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 static 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_recvspace) = 1024*64;
187 #define V_tcp_recvspace VNET(tcp_recvspace)
188 SYSCTL_VNET_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
189 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size");
191 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1;
192 #define V_tcp_do_autorcvbuf VNET(tcp_do_autorcvbuf)
193 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_RW,
194 &VNET_NAME(tcp_do_autorcvbuf), 0,
195 "Enable automatic receive buffer sizing");
197 VNET_DEFINE(int, tcp_autorcvbuf_inc) = 16*1024;
198 #define V_tcp_autorcvbuf_inc VNET(tcp_autorcvbuf_inc)
199 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_RW,
200 &VNET_NAME(tcp_autorcvbuf_inc), 0,
201 "Incrementor step size of automatic receive buffer");
203 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024;
204 #define V_tcp_autorcvbuf_max VNET(tcp_autorcvbuf_max)
205 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_RW,
206 &VNET_NAME(tcp_autorcvbuf_max), 0,
207 "Max size of automatic receive buffer");
209 VNET_DEFINE(struct inpcbhead, tcb);
210 #define tcb6 tcb /* for KAME src sync over BSD*'s */
211 VNET_DEFINE(struct inpcbinfo, tcbinfo);
213 static void tcp_dooptions(struct tcpopt *, u_char *, int, int);
214 static void tcp_do_segment(struct mbuf *, struct tcphdr *,
215 struct socket *, struct tcpcb *, int, int, uint8_t,
217 static void tcp_dropwithreset(struct mbuf *, struct tcphdr *,
218 struct tcpcb *, int, int);
219 static void tcp_pulloutofband(struct socket *,
220 struct tcphdr *, struct mbuf *, int);
221 static void tcp_xmit_timer(struct tcpcb *, int);
222 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *);
223 static void inline tcp_fields_to_host(struct tcphdr *);
225 static void inline tcp_fields_to_net(struct tcphdr *);
226 static int inline tcp_signature_verify_input(struct mbuf *, int, int,
227 int, struct tcpopt *, struct tcphdr *, u_int);
229 static void inline cc_ack_received(struct tcpcb *tp, struct tcphdr *th,
231 static void inline cc_conn_init(struct tcpcb *tp);
232 static void inline cc_post_recovery(struct tcpcb *tp, struct tcphdr *th);
233 static void inline hhook_run_tcp_est_in(struct tcpcb *tp,
234 struct tcphdr *th, struct tcpopt *to);
237 * Kernel module interface for updating tcpstat. The argument is an index
238 * into tcpstat treated as an array of u_long. While this encodes the
239 * general layout of tcpstat into the caller, it doesn't encode its location,
240 * so that future changes to add, for example, per-CPU stats support won't
241 * cause binary compatibility problems for kernel modules.
244 kmod_tcpstat_inc(int statnum)
247 (*((u_long *)&V_tcpstat + statnum))++;
251 * Wrapper for the TCP established input helper hook.
254 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to)
256 struct tcp_hhook_data hhook_data;
258 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) {
263 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data,
269 * CC wrapper hook functions
272 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t type)
274 INP_WLOCK_ASSERT(tp->t_inpcb);
276 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th);
277 if (tp->snd_cwnd == min(tp->snd_cwnd, tp->snd_wnd))
278 tp->ccv->flags |= CCF_CWND_LIMITED;
280 tp->ccv->flags &= ~CCF_CWND_LIMITED;
282 if (type == CC_ACK) {
283 if (tp->snd_cwnd > tp->snd_ssthresh) {
284 tp->t_bytes_acked += min(tp->ccv->bytes_this_ack,
285 V_tcp_abc_l_var * tp->t_maxseg);
286 if (tp->t_bytes_acked >= tp->snd_cwnd) {
287 tp->t_bytes_acked -= tp->snd_cwnd;
288 tp->ccv->flags |= CCF_ABC_SENTAWND;
291 tp->ccv->flags &= ~CCF_ABC_SENTAWND;
292 tp->t_bytes_acked = 0;
296 if (CC_ALGO(tp)->ack_received != NULL) {
297 /* XXXLAS: Find a way to live without this */
298 tp->ccv->curack = th->th_ack;
299 CC_ALGO(tp)->ack_received(tp->ccv, type);
304 cc_conn_init(struct tcpcb *tp)
306 struct hc_metrics_lite metrics;
307 struct inpcb *inp = tp->t_inpcb;
310 INP_WLOCK_ASSERT(tp->t_inpcb);
312 tcp_hc_get(&inp->inp_inc, &metrics);
314 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) {
316 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
317 TCPSTAT_INC(tcps_usedrtt);
318 if (metrics.rmx_rttvar) {
319 tp->t_rttvar = metrics.rmx_rttvar;
320 TCPSTAT_INC(tcps_usedrttvar);
322 /* default variation is +- 1 rtt */
324 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
326 TCPT_RANGESET(tp->t_rxtcur,
327 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
328 tp->t_rttmin, TCPTV_REXMTMAX);
330 if (metrics.rmx_ssthresh) {
332 * There's some sort of gateway or interface
333 * buffer limit on the path. Use this to set
334 * the slow start threshhold, but set the
335 * threshold to no less than 2*mss.
337 tp->snd_ssthresh = max(2 * tp->t_maxseg, metrics.rmx_ssthresh);
338 TCPSTAT_INC(tcps_usedssthresh);
342 * Set the initial slow-start flight size.
344 * RFC3390 says only do this if SYN or SYN/ACK didn't got lost.
345 * XXX: We currently check only in syncache_socket for that.
347 if (V_tcp_do_rfc3390)
348 tp->snd_cwnd = min(4 * tp->t_maxseg,
349 max(2 * tp->t_maxseg, 4380));
351 tp->snd_cwnd = tp->t_maxseg;
353 if (CC_ALGO(tp)->conn_init != NULL)
354 CC_ALGO(tp)->conn_init(tp->ccv);
358 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type)
360 INP_WLOCK_ASSERT(tp->t_inpcb);
364 if (!IN_FASTRECOVERY(tp->t_flags)) {
365 tp->snd_recover = tp->snd_max;
366 if (tp->t_flags & TF_ECN_PERMIT)
367 tp->t_flags |= TF_ECN_SND_CWR;
371 if (!IN_CONGRECOVERY(tp->t_flags)) {
372 TCPSTAT_INC(tcps_ecn_rcwnd);
373 tp->snd_recover = tp->snd_max;
374 if (tp->t_flags & TF_ECN_PERMIT)
375 tp->t_flags |= TF_ECN_SND_CWR;
380 tp->t_bytes_acked = 0;
381 EXIT_RECOVERY(tp->t_flags);
382 tp->snd_ssthresh = max(2, min(tp->snd_wnd, tp->snd_cwnd) / 2 /
383 tp->t_maxseg) * tp->t_maxseg;
384 tp->snd_cwnd = tp->t_maxseg;
387 TCPSTAT_INC(tcps_sndrexmitbad);
388 /* RTO was unnecessary, so reset everything. */
389 tp->snd_cwnd = tp->snd_cwnd_prev;
390 tp->snd_ssthresh = tp->snd_ssthresh_prev;
391 tp->snd_recover = tp->snd_recover_prev;
392 if (tp->t_flags & TF_WASFRECOVERY)
393 ENTER_FASTRECOVERY(tp->t_flags);
394 if (tp->t_flags & TF_WASCRECOVERY)
395 ENTER_CONGRECOVERY(tp->t_flags);
396 tp->snd_nxt = tp->snd_max;
397 tp->t_flags &= ~TF_PREVVALID;
402 if (CC_ALGO(tp)->cong_signal != NULL) {
404 tp->ccv->curack = th->th_ack;
405 CC_ALGO(tp)->cong_signal(tp->ccv, type);
410 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th)
412 INP_WLOCK_ASSERT(tp->t_inpcb);
414 /* XXXLAS: KASSERT that we're in recovery? */
416 if (CC_ALGO(tp)->post_recovery != NULL) {
417 tp->ccv->curack = th->th_ack;
418 CC_ALGO(tp)->post_recovery(tp->ccv);
420 /* XXXLAS: EXIT_RECOVERY ? */
421 tp->t_bytes_acked = 0;
425 tcp_fields_to_host(struct tcphdr *th)
428 th->th_seq = ntohl(th->th_seq);
429 th->th_ack = ntohl(th->th_ack);
430 th->th_win = ntohs(th->th_win);
431 th->th_urp = ntohs(th->th_urp);
436 tcp_fields_to_net(struct tcphdr *th)
439 th->th_seq = htonl(th->th_seq);
440 th->th_ack = htonl(th->th_ack);
441 th->th_win = htons(th->th_win);
442 th->th_urp = htons(th->th_urp);
446 tcp_signature_verify_input(struct mbuf *m, int off0, int tlen, int optlen,
447 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag)
451 tcp_fields_to_net(th);
452 ret = tcp_signature_verify(m, off0, tlen, optlen, to, th, tcpbflag);
453 tcp_fields_to_host(th);
458 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
460 #define ND6_HINT(tp) \
462 if ((tp) && (tp)->t_inpcb && \
463 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \
464 nd6_nud_hint(NULL, NULL, 0); \
471 * Indicate whether this ack should be delayed. We can delay the ack if
472 * - there is no delayed ack timer in progress and
473 * - our last ack wasn't a 0-sized window. We never want to delay
474 * the ack that opens up a 0-sized window and
475 * - delayed acks are enabled or
476 * - this is a half-synchronized T/TCP connection.
478 #define DELAY_ACK(tp) \
479 ((!tcp_timer_active(tp, TT_DELACK) && \
480 (tp->t_flags & TF_RXWIN0SENT) == 0) && \
481 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
484 * TCP input handling is split into multiple parts:
485 * tcp6_input is a thin wrapper around tcp_input for the extended
486 * ip6_protox[] call format in ip6_input
487 * tcp_input handles primary segment validation, inpcb lookup and
488 * SYN processing on listen sockets
489 * tcp_do_segment processes the ACK and text of the segment for
490 * establishing, established and closing connections
494 tcp6_input(struct mbuf **mp, int *offp, int proto)
496 struct mbuf *m = *mp;
497 struct in6_ifaddr *ia6;
499 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
502 * draft-itojun-ipv6-tcp-to-anycast
503 * better place to put this in?
505 ia6 = ip6_getdstifaddr(m);
506 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
509 ifa_free(&ia6->ia_ifa);
510 ip6 = mtod(m, struct ip6_hdr *);
511 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
512 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
522 tcp_input(struct mbuf *m, int off0)
524 struct tcphdr *th = NULL;
525 struct ip *ip = NULL;
529 struct inpcb *inp = NULL;
530 struct tcpcb *tp = NULL;
531 struct socket *so = NULL;
540 int rstreason = 0; /* For badport_bandlim accounting purposes */
542 uint8_t sig_checked = 0;
545 #ifdef IPFIREWALL_FORWARD
546 struct m_tag *fwd_tag;
549 struct ip6_hdr *ip6 = NULL;
552 const void *ip6 = NULL;
554 struct tcpopt to; /* options in this segment */
555 char *s = NULL; /* address and port logging */
557 #define TI_UNLOCKED 1
562 * The size of tcp_saveipgen must be the size of the max ip header,
565 u_char tcp_saveipgen[IP6_HDR_LEN];
566 struct tcphdr tcp_savetcp;
571 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
575 TCPSTAT_INC(tcps_rcvtotal);
579 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */
580 ip6 = mtod(m, struct ip6_hdr *);
581 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
582 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) {
583 TCPSTAT_INC(tcps_rcvbadsum);
586 th = (struct tcphdr *)((caddr_t)ip6 + off0);
589 * Be proactive about unspecified IPv6 address in source.
590 * As we use all-zero to indicate unbounded/unconnected pcb,
591 * unspecified IPv6 address can be used to confuse us.
593 * Note that packets with unspecified IPv6 destination is
594 * already dropped in ip6_input.
596 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
602 #if defined(INET) && defined(INET6)
608 * Get IP and TCP header together in first mbuf.
609 * Note: IP leaves IP header in first mbuf.
611 if (off0 > sizeof (struct ip)) {
612 ip_stripoptions(m, (struct mbuf *)0);
613 off0 = sizeof(struct ip);
615 if (m->m_len < sizeof (struct tcpiphdr)) {
616 if ((m = m_pullup(m, sizeof (struct tcpiphdr)))
618 TCPSTAT_INC(tcps_rcvshort);
622 ip = mtod(m, struct ip *);
623 ipov = (struct ipovly *)ip;
624 th = (struct tcphdr *)((caddr_t)ip + off0);
627 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
628 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
629 th->th_sum = m->m_pkthdr.csum_data;
631 th->th_sum = in_pseudo(ip->ip_src.s_addr,
633 htonl(m->m_pkthdr.csum_data +
636 th->th_sum ^= 0xffff;
638 ipov->ih_len = (u_short)tlen;
639 ipov->ih_len = htons(ipov->ih_len);
643 * Checksum extended TCP header and data.
645 len = sizeof (struct ip) + tlen;
646 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
647 ipov->ih_len = (u_short)tlen;
648 ipov->ih_len = htons(ipov->ih_len);
649 th->th_sum = in_cksum(m, len);
652 TCPSTAT_INC(tcps_rcvbadsum);
655 /* Re-initialization for later version check */
656 ip->ip_v = IPVERSION;
662 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
664 #if defined(INET) && defined(INET6)
672 * Check that TCP offset makes sense,
673 * pull out TCP options and adjust length. XXX
675 off = th->th_off << 2;
676 if (off < sizeof (struct tcphdr) || off > tlen) {
677 TCPSTAT_INC(tcps_rcvbadoff);
680 tlen -= off; /* tlen is used instead of ti->ti_len */
681 if (off > sizeof (struct tcphdr)) {
684 IP6_EXTHDR_CHECK(m, off0, off, );
685 ip6 = mtod(m, struct ip6_hdr *);
686 th = (struct tcphdr *)((caddr_t)ip6 + off0);
689 #if defined(INET) && defined(INET6)
694 if (m->m_len < sizeof(struct ip) + off) {
695 if ((m = m_pullup(m, sizeof (struct ip) + off))
697 TCPSTAT_INC(tcps_rcvshort);
700 ip = mtod(m, struct ip *);
701 ipov = (struct ipovly *)ip;
702 th = (struct tcphdr *)((caddr_t)ip + off0);
706 optlen = off - sizeof (struct tcphdr);
707 optp = (u_char *)(th + 1);
709 thflags = th->th_flags;
712 * Convert TCP protocol specific fields to host format.
714 tcp_fields_to_host(th);
717 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options.
719 drop_hdrlen = off0 + off;
722 * Locate pcb for segment; if we're likely to add or remove a
723 * connection then first acquire pcbinfo lock. There are two cases
724 * where we might discover later we need a write lock despite the
725 * flags: ACKs moving a connection out of the syncache, and ACKs for
726 * a connection in TIMEWAIT.
728 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0) {
729 INP_INFO_WLOCK(&V_tcbinfo);
730 ti_locked = TI_WLOCKED;
732 ti_locked = TI_UNLOCKED;
736 if (ti_locked == TI_WLOCKED) {
737 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
739 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
743 #ifdef IPFIREWALL_FORWARD
745 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
747 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
748 #endif /* IPFIREWALL_FORWARD */
751 #ifdef IPFIREWALL_FORWARD
752 if (isipv6 && fwd_tag != NULL) {
753 struct sockaddr_in6 *next_hop6;
755 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1);
757 * Transparently forwarded. Pretend to be the destination.
758 * Already got one like this?
760 inp = in6_pcblookup_mbuf(&V_tcbinfo,
761 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport,
762 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m);
765 * It's new. Try to find the ambushing socket.
766 * Because we've rewritten the destination address,
767 * any hardware-generated hash is ignored.
769 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src,
770 th->th_sport, &next_hop6->sin6_addr,
771 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) :
772 th->th_dport, INPLOOKUP_WILDCARD |
773 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
775 /* Remove the tag from the packet. We don't need it anymore. */
776 m_tag_delete(m, fwd_tag);
778 #endif /* IPFIREWALL_FORWARD */
780 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src,
781 th->th_sport, &ip6->ip6_dst, th->th_dport,
782 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
783 m->m_pkthdr.rcvif, m);
786 #if defined(INET6) && defined(INET)
790 #ifdef IPFIREWALL_FORWARD
791 if (fwd_tag != NULL) {
792 struct sockaddr_in *next_hop;
794 next_hop = (struct sockaddr_in *)(fwd_tag+1);
796 * Transparently forwarded. Pretend to be the destination.
797 * already got one like this?
799 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport,
800 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB,
801 m->m_pkthdr.rcvif, m);
804 * It's new. Try to find the ambushing socket.
805 * Because we've rewritten the destination address,
806 * any hardware-generated hash is ignored.
808 inp = in_pcblookup(&V_tcbinfo, ip->ip_src,
809 th->th_sport, next_hop->sin_addr,
810 next_hop->sin_port ? ntohs(next_hop->sin_port) :
811 th->th_dport, INPLOOKUP_WILDCARD |
812 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
814 /* Remove the tag from the packet. We don't need it anymore. */
815 m_tag_delete(m, fwd_tag);
817 #endif /* IPFIREWALL_FORWARD */
818 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src,
819 th->th_sport, ip->ip_dst, th->th_dport,
820 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
821 m->m_pkthdr.rcvif, m);
825 * If the INPCB does not exist then all data in the incoming
826 * segment is discarded and an appropriate RST is sent back.
827 * XXX MRT Send RST using which routing table?
831 * Log communication attempts to ports that are not
834 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) ||
835 tcp_log_in_vain == 2) {
836 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6)))
837 log(LOG_INFO, "%s; %s: Connection attempt "
838 "to closed port\n", s, __func__);
841 * When blackholing do not respond with a RST but
842 * completely ignore the segment and drop it.
844 if ((V_blackhole == 1 && (thflags & TH_SYN)) ||
848 rstreason = BANDLIM_RST_CLOSEDPORT;
851 INP_WLOCK_ASSERT(inp);
852 if (!(inp->inp_flags & INP_HW_FLOWID)
853 && (m->m_flags & M_FLOWID)
854 && ((inp->inp_socket == NULL)
855 || !(inp->inp_socket->so_options & SO_ACCEPTCONN))) {
856 inp->inp_flags |= INP_HW_FLOWID;
857 inp->inp_flags &= ~INP_SW_FLOWID;
858 inp->inp_flowid = m->m_pkthdr.flowid;
862 if (isipv6 && ipsec6_in_reject(m, inp)) {
863 V_ipsec6stat.in_polvio++;
867 if (ipsec4_in_reject(m, inp) != 0) {
868 V_ipsec4stat.in_polvio++;
874 * Check the minimum TTL for socket.
876 if (inp->inp_ip_minttl != 0) {
878 if (isipv6 && inp->inp_ip_minttl > ip6->ip6_hlim)
882 if (inp->inp_ip_minttl > ip->ip_ttl)
887 * A previous connection in TIMEWAIT state is supposed to catch stray
888 * or duplicate segments arriving late. If this segment was a
889 * legitimate new connection attempt the old INPCB gets removed and
890 * we can try again to find a listening socket.
892 * At this point, due to earlier optimism, we may hold only an inpcb
893 * lock, and not the inpcbinfo write lock. If so, we need to try to
894 * acquire it, or if that fails, acquire a reference on the inpcb,
895 * drop all locks, acquire a global write lock, and then re-acquire
896 * the inpcb lock. We may at that point discover that another thread
897 * has tried to free the inpcb, in which case we need to loop back
898 * and try to find a new inpcb to deliver to.
900 * XXXRW: It may be time to rethink timewait locking.
903 if (inp->inp_flags & INP_TIMEWAIT) {
904 if (ti_locked == TI_UNLOCKED) {
905 if (INP_INFO_TRY_WLOCK(&V_tcbinfo) == 0) {
908 INP_INFO_WLOCK(&V_tcbinfo);
909 ti_locked = TI_WLOCKED;
911 if (in_pcbrele_wlocked(inp)) {
916 ti_locked = TI_WLOCKED;
918 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
920 if (thflags & TH_SYN)
921 tcp_dooptions(&to, optp, optlen, TO_SYN);
923 * NB: tcp_twcheck unlocks the INP and frees the mbuf.
925 if (tcp_twcheck(inp, &to, th, m, tlen))
927 INP_INFO_WUNLOCK(&V_tcbinfo);
931 * The TCPCB may no longer exist if the connection is winding
932 * down or it is in the CLOSED state. Either way we drop the
933 * segment and send an appropriate response.
936 if (tp == NULL || tp->t_state == TCPS_CLOSED) {
937 rstreason = BANDLIM_RST_CLOSEDPORT;
942 * We've identified a valid inpcb, but it could be that we need an
943 * inpcbinfo write lock but don't hold it. In this case, attempt to
944 * acquire using the same strategy as the TIMEWAIT case above. If we
945 * relock, we have to jump back to 'relocked' as the connection might
946 * now be in TIMEWAIT.
949 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0)
950 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
952 if (tp->t_state != TCPS_ESTABLISHED) {
953 if (ti_locked == TI_UNLOCKED) {
954 if (INP_INFO_TRY_WLOCK(&V_tcbinfo) == 0) {
957 INP_INFO_WLOCK(&V_tcbinfo);
958 ti_locked = TI_WLOCKED;
960 if (in_pcbrele_wlocked(inp)) {
966 ti_locked = TI_WLOCKED;
968 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
972 INP_WLOCK_ASSERT(inp);
973 if (mac_inpcb_check_deliver(inp, m))
976 so = inp->inp_socket;
977 KASSERT(so != NULL, ("%s: so == NULL", __func__));
979 if (so->so_options & SO_DEBUG) {
980 ostate = tp->t_state;
983 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
986 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
989 #endif /* TCPDEBUG */
991 * When the socket is accepting connections (the INPCB is in LISTEN
992 * state) we look into the SYN cache if this is a new connection
993 * attempt or the completion of a previous one. Because listen
994 * sockets are never in TCPS_ESTABLISHED, the V_tcbinfo lock will be
997 if (so->so_options & SO_ACCEPTCONN) {
998 struct in_conninfo inc;
1000 KASSERT(tp->t_state == TCPS_LISTEN, ("%s: so accepting but "
1001 "tp not listening", __func__));
1002 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1004 bzero(&inc, sizeof(inc));
1007 inc.inc_flags |= INC_ISIPV6;
1008 inc.inc6_faddr = ip6->ip6_src;
1009 inc.inc6_laddr = ip6->ip6_dst;
1013 inc.inc_faddr = ip->ip_src;
1014 inc.inc_laddr = ip->ip_dst;
1016 inc.inc_fport = th->th_sport;
1017 inc.inc_lport = th->th_dport;
1018 inc.inc_fibnum = so->so_fibnum;
1021 * Check for an existing connection attempt in syncache if
1022 * the flag is only ACK. A successful lookup creates a new
1023 * socket appended to the listen queue in SYN_RECEIVED state.
1025 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
1027 * Parse the TCP options here because
1028 * syncookies need access to the reflected
1031 tcp_dooptions(&to, optp, optlen, 0);
1033 * NB: syncache_expand() doesn't unlock
1034 * inp and tcpinfo locks.
1036 if (!syncache_expand(&inc, &to, th, &so, m)) {
1038 * No syncache entry or ACK was not
1039 * for our SYN/ACK. Send a RST.
1040 * NB: syncache did its own logging
1041 * of the failure cause.
1043 rstreason = BANDLIM_RST_OPENPORT;
1048 * We completed the 3-way handshake
1049 * but could not allocate a socket
1050 * either due to memory shortage,
1051 * listen queue length limits or
1052 * global socket limits. Send RST
1053 * or wait and have the remote end
1054 * retransmit the ACK for another
1057 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1058 log(LOG_DEBUG, "%s; %s: Listen socket: "
1059 "Socket allocation failed due to "
1060 "limits or memory shortage, %s\n",
1062 V_tcp_sc_rst_sock_fail ?
1063 "sending RST" : "try again");
1064 if (V_tcp_sc_rst_sock_fail) {
1065 rstreason = BANDLIM_UNLIMITED;
1071 * Socket is created in state SYN_RECEIVED.
1072 * Unlock the listen socket, lock the newly
1073 * created socket and update the tp variable.
1075 INP_WUNLOCK(inp); /* listen socket */
1076 inp = sotoinpcb(so);
1077 INP_WLOCK(inp); /* new connection */
1078 tp = intotcpcb(inp);
1079 KASSERT(tp->t_state == TCPS_SYN_RECEIVED,
1080 ("%s: ", __func__));
1081 #ifdef TCP_SIGNATURE
1082 if (sig_checked == 0) {
1083 tcp_dooptions(&to, optp, optlen,
1084 (thflags & TH_SYN) ? TO_SYN : 0);
1085 if (!tcp_signature_verify_input(m, off0, tlen,
1086 optlen, &to, th, tp->t_flags)) {
1089 * In SYN_SENT state if it receives an
1090 * RST, it is allowed for further
1093 if ((thflags & TH_RST) == 0 ||
1094 (tp->t_state == TCPS_SYN_SENT) == 0)
1102 * Process the segment and the data it
1103 * contains. tcp_do_segment() consumes
1104 * the mbuf chain and unlocks the inpcb.
1106 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen,
1108 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1112 * Segment flag validation for new connection attempts:
1114 * Our (SYN|ACK) response was rejected.
1115 * Check with syncache and remove entry to prevent
1118 * NB: syncache_chkrst does its own logging of failure
1121 if (thflags & TH_RST) {
1122 syncache_chkrst(&inc, th);
1126 * We can't do anything without SYN.
1128 if ((thflags & TH_SYN) == 0) {
1129 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1130 log(LOG_DEBUG, "%s; %s: Listen socket: "
1131 "SYN is missing, segment ignored\n",
1133 TCPSTAT_INC(tcps_badsyn);
1137 * (SYN|ACK) is bogus on a listen socket.
1139 if (thflags & TH_ACK) {
1140 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1141 log(LOG_DEBUG, "%s; %s: Listen socket: "
1142 "SYN|ACK invalid, segment rejected\n",
1144 syncache_badack(&inc); /* XXX: Not needed! */
1145 TCPSTAT_INC(tcps_badsyn);
1146 rstreason = BANDLIM_RST_OPENPORT;
1150 * If the drop_synfin option is enabled, drop all
1151 * segments with both the SYN and FIN bits set.
1152 * This prevents e.g. nmap from identifying the
1154 * XXX: Poor reasoning. nmap has other methods
1155 * and is constantly refining its stack detection
1157 * XXX: This is a violation of the TCP specification
1158 * and was used by RFC1644.
1160 if ((thflags & TH_FIN) && V_drop_synfin) {
1161 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1162 log(LOG_DEBUG, "%s; %s: Listen socket: "
1163 "SYN|FIN segment ignored (based on "
1164 "sysctl setting)\n", s, __func__);
1165 TCPSTAT_INC(tcps_badsyn);
1169 * Segment's flags are (SYN) or (SYN|FIN).
1171 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored
1172 * as they do not affect the state of the TCP FSM.
1173 * The data pointed to by TH_URG and th_urp is ignored.
1175 KASSERT((thflags & (TH_RST|TH_ACK)) == 0,
1176 ("%s: Listen socket: TH_RST or TH_ACK set", __func__));
1177 KASSERT(thflags & (TH_SYN),
1178 ("%s: Listen socket: TH_SYN not set", __func__));
1181 * If deprecated address is forbidden,
1182 * we do not accept SYN to deprecated interface
1183 * address to prevent any new inbound connection from
1184 * getting established.
1185 * When we do not accept SYN, we send a TCP RST,
1186 * with deprecated source address (instead of dropping
1187 * it). We compromise it as it is much better for peer
1188 * to send a RST, and RST will be the final packet
1191 * If we do not forbid deprecated addresses, we accept
1192 * the SYN packet. RFC2462 does not suggest dropping
1194 * If we decipher RFC2462 5.5.4, it says like this:
1195 * 1. use of deprecated addr with existing
1196 * communication is okay - "SHOULD continue to be
1198 * 2. use of it with new communication:
1199 * (2a) "SHOULD NOT be used if alternate address
1200 * with sufficient scope is available"
1201 * (2b) nothing mentioned otherwise.
1202 * Here we fall into (2b) case as we have no choice in
1203 * our source address selection - we must obey the peer.
1205 * The wording in RFC2462 is confusing, and there are
1206 * multiple description text for deprecated address
1207 * handling - worse, they are not exactly the same.
1208 * I believe 5.5.4 is the best one, so we follow 5.5.4.
1210 if (isipv6 && !V_ip6_use_deprecated) {
1211 struct in6_ifaddr *ia6;
1213 ia6 = ip6_getdstifaddr(m);
1215 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1216 ifa_free(&ia6->ia_ifa);
1217 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1218 log(LOG_DEBUG, "%s; %s: Listen socket: "
1219 "Connection attempt to deprecated "
1220 "IPv6 address rejected\n",
1222 rstreason = BANDLIM_RST_OPENPORT;
1225 ifa_free(&ia6->ia_ifa);
1229 * Basic sanity checks on incoming SYN requests:
1230 * Don't respond if the destination is a link layer
1231 * broadcast according to RFC1122 4.2.3.10, p. 104.
1232 * If it is from this socket it must be forged.
1233 * Don't respond if the source or destination is a
1234 * global or subnet broad- or multicast address.
1235 * Note that it is quite possible to receive unicast
1236 * link-layer packets with a broadcast IP address. Use
1237 * in_broadcast() to find them.
1239 if (m->m_flags & (M_BCAST|M_MCAST)) {
1240 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1241 log(LOG_DEBUG, "%s; %s: Listen socket: "
1242 "Connection attempt from broad- or multicast "
1243 "link layer address ignored\n", s, __func__);
1248 if (th->th_dport == th->th_sport &&
1249 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) {
1250 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1251 log(LOG_DEBUG, "%s; %s: Listen socket: "
1252 "Connection attempt to/from self "
1253 "ignored\n", s, __func__);
1256 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1257 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) {
1258 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1259 log(LOG_DEBUG, "%s; %s: Listen socket: "
1260 "Connection attempt from/to multicast "
1261 "address ignored\n", s, __func__);
1266 #if defined(INET) && defined(INET6)
1271 if (th->th_dport == th->th_sport &&
1272 ip->ip_dst.s_addr == ip->ip_src.s_addr) {
1273 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1274 log(LOG_DEBUG, "%s; %s: Listen socket: "
1275 "Connection attempt from/to self "
1276 "ignored\n", s, __func__);
1279 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1280 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1281 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1282 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
1283 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1284 log(LOG_DEBUG, "%s; %s: Listen socket: "
1285 "Connection attempt from/to broad- "
1286 "or multicast address ignored\n",
1293 * SYN appears to be valid. Create compressed TCP state
1297 if (so->so_options & SO_DEBUG)
1298 tcp_trace(TA_INPUT, ostate, tp,
1299 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1301 tcp_dooptions(&to, optp, optlen, TO_SYN);
1302 syncache_add(&inc, &to, th, inp, &so, m);
1304 * Entry added to syncache and mbuf consumed.
1305 * Everything already unlocked by syncache_add().
1307 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1311 #ifdef TCP_SIGNATURE
1312 if (sig_checked == 0) {
1313 tcp_dooptions(&to, optp, optlen,
1314 (thflags & TH_SYN) ? TO_SYN : 0);
1315 if (!tcp_signature_verify_input(m, off0, tlen, optlen, &to,
1319 * In SYN_SENT state if it receives an RST, it is
1320 * allowed for further processing.
1322 if ((thflags & TH_RST) == 0 ||
1323 (tp->t_state == TCPS_SYN_SENT) == 0)
1331 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later
1332 * state. tcp_do_segment() always consumes the mbuf chain, unlocks
1333 * the inpcb, and unlocks pcbinfo.
1335 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos, ti_locked);
1336 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1340 if (ti_locked == TI_WLOCKED) {
1341 INP_INFO_WUNLOCK(&V_tcbinfo);
1342 ti_locked = TI_UNLOCKED;
1346 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset "
1347 "ti_locked: %d", __func__, ti_locked));
1348 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1353 tcp_dropwithreset(m, th, tp, tlen, rstreason);
1356 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
1357 m = NULL; /* mbuf chain got consumed. */
1361 if (ti_locked == TI_WLOCKED) {
1362 INP_INFO_WUNLOCK(&V_tcbinfo);
1363 ti_locked = TI_UNLOCKED;
1367 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock "
1368 "ti_locked: %d", __func__, ti_locked));
1369 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1377 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1385 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so,
1386 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos,
1389 int thflags, acked, ourfinisacked, needoutput = 0;
1390 int rstreason, todrop, win;
1396 * The size of tcp_saveipgen must be the size of the max ip header,
1399 u_char tcp_saveipgen[IP6_HDR_LEN];
1400 struct tcphdr tcp_savetcp;
1403 thflags = th->th_flags;
1404 tp->sackhint.last_sack_ack = 0;
1407 * If this is either a state-changing packet or current state isn't
1408 * established, we require a write lock on tcbinfo. Otherwise, we
1409 * allow either a read lock or a write lock, as we may have acquired
1410 * a write lock due to a race.
1412 * Require a global write lock for SYN/FIN/RST segments or
1413 * non-established connections; otherwise accept either a read or
1414 * write lock, as we may have conservatively acquired a write lock in
1415 * certain cases in tcp_input() (is this still true?). Currently we
1416 * will never enter with no lock, so we try to drop it quickly in the
1417 * common pure ack/pure data cases.
1419 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 ||
1420 tp->t_state != TCPS_ESTABLISHED) {
1421 KASSERT(ti_locked == TI_WLOCKED, ("%s ti_locked %d for "
1422 "SYN/FIN/RST/!EST", __func__, ti_locked));
1423 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1426 if (ti_locked == TI_WLOCKED)
1427 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1429 KASSERT(ti_locked == TI_UNLOCKED, ("%s: EST "
1430 "ti_locked: %d", __func__, ti_locked));
1431 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1435 INP_WLOCK_ASSERT(tp->t_inpcb);
1436 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN",
1438 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT",
1442 * Segment received on connection.
1443 * Reset idle time and keep-alive timer.
1444 * XXX: This should be done after segment
1445 * validation to ignore broken/spoofed segs.
1447 tp->t_rcvtime = ticks;
1448 if (TCPS_HAVEESTABLISHED(tp->t_state))
1449 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
1452 * Unscale the window into a 32-bit value.
1453 * For the SYN_SENT state the scale is zero.
1455 tiwin = th->th_win << tp->snd_scale;
1458 * TCP ECN processing.
1460 if (tp->t_flags & TF_ECN_PERMIT) {
1461 if (thflags & TH_CWR)
1462 tp->t_flags &= ~TF_ECN_SND_ECE;
1463 switch (iptos & IPTOS_ECN_MASK) {
1465 tp->t_flags |= TF_ECN_SND_ECE;
1466 TCPSTAT_INC(tcps_ecn_ce);
1468 case IPTOS_ECN_ECT0:
1469 TCPSTAT_INC(tcps_ecn_ect0);
1471 case IPTOS_ECN_ECT1:
1472 TCPSTAT_INC(tcps_ecn_ect1);
1475 /* Congestion experienced. */
1476 if (thflags & TH_ECE) {
1477 cc_cong_signal(tp, th, CC_ECN);
1482 * Parse options on any incoming segment.
1484 tcp_dooptions(&to, (u_char *)(th + 1),
1485 (th->th_off << 2) - sizeof(struct tcphdr),
1486 (thflags & TH_SYN) ? TO_SYN : 0);
1489 * If echoed timestamp is later than the current time,
1490 * fall back to non RFC1323 RTT calculation. Normalize
1491 * timestamp if syncookies were used when this connection
1494 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1495 to.to_tsecr -= tp->ts_offset;
1496 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks()))
1501 * Process options only when we get SYN/ACK back. The SYN case
1502 * for incoming connections is handled in tcp_syncache.
1503 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
1504 * or <SYN,ACK>) segment itself is never scaled.
1505 * XXX this is traditional behavior, may need to be cleaned up.
1507 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1508 if ((to.to_flags & TOF_SCALE) &&
1509 (tp->t_flags & TF_REQ_SCALE)) {
1510 tp->t_flags |= TF_RCVD_SCALE;
1511 tp->snd_scale = to.to_wscale;
1514 * Initial send window. It will be updated with
1515 * the next incoming segment to the scaled value.
1517 tp->snd_wnd = th->th_win;
1518 if (to.to_flags & TOF_TS) {
1519 tp->t_flags |= TF_RCVD_TSTMP;
1520 tp->ts_recent = to.to_tsval;
1521 tp->ts_recent_age = tcp_ts_getticks();
1523 if (to.to_flags & TOF_MSS)
1524 tcp_mss(tp, to.to_mss);
1525 if ((tp->t_flags & TF_SACK_PERMIT) &&
1526 (to.to_flags & TOF_SACKPERM) == 0)
1527 tp->t_flags &= ~TF_SACK_PERMIT;
1531 * Header prediction: check for the two common cases
1532 * of a uni-directional data xfer. If the packet has
1533 * no control flags, is in-sequence, the window didn't
1534 * change and we're not retransmitting, it's a
1535 * candidate. If the length is zero and the ack moved
1536 * forward, we're the sender side of the xfer. Just
1537 * free the data acked & wake any higher level process
1538 * that was blocked waiting for space. If the length
1539 * is non-zero and the ack didn't move, we're the
1540 * receiver side. If we're getting packets in-order
1541 * (the reassembly queue is empty), add the data to
1542 * the socket buffer and note that we need a delayed ack.
1543 * Make sure that the hidden state-flags are also off.
1544 * Since we check for TCPS_ESTABLISHED first, it can only
1547 if (tp->t_state == TCPS_ESTABLISHED &&
1548 th->th_seq == tp->rcv_nxt &&
1549 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1550 tp->snd_nxt == tp->snd_max &&
1551 tiwin && tiwin == tp->snd_wnd &&
1552 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1553 LIST_EMPTY(&tp->t_segq) &&
1554 ((to.to_flags & TOF_TS) == 0 ||
1555 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) {
1558 * If last ACK falls within this segment's sequence numbers,
1559 * record the timestamp.
1560 * NOTE that the test is modified according to the latest
1561 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1563 if ((to.to_flags & TOF_TS) != 0 &&
1564 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1565 tp->ts_recent_age = tcp_ts_getticks();
1566 tp->ts_recent = to.to_tsval;
1570 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1571 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1572 !IN_RECOVERY(tp->t_flags) &&
1573 (to.to_flags & TOF_SACK) == 0 &&
1574 TAILQ_EMPTY(&tp->snd_holes)) {
1576 * This is a pure ack for outstanding data.
1578 if (ti_locked == TI_WLOCKED)
1579 INP_INFO_WUNLOCK(&V_tcbinfo);
1580 ti_locked = TI_UNLOCKED;
1582 TCPSTAT_INC(tcps_predack);
1585 * "bad retransmit" recovery.
1587 if (tp->t_rxtshift == 1 &&
1588 tp->t_flags & TF_PREVVALID &&
1589 (int)(ticks - tp->t_badrxtwin) < 0) {
1590 cc_cong_signal(tp, th, CC_RTO_ERR);
1594 * Recalculate the transmit timer / rtt.
1596 * Some boxes send broken timestamp replies
1597 * during the SYN+ACK phase, ignore
1598 * timestamps of 0 or we could calculate a
1599 * huge RTT and blow up the retransmit timer.
1601 if ((to.to_flags & TOF_TS) != 0 &&
1605 t = tcp_ts_getticks() - to.to_tsecr;
1606 if (!tp->t_rttlow || tp->t_rttlow > t)
1609 TCP_TS_TO_TICKS(t) + 1);
1610 } else if (tp->t_rtttime &&
1611 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1612 if (!tp->t_rttlow ||
1613 tp->t_rttlow > ticks - tp->t_rtttime)
1614 tp->t_rttlow = ticks - tp->t_rtttime;
1616 ticks - tp->t_rtttime);
1618 acked = BYTES_THIS_ACK(tp, th);
1620 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
1621 hhook_run_tcp_est_in(tp, th, &to);
1623 TCPSTAT_INC(tcps_rcvackpack);
1624 TCPSTAT_ADD(tcps_rcvackbyte, acked);
1625 sbdrop(&so->so_snd, acked);
1626 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1627 SEQ_LEQ(th->th_ack, tp->snd_recover))
1628 tp->snd_recover = th->th_ack - 1;
1631 * Let the congestion control algorithm update
1632 * congestion control related information. This
1633 * typically means increasing the congestion
1636 cc_ack_received(tp, th, CC_ACK);
1638 tp->snd_una = th->th_ack;
1640 * Pull snd_wl2 up to prevent seq wrap relative
1643 tp->snd_wl2 = th->th_ack;
1646 ND6_HINT(tp); /* Some progress has been made. */
1649 * If all outstanding data are acked, stop
1650 * retransmit timer, otherwise restart timer
1651 * using current (possibly backed-off) value.
1652 * If process is waiting for space,
1653 * wakeup/selwakeup/signal. If data
1654 * are ready to send, let tcp_output
1655 * decide between more output or persist.
1658 if (so->so_options & SO_DEBUG)
1659 tcp_trace(TA_INPUT, ostate, tp,
1660 (void *)tcp_saveipgen,
1663 if (tp->snd_una == tp->snd_max)
1664 tcp_timer_activate(tp, TT_REXMT, 0);
1665 else if (!tcp_timer_active(tp, TT_PERSIST))
1666 tcp_timer_activate(tp, TT_REXMT,
1669 if (so->so_snd.sb_cc)
1670 (void) tcp_output(tp);
1673 } else if (th->th_ack == tp->snd_una &&
1674 tlen <= sbspace(&so->so_rcv)) {
1675 int newsize = 0; /* automatic sockbuf scaling */
1678 * This is a pure, in-sequence data packet with
1679 * nothing on the reassembly queue and we have enough
1680 * buffer space to take it.
1682 if (ti_locked == TI_WLOCKED)
1683 INP_INFO_WUNLOCK(&V_tcbinfo);
1684 ti_locked = TI_UNLOCKED;
1686 /* Clean receiver SACK report if present */
1687 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks)
1688 tcp_clean_sackreport(tp);
1689 TCPSTAT_INC(tcps_preddat);
1690 tp->rcv_nxt += tlen;
1692 * Pull snd_wl1 up to prevent seq wrap relative to
1695 tp->snd_wl1 = th->th_seq;
1697 * Pull rcv_up up to prevent seq wrap relative to
1700 tp->rcv_up = tp->rcv_nxt;
1701 TCPSTAT_INC(tcps_rcvpack);
1702 TCPSTAT_ADD(tcps_rcvbyte, tlen);
1703 ND6_HINT(tp); /* Some progress has been made */
1705 if (so->so_options & SO_DEBUG)
1706 tcp_trace(TA_INPUT, ostate, tp,
1707 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1710 * Automatic sizing of receive socket buffer. Often the send
1711 * buffer size is not optimally adjusted to the actual network
1712 * conditions at hand (delay bandwidth product). Setting the
1713 * buffer size too small limits throughput on links with high
1714 * bandwidth and high delay (eg. trans-continental/oceanic links).
1716 * On the receive side the socket buffer memory is only rarely
1717 * used to any significant extent. This allows us to be much
1718 * more aggressive in scaling the receive socket buffer. For
1719 * the case that the buffer space is actually used to a large
1720 * extent and we run out of kernel memory we can simply drop
1721 * the new segments; TCP on the sender will just retransmit it
1722 * later. Setting the buffer size too big may only consume too
1723 * much kernel memory if the application doesn't read() from
1724 * the socket or packet loss or reordering makes use of the
1727 * The criteria to step up the receive buffer one notch are:
1728 * 1. the number of bytes received during the time it takes
1729 * one timestamp to be reflected back to us (the RTT);
1730 * 2. received bytes per RTT is within seven eighth of the
1731 * current socket buffer size;
1732 * 3. receive buffer size has not hit maximal automatic size;
1734 * This algorithm does one step per RTT at most and only if
1735 * we receive a bulk stream w/o packet losses or reorderings.
1736 * Shrinking the buffer during idle times is not necessary as
1737 * it doesn't consume any memory when idle.
1739 * TODO: Only step up if the application is actually serving
1740 * the buffer to better manage the socket buffer resources.
1742 if (V_tcp_do_autorcvbuf &&
1744 (so->so_rcv.sb_flags & SB_AUTOSIZE)) {
1745 if (TSTMP_GT(to.to_tsecr, tp->rfbuf_ts) &&
1746 to.to_tsecr - tp->rfbuf_ts < hz) {
1748 (so->so_rcv.sb_hiwat / 8 * 7) &&
1749 so->so_rcv.sb_hiwat <
1750 V_tcp_autorcvbuf_max) {
1752 min(so->so_rcv.sb_hiwat +
1753 V_tcp_autorcvbuf_inc,
1754 V_tcp_autorcvbuf_max);
1756 /* Start over with next RTT. */
1760 tp->rfbuf_cnt += tlen; /* add up */
1763 /* Add data to socket buffer. */
1764 SOCKBUF_LOCK(&so->so_rcv);
1765 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1769 * Set new socket buffer size.
1770 * Give up when limit is reached.
1773 if (!sbreserve_locked(&so->so_rcv,
1775 so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1776 m_adj(m, drop_hdrlen); /* delayed header drop */
1777 sbappendstream_locked(&so->so_rcv, m);
1779 /* NB: sorwakeup_locked() does an implicit unlock. */
1780 sorwakeup_locked(so);
1781 if (DELAY_ACK(tp)) {
1782 tp->t_flags |= TF_DELACK;
1784 tp->t_flags |= TF_ACKNOW;
1792 * Calculate amount of space in receive window,
1793 * and then do TCP input processing.
1794 * Receive window is amount of space in rcv queue,
1795 * but not less than advertised window.
1797 win = sbspace(&so->so_rcv);
1800 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1802 /* Reset receive buffer auto scaling when not in bulk receive mode. */
1806 switch (tp->t_state) {
1809 * If the state is SYN_RECEIVED:
1810 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1812 case TCPS_SYN_RECEIVED:
1813 if ((thflags & TH_ACK) &&
1814 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1815 SEQ_GT(th->th_ack, tp->snd_max))) {
1816 rstreason = BANDLIM_RST_OPENPORT;
1822 * If the state is SYN_SENT:
1823 * if seg contains an ACK, but not for our SYN, drop the input.
1824 * if seg contains a RST, then drop the connection.
1825 * if seg does not contain SYN, then drop it.
1826 * Otherwise this is an acceptable SYN segment
1827 * initialize tp->rcv_nxt and tp->irs
1828 * if seg contains ack then advance tp->snd_una
1829 * if seg contains an ECE and ECN support is enabled, the stream
1831 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1832 * arrange for segment to be acked (eventually)
1833 * continue processing rest of data/controls, beginning with URG
1836 if ((thflags & TH_ACK) &&
1837 (SEQ_LEQ(th->th_ack, tp->iss) ||
1838 SEQ_GT(th->th_ack, tp->snd_max))) {
1839 rstreason = BANDLIM_UNLIMITED;
1842 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST))
1843 tp = tcp_drop(tp, ECONNREFUSED);
1844 if (thflags & TH_RST)
1846 if (!(thflags & TH_SYN))
1849 tp->irs = th->th_seq;
1851 if (thflags & TH_ACK) {
1852 TCPSTAT_INC(tcps_connects);
1855 mac_socketpeer_set_from_mbuf(m, so);
1857 /* Do window scaling on this connection? */
1858 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1859 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1860 tp->rcv_scale = tp->request_r_scale;
1862 tp->rcv_adv += imin(tp->rcv_wnd,
1863 TCP_MAXWIN << tp->rcv_scale);
1864 tp->snd_una++; /* SYN is acked */
1866 * If there's data, delay ACK; if there's also a FIN
1867 * ACKNOW will be turned on later.
1869 if (DELAY_ACK(tp) && tlen != 0)
1870 tcp_timer_activate(tp, TT_DELACK,
1873 tp->t_flags |= TF_ACKNOW;
1875 if ((thflags & TH_ECE) && V_tcp_do_ecn) {
1876 tp->t_flags |= TF_ECN_PERMIT;
1877 TCPSTAT_INC(tcps_ecn_shs);
1881 * Received <SYN,ACK> in SYN_SENT[*] state.
1883 * SYN_SENT --> ESTABLISHED
1884 * SYN_SENT* --> FIN_WAIT_1
1886 tp->t_starttime = ticks;
1887 if (tp->t_flags & TF_NEEDFIN) {
1888 tp->t_state = TCPS_FIN_WAIT_1;
1889 tp->t_flags &= ~TF_NEEDFIN;
1892 tp->t_state = TCPS_ESTABLISHED;
1894 tcp_timer_activate(tp, TT_KEEP,
1899 * Received initial SYN in SYN-SENT[*] state =>
1900 * simultaneous open. If segment contains CC option
1901 * and there is a cached CC, apply TAO test.
1902 * If it succeeds, connection is * half-synchronized.
1903 * Otherwise, do 3-way handshake:
1904 * SYN-SENT -> SYN-RECEIVED
1905 * SYN-SENT* -> SYN-RECEIVED*
1906 * If there was no CC option, clear cached CC value.
1908 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
1909 tcp_timer_activate(tp, TT_REXMT, 0);
1910 tp->t_state = TCPS_SYN_RECEIVED;
1913 KASSERT(ti_locked == TI_WLOCKED, ("%s: trimthenstep6: "
1914 "ti_locked %d", __func__, ti_locked));
1915 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1916 INP_WLOCK_ASSERT(tp->t_inpcb);
1919 * Advance th->th_seq to correspond to first data byte.
1920 * If data, trim to stay within window,
1921 * dropping FIN if necessary.
1924 if (tlen > tp->rcv_wnd) {
1925 todrop = tlen - tp->rcv_wnd;
1929 TCPSTAT_INC(tcps_rcvpackafterwin);
1930 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
1932 tp->snd_wl1 = th->th_seq - 1;
1933 tp->rcv_up = th->th_seq;
1935 * Client side of transaction: already sent SYN and data.
1936 * If the remote host used T/TCP to validate the SYN,
1937 * our data will be ACK'd; if so, enter normal data segment
1938 * processing in the middle of step 5, ack processing.
1939 * Otherwise, goto step 6.
1941 if (thflags & TH_ACK)
1947 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
1948 * do normal processing.
1950 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later.
1954 break; /* continue normal processing */
1958 * States other than LISTEN or SYN_SENT.
1959 * First check the RST flag and sequence number since reset segments
1960 * are exempt from the timestamp and connection count tests. This
1961 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
1962 * below which allowed reset segments in half the sequence space
1963 * to fall though and be processed (which gives forged reset
1964 * segments with a random sequence number a 50 percent chance of
1965 * killing a connection).
1966 * Then check timestamp, if present.
1967 * Then check the connection count, if present.
1968 * Then check that at least some bytes of segment are within
1969 * receive window. If segment begins before rcv_nxt,
1970 * drop leading data (and SYN); if nothing left, just ack.
1973 * If the RST bit is set, check the sequence number to see
1974 * if this is a valid reset segment.
1976 * In all states except SYN-SENT, all reset (RST) segments
1977 * are validated by checking their SEQ-fields. A reset is
1978 * valid if its sequence number is in the window.
1979 * Note: this does not take into account delayed ACKs, so
1980 * we should test against last_ack_sent instead of rcv_nxt.
1981 * The sequence number in the reset segment is normally an
1982 * echo of our outgoing acknowlegement numbers, but some hosts
1983 * send a reset with the sequence number at the rightmost edge
1984 * of our receive window, and we have to handle this case.
1985 * Note 2: Paul Watson's paper "Slipping in the Window" has shown
1986 * that brute force RST attacks are possible. To combat this,
1987 * we use a much stricter check while in the ESTABLISHED state,
1988 * only accepting RSTs where the sequence number is equal to
1989 * last_ack_sent. In all other states (the states in which a
1990 * RST is more likely), the more permissive check is used.
1991 * If we have multiple segments in flight, the initial reset
1992 * segment sequence numbers will be to the left of last_ack_sent,
1993 * but they will eventually catch up.
1994 * In any case, it never made sense to trim reset segments to
1995 * fit the receive window since RFC 1122 says:
1996 * 4.2.2.12 RST Segment: RFC-793 Section 3.4
1998 * A TCP SHOULD allow a received RST segment to include data.
2001 * It has been suggested that a RST segment could contain
2002 * ASCII text that encoded and explained the cause of the
2003 * RST. No standard has yet been established for such
2006 * If the reset segment passes the sequence number test examine
2008 * SYN_RECEIVED STATE:
2009 * If passive open, return to LISTEN state.
2010 * If active open, inform user that connection was refused.
2011 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
2012 * Inform user that connection was reset, and close tcb.
2013 * CLOSING, LAST_ACK STATES:
2016 * Drop the segment - see Stevens, vol. 2, p. 964 and
2019 if (thflags & TH_RST) {
2020 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) &&
2021 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
2022 switch (tp->t_state) {
2024 case TCPS_SYN_RECEIVED:
2025 so->so_error = ECONNREFUSED;
2028 case TCPS_ESTABLISHED:
2029 if (V_tcp_insecure_rst == 0 &&
2030 !(SEQ_GEQ(th->th_seq, tp->rcv_nxt - 1) &&
2031 SEQ_LEQ(th->th_seq, tp->rcv_nxt + 1)) &&
2032 !(SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) &&
2033 SEQ_LEQ(th->th_seq, tp->last_ack_sent + 1))) {
2034 TCPSTAT_INC(tcps_badrst);
2038 case TCPS_FIN_WAIT_1:
2039 case TCPS_FIN_WAIT_2:
2040 case TCPS_CLOSE_WAIT:
2041 so->so_error = ECONNRESET;
2043 KASSERT(ti_locked == TI_WLOCKED,
2044 ("tcp_do_segment: TH_RST 1 ti_locked %d",
2046 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2048 tp->t_state = TCPS_CLOSED;
2049 TCPSTAT_INC(tcps_drops);
2055 KASSERT(ti_locked == TI_WLOCKED,
2056 ("tcp_do_segment: TH_RST 2 ti_locked %d",
2058 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2068 * RFC 1323 PAWS: If we have a timestamp reply on this segment
2069 * and it's less than ts_recent, drop it.
2071 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
2072 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
2074 /* Check to see if ts_recent is over 24 days old. */
2075 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) {
2077 * Invalidate ts_recent. If this segment updates
2078 * ts_recent, the age will be reset later and ts_recent
2079 * will get a valid value. If it does not, setting
2080 * ts_recent to zero will at least satisfy the
2081 * requirement that zero be placed in the timestamp
2082 * echo reply when ts_recent isn't valid. The
2083 * age isn't reset until we get a valid ts_recent
2084 * because we don't want out-of-order segments to be
2085 * dropped when ts_recent is old.
2089 TCPSTAT_INC(tcps_rcvduppack);
2090 TCPSTAT_ADD(tcps_rcvdupbyte, tlen);
2091 TCPSTAT_INC(tcps_pawsdrop);
2099 * In the SYN-RECEIVED state, validate that the packet belongs to
2100 * this connection before trimming the data to fit the receive
2101 * window. Check the sequence number versus IRS since we know
2102 * the sequence numbers haven't wrapped. This is a partial fix
2103 * for the "LAND" DoS attack.
2105 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
2106 rstreason = BANDLIM_RST_OPENPORT;
2110 todrop = tp->rcv_nxt - th->th_seq;
2113 * If this is a duplicate SYN for our current connection,
2114 * advance over it and pretend and it's not a SYN.
2116 if (thflags & TH_SYN && th->th_seq == tp->irs) {
2126 * Following if statement from Stevens, vol. 2, p. 960.
2129 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
2131 * Any valid FIN must be to the left of the window.
2132 * At this point the FIN must be a duplicate or out
2133 * of sequence; drop it.
2138 * Send an ACK to resynchronize and drop any data.
2139 * But keep on processing for RST or ACK.
2141 tp->t_flags |= TF_ACKNOW;
2143 TCPSTAT_INC(tcps_rcvduppack);
2144 TCPSTAT_ADD(tcps_rcvdupbyte, todrop);
2146 TCPSTAT_INC(tcps_rcvpartduppack);
2147 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop);
2149 drop_hdrlen += todrop; /* drop from the top afterwards */
2150 th->th_seq += todrop;
2152 if (th->th_urp > todrop)
2153 th->th_urp -= todrop;
2161 * If new data are received on a connection after the
2162 * user processes are gone, then RST the other end.
2164 if ((so->so_state & SS_NOFDREF) &&
2165 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
2168 KASSERT(ti_locked == TI_WLOCKED, ("%s: SS_NOFDEREF && "
2169 "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked));
2170 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2172 if ((s = tcp_log_addrs(&tp->t_inpcb->inp_inc, th, NULL, NULL))) {
2173 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data after socket "
2174 "was closed, sending RST and removing tcpcb\n",
2175 s, __func__, tcpstates[tp->t_state], tlen);
2179 TCPSTAT_INC(tcps_rcvafterclose);
2180 rstreason = BANDLIM_UNLIMITED;
2185 * If segment ends after window, drop trailing data
2186 * (and PUSH and FIN); if nothing left, just ACK.
2188 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
2190 TCPSTAT_INC(tcps_rcvpackafterwin);
2191 if (todrop >= tlen) {
2192 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen);
2194 * If window is closed can only take segments at
2195 * window edge, and have to drop data and PUSH from
2196 * incoming segments. Continue processing, but
2197 * remember to ack. Otherwise, drop segment
2200 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2201 tp->t_flags |= TF_ACKNOW;
2202 TCPSTAT_INC(tcps_rcvwinprobe);
2206 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2209 thflags &= ~(TH_PUSH|TH_FIN);
2213 * If last ACK falls within this segment's sequence numbers,
2214 * record its timestamp.
2216 * 1) That the test incorporates suggestions from the latest
2217 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
2218 * 2) That updating only on newer timestamps interferes with
2219 * our earlier PAWS tests, so this check should be solely
2220 * predicated on the sequence space of this segment.
2221 * 3) That we modify the segment boundary check to be
2222 * Last.ACK.Sent <= SEG.SEQ + SEG.Len
2223 * instead of RFC1323's
2224 * Last.ACK.Sent < SEG.SEQ + SEG.Len,
2225 * This modified check allows us to overcome RFC1323's
2226 * limitations as described in Stevens TCP/IP Illustrated
2227 * Vol. 2 p.869. In such cases, we can still calculate the
2228 * RTT correctly when RCV.NXT == Last.ACK.Sent.
2230 if ((to.to_flags & TOF_TS) != 0 &&
2231 SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2232 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
2233 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
2234 tp->ts_recent_age = tcp_ts_getticks();
2235 tp->ts_recent = to.to_tsval;
2239 * If a SYN is in the window, then this is an
2240 * error and we send an RST and drop the connection.
2242 if (thflags & TH_SYN) {
2243 KASSERT(ti_locked == TI_WLOCKED,
2244 ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked));
2245 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2247 tp = tcp_drop(tp, ECONNRESET);
2248 rstreason = BANDLIM_UNLIMITED;
2253 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
2254 * flag is on (half-synchronized state), then queue data for
2255 * later processing; else drop segment and return.
2257 if ((thflags & TH_ACK) == 0) {
2258 if (tp->t_state == TCPS_SYN_RECEIVED ||
2259 (tp->t_flags & TF_NEEDSYN))
2261 else if (tp->t_flags & TF_ACKNOW)
2270 switch (tp->t_state) {
2273 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
2274 * ESTABLISHED state and continue processing.
2275 * The ACK was checked above.
2277 case TCPS_SYN_RECEIVED:
2279 TCPSTAT_INC(tcps_connects);
2281 /* Do window scaling? */
2282 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2283 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2284 tp->rcv_scale = tp->request_r_scale;
2285 tp->snd_wnd = tiwin;
2289 * SYN-RECEIVED -> ESTABLISHED
2290 * SYN-RECEIVED* -> FIN-WAIT-1
2292 tp->t_starttime = ticks;
2293 if (tp->t_flags & TF_NEEDFIN) {
2294 tp->t_state = TCPS_FIN_WAIT_1;
2295 tp->t_flags &= ~TF_NEEDFIN;
2297 tp->t_state = TCPS_ESTABLISHED;
2299 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
2302 * If segment contains data or ACK, will call tcp_reass()
2303 * later; if not, do so now to pass queued data to user.
2305 if (tlen == 0 && (thflags & TH_FIN) == 0)
2306 (void) tcp_reass(tp, (struct tcphdr *)0, 0,
2308 tp->snd_wl1 = th->th_seq - 1;
2312 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2313 * ACKs. If the ack is in the range
2314 * tp->snd_una < th->th_ack <= tp->snd_max
2315 * then advance tp->snd_una to th->th_ack and drop
2316 * data from the retransmission queue. If this ACK reflects
2317 * more up to date window information we update our window information.
2319 case TCPS_ESTABLISHED:
2320 case TCPS_FIN_WAIT_1:
2321 case TCPS_FIN_WAIT_2:
2322 case TCPS_CLOSE_WAIT:
2325 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2326 TCPSTAT_INC(tcps_rcvacktoomuch);
2329 if ((tp->t_flags & TF_SACK_PERMIT) &&
2330 ((to.to_flags & TOF_SACK) ||
2331 !TAILQ_EMPTY(&tp->snd_holes)))
2332 tcp_sack_doack(tp, &to, th->th_ack);
2334 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
2335 hhook_run_tcp_est_in(tp, th, &to);
2337 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2338 if (tlen == 0 && tiwin == tp->snd_wnd) {
2339 TCPSTAT_INC(tcps_rcvdupack);
2341 * If we have outstanding data (other than
2342 * a window probe), this is a completely
2343 * duplicate ack (ie, window info didn't
2344 * change), the ack is the biggest we've
2345 * seen and we've seen exactly our rexmt
2346 * threshhold of them, assume a packet
2347 * has been dropped and retransmit it.
2348 * Kludge snd_nxt & the congestion
2349 * window so we send only this one
2352 * We know we're losing at the current
2353 * window size so do congestion avoidance
2354 * (set ssthresh to half the current window
2355 * and pull our congestion window back to
2356 * the new ssthresh).
2358 * Dup acks mean that packets have left the
2359 * network (they're now cached at the receiver)
2360 * so bump cwnd by the amount in the receiver
2361 * to keep a constant cwnd packets in the
2364 * When using TCP ECN, notify the peer that
2365 * we reduced the cwnd.
2367 if (!tcp_timer_active(tp, TT_REXMT) ||
2368 th->th_ack != tp->snd_una)
2370 else if (++tp->t_dupacks > tcprexmtthresh ||
2371 IN_FASTRECOVERY(tp->t_flags)) {
2372 cc_ack_received(tp, th, CC_DUPACK);
2373 if ((tp->t_flags & TF_SACK_PERMIT) &&
2374 IN_FASTRECOVERY(tp->t_flags)) {
2378 * Compute the amount of data in flight first.
2379 * We can inject new data into the pipe iff
2380 * we have less than 1/2 the original window's
2381 * worth of data in flight.
2383 awnd = (tp->snd_nxt - tp->snd_fack) +
2384 tp->sackhint.sack_bytes_rexmit;
2385 if (awnd < tp->snd_ssthresh) {
2386 tp->snd_cwnd += tp->t_maxseg;
2387 if (tp->snd_cwnd > tp->snd_ssthresh)
2388 tp->snd_cwnd = tp->snd_ssthresh;
2391 tp->snd_cwnd += tp->t_maxseg;
2392 (void) tcp_output(tp);
2394 } else if (tp->t_dupacks == tcprexmtthresh) {
2395 tcp_seq onxt = tp->snd_nxt;
2398 * If we're doing sack, check to
2399 * see if we're already in sack
2400 * recovery. If we're not doing sack,
2401 * check to see if we're in newreno
2404 if (tp->t_flags & TF_SACK_PERMIT) {
2405 if (IN_FASTRECOVERY(tp->t_flags)) {
2410 if (SEQ_LEQ(th->th_ack,
2416 /* Congestion signal before ack. */
2417 cc_cong_signal(tp, th, CC_NDUPACK);
2418 cc_ack_received(tp, th, CC_DUPACK);
2419 tcp_timer_activate(tp, TT_REXMT, 0);
2421 if (tp->t_flags & TF_SACK_PERMIT) {
2423 tcps_sack_recovery_episode);
2424 tp->sack_newdata = tp->snd_nxt;
2425 tp->snd_cwnd = tp->t_maxseg;
2426 (void) tcp_output(tp);
2429 tp->snd_nxt = th->th_ack;
2430 tp->snd_cwnd = tp->t_maxseg;
2431 (void) tcp_output(tp);
2432 KASSERT(tp->snd_limited <= 2,
2433 ("%s: tp->snd_limited too big",
2435 tp->snd_cwnd = tp->snd_ssthresh +
2437 (tp->t_dupacks - tp->snd_limited);
2438 if (SEQ_GT(onxt, tp->snd_nxt))
2441 } else if (V_tcp_do_rfc3042) {
2442 cc_ack_received(tp, th, CC_DUPACK);
2443 u_long oldcwnd = tp->snd_cwnd;
2444 tcp_seq oldsndmax = tp->snd_max;
2447 KASSERT(tp->t_dupacks == 1 ||
2449 ("%s: dupacks not 1 or 2",
2451 if (tp->t_dupacks == 1)
2452 tp->snd_limited = 0;
2454 (tp->snd_nxt - tp->snd_una) +
2455 (tp->t_dupacks - tp->snd_limited) *
2457 (void) tcp_output(tp);
2458 sent = tp->snd_max - oldsndmax;
2459 if (sent > tp->t_maxseg) {
2460 KASSERT((tp->t_dupacks == 2 &&
2461 tp->snd_limited == 0) ||
2462 (sent == tp->t_maxseg + 1 &&
2463 tp->t_flags & TF_SENTFIN),
2464 ("%s: sent too much",
2466 tp->snd_limited = 2;
2467 } else if (sent > 0)
2469 tp->snd_cwnd = oldcwnd;
2477 KASSERT(SEQ_GT(th->th_ack, tp->snd_una),
2478 ("%s: th_ack <= snd_una", __func__));
2481 * If the congestion window was inflated to account
2482 * for the other side's cached packets, retract it.
2484 if (IN_FASTRECOVERY(tp->t_flags)) {
2485 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2486 if (tp->t_flags & TF_SACK_PERMIT)
2487 tcp_sack_partialack(tp, th);
2489 tcp_newreno_partial_ack(tp, th);
2491 cc_post_recovery(tp, th);
2495 * If we reach this point, ACK is not a duplicate,
2496 * i.e., it ACKs something we sent.
2498 if (tp->t_flags & TF_NEEDSYN) {
2500 * T/TCP: Connection was half-synchronized, and our
2501 * SYN has been ACK'd (so connection is now fully
2502 * synchronized). Go to non-starred state,
2503 * increment snd_una for ACK of SYN, and check if
2504 * we can do window scaling.
2506 tp->t_flags &= ~TF_NEEDSYN;
2508 /* Do window scaling? */
2509 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2510 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2511 tp->rcv_scale = tp->request_r_scale;
2512 /* Send window already scaled. */
2517 INP_WLOCK_ASSERT(tp->t_inpcb);
2519 acked = BYTES_THIS_ACK(tp, th);
2520 TCPSTAT_INC(tcps_rcvackpack);
2521 TCPSTAT_ADD(tcps_rcvackbyte, acked);
2524 * If we just performed our first retransmit, and the ACK
2525 * arrives within our recovery window, then it was a mistake
2526 * to do the retransmit in the first place. Recover our
2527 * original cwnd and ssthresh, and proceed to transmit where
2530 if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID &&
2531 (int)(ticks - tp->t_badrxtwin) < 0)
2532 cc_cong_signal(tp, th, CC_RTO_ERR);
2535 * If we have a timestamp reply, update smoothed
2536 * round trip time. If no timestamp is present but
2537 * transmit timer is running and timed sequence
2538 * number was acked, update smoothed round trip time.
2539 * Since we now have an rtt measurement, cancel the
2540 * timer backoff (cf., Phil Karn's retransmit alg.).
2541 * Recompute the initial retransmit timer.
2543 * Some boxes send broken timestamp replies
2544 * during the SYN+ACK phase, ignore
2545 * timestamps of 0 or we could calculate a
2546 * huge RTT and blow up the retransmit timer.
2548 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) {
2551 t = tcp_ts_getticks() - to.to_tsecr;
2552 if (!tp->t_rttlow || tp->t_rttlow > t)
2554 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1);
2555 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2556 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2557 tp->t_rttlow = ticks - tp->t_rtttime;
2558 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2562 * If all outstanding data is acked, stop retransmit
2563 * timer and remember to restart (more output or persist).
2564 * If there is more data to be acked, restart retransmit
2565 * timer, using current (possibly backed-off) value.
2567 if (th->th_ack == tp->snd_max) {
2568 tcp_timer_activate(tp, TT_REXMT, 0);
2570 } else if (!tcp_timer_active(tp, TT_PERSIST))
2571 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
2574 * If no data (only SYN) was ACK'd,
2575 * skip rest of ACK processing.
2581 * Let the congestion control algorithm update congestion
2582 * control related information. This typically means increasing
2583 * the congestion window.
2585 cc_ack_received(tp, th, CC_ACK);
2587 SOCKBUF_LOCK(&so->so_snd);
2588 if (acked > so->so_snd.sb_cc) {
2589 tp->snd_wnd -= so->so_snd.sb_cc;
2590 sbdrop_locked(&so->so_snd, (int)so->so_snd.sb_cc);
2593 sbdrop_locked(&so->so_snd, acked);
2594 tp->snd_wnd -= acked;
2597 /* NB: sowwakeup_locked() does an implicit unlock. */
2598 sowwakeup_locked(so);
2599 /* Detect una wraparound. */
2600 if (!IN_RECOVERY(tp->t_flags) &&
2601 SEQ_GT(tp->snd_una, tp->snd_recover) &&
2602 SEQ_LEQ(th->th_ack, tp->snd_recover))
2603 tp->snd_recover = th->th_ack - 1;
2604 /* XXXLAS: Can this be moved up into cc_post_recovery? */
2605 if (IN_RECOVERY(tp->t_flags) &&
2606 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2607 EXIT_RECOVERY(tp->t_flags);
2609 tp->snd_una = th->th_ack;
2610 if (tp->t_flags & TF_SACK_PERMIT) {
2611 if (SEQ_GT(tp->snd_una, tp->snd_recover))
2612 tp->snd_recover = tp->snd_una;
2614 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2615 tp->snd_nxt = tp->snd_una;
2617 switch (tp->t_state) {
2620 * In FIN_WAIT_1 STATE in addition to the processing
2621 * for the ESTABLISHED state if our FIN is now acknowledged
2622 * then enter FIN_WAIT_2.
2624 case TCPS_FIN_WAIT_1:
2625 if (ourfinisacked) {
2627 * If we can't receive any more
2628 * data, then closing user can proceed.
2629 * Starting the timer is contrary to the
2630 * specification, but if we don't get a FIN
2631 * we'll hang forever.
2634 * we should release the tp also, and use a
2637 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2638 soisdisconnected(so);
2639 tcp_timer_activate(tp, TT_2MSL,
2640 (tcp_fast_finwait2_recycle ?
2641 tcp_finwait2_timeout :
2644 tp->t_state = TCPS_FIN_WAIT_2;
2649 * In CLOSING STATE in addition to the processing for
2650 * the ESTABLISHED state if the ACK acknowledges our FIN
2651 * then enter the TIME-WAIT state, otherwise ignore
2655 if (ourfinisacked) {
2656 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2658 INP_INFO_WUNLOCK(&V_tcbinfo);
2665 * In LAST_ACK, we may still be waiting for data to drain
2666 * and/or to be acked, as well as for the ack of our FIN.
2667 * If our FIN is now acknowledged, delete the TCB,
2668 * enter the closed state and return.
2671 if (ourfinisacked) {
2672 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2681 INP_WLOCK_ASSERT(tp->t_inpcb);
2684 * Update window information.
2685 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2687 if ((thflags & TH_ACK) &&
2688 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2689 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2690 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2691 /* keep track of pure window updates */
2693 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2694 TCPSTAT_INC(tcps_rcvwinupd);
2695 tp->snd_wnd = tiwin;
2696 tp->snd_wl1 = th->th_seq;
2697 tp->snd_wl2 = th->th_ack;
2698 if (tp->snd_wnd > tp->max_sndwnd)
2699 tp->max_sndwnd = tp->snd_wnd;
2704 * Process segments with URG.
2706 if ((thflags & TH_URG) && th->th_urp &&
2707 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2709 * This is a kludge, but if we receive and accept
2710 * random urgent pointers, we'll crash in
2711 * soreceive. It's hard to imagine someone
2712 * actually wanting to send this much urgent data.
2714 SOCKBUF_LOCK(&so->so_rcv);
2715 if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2716 th->th_urp = 0; /* XXX */
2717 thflags &= ~TH_URG; /* XXX */
2718 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */
2719 goto dodata; /* XXX */
2722 * If this segment advances the known urgent pointer,
2723 * then mark the data stream. This should not happen
2724 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2725 * a FIN has been received from the remote side.
2726 * In these states we ignore the URG.
2728 * According to RFC961 (Assigned Protocols),
2729 * the urgent pointer points to the last octet
2730 * of urgent data. We continue, however,
2731 * to consider it to indicate the first octet
2732 * of data past the urgent section as the original
2733 * spec states (in one of two places).
2735 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2736 tp->rcv_up = th->th_seq + th->th_urp;
2737 so->so_oobmark = so->so_rcv.sb_cc +
2738 (tp->rcv_up - tp->rcv_nxt) - 1;
2739 if (so->so_oobmark == 0)
2740 so->so_rcv.sb_state |= SBS_RCVATMARK;
2742 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2744 SOCKBUF_UNLOCK(&so->so_rcv);
2746 * Remove out of band data so doesn't get presented to user.
2747 * This can happen independent of advancing the URG pointer,
2748 * but if two URG's are pending at once, some out-of-band
2749 * data may creep in... ick.
2751 if (th->th_urp <= (u_long)tlen &&
2752 !(so->so_options & SO_OOBINLINE)) {
2753 /* hdr drop is delayed */
2754 tcp_pulloutofband(so, th, m, drop_hdrlen);
2758 * If no out of band data is expected,
2759 * pull receive urgent pointer along
2760 * with the receive window.
2762 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2763 tp->rcv_up = tp->rcv_nxt;
2766 INP_WLOCK_ASSERT(tp->t_inpcb);
2769 * Process the segment text, merging it into the TCP sequencing queue,
2770 * and arranging for acknowledgment of receipt if necessary.
2771 * This process logically involves adjusting tp->rcv_wnd as data
2772 * is presented to the user (this happens in tcp_usrreq.c,
2773 * case PRU_RCVD). If a FIN has already been received on this
2774 * connection then we just ignore the text.
2776 if ((tlen || (thflags & TH_FIN)) &&
2777 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2778 tcp_seq save_start = th->th_seq;
2779 m_adj(m, drop_hdrlen); /* delayed header drop */
2781 * Insert segment which includes th into TCP reassembly queue
2782 * with control block tp. Set thflags to whether reassembly now
2783 * includes a segment with FIN. This handles the common case
2784 * inline (segment is the next to be received on an established
2785 * connection, and the queue is empty), avoiding linkage into
2786 * and removal from the queue and repetition of various
2788 * Set DELACK for segments received in order, but ack
2789 * immediately when segments are out of order (so
2790 * fast retransmit can work).
2792 if (th->th_seq == tp->rcv_nxt &&
2793 LIST_EMPTY(&tp->t_segq) &&
2794 TCPS_HAVEESTABLISHED(tp->t_state)) {
2796 tp->t_flags |= TF_DELACK;
2798 tp->t_flags |= TF_ACKNOW;
2799 tp->rcv_nxt += tlen;
2800 thflags = th->th_flags & TH_FIN;
2801 TCPSTAT_INC(tcps_rcvpack);
2802 TCPSTAT_ADD(tcps_rcvbyte, tlen);
2804 SOCKBUF_LOCK(&so->so_rcv);
2805 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
2808 sbappendstream_locked(&so->so_rcv, m);
2809 /* NB: sorwakeup_locked() does an implicit unlock. */
2810 sorwakeup_locked(so);
2813 * XXX: Due to the header drop above "th" is
2814 * theoretically invalid by now. Fortunately
2815 * m_adj() doesn't actually frees any mbufs
2816 * when trimming from the head.
2818 thflags = tcp_reass(tp, th, &tlen, m);
2819 tp->t_flags |= TF_ACKNOW;
2821 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT))
2822 tcp_update_sack_list(tp, save_start, save_start + tlen);
2825 * Note the amount of data that peer has sent into
2826 * our window, in order to estimate the sender's
2830 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
2831 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2833 len = so->so_rcv.sb_hiwat;
2841 * If FIN is received ACK the FIN and let the user know
2842 * that the connection is closing.
2844 if (thflags & TH_FIN) {
2845 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2848 * If connection is half-synchronized
2849 * (ie NEEDSYN flag on) then delay ACK,
2850 * so it may be piggybacked when SYN is sent.
2851 * Otherwise, since we received a FIN then no
2852 * more input can be expected, send ACK now.
2854 if (tp->t_flags & TF_NEEDSYN)
2855 tp->t_flags |= TF_DELACK;
2857 tp->t_flags |= TF_ACKNOW;
2860 switch (tp->t_state) {
2863 * In SYN_RECEIVED and ESTABLISHED STATES
2864 * enter the CLOSE_WAIT state.
2866 case TCPS_SYN_RECEIVED:
2867 tp->t_starttime = ticks;
2869 case TCPS_ESTABLISHED:
2870 tp->t_state = TCPS_CLOSE_WAIT;
2874 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2875 * enter the CLOSING state.
2877 case TCPS_FIN_WAIT_1:
2878 tp->t_state = TCPS_CLOSING;
2882 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2883 * starting the time-wait timer, turning off the other
2886 case TCPS_FIN_WAIT_2:
2887 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2888 KASSERT(ti_locked == TI_WLOCKED, ("%s: dodata "
2889 "TCP_FIN_WAIT_2 ti_locked: %d", __func__,
2893 INP_INFO_WUNLOCK(&V_tcbinfo);
2897 if (ti_locked == TI_WLOCKED)
2898 INP_INFO_WUNLOCK(&V_tcbinfo);
2899 ti_locked = TI_UNLOCKED;
2902 if (so->so_options & SO_DEBUG)
2903 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
2908 * Return any desired output.
2910 if (needoutput || (tp->t_flags & TF_ACKNOW))
2911 (void) tcp_output(tp);
2914 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d",
2915 __func__, ti_locked));
2916 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
2917 INP_WLOCK_ASSERT(tp->t_inpcb);
2919 if (tp->t_flags & TF_DELACK) {
2920 tp->t_flags &= ~TF_DELACK;
2921 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
2923 INP_WUNLOCK(tp->t_inpcb);
2928 * Generate an ACK dropping incoming segment if it occupies
2929 * sequence space, where the ACK reflects our state.
2931 * We can now skip the test for the RST flag since all
2932 * paths to this code happen after packets containing
2933 * RST have been dropped.
2935 * In the SYN-RECEIVED state, don't send an ACK unless the
2936 * segment we received passes the SYN-RECEIVED ACK test.
2937 * If it fails send a RST. This breaks the loop in the
2938 * "LAND" DoS attack, and also prevents an ACK storm
2939 * between two listening ports that have been sent forged
2940 * SYN segments, each with the source address of the other.
2942 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
2943 (SEQ_GT(tp->snd_una, th->th_ack) ||
2944 SEQ_GT(th->th_ack, tp->snd_max)) ) {
2945 rstreason = BANDLIM_RST_OPENPORT;
2949 if (so->so_options & SO_DEBUG)
2950 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2953 if (ti_locked == TI_WLOCKED)
2954 INP_INFO_WUNLOCK(&V_tcbinfo);
2955 ti_locked = TI_UNLOCKED;
2957 tp->t_flags |= TF_ACKNOW;
2958 (void) tcp_output(tp);
2959 INP_WUNLOCK(tp->t_inpcb);
2964 if (ti_locked == TI_WLOCKED)
2965 INP_INFO_WUNLOCK(&V_tcbinfo);
2966 ti_locked = TI_UNLOCKED;
2969 tcp_dropwithreset(m, th, tp, tlen, rstreason);
2970 INP_WUNLOCK(tp->t_inpcb);
2972 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
2976 if (ti_locked == TI_WLOCKED) {
2977 INP_INFO_WUNLOCK(&V_tcbinfo);
2978 ti_locked = TI_UNLOCKED;
2982 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
2986 * Drop space held by incoming segment and return.
2989 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2990 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2994 INP_WUNLOCK(tp->t_inpcb);
2999 * Issue RST and make ACK acceptable to originator of segment.
3000 * The mbuf must still include the original packet header.
3004 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
3005 int tlen, int rstreason)
3011 struct ip6_hdr *ip6;
3015 INP_WLOCK_ASSERT(tp->t_inpcb);
3018 /* Don't bother if destination was broadcast/multicast. */
3019 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
3022 if (mtod(m, struct ip *)->ip_v == 6) {
3023 ip6 = mtod(m, struct ip6_hdr *);
3024 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3025 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3027 /* IPv6 anycast check is done at tcp6_input() */
3030 #if defined(INET) && defined(INET6)
3035 ip = mtod(m, struct ip *);
3036 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3037 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3038 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3039 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3044 /* Perform bandwidth limiting. */
3045 if (badport_bandlim(rstreason) < 0)
3048 /* tcp_respond consumes the mbuf chain. */
3049 if (th->th_flags & TH_ACK) {
3050 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0,
3051 th->th_ack, TH_RST);
3053 if (th->th_flags & TH_SYN)
3055 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
3056 (tcp_seq)0, TH_RST|TH_ACK);
3064 * Parse TCP options and place in tcpopt.
3067 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags)
3072 for (; cnt > 0; cnt -= optlen, cp += optlen) {
3074 if (opt == TCPOPT_EOL)
3076 if (opt == TCPOPT_NOP)
3082 if (optlen < 2 || optlen > cnt)
3087 if (optlen != TCPOLEN_MAXSEG)
3089 if (!(flags & TO_SYN))
3091 to->to_flags |= TOF_MSS;
3092 bcopy((char *)cp + 2,
3093 (char *)&to->to_mss, sizeof(to->to_mss));
3094 to->to_mss = ntohs(to->to_mss);
3097 if (optlen != TCPOLEN_WINDOW)
3099 if (!(flags & TO_SYN))
3101 to->to_flags |= TOF_SCALE;
3102 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT);
3104 case TCPOPT_TIMESTAMP:
3105 if (optlen != TCPOLEN_TIMESTAMP)
3107 to->to_flags |= TOF_TS;
3108 bcopy((char *)cp + 2,
3109 (char *)&to->to_tsval, sizeof(to->to_tsval));
3110 to->to_tsval = ntohl(to->to_tsval);
3111 bcopy((char *)cp + 6,
3112 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
3113 to->to_tsecr = ntohl(to->to_tsecr);
3115 #ifdef TCP_SIGNATURE
3117 * XXX In order to reply to a host which has set the
3118 * TCP_SIGNATURE option in its initial SYN, we have to
3119 * record the fact that the option was observed here
3120 * for the syncache code to perform the correct response.
3122 case TCPOPT_SIGNATURE:
3123 if (optlen != TCPOLEN_SIGNATURE)
3125 to->to_flags |= TOF_SIGNATURE;
3126 to->to_signature = cp + 2;
3129 case TCPOPT_SACK_PERMITTED:
3130 if (optlen != TCPOLEN_SACK_PERMITTED)
3132 if (!(flags & TO_SYN))
3136 to->to_flags |= TOF_SACKPERM;
3139 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
3143 to->to_flags |= TOF_SACK;
3144 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
3145 to->to_sacks = cp + 2;
3146 TCPSTAT_INC(tcps_sack_rcv_blocks);
3155 * Pull out of band byte out of a segment so
3156 * it doesn't appear in the user's data queue.
3157 * It is still reflected in the segment length for
3158 * sequencing purposes.
3161 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m,
3164 int cnt = off + th->th_urp - 1;
3167 if (m->m_len > cnt) {
3168 char *cp = mtod(m, caddr_t) + cnt;
3169 struct tcpcb *tp = sototcpcb(so);
3171 INP_WLOCK_ASSERT(tp->t_inpcb);
3174 tp->t_oobflags |= TCPOOB_HAVEDATA;
3175 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3177 if (m->m_flags & M_PKTHDR)
3186 panic("tcp_pulloutofband");
3190 * Collect new round-trip time estimate
3191 * and update averages and current timeout.
3194 tcp_xmit_timer(struct tcpcb *tp, int rtt)
3198 INP_WLOCK_ASSERT(tp->t_inpcb);
3200 TCPSTAT_INC(tcps_rttupdated);
3202 if (tp->t_srtt != 0) {
3204 * srtt is stored as fixed point with 5 bits after the
3205 * binary point (i.e., scaled by 8). The following magic
3206 * is equivalent to the smoothing algorithm in rfc793 with
3207 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3208 * point). Adjust rtt to origin 0.
3210 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
3211 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
3213 if ((tp->t_srtt += delta) <= 0)
3217 * We accumulate a smoothed rtt variance (actually, a
3218 * smoothed mean difference), then set the retransmit
3219 * timer to smoothed rtt + 4 times the smoothed variance.
3220 * rttvar is stored as fixed point with 4 bits after the
3221 * binary point (scaled by 16). The following is
3222 * equivalent to rfc793 smoothing with an alpha of .75
3223 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
3224 * rfc793's wired-in beta.
3228 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
3229 if ((tp->t_rttvar += delta) <= 0)
3231 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
3232 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3235 * No rtt measurement yet - use the unsmoothed rtt.
3236 * Set the variance to half the rtt (so our first
3237 * retransmit happens at 3*rtt).
3239 tp->t_srtt = rtt << TCP_RTT_SHIFT;
3240 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
3241 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3247 * the retransmit should happen at rtt + 4 * rttvar.
3248 * Because of the way we do the smoothing, srtt and rttvar
3249 * will each average +1/2 tick of bias. When we compute
3250 * the retransmit timer, we want 1/2 tick of rounding and
3251 * 1 extra tick because of +-1/2 tick uncertainty in the
3252 * firing of the timer. The bias will give us exactly the
3253 * 1.5 tick we need. But, because the bias is
3254 * statistical, we have to test that we don't drop below
3255 * the minimum feasible timer (which is 2 ticks).
3257 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3258 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3261 * We received an ack for a packet that wasn't retransmitted;
3262 * it is probably safe to discard any error indications we've
3263 * received recently. This isn't quite right, but close enough
3264 * for now (a route might have failed after we sent a segment,
3265 * and the return path might not be symmetrical).
3267 tp->t_softerror = 0;
3271 * Determine a reasonable value for maxseg size.
3272 * If the route is known, check route for mtu.
3273 * If none, use an mss that can be handled on the outgoing
3274 * interface without forcing IP to fragment; if bigger than
3275 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
3276 * to utilize large mbufs. If no route is found, route has no mtu,
3277 * or the destination isn't local, use a default, hopefully conservative
3278 * size (usually 512 or the default IP max size, but no more than the mtu
3279 * of the interface), as we can't discover anything about intervening
3280 * gateways or networks. We also initialize the congestion/slow start
3281 * window to be a single segment if the destination isn't local.
3282 * While looking at the routing entry, we also initialize other path-dependent
3283 * parameters from pre-set or cached values in the routing entry.
3285 * Also take into account the space needed for options that we
3286 * send regularly. Make maxseg shorter by that amount to assure
3287 * that we can send maxseg amount of data even when the options
3288 * are present. Store the upper limit of the length of options plus
3291 * NOTE that this routine is only called when we process an incoming
3292 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS
3293 * settings are handled in tcp_mssopt().
3296 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer,
3297 struct hc_metrics_lite *metricptr, int *mtuflags)
3301 struct inpcb *inp = tp->t_inpcb;
3302 struct hc_metrics_lite metrics;
3305 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
3306 size_t min_protoh = isipv6 ?
3307 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
3308 sizeof (struct tcpiphdr);
3310 const size_t min_protoh = sizeof(struct tcpiphdr);
3313 INP_WLOCK_ASSERT(tp->t_inpcb);
3315 if (mtuoffer != -1) {
3316 KASSERT(offer == -1, ("%s: conflict", __func__));
3317 offer = mtuoffer - min_protoh;
3324 maxmtu = tcp_maxmtu6(&inp->inp_inc, mtuflags);
3325 tp->t_maxopd = tp->t_maxseg = V_tcp_v6mssdflt;
3328 #if defined(INET) && defined(INET6)
3333 maxmtu = tcp_maxmtu(&inp->inp_inc, mtuflags);
3334 tp->t_maxopd = tp->t_maxseg = V_tcp_mssdflt;
3339 * No route to sender, stay with default mss and return.
3343 * In case we return early we need to initialize metrics
3344 * to a defined state as tcp_hc_get() would do for us
3345 * if there was no cache hit.
3347 if (metricptr != NULL)
3348 bzero(metricptr, sizeof(struct hc_metrics_lite));
3352 /* What have we got? */
3356 * Offer == 0 means that there was no MSS on the SYN
3357 * segment, in this case we use tcp_mssdflt as
3358 * already assigned to t_maxopd above.
3360 offer = tp->t_maxopd;
3365 * Offer == -1 means that we didn't receive SYN yet.
3371 * Prevent DoS attack with too small MSS. Round up
3372 * to at least minmss.
3374 offer = max(offer, V_tcp_minmss);
3378 * rmx information is now retrieved from tcp_hostcache.
3380 tcp_hc_get(&inp->inp_inc, &metrics);
3381 if (metricptr != NULL)
3382 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite));
3385 * If there's a discovered mtu int tcp hostcache, use it
3386 * else, use the link mtu.
3388 if (metrics.rmx_mtu)
3389 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
3393 mss = maxmtu - min_protoh;
3394 if (!V_path_mtu_discovery &&
3395 !in6_localaddr(&inp->in6p_faddr))
3396 mss = min(mss, V_tcp_v6mssdflt);
3399 #if defined(INET) && defined(INET6)
3404 mss = maxmtu - min_protoh;
3405 if (!V_path_mtu_discovery &&
3406 !in_localaddr(inp->inp_faddr))
3407 mss = min(mss, V_tcp_mssdflt);
3411 * XXX - The above conditional (mss = maxmtu - min_protoh)
3412 * probably violates the TCP spec.
3413 * The problem is that, since we don't know the
3414 * other end's MSS, we are supposed to use a conservative
3415 * default. But, if we do that, then MTU discovery will
3416 * never actually take place, because the conservative
3417 * default is much less than the MTUs typically seen
3418 * on the Internet today. For the moment, we'll sweep
3419 * this under the carpet.
3421 * The conservative default might not actually be a problem
3422 * if the only case this occurs is when sending an initial
3423 * SYN with options and data to a host we've never talked
3424 * to before. Then, they will reply with an MSS value which
3425 * will get recorded and the new parameters should get
3426 * recomputed. For Further Study.
3429 mss = min(mss, offer);
3432 * Sanity check: make sure that maxopd will be large
3433 * enough to allow some data on segments even if the
3434 * all the option space is used (40bytes). Otherwise
3435 * funny things may happen in tcp_output.
3440 * maxopd stores the maximum length of data AND options
3441 * in a segment; maxseg is the amount of data in a normal
3442 * segment. We need to store this value (maxopd) apart
3443 * from maxseg, because now every segment carries options
3444 * and thus we normally have somewhat less data in segments.
3449 * origoffer==-1 indicates that no segments were received yet.
3450 * In this case we just guess.
3452 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
3454 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
3455 mss -= TCPOLEN_TSTAMP_APPA;
3457 #if (MCLBYTES & (MCLBYTES - 1)) == 0
3459 mss &= ~(MCLBYTES-1);
3462 mss = mss / MCLBYTES * MCLBYTES;
3468 tcp_mss(struct tcpcb *tp, int offer)
3474 struct hc_metrics_lite metrics;
3477 KASSERT(tp != NULL, ("%s: tp == NULL", __func__));
3479 tcp_mss_update(tp, offer, -1, &metrics, &mtuflags);
3485 * If there's a pipesize, change the socket buffer to that size,
3486 * don't change if sb_hiwat is different than default (then it
3487 * has been changed on purpose with setsockopt).
3488 * Make the socket buffers an integral number of mss units;
3489 * if the mss is larger than the socket buffer, decrease the mss.
3491 so = inp->inp_socket;
3492 SOCKBUF_LOCK(&so->so_snd);
3493 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe)
3494 bufsize = metrics.rmx_sendpipe;
3496 bufsize = so->so_snd.sb_hiwat;
3500 bufsize = roundup(bufsize, mss);
3501 if (bufsize > sb_max)
3503 if (bufsize > so->so_snd.sb_hiwat)
3504 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
3506 SOCKBUF_UNLOCK(&so->so_snd);
3509 SOCKBUF_LOCK(&so->so_rcv);
3510 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe)
3511 bufsize = metrics.rmx_recvpipe;
3513 bufsize = so->so_rcv.sb_hiwat;
3514 if (bufsize > mss) {
3515 bufsize = roundup(bufsize, mss);
3516 if (bufsize > sb_max)
3518 if (bufsize > so->so_rcv.sb_hiwat)
3519 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
3521 SOCKBUF_UNLOCK(&so->so_rcv);
3523 /* Check the interface for TSO capabilities. */
3524 if (mtuflags & CSUM_TSO)
3525 tp->t_flags |= TF_TSO;
3529 * Determine the MSS option to send on an outgoing SYN.
3532 tcp_mssopt(struct in_conninfo *inc)
3539 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3542 if (inc->inc_flags & INC_ISIPV6) {
3543 mss = V_tcp_v6mssdflt;
3544 maxmtu = tcp_maxmtu6(inc, NULL);
3545 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3546 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3549 #if defined(INET) && defined(INET6)
3554 mss = V_tcp_mssdflt;
3555 maxmtu = tcp_maxmtu(inc, NULL);
3556 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3557 min_protoh = sizeof(struct tcpiphdr);
3560 if (maxmtu && thcmtu)
3561 mss = min(maxmtu, thcmtu) - min_protoh;
3562 else if (maxmtu || thcmtu)
3563 mss = max(maxmtu, thcmtu) - min_protoh;
3570 * On a partial ack arrives, force the retransmission of the
3571 * next unacknowledged segment. Do not clear tp->t_dupacks.
3572 * By setting snd_nxt to ti_ack, this forces retransmission timer to
3576 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th)
3578 tcp_seq onxt = tp->snd_nxt;
3579 u_long ocwnd = tp->snd_cwnd;
3581 INP_WLOCK_ASSERT(tp->t_inpcb);
3583 tcp_timer_activate(tp, TT_REXMT, 0);
3585 tp->snd_nxt = th->th_ack;
3587 * Set snd_cwnd to one segment beyond acknowledged offset.
3588 * (tp->snd_una has not yet been updated when this function is called.)
3590 tp->snd_cwnd = tp->t_maxseg + BYTES_THIS_ACK(tp, th);
3591 tp->t_flags |= TF_ACKNOW;
3592 (void) tcp_output(tp);
3593 tp->snd_cwnd = ocwnd;
3594 if (SEQ_GT(onxt, tp->snd_nxt))
3597 * Partial window deflation. Relies on fact that tp->snd_una
3600 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th))
3601 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th);
3604 tp->snd_cwnd += tp->t_maxseg;