2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
3 * The Regents of the University of California. All rights reserved.
4 * Copyright (c) 2007-2008,2010
5 * Swinburne University of Technology, Melbourne, Australia.
6 * Copyright (c) 2009-2010 Lawrence Stewart <lstewart@freebsd.org>
7 * Copyright (c) 2010 The FreeBSD Foundation
8 * Copyright (c) 2010-2011 Juniper Networks, Inc.
11 * Portions of this software were developed at the Centre for Advanced Internet
12 * Architectures, Swinburne University of Technology, by Lawrence Stewart,
13 * James Healy and David Hayes, made possible in part by a grant from the Cisco
14 * University Research Program Fund at Community Foundation Silicon Valley.
16 * Portions of this software were developed at the Centre for Advanced
17 * Internet Architectures, Swinburne University of Technology, Melbourne,
18 * Australia by David Hayes under sponsorship from the FreeBSD Foundation.
20 * Portions of this software were developed by Robert N. M. Watson under
21 * contract to Juniper Networks, Inc.
23 * Redistribution and use in source and binary forms, with or without
24 * modification, are permitted provided that the following conditions
26 * 1. Redistributions of source code must retain the above copyright
27 * notice, this list of conditions and the following disclaimer.
28 * 2. Redistributions in binary form must reproduce the above copyright
29 * notice, this list of conditions and the following disclaimer in the
30 * documentation and/or other materials provided with the distribution.
31 * 4. Neither the name of the University nor the names of its contributors
32 * may be used to endorse or promote products derived from this software
33 * without specific prior written permission.
35 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
36 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
37 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
38 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
39 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
40 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
41 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
42 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
43 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
44 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
47 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95
50 #include <sys/cdefs.h>
51 __FBSDID("$FreeBSD$");
53 #include "opt_ipfw.h" /* for ipfw_fwd */
55 #include "opt_inet6.h"
56 #include "opt_ipsec.h"
57 #include "opt_tcpdebug.h"
59 #include <sys/param.h>
60 #include <sys/kernel.h>
61 #include <sys/hhook.h>
62 #include <sys/malloc.h>
64 #include <sys/proc.h> /* for proc0 declaration */
65 #include <sys/protosw.h>
67 #include <sys/signalvar.h>
68 #include <sys/socket.h>
69 #include <sys/socketvar.h>
70 #include <sys/sysctl.h>
71 #include <sys/syslog.h>
72 #include <sys/systm.h>
74 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
79 #include <net/if_var.h>
80 #include <net/route.h>
83 #define TCPSTATES /* for logging */
85 #include <netinet/cc.h>
86 #include <netinet/in.h>
87 #include <netinet/in_kdtrace.h>
88 #include <netinet/in_pcb.h>
89 #include <netinet/in_systm.h>
90 #include <netinet/in_var.h>
91 #include <netinet/ip.h>
92 #include <netinet/ip_icmp.h> /* required for icmp_var.h */
93 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */
94 #include <netinet/ip_var.h>
95 #include <netinet/ip_options.h>
96 #include <netinet/ip6.h>
97 #include <netinet/icmp6.h>
98 #include <netinet6/in6_pcb.h>
99 #include <netinet6/ip6_var.h>
100 #include <netinet6/nd6.h>
101 #include <netinet/tcp_fsm.h>
102 #include <netinet/tcp_seq.h>
103 #include <netinet/tcp_timer.h>
104 #include <netinet/tcp_var.h>
105 #include <netinet6/tcp6_var.h>
106 #include <netinet/tcpip.h>
108 #include <netinet/tcp_pcap.h>
110 #include <netinet/tcp_syncache.h>
112 #include <netinet/tcp_debug.h>
113 #endif /* TCPDEBUG */
115 #include <netinet/tcp_offload.h>
119 #include <netipsec/ipsec.h>
120 #include <netipsec/ipsec6.h>
123 #include <machine/in_cksum.h>
125 #include <security/mac/mac_framework.h>
127 const int tcprexmtthresh = 3;
129 int tcp_log_in_vain = 0;
130 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
132 "Log all incoming TCP segments to closed ports");
134 VNET_DEFINE(int, blackhole) = 0;
135 #define V_blackhole VNET(blackhole)
136 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW,
137 &VNET_NAME(blackhole), 0,
138 "Do not send RST on segments to closed ports");
140 VNET_DEFINE(int, tcp_delack_enabled) = 1;
141 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW,
142 &VNET_NAME(tcp_delack_enabled), 0,
143 "Delay ACK to try and piggyback it onto a data packet");
145 VNET_DEFINE(int, drop_synfin) = 0;
146 #define V_drop_synfin VNET(drop_synfin)
147 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_VNET | CTLFLAG_RW,
148 &VNET_NAME(drop_synfin), 0,
149 "Drop TCP packets with SYN+FIN set");
151 VNET_DEFINE(int, tcp_do_rfc6675_pipe) = 0;
152 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc6675_pipe, CTLFLAG_VNET | CTLFLAG_RW,
153 &VNET_NAME(tcp_do_rfc6675_pipe), 0,
154 "Use calculated pipe/in-flight bytes per RFC 6675");
156 VNET_DEFINE(int, tcp_do_rfc3042) = 1;
157 #define V_tcp_do_rfc3042 VNET(tcp_do_rfc3042)
158 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW,
159 &VNET_NAME(tcp_do_rfc3042), 0,
160 "Enable RFC 3042 (Limited Transmit)");
162 VNET_DEFINE(int, tcp_do_rfc3390) = 1;
163 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW,
164 &VNET_NAME(tcp_do_rfc3390), 0,
165 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
167 VNET_DEFINE(int, tcp_initcwnd_segments) = 10;
168 SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments,
169 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0,
170 "Slow-start flight size (initial congestion window) in number of segments");
172 VNET_DEFINE(int, tcp_do_rfc3465) = 1;
173 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW,
174 &VNET_NAME(tcp_do_rfc3465), 0,
175 "Enable RFC 3465 (Appropriate Byte Counting)");
177 VNET_DEFINE(int, tcp_abc_l_var) = 2;
178 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW,
179 &VNET_NAME(tcp_abc_l_var), 2,
180 "Cap the max cwnd increment during slow-start to this number of segments");
182 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, CTLFLAG_RW, 0, "TCP ECN");
184 VNET_DEFINE(int, tcp_do_ecn) = 0;
185 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW,
186 &VNET_NAME(tcp_do_ecn), 0,
189 VNET_DEFINE(int, tcp_ecn_maxretries) = 1;
190 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_VNET | CTLFLAG_RW,
191 &VNET_NAME(tcp_ecn_maxretries), 0,
192 "Max retries before giving up on ECN");
194 VNET_DEFINE(int, tcp_insecure_syn) = 0;
195 #define V_tcp_insecure_syn VNET(tcp_insecure_syn)
196 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW,
197 &VNET_NAME(tcp_insecure_syn), 0,
198 "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets");
200 VNET_DEFINE(int, tcp_insecure_rst) = 0;
201 #define V_tcp_insecure_rst VNET(tcp_insecure_rst)
202 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW,
203 &VNET_NAME(tcp_insecure_rst), 0,
204 "Follow RFC793 instead of RFC5961 criteria for accepting RST packets");
206 VNET_DEFINE(int, tcp_recvspace) = 1024*64;
207 #define V_tcp_recvspace VNET(tcp_recvspace)
208 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW,
209 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size");
211 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1;
212 #define V_tcp_do_autorcvbuf VNET(tcp_do_autorcvbuf)
213 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW,
214 &VNET_NAME(tcp_do_autorcvbuf), 0,
215 "Enable automatic receive buffer sizing");
217 VNET_DEFINE(int, tcp_autorcvbuf_inc) = 16*1024;
218 #define V_tcp_autorcvbuf_inc VNET(tcp_autorcvbuf_inc)
219 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_VNET | CTLFLAG_RW,
220 &VNET_NAME(tcp_autorcvbuf_inc), 0,
221 "Incrementor step size of automatic receive buffer");
223 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024;
224 #define V_tcp_autorcvbuf_max VNET(tcp_autorcvbuf_max)
225 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW,
226 &VNET_NAME(tcp_autorcvbuf_max), 0,
227 "Max size of automatic receive buffer");
229 VNET_DEFINE(struct inpcbhead, tcb);
230 #define tcb6 tcb /* for KAME src sync over BSD*'s */
231 VNET_DEFINE(struct inpcbinfo, tcbinfo);
233 static void tcp_dooptions(struct tcpopt *, u_char *, int, int);
234 static void tcp_do_segment(struct mbuf *, struct tcphdr *,
235 struct socket *, struct tcpcb *, int, int, uint8_t,
237 static void tcp_dropwithreset(struct mbuf *, struct tcphdr *,
238 struct tcpcb *, int, int);
239 static void tcp_pulloutofband(struct socket *,
240 struct tcphdr *, struct mbuf *, int);
241 static void tcp_xmit_timer(struct tcpcb *, int);
242 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *);
243 static void inline cc_ack_received(struct tcpcb *tp, struct tcphdr *th,
245 static void inline cc_conn_init(struct tcpcb *tp);
246 static void inline cc_post_recovery(struct tcpcb *tp, struct tcphdr *th);
247 static void inline hhook_run_tcp_est_in(struct tcpcb *tp,
248 struct tcphdr *th, struct tcpopt *to);
251 * TCP statistics are stored in an "array" of counter(9)s.
253 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat);
254 VNET_PCPUSTAT_SYSINIT(tcpstat);
255 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat,
256 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
259 VNET_PCPUSTAT_SYSUNINIT(tcpstat);
262 * Kernel module interface for updating tcpstat. The argument is an index
263 * into tcpstat treated as an array.
266 kmod_tcpstat_inc(int statnum)
269 counter_u64_add(VNET(tcpstat)[statnum], 1);
273 * Wrapper for the TCP established input helper hook.
276 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to)
278 struct tcp_hhook_data hhook_data;
280 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) {
285 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data,
291 * CC wrapper hook functions
294 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t type)
296 INP_WLOCK_ASSERT(tp->t_inpcb);
298 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th);
299 if (tp->snd_cwnd <= tp->snd_wnd)
300 tp->ccv->flags |= CCF_CWND_LIMITED;
302 tp->ccv->flags &= ~CCF_CWND_LIMITED;
304 if (type == CC_ACK) {
305 if (tp->snd_cwnd > tp->snd_ssthresh) {
306 tp->t_bytes_acked += min(tp->ccv->bytes_this_ack,
307 V_tcp_abc_l_var * tp->t_maxseg);
308 if (tp->t_bytes_acked >= tp->snd_cwnd) {
309 tp->t_bytes_acked -= tp->snd_cwnd;
310 tp->ccv->flags |= CCF_ABC_SENTAWND;
313 tp->ccv->flags &= ~CCF_ABC_SENTAWND;
314 tp->t_bytes_acked = 0;
318 if (CC_ALGO(tp)->ack_received != NULL) {
319 /* XXXLAS: Find a way to live without this */
320 tp->ccv->curack = th->th_ack;
321 CC_ALGO(tp)->ack_received(tp->ccv, type);
326 cc_conn_init(struct tcpcb *tp)
328 struct hc_metrics_lite metrics;
329 struct inpcb *inp = tp->t_inpcb;
332 INP_WLOCK_ASSERT(tp->t_inpcb);
334 tcp_hc_get(&inp->inp_inc, &metrics);
336 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) {
338 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
339 TCPSTAT_INC(tcps_usedrtt);
340 if (metrics.rmx_rttvar) {
341 tp->t_rttvar = metrics.rmx_rttvar;
342 TCPSTAT_INC(tcps_usedrttvar);
344 /* default variation is +- 1 rtt */
346 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
348 TCPT_RANGESET(tp->t_rxtcur,
349 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
350 tp->t_rttmin, TCPTV_REXMTMAX);
352 if (metrics.rmx_ssthresh) {
354 * There's some sort of gateway or interface
355 * buffer limit on the path. Use this to set
356 * the slow start threshhold, but set the
357 * threshold to no less than 2*mss.
359 tp->snd_ssthresh = max(2 * tp->t_maxseg, metrics.rmx_ssthresh);
360 TCPSTAT_INC(tcps_usedssthresh);
364 * Set the initial slow-start flight size.
366 * RFC5681 Section 3.1 specifies the default conservative values.
367 * RFC3390 specifies slightly more aggressive values.
368 * RFC6928 increases it to ten segments.
369 * Support for user specified value for initial flight size.
371 * If a SYN or SYN/ACK was lost and retransmitted, we have to
372 * reduce the initial CWND to one segment as congestion is likely
373 * requiring us to be cautious.
375 if (tp->snd_cwnd == 1)
376 tp->snd_cwnd = tp->t_maxseg; /* SYN(-ACK) lost */
377 else if (V_tcp_initcwnd_segments)
378 tp->snd_cwnd = min(V_tcp_initcwnd_segments * tp->t_maxseg,
379 max(2 * tp->t_maxseg, V_tcp_initcwnd_segments * 1460));
380 else if (V_tcp_do_rfc3390)
381 tp->snd_cwnd = min(4 * tp->t_maxseg,
382 max(2 * tp->t_maxseg, 4380));
384 /* Per RFC5681 Section 3.1 */
385 if (tp->t_maxseg > 2190)
386 tp->snd_cwnd = 2 * tp->t_maxseg;
387 else if (tp->t_maxseg > 1095)
388 tp->snd_cwnd = 3 * tp->t_maxseg;
390 tp->snd_cwnd = 4 * tp->t_maxseg;
393 if (CC_ALGO(tp)->conn_init != NULL)
394 CC_ALGO(tp)->conn_init(tp->ccv);
398 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type)
400 INP_WLOCK_ASSERT(tp->t_inpcb);
404 if (!IN_FASTRECOVERY(tp->t_flags)) {
405 tp->snd_recover = tp->snd_max;
406 if (tp->t_flags & TF_ECN_PERMIT)
407 tp->t_flags |= TF_ECN_SND_CWR;
411 if (!IN_CONGRECOVERY(tp->t_flags)) {
412 TCPSTAT_INC(tcps_ecn_rcwnd);
413 tp->snd_recover = tp->snd_max;
414 if (tp->t_flags & TF_ECN_PERMIT)
415 tp->t_flags |= TF_ECN_SND_CWR;
420 tp->t_bytes_acked = 0;
421 EXIT_RECOVERY(tp->t_flags);
422 tp->snd_ssthresh = max(2, min(tp->snd_wnd, tp->snd_cwnd) / 2 /
423 tp->t_maxseg) * tp->t_maxseg;
424 tp->snd_cwnd = tp->t_maxseg;
427 TCPSTAT_INC(tcps_sndrexmitbad);
428 /* RTO was unnecessary, so reset everything. */
429 tp->snd_cwnd = tp->snd_cwnd_prev;
430 tp->snd_ssthresh = tp->snd_ssthresh_prev;
431 tp->snd_recover = tp->snd_recover_prev;
432 if (tp->t_flags & TF_WASFRECOVERY)
433 ENTER_FASTRECOVERY(tp->t_flags);
434 if (tp->t_flags & TF_WASCRECOVERY)
435 ENTER_CONGRECOVERY(tp->t_flags);
436 tp->snd_nxt = tp->snd_max;
437 tp->t_flags &= ~TF_PREVVALID;
442 if (CC_ALGO(tp)->cong_signal != NULL) {
444 tp->ccv->curack = th->th_ack;
445 CC_ALGO(tp)->cong_signal(tp->ccv, type);
450 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th)
452 INP_WLOCK_ASSERT(tp->t_inpcb);
454 /* XXXLAS: KASSERT that we're in recovery? */
456 if (CC_ALGO(tp)->post_recovery != NULL) {
457 tp->ccv->curack = th->th_ack;
458 CC_ALGO(tp)->post_recovery(tp->ccv);
460 /* XXXLAS: EXIT_RECOVERY ? */
461 tp->t_bytes_acked = 0;
466 tcp_signature_verify_input(struct mbuf *m, int off0, int tlen, int optlen,
467 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag)
471 tcp_fields_to_net(th);
472 ret = tcp_signature_verify(m, off0, tlen, optlen, to, th, tcpbflag);
473 tcp_fields_to_host(th);
479 * Indicate whether this ack should be delayed. We can delay the ack if
480 * following conditions are met:
481 * - There is no delayed ack timer in progress.
482 * - Our last ack wasn't a 0-sized window. We never want to delay
483 * the ack that opens up a 0-sized window.
484 * - LRO wasn't used for this segment. We make sure by checking that the
485 * segment size is not larger than the MSS.
486 * - Delayed acks are enabled or this is a half-synchronized T/TCP
489 #define DELAY_ACK(tp, tlen) \
490 ((!tcp_timer_active(tp, TT_DELACK) && \
491 (tp->t_flags & TF_RXWIN0SENT) == 0) && \
492 (tlen <= tp->t_maxopd) && \
493 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
496 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos)
498 INP_WLOCK_ASSERT(tp->t_inpcb);
500 if (CC_ALGO(tp)->ecnpkt_handler != NULL) {
501 switch (iptos & IPTOS_ECN_MASK) {
503 tp->ccv->flags |= CCF_IPHDR_CE;
506 tp->ccv->flags &= ~CCF_IPHDR_CE;
509 tp->ccv->flags &= ~CCF_IPHDR_CE;
513 if (th->th_flags & TH_CWR)
514 tp->ccv->flags |= CCF_TCPHDR_CWR;
516 tp->ccv->flags &= ~CCF_TCPHDR_CWR;
518 if (tp->t_flags & TF_DELACK)
519 tp->ccv->flags |= CCF_DELACK;
521 tp->ccv->flags &= ~CCF_DELACK;
523 CC_ALGO(tp)->ecnpkt_handler(tp->ccv);
525 if (tp->ccv->flags & CCF_ACKNOW)
526 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
531 * TCP input handling is split into multiple parts:
532 * tcp6_input is a thin wrapper around tcp_input for the extended
533 * ip6_protox[] call format in ip6_input
534 * tcp_input handles primary segment validation, inpcb lookup and
535 * SYN processing on listen sockets
536 * tcp_do_segment processes the ACK and text of the segment for
537 * establishing, established and closing connections
541 tcp6_input(struct mbuf **mp, int *offp, int proto)
543 struct mbuf *m = *mp;
544 struct in6_ifaddr *ia6;
547 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
550 * draft-itojun-ipv6-tcp-to-anycast
551 * better place to put this in?
553 ip6 = mtod(m, struct ip6_hdr *);
554 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */);
555 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
558 ifa_free(&ia6->ia_ifa);
559 ip6 = mtod(m, struct ip6_hdr *);
560 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
561 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
562 return (IPPROTO_DONE);
565 ifa_free(&ia6->ia_ifa);
567 return (tcp_input(mp, offp, proto));
572 tcp_input(struct mbuf **mp, int *offp, int proto)
574 struct mbuf *m = *mp;
575 struct tcphdr *th = NULL;
576 struct ip *ip = NULL;
577 struct inpcb *inp = NULL;
578 struct tcpcb *tp = NULL;
579 struct socket *so = NULL;
589 int rstreason = 0; /* For badport_bandlim accounting purposes */
591 uint8_t sig_checked = 0;
594 struct m_tag *fwd_tag = NULL;
596 struct ip6_hdr *ip6 = NULL;
599 const void *ip6 = NULL;
601 struct tcpopt to; /* options in this segment */
602 char *s = NULL; /* address and port logging */
604 #define TI_UNLOCKED 1
609 * The size of tcp_saveipgen must be the size of the max ip header,
612 u_char tcp_saveipgen[IP6_HDR_LEN];
613 struct tcphdr tcp_savetcp;
618 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
625 TCPSTAT_INC(tcps_rcvtotal);
629 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */
631 if (m->m_len < (sizeof(*ip6) + sizeof(*th))) {
632 m = m_pullup(m, sizeof(*ip6) + sizeof(*th));
634 TCPSTAT_INC(tcps_rcvshort);
635 return (IPPROTO_DONE);
639 ip6 = mtod(m, struct ip6_hdr *);
640 th = (struct tcphdr *)((caddr_t)ip6 + off0);
641 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
642 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) {
643 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
644 th->th_sum = m->m_pkthdr.csum_data;
646 th->th_sum = in6_cksum_pseudo(ip6, tlen,
647 IPPROTO_TCP, m->m_pkthdr.csum_data);
648 th->th_sum ^= 0xffff;
650 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen);
652 TCPSTAT_INC(tcps_rcvbadsum);
657 * Be proactive about unspecified IPv6 address in source.
658 * As we use all-zero to indicate unbounded/unconnected pcb,
659 * unspecified IPv6 address can be used to confuse us.
661 * Note that packets with unspecified IPv6 destination is
662 * already dropped in ip6_input.
664 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
670 #if defined(INET) && defined(INET6)
676 * Get IP and TCP header together in first mbuf.
677 * Note: IP leaves IP header in first mbuf.
679 if (off0 > sizeof (struct ip)) {
681 off0 = sizeof(struct ip);
683 if (m->m_len < sizeof (struct tcpiphdr)) {
684 if ((m = m_pullup(m, sizeof (struct tcpiphdr)))
686 TCPSTAT_INC(tcps_rcvshort);
687 return (IPPROTO_DONE);
690 ip = mtod(m, struct ip *);
691 th = (struct tcphdr *)((caddr_t)ip + off0);
692 tlen = ntohs(ip->ip_len) - off0;
694 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
695 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
696 th->th_sum = m->m_pkthdr.csum_data;
698 th->th_sum = in_pseudo(ip->ip_src.s_addr,
700 htonl(m->m_pkthdr.csum_data + tlen +
702 th->th_sum ^= 0xffff;
704 struct ipovly *ipov = (struct ipovly *)ip;
707 * Checksum extended TCP header and data.
710 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
711 ipov->ih_len = htons(tlen);
712 th->th_sum = in_cksum(m, len);
713 /* Reset length for SDT probes. */
714 ip->ip_len = htons(tlen + off0);
718 TCPSTAT_INC(tcps_rcvbadsum);
721 /* Re-initialization for later version check */
722 ip->ip_v = IPVERSION;
728 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
730 #if defined(INET) && defined(INET6)
738 * Check that TCP offset makes sense,
739 * pull out TCP options and adjust length. XXX
741 off = th->th_off << 2;
742 if (off < sizeof (struct tcphdr) || off > tlen) {
743 TCPSTAT_INC(tcps_rcvbadoff);
746 tlen -= off; /* tlen is used instead of ti->ti_len */
747 if (off > sizeof (struct tcphdr)) {
750 IP6_EXTHDR_CHECK(m, off0, off, IPPROTO_DONE);
751 ip6 = mtod(m, struct ip6_hdr *);
752 th = (struct tcphdr *)((caddr_t)ip6 + off0);
755 #if defined(INET) && defined(INET6)
760 if (m->m_len < sizeof(struct ip) + off) {
761 if ((m = m_pullup(m, sizeof (struct ip) + off))
763 TCPSTAT_INC(tcps_rcvshort);
764 return (IPPROTO_DONE);
766 ip = mtod(m, struct ip *);
767 th = (struct tcphdr *)((caddr_t)ip + off0);
771 optlen = off - sizeof (struct tcphdr);
772 optp = (u_char *)(th + 1);
774 thflags = th->th_flags;
777 * Convert TCP protocol specific fields to host format.
779 tcp_fields_to_host(th);
782 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options.
784 drop_hdrlen = off0 + off;
787 * Locate pcb for segment; if we're likely to add or remove a
788 * connection then first acquire pcbinfo lock. There are three cases
789 * where we might discover later we need a write lock despite the
790 * flags: ACKs moving a connection out of the syncache, ACKs for a
791 * connection in TIMEWAIT and SYNs not targeting a listening socket.
793 if ((thflags & (TH_FIN | TH_RST)) != 0) {
794 INP_INFO_RLOCK(&V_tcbinfo);
795 ti_locked = TI_RLOCKED;
797 ti_locked = TI_UNLOCKED;
800 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
804 (isipv6 && (m->m_flags & M_IP6_NEXTHOP))
806 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP))
809 #if defined(INET) && !defined(INET6)
810 (m->m_flags & M_IP_NEXTHOP)
813 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
817 if (ti_locked == TI_RLOCKED) {
818 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
820 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
824 if (isipv6 && fwd_tag != NULL) {
825 struct sockaddr_in6 *next_hop6;
827 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1);
829 * Transparently forwarded. Pretend to be the destination.
830 * Already got one like this?
832 inp = in6_pcblookup_mbuf(&V_tcbinfo,
833 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport,
834 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m);
837 * It's new. Try to find the ambushing socket.
838 * Because we've rewritten the destination address,
839 * any hardware-generated hash is ignored.
841 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src,
842 th->th_sport, &next_hop6->sin6_addr,
843 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) :
844 th->th_dport, INPLOOKUP_WILDCARD |
845 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
848 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src,
849 th->th_sport, &ip6->ip6_dst, th->th_dport,
850 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
851 m->m_pkthdr.rcvif, m);
854 #if defined(INET6) && defined(INET)
858 if (fwd_tag != NULL) {
859 struct sockaddr_in *next_hop;
861 next_hop = (struct sockaddr_in *)(fwd_tag+1);
863 * Transparently forwarded. Pretend to be the destination.
864 * already got one like this?
866 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport,
867 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB,
868 m->m_pkthdr.rcvif, m);
871 * It's new. Try to find the ambushing socket.
872 * Because we've rewritten the destination address,
873 * any hardware-generated hash is ignored.
875 inp = in_pcblookup(&V_tcbinfo, ip->ip_src,
876 th->th_sport, next_hop->sin_addr,
877 next_hop->sin_port ? ntohs(next_hop->sin_port) :
878 th->th_dport, INPLOOKUP_WILDCARD |
879 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
882 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src,
883 th->th_sport, ip->ip_dst, th->th_dport,
884 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
885 m->m_pkthdr.rcvif, m);
889 * If the INPCB does not exist then all data in the incoming
890 * segment is discarded and an appropriate RST is sent back.
891 * XXX MRT Send RST using which routing table?
895 * Log communication attempts to ports that are not
898 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) ||
899 tcp_log_in_vain == 2) {
900 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6)))
901 log(LOG_INFO, "%s; %s: Connection attempt "
902 "to closed port\n", s, __func__);
905 * When blackholing do not respond with a RST but
906 * completely ignore the segment and drop it.
908 if ((V_blackhole == 1 && (thflags & TH_SYN)) ||
912 rstreason = BANDLIM_RST_CLOSEDPORT;
915 INP_WLOCK_ASSERT(inp);
916 if ((inp->inp_flowtype == M_HASHTYPE_NONE) &&
917 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) &&
918 ((inp->inp_socket == NULL) ||
919 (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0)) {
920 inp->inp_flowid = m->m_pkthdr.flowid;
921 inp->inp_flowtype = M_HASHTYPE_GET(m);
925 if (isipv6 && ipsec6_in_reject(m, inp)) {
929 if (ipsec4_in_reject(m, inp) != 0) {
935 * Check the minimum TTL for socket.
937 if (inp->inp_ip_minttl != 0) {
939 if (isipv6 && inp->inp_ip_minttl > ip6->ip6_hlim)
943 if (inp->inp_ip_minttl > ip->ip_ttl)
948 * A previous connection in TIMEWAIT state is supposed to catch stray
949 * or duplicate segments arriving late. If this segment was a
950 * legitimate new connection attempt, the old INPCB gets removed and
951 * we can try again to find a listening socket.
953 * At this point, due to earlier optimism, we may hold only an inpcb
954 * lock, and not the inpcbinfo write lock. If so, we need to try to
955 * acquire it, or if that fails, acquire a reference on the inpcb,
956 * drop all locks, acquire a global write lock, and then re-acquire
957 * the inpcb lock. We may at that point discover that another thread
958 * has tried to free the inpcb, in which case we need to loop back
959 * and try to find a new inpcb to deliver to.
961 * XXXRW: It may be time to rethink timewait locking.
964 if (inp->inp_flags & INP_TIMEWAIT) {
965 if (ti_locked == TI_UNLOCKED) {
966 if (INP_INFO_TRY_RLOCK(&V_tcbinfo) == 0) {
969 INP_INFO_RLOCK(&V_tcbinfo);
970 ti_locked = TI_RLOCKED;
972 if (in_pcbrele_wlocked(inp)) {
977 ti_locked = TI_RLOCKED;
979 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
981 if (thflags & TH_SYN)
982 tcp_dooptions(&to, optp, optlen, TO_SYN);
984 * NB: tcp_twcheck unlocks the INP and frees the mbuf.
986 if (tcp_twcheck(inp, &to, th, m, tlen))
988 INP_INFO_RUNLOCK(&V_tcbinfo);
989 return (IPPROTO_DONE);
992 * The TCPCB may no longer exist if the connection is winding
993 * down or it is in the CLOSED state. Either way we drop the
994 * segment and send an appropriate response.
997 if (tp == NULL || tp->t_state == TCPS_CLOSED) {
998 rstreason = BANDLIM_RST_CLOSEDPORT;
1003 if (tp->t_flags & TF_TOE) {
1004 tcp_offload_input(tp, m);
1005 m = NULL; /* consumed by the TOE driver */
1011 * We've identified a valid inpcb, but it could be that we need an
1012 * inpcbinfo write lock but don't hold it. In this case, attempt to
1013 * acquire using the same strategy as the TIMEWAIT case above. If we
1014 * relock, we have to jump back to 'relocked' as the connection might
1015 * now be in TIMEWAIT.
1018 if ((thflags & (TH_FIN | TH_RST)) != 0)
1019 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1021 if (!((tp->t_state == TCPS_ESTABLISHED && (thflags & TH_SYN) == 0) ||
1022 (tp->t_state == TCPS_LISTEN && (thflags & TH_SYN)))) {
1023 if (ti_locked == TI_UNLOCKED) {
1024 if (INP_INFO_TRY_RLOCK(&V_tcbinfo) == 0) {
1027 INP_INFO_RLOCK(&V_tcbinfo);
1028 ti_locked = TI_RLOCKED;
1030 if (in_pcbrele_wlocked(inp)) {
1036 ti_locked = TI_RLOCKED;
1038 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1042 INP_WLOCK_ASSERT(inp);
1043 if (mac_inpcb_check_deliver(inp, m))
1046 so = inp->inp_socket;
1047 KASSERT(so != NULL, ("%s: so == NULL", __func__));
1049 if (so->so_options & SO_DEBUG) {
1050 ostate = tp->t_state;
1053 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
1056 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
1059 #endif /* TCPDEBUG */
1061 * When the socket is accepting connections (the INPCB is in LISTEN
1062 * state) we look into the SYN cache if this is a new connection
1063 * attempt or the completion of a previous one.
1065 if (so->so_options & SO_ACCEPTCONN) {
1066 struct in_conninfo inc;
1068 KASSERT(tp->t_state == TCPS_LISTEN, ("%s: so accepting but "
1069 "tp not listening", __func__));
1070 bzero(&inc, sizeof(inc));
1073 inc.inc_flags |= INC_ISIPV6;
1074 inc.inc6_faddr = ip6->ip6_src;
1075 inc.inc6_laddr = ip6->ip6_dst;
1079 inc.inc_faddr = ip->ip_src;
1080 inc.inc_laddr = ip->ip_dst;
1082 inc.inc_fport = th->th_sport;
1083 inc.inc_lport = th->th_dport;
1084 inc.inc_fibnum = so->so_fibnum;
1087 * Check for an existing connection attempt in syncache if
1088 * the flag is only ACK. A successful lookup creates a new
1089 * socket appended to the listen queue in SYN_RECEIVED state.
1091 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
1093 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1095 * Parse the TCP options here because
1096 * syncookies need access to the reflected
1099 tcp_dooptions(&to, optp, optlen, 0);
1101 * NB: syncache_expand() doesn't unlock
1102 * inp and tcpinfo locks.
1104 if (!syncache_expand(&inc, &to, th, &so, m)) {
1106 * No syncache entry or ACK was not
1107 * for our SYN/ACK. Send a RST.
1108 * NB: syncache did its own logging
1109 * of the failure cause.
1111 rstreason = BANDLIM_RST_OPENPORT;
1116 * We completed the 3-way handshake
1117 * but could not allocate a socket
1118 * either due to memory shortage,
1119 * listen queue length limits or
1120 * global socket limits. Send RST
1121 * or wait and have the remote end
1122 * retransmit the ACK for another
1125 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1126 log(LOG_DEBUG, "%s; %s: Listen socket: "
1127 "Socket allocation failed due to "
1128 "limits or memory shortage, %s\n",
1130 V_tcp_sc_rst_sock_fail ?
1131 "sending RST" : "try again");
1132 if (V_tcp_sc_rst_sock_fail) {
1133 rstreason = BANDLIM_UNLIMITED;
1139 * Socket is created in state SYN_RECEIVED.
1140 * Unlock the listen socket, lock the newly
1141 * created socket and update the tp variable.
1143 INP_WUNLOCK(inp); /* listen socket */
1144 inp = sotoinpcb(so);
1146 * New connection inpcb is already locked by
1147 * syncache_expand().
1149 INP_WLOCK_ASSERT(inp);
1150 tp = intotcpcb(inp);
1151 KASSERT(tp->t_state == TCPS_SYN_RECEIVED,
1152 ("%s: ", __func__));
1153 #ifdef TCP_SIGNATURE
1154 if (sig_checked == 0) {
1155 tcp_dooptions(&to, optp, optlen,
1156 (thflags & TH_SYN) ? TO_SYN : 0);
1157 if (!tcp_signature_verify_input(m, off0, tlen,
1158 optlen, &to, th, tp->t_flags)) {
1161 * In SYN_SENT state if it receives an
1162 * RST, it is allowed for further
1165 if ((thflags & TH_RST) == 0 ||
1166 (tp->t_state == TCPS_SYN_SENT) == 0)
1174 * Process the segment and the data it
1175 * contains. tcp_do_segment() consumes
1176 * the mbuf chain and unlocks the inpcb.
1178 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen,
1180 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1181 return (IPPROTO_DONE);
1184 * Segment flag validation for new connection attempts:
1186 * Our (SYN|ACK) response was rejected.
1187 * Check with syncache and remove entry to prevent
1190 * NB: syncache_chkrst does its own logging of failure
1193 if (thflags & TH_RST) {
1194 syncache_chkrst(&inc, th);
1198 * We can't do anything without SYN.
1200 if ((thflags & TH_SYN) == 0) {
1201 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1202 log(LOG_DEBUG, "%s; %s: Listen socket: "
1203 "SYN is missing, segment ignored\n",
1205 TCPSTAT_INC(tcps_badsyn);
1209 * (SYN|ACK) is bogus on a listen socket.
1211 if (thflags & TH_ACK) {
1212 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1213 log(LOG_DEBUG, "%s; %s: Listen socket: "
1214 "SYN|ACK invalid, segment rejected\n",
1216 syncache_badack(&inc); /* XXX: Not needed! */
1217 TCPSTAT_INC(tcps_badsyn);
1218 rstreason = BANDLIM_RST_OPENPORT;
1222 * If the drop_synfin option is enabled, drop all
1223 * segments with both the SYN and FIN bits set.
1224 * This prevents e.g. nmap from identifying the
1226 * XXX: Poor reasoning. nmap has other methods
1227 * and is constantly refining its stack detection
1229 * XXX: This is a violation of the TCP specification
1230 * and was used by RFC1644.
1232 if ((thflags & TH_FIN) && V_drop_synfin) {
1233 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1234 log(LOG_DEBUG, "%s; %s: Listen socket: "
1235 "SYN|FIN segment ignored (based on "
1236 "sysctl setting)\n", s, __func__);
1237 TCPSTAT_INC(tcps_badsyn);
1241 * Segment's flags are (SYN) or (SYN|FIN).
1243 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored
1244 * as they do not affect the state of the TCP FSM.
1245 * The data pointed to by TH_URG and th_urp is ignored.
1247 KASSERT((thflags & (TH_RST|TH_ACK)) == 0,
1248 ("%s: Listen socket: TH_RST or TH_ACK set", __func__));
1249 KASSERT(thflags & (TH_SYN),
1250 ("%s: Listen socket: TH_SYN not set", __func__));
1253 * If deprecated address is forbidden,
1254 * we do not accept SYN to deprecated interface
1255 * address to prevent any new inbound connection from
1256 * getting established.
1257 * When we do not accept SYN, we send a TCP RST,
1258 * with deprecated source address (instead of dropping
1259 * it). We compromise it as it is much better for peer
1260 * to send a RST, and RST will be the final packet
1263 * If we do not forbid deprecated addresses, we accept
1264 * the SYN packet. RFC2462 does not suggest dropping
1266 * If we decipher RFC2462 5.5.4, it says like this:
1267 * 1. use of deprecated addr with existing
1268 * communication is okay - "SHOULD continue to be
1270 * 2. use of it with new communication:
1271 * (2a) "SHOULD NOT be used if alternate address
1272 * with sufficient scope is available"
1273 * (2b) nothing mentioned otherwise.
1274 * Here we fall into (2b) case as we have no choice in
1275 * our source address selection - we must obey the peer.
1277 * The wording in RFC2462 is confusing, and there are
1278 * multiple description text for deprecated address
1279 * handling - worse, they are not exactly the same.
1280 * I believe 5.5.4 is the best one, so we follow 5.5.4.
1282 if (isipv6 && !V_ip6_use_deprecated) {
1283 struct in6_ifaddr *ia6;
1285 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */);
1287 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1288 ifa_free(&ia6->ia_ifa);
1289 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1290 log(LOG_DEBUG, "%s; %s: Listen socket: "
1291 "Connection attempt to deprecated "
1292 "IPv6 address rejected\n",
1294 rstreason = BANDLIM_RST_OPENPORT;
1298 ifa_free(&ia6->ia_ifa);
1302 * Basic sanity checks on incoming SYN requests:
1303 * Don't respond if the destination is a link layer
1304 * broadcast according to RFC1122 4.2.3.10, p. 104.
1305 * If it is from this socket it must be forged.
1306 * Don't respond if the source or destination is a
1307 * global or subnet broad- or multicast address.
1308 * Note that it is quite possible to receive unicast
1309 * link-layer packets with a broadcast IP address. Use
1310 * in_broadcast() to find them.
1312 if (m->m_flags & (M_BCAST|M_MCAST)) {
1313 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1314 log(LOG_DEBUG, "%s; %s: Listen socket: "
1315 "Connection attempt from broad- or multicast "
1316 "link layer address ignored\n", s, __func__);
1321 if (th->th_dport == th->th_sport &&
1322 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) {
1323 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1324 log(LOG_DEBUG, "%s; %s: Listen socket: "
1325 "Connection attempt to/from self "
1326 "ignored\n", s, __func__);
1329 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1330 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) {
1331 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1332 log(LOG_DEBUG, "%s; %s: Listen socket: "
1333 "Connection attempt from/to multicast "
1334 "address ignored\n", s, __func__);
1339 #if defined(INET) && defined(INET6)
1344 if (th->th_dport == th->th_sport &&
1345 ip->ip_dst.s_addr == ip->ip_src.s_addr) {
1346 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1347 log(LOG_DEBUG, "%s; %s: Listen socket: "
1348 "Connection attempt from/to self "
1349 "ignored\n", s, __func__);
1352 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1353 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1354 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1355 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
1356 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1357 log(LOG_DEBUG, "%s; %s: Listen socket: "
1358 "Connection attempt from/to broad- "
1359 "or multicast address ignored\n",
1366 * SYN appears to be valid. Create compressed TCP state
1370 if (so->so_options & SO_DEBUG)
1371 tcp_trace(TA_INPUT, ostate, tp,
1372 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1374 TCP_PROBE3(debug__input, tp, th, mtod(m, const char *));
1375 tcp_dooptions(&to, optp, optlen, TO_SYN);
1376 syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL);
1378 * Entry added to syncache and mbuf consumed.
1379 * Only the listen socket is unlocked by syncache_add().
1381 if (ti_locked == TI_RLOCKED) {
1382 INP_INFO_RUNLOCK(&V_tcbinfo);
1383 ti_locked = TI_UNLOCKED;
1385 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1386 return (IPPROTO_DONE);
1387 } else if (tp->t_state == TCPS_LISTEN) {
1389 * When a listen socket is torn down the SO_ACCEPTCONN
1390 * flag is removed first while connections are drained
1391 * from the accept queue in a unlock/lock cycle of the
1392 * ACCEPT_LOCK, opening a race condition allowing a SYN
1393 * attempt go through unhandled.
1398 #ifdef TCP_SIGNATURE
1399 if (sig_checked == 0) {
1400 tcp_dooptions(&to, optp, optlen,
1401 (thflags & TH_SYN) ? TO_SYN : 0);
1402 if (!tcp_signature_verify_input(m, off0, tlen, optlen, &to,
1406 * In SYN_SENT state if it receives an RST, it is
1407 * allowed for further processing.
1409 if ((thflags & TH_RST) == 0 ||
1410 (tp->t_state == TCPS_SYN_SENT) == 0)
1417 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th);
1420 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later
1421 * state. tcp_do_segment() always consumes the mbuf chain, unlocks
1422 * the inpcb, and unlocks pcbinfo.
1424 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos, ti_locked);
1425 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1426 return (IPPROTO_DONE);
1429 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th);
1431 if (ti_locked == TI_RLOCKED) {
1432 INP_INFO_RUNLOCK(&V_tcbinfo);
1433 ti_locked = TI_UNLOCKED;
1437 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset "
1438 "ti_locked: %d", __func__, ti_locked));
1439 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1444 tcp_dropwithreset(m, th, tp, tlen, rstreason);
1447 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
1448 m = NULL; /* mbuf chain got consumed. */
1453 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th);
1455 if (ti_locked == TI_RLOCKED) {
1456 INP_INFO_RUNLOCK(&V_tcbinfo);
1457 ti_locked = TI_UNLOCKED;
1461 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock "
1462 "ti_locked: %d", __func__, ti_locked));
1463 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1471 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1476 return (IPPROTO_DONE);
1480 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so,
1481 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos,
1484 int thflags, acked, ourfinisacked, needoutput = 0;
1485 int rstreason, todrop, win;
1488 struct in_conninfo *inc;
1494 * The size of tcp_saveipgen must be the size of the max ip header,
1497 u_char tcp_saveipgen[IP6_HDR_LEN];
1498 struct tcphdr tcp_savetcp;
1501 thflags = th->th_flags;
1502 inc = &tp->t_inpcb->inp_inc;
1503 tp->sackhint.last_sack_ack = 0;
1506 * If this is either a state-changing packet or current state isn't
1507 * established, we require a write lock on tcbinfo. Otherwise, we
1508 * allow the tcbinfo to be in either alocked or unlocked, as the
1509 * caller may have unnecessarily acquired a write lock due to a race.
1511 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 ||
1512 tp->t_state != TCPS_ESTABLISHED) {
1513 KASSERT(ti_locked == TI_RLOCKED, ("%s ti_locked %d for "
1514 "SYN/FIN/RST/!EST", __func__, ti_locked));
1515 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1518 if (ti_locked == TI_RLOCKED)
1519 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1521 KASSERT(ti_locked == TI_UNLOCKED, ("%s: EST "
1522 "ti_locked: %d", __func__, ti_locked));
1523 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1527 INP_WLOCK_ASSERT(tp->t_inpcb);
1528 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN",
1530 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT",
1534 /* Save segment, if requested. */
1535 tcp_pcap_add(th, m, &(tp->t_inpkts));
1539 * Segment received on connection.
1540 * Reset idle time and keep-alive timer.
1541 * XXX: This should be done after segment
1542 * validation to ignore broken/spoofed segs.
1544 tp->t_rcvtime = ticks;
1545 if (TCPS_HAVEESTABLISHED(tp->t_state))
1546 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
1549 * Scale up the window into a 32-bit value.
1550 * For the SYN_SENT state the scale is zero.
1552 tiwin = th->th_win << tp->snd_scale;
1555 * TCP ECN processing.
1557 if (tp->t_flags & TF_ECN_PERMIT) {
1558 if (thflags & TH_CWR)
1559 tp->t_flags &= ~TF_ECN_SND_ECE;
1560 switch (iptos & IPTOS_ECN_MASK) {
1562 tp->t_flags |= TF_ECN_SND_ECE;
1563 TCPSTAT_INC(tcps_ecn_ce);
1565 case IPTOS_ECN_ECT0:
1566 TCPSTAT_INC(tcps_ecn_ect0);
1568 case IPTOS_ECN_ECT1:
1569 TCPSTAT_INC(tcps_ecn_ect1);
1573 /* Process a packet differently from RFC3168. */
1574 cc_ecnpkt_handler(tp, th, iptos);
1576 /* Congestion experienced. */
1577 if (thflags & TH_ECE) {
1578 cc_cong_signal(tp, th, CC_ECN);
1583 * Parse options on any incoming segment.
1585 tcp_dooptions(&to, (u_char *)(th + 1),
1586 (th->th_off << 2) - sizeof(struct tcphdr),
1587 (thflags & TH_SYN) ? TO_SYN : 0);
1590 * If echoed timestamp is later than the current time,
1591 * fall back to non RFC1323 RTT calculation. Normalize
1592 * timestamp if syncookies were used when this connection
1595 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1596 to.to_tsecr -= tp->ts_offset;
1597 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks()))
1601 * If timestamps were negotiated during SYN/ACK they should
1602 * appear on every segment during this session and vice versa.
1604 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) {
1605 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1606 log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1607 "no action\n", s, __func__);
1611 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) {
1612 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1613 log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
1614 "no action\n", s, __func__);
1620 * Process options only when we get SYN/ACK back. The SYN case
1621 * for incoming connections is handled in tcp_syncache.
1622 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
1623 * or <SYN,ACK>) segment itself is never scaled.
1624 * XXX this is traditional behavior, may need to be cleaned up.
1626 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1627 if ((to.to_flags & TOF_SCALE) &&
1628 (tp->t_flags & TF_REQ_SCALE)) {
1629 tp->t_flags |= TF_RCVD_SCALE;
1630 tp->snd_scale = to.to_wscale;
1633 * Initial send window. It will be updated with
1634 * the next incoming segment to the scaled value.
1636 tp->snd_wnd = th->th_win;
1637 if (to.to_flags & TOF_TS) {
1638 tp->t_flags |= TF_RCVD_TSTMP;
1639 tp->ts_recent = to.to_tsval;
1640 tp->ts_recent_age = tcp_ts_getticks();
1642 if (to.to_flags & TOF_MSS)
1643 tcp_mss(tp, to.to_mss);
1644 if ((tp->t_flags & TF_SACK_PERMIT) &&
1645 (to.to_flags & TOF_SACKPERM) == 0)
1646 tp->t_flags &= ~TF_SACK_PERMIT;
1650 * Header prediction: check for the two common cases
1651 * of a uni-directional data xfer. If the packet has
1652 * no control flags, is in-sequence, the window didn't
1653 * change and we're not retransmitting, it's a
1654 * candidate. If the length is zero and the ack moved
1655 * forward, we're the sender side of the xfer. Just
1656 * free the data acked & wake any higher level process
1657 * that was blocked waiting for space. If the length
1658 * is non-zero and the ack didn't move, we're the
1659 * receiver side. If we're getting packets in-order
1660 * (the reassembly queue is empty), add the data to
1661 * the socket buffer and note that we need a delayed ack.
1662 * Make sure that the hidden state-flags are also off.
1663 * Since we check for TCPS_ESTABLISHED first, it can only
1666 if (tp->t_state == TCPS_ESTABLISHED &&
1667 th->th_seq == tp->rcv_nxt &&
1668 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1669 tp->snd_nxt == tp->snd_max &&
1670 tiwin && tiwin == tp->snd_wnd &&
1671 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1672 LIST_EMPTY(&tp->t_segq) &&
1673 ((to.to_flags & TOF_TS) == 0 ||
1674 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) {
1677 * If last ACK falls within this segment's sequence numbers,
1678 * record the timestamp.
1679 * NOTE that the test is modified according to the latest
1680 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1682 if ((to.to_flags & TOF_TS) != 0 &&
1683 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1684 tp->ts_recent_age = tcp_ts_getticks();
1685 tp->ts_recent = to.to_tsval;
1689 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1690 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1691 !IN_RECOVERY(tp->t_flags) &&
1692 (to.to_flags & TOF_SACK) == 0 &&
1693 TAILQ_EMPTY(&tp->snd_holes)) {
1695 * This is a pure ack for outstanding data.
1697 if (ti_locked == TI_RLOCKED)
1698 INP_INFO_RUNLOCK(&V_tcbinfo);
1699 ti_locked = TI_UNLOCKED;
1701 TCPSTAT_INC(tcps_predack);
1704 * "bad retransmit" recovery.
1706 if (tp->t_rxtshift == 1 &&
1707 tp->t_flags & TF_PREVVALID &&
1708 (int)(ticks - tp->t_badrxtwin) < 0) {
1709 cc_cong_signal(tp, th, CC_RTO_ERR);
1713 * Recalculate the transmit timer / rtt.
1715 * Some boxes send broken timestamp replies
1716 * during the SYN+ACK phase, ignore
1717 * timestamps of 0 or we could calculate a
1718 * huge RTT and blow up the retransmit timer.
1720 if ((to.to_flags & TOF_TS) != 0 &&
1724 t = tcp_ts_getticks() - to.to_tsecr;
1725 if (!tp->t_rttlow || tp->t_rttlow > t)
1728 TCP_TS_TO_TICKS(t) + 1);
1729 } else if (tp->t_rtttime &&
1730 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1731 if (!tp->t_rttlow ||
1732 tp->t_rttlow > ticks - tp->t_rtttime)
1733 tp->t_rttlow = ticks - tp->t_rtttime;
1735 ticks - tp->t_rtttime);
1737 acked = BYTES_THIS_ACK(tp, th);
1739 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
1740 hhook_run_tcp_est_in(tp, th, &to);
1742 TCPSTAT_INC(tcps_rcvackpack);
1743 TCPSTAT_ADD(tcps_rcvackbyte, acked);
1744 sbdrop(&so->so_snd, acked);
1745 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1746 SEQ_LEQ(th->th_ack, tp->snd_recover))
1747 tp->snd_recover = th->th_ack - 1;
1750 * Let the congestion control algorithm update
1751 * congestion control related information. This
1752 * typically means increasing the congestion
1755 cc_ack_received(tp, th, CC_ACK);
1757 tp->snd_una = th->th_ack;
1759 * Pull snd_wl2 up to prevent seq wrap relative
1762 tp->snd_wl2 = th->th_ack;
1767 * If all outstanding data are acked, stop
1768 * retransmit timer, otherwise restart timer
1769 * using current (possibly backed-off) value.
1770 * If process is waiting for space,
1771 * wakeup/selwakeup/signal. If data
1772 * are ready to send, let tcp_output
1773 * decide between more output or persist.
1776 if (so->so_options & SO_DEBUG)
1777 tcp_trace(TA_INPUT, ostate, tp,
1778 (void *)tcp_saveipgen,
1781 TCP_PROBE3(debug__input, tp, th,
1782 mtod(m, const char *));
1783 if (tp->snd_una == tp->snd_max)
1784 tcp_timer_activate(tp, TT_REXMT, 0);
1785 else if (!tcp_timer_active(tp, TT_PERSIST))
1786 tcp_timer_activate(tp, TT_REXMT,
1789 if (sbavail(&so->so_snd))
1790 (void) tcp_output(tp);
1793 } else if (th->th_ack == tp->snd_una &&
1794 tlen <= sbspace(&so->so_rcv)) {
1795 int newsize = 0; /* automatic sockbuf scaling */
1798 * This is a pure, in-sequence data packet with
1799 * nothing on the reassembly queue and we have enough
1800 * buffer space to take it.
1802 if (ti_locked == TI_RLOCKED)
1803 INP_INFO_RUNLOCK(&V_tcbinfo);
1804 ti_locked = TI_UNLOCKED;
1806 /* Clean receiver SACK report if present */
1807 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks)
1808 tcp_clean_sackreport(tp);
1809 TCPSTAT_INC(tcps_preddat);
1810 tp->rcv_nxt += tlen;
1812 * Pull snd_wl1 up to prevent seq wrap relative to
1815 tp->snd_wl1 = th->th_seq;
1817 * Pull rcv_up up to prevent seq wrap relative to
1820 tp->rcv_up = tp->rcv_nxt;
1821 TCPSTAT_INC(tcps_rcvpack);
1822 TCPSTAT_ADD(tcps_rcvbyