211 if (m->m_flags & M_PKTHDR)
3220 panic("tcp_pulloutofband");
3224 * Collect new round-trip time estimate
3225 * and update averages and current timeout.
3228 tcp_xmit_timer(struct tcpcb *tp, int rtt)
3232 INP_WLOCK_ASSERT(tp->t_inpcb);
3234 TCPSTAT_INC(tcps_rttupdated);
3236 if (tp->t_srtt != 0) {
3238 * srtt is stored as fixed point with 5 bits after the
3239 * binary point (i.e., scaled by 8). The following magic
3240 * is equivalent to the smoothing algorithm in rfc793 with
3241 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3242 * point). Adjust rtt to origin 0.
3244 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
3245 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
3247 if ((tp->t_srtt += delta) <= 0)
3251 * We accumulate a smoothed rtt variance (actually, a
3252 * smoothed mean difference), then set the retransmit
3253 * timer to smoothed rtt + 4 times the smoothed variance.
3254 * rttvar is stored as fixed point with 4 bits after the
3255 * binary point (scaled by 16). The following is
3256 * equivalent to rfc793 smoothing with an alpha of .75
3257 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
3258 * rfc793's wired-in beta.
3262 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
3263 if ((tp->t_rttvar += delta) <= 0)
3265 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
3266 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3269 * No rtt measurement yet - use the unsmoothed rtt.
3270 * Set the variance to half the rtt (so our first
3271 * retransmit happens at 3*rtt).
3273 tp->t_srtt = rtt << TCP_RTT_SHIFT;
3274 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
3275 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3281 * the retransmit should happen at rtt + 4 * rttvar.
3282 * Because of the way we do the smoothing, srtt and rttvar
3283 * will each average +1/2 tick of bias. When we compute
3284 * the retransmit timer, we want 1/2 tick of rounding and
3285 * 1 extra tick because of +-1/2 tick uncertainty in the
3286 * firing of the timer. The bias will give us exactly the
3287 * 1.5 tick we need. But, because the bias is
3288 * statistical, we have to test that we don't drop below
3289 * the minimum feasible timer (which is 2 ticks).
3291 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3292 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3295 * We received an ack for a packet that wasn't retransmitted;
3296 * it is probably safe to discard any error indications we've
3297 * received recently. This isn't quite right, but close enough
3298 * for now (a route might have failed after we sent a segment,
3299 * and the return path might not be symmetrical).
3301 tp->t_softerror = 0;
3305 * Determine a reasonable value for maxseg size.
3306 * If the route is known, check route for mtu.
3307 * If none, use an mss that can be handled on the outgoing
3308 * interface without forcing IP to fragment; if bigger than
3309 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
3310 * to utilize large mbufs. If no route is found, route has no mtu,
3311 * or the destination isn't local, use a default, hopefully conservative
3312 * size (usually 512 or the default IP max size, but no more than the mtu
3313 * of the interface), as we can't discover anything about intervening
3314 * gateways or networks. We also initialize the congestion/slow start
3315 * window to be a single segment if the destination isn't local.
3316 * While looking at the routing entry, we also initialize other path-dependent
3317 * parameters from pre-set or cached values in the routing entry.
3319 * Also take into account the space needed for options that we
3320 * send regularly. Make maxseg shorter by that amount to assure
3321 * that we can send maxseg amount of data even when the options
3322 * are present. Store the upper limit of the length of options plus
3325 * NOTE that this routine is only called when we process an incoming
3326 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS
3327 * settings are handled in tcp_mssopt().
3330 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer,
3331 struct hc_metrics_lite *metricptr, int *mtuflags)
3335 struct inpcb *inp = tp->t_inpcb;
3336 struct hc_metrics_lite metrics;
3339 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
3340 size_t min_protoh = isipv6 ?
3341 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
3342 sizeof (struct tcpiphdr);
3344 const size_t min_protoh = sizeof(struct tcpiphdr);
3347 INP_WLOCK_ASSERT(tp->t_inpcb);
3349 if (mtuoffer != -1) {
3350 KASSERT(offer == -1, ("%s: conflict", __func__));
3351 offer = mtuoffer - min_protoh;
3358 maxmtu = tcp_maxmtu6(&inp->inp_inc, mtuflags);
3359 tp->t_maxopd = tp->t_maxseg = V_tcp_v6mssdflt;
3362 #if defined(INET) && defined(INET6)
3367 maxmtu = tcp_maxmtu(&inp->inp_inc, mtuflags);
3368 tp->t_maxopd = tp->t_maxseg = V_tcp_mssdflt;
3373 * No route to sender, stay with default mss and return.
3377 * In case we return early we need to initialize metrics
3378 * to a defined state as tcp_hc_get() would do for us
3379 * if there was no cache hit.
3381 if (metricptr != NULL)
3382 bzero(metricptr, sizeof(struct hc_metrics_lite));
3386 /* What have we got? */
3390 * Offer == 0 means that there was no MSS on the SYN
3391 * segment, in this case we use tcp_mssdflt as
3392 * already assigned to t_maxopd above.
3394 offer = tp->t_maxopd;
3399 * Offer == -1 means that we didn't receive SYN yet.
3405 * Prevent DoS attack with too small MSS. Round up
3406 * to at least minmss.
3408 offer = max(offer, V_tcp_minmss);
3412 * rmx information is now retrieved from tcp_hostcache.
3414 tcp_hc_get(&inp->inp_inc, &metrics);
3415 if (metricptr != NULL)
3416 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite));
3419 * If there's a discovered mtu int tcp hostcache, use it
3420 * else, use the link mtu.
3422 if (metrics.rmx_mtu)
3423 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
3427 mss = maxmtu - min_protoh;
3428 if (!V_path_mtu_discovery &&
3429 !in6_localaddr(&inp->in6p_faddr))
3430 mss = min(mss, V_tcp_v6mssdflt);
3433 #if defined(INET) && defined(INET6)
3438 mss = maxmtu - min_protoh;
3439 if (!V_path_mtu_discovery &&
3440 !in_localaddr(inp->inp_faddr))
3441 mss = min(mss, V_tcp_mssdflt);
3445 * XXX - The above conditional (mss = maxmtu - min_protoh)
3446 * probably violates the TCP spec.
3447 * The problem is that, since we don't know the
3448 * other end's MSS, we are supposed to use a conservative
3449 * default. But, if we do that, then MTU discovery will
3450 * never actually take place, because the conservative
3451 * default is much less than the MTUs typically seen
3452 * on the Internet today. For the moment, we'll sweep
3453 * this under the carpet.
3455 * The conservative default might not actually be a problem
3456 * if the only case this occurs is when sending an initial
3457 * SYN with options and data to a host we've never talked
3458 * to before. Then, they will reply with an MSS value which
3459 * will get recorded and the new parameters should get
3460 * recomputed. For Further Study.
3463 mss = min(mss, offer);
3466 * Sanity check: make sure that maxopd will be large
3467 * enough to allow some data on segments even if the
3468 * all the option space is used (40bytes). Otherwise
3469 * funny things may happen in tcp_output.
3474 * maxopd stores the maximum length of data AND options
3475 * in a segment; maxseg is the amount of data in a normal
3476 * segment. We need to store this value (maxopd) apart
3477 * from maxseg, because now every segment carries options
3478 * and thus we normally have somewhat less data in segments.
3483 * origoffer==-1 indicates that no segments were received yet.
3484 * In this case we just guess.
3486 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
3488 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
3489 mss -= TCPOLEN_TSTAMP_APPA;
3491 #if (MCLBYTES & (MCLBYTES - 1)) == 0
3493 mss &= ~(MCLBYTES-1);
3496 mss = mss / MCLBYTES * MCLBYTES;
3502 tcp_mss(struct tcpcb *tp, int offer)
3508 struct hc_metrics_lite metrics;
3511 KASSERT(tp != NULL, ("%s: tp == NULL", __func__));
3513 tcp_mss_update(tp, offer, -1, &metrics, &mtuflags);
3519 * If there's a pipesize, change the socket buffer to that size,
3520 * don't change if sb_hiwat is different than default (then it
3521 * has been changed on purpose with setsockopt).
3522 * Make the socket buffers an integral number of mss units;
3523 * if the mss is larger than the socket buffer, decrease the mss.
3525 so = inp->inp_socket;
3526 SOCKBUF_LOCK(&so->so_snd);
3527 if ((so->so_snd.sb_hiwat == tcp_sendspace) && metrics.rmx_sendpipe)
3528 bufsize = metrics.rmx_sendpipe;
3530 bufsize = so->so_snd.sb_hiwat;
3534 bufsize = roundup(bufsize, mss);
3535 if (bufsize > sb_max)
3537 if (bufsize > so->so_snd.sb_hiwat)
3538 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
3540 SOCKBUF_UNLOCK(&so->so_snd);
3543 SOCKBUF_LOCK(&so->so_rcv);
3544 if ((so->so_rcv.sb_hiwat == tcp_recvspace) && metrics.rmx_recvpipe)
3545 bufsize = metrics.rmx_recvpipe;
3547 bufsize = so->so_rcv.sb_hiwat;
3548 if (bufsize > mss) {
3549 bufsize = roundup(bufsize, mss);
3550 if (bufsize > sb_max)
3552 if (bufsize > so->so_rcv.sb_hiwat)
3553 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
3555 SOCKBUF_UNLOCK(&so->so_rcv);
3557 /* Check the interface for TSO capabilities. */
3558 if (mtuflags & CSUM_TSO)
3559 tp->t_flags |= TF_TSO;
3563 * Determine the MSS option to send on an outgoing SYN.
3566 tcp_mssopt(struct in_conninfo *inc)
3573 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3576 if (inc->inc_flags & INC_ISIPV6) {
3577 mss = V_tcp_v6mssdflt;
3578 maxmtu = tcp_maxmtu6(inc, NULL);
3579 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3580 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3583 #if defined(INET) && defined(INET6)
3588 mss = V_tcp_mssdflt;
3589 maxmtu = tcp_maxmtu(inc, NULL);
3590 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3591 min_protoh = sizeof(struct tcpiphdr);
3594 if (maxmtu && thcmtu)
3595 mss = min(maxmtu, thcmtu) - min_protoh;
3596 else if (maxmtu || thcmtu)
3597 mss = max(maxmtu, thcmtu) - min_protoh;
3604 * On a partial ack arrives, force the retransmission of the
3605 * next unacknowledged segment. Do not clear tp->t_dupacks.
3606 * By setting snd_nxt to ti_ack, this forces retransmission timer to
3610 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th)
3612 tcp_seq onxt = tp->snd_nxt;
3613 u_long ocwnd = tp->snd_cwnd;
3615 INP_WLOCK_ASSERT(tp->t_inpcb);
3617 tcp_timer_activate(tp, TT_REXMT, 0);
3619 tp->snd_nxt = th->th_ack;
3621 * Set snd_cwnd to one segment beyond acknowledged offset.
3622 * (tp->snd_una has not yet been updated when this function is called.)
3624 tp->snd_cwnd = tp->t_maxseg + BYTES_THIS_ACK(tp, th);
3625 tp->t_flags |= TF_ACKNOW;
3626 (void) tcp_output(tp);
3627 tp->snd_cwnd = ocwnd;
3628 if (SEQ_GT(onxt, tp->snd_nxt))
3631 * Partial window deflation. Relies on fact that tp->snd_una
3634 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th))
3635 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th);
3638 tp->snd_cwnd += tp->t_maxseg;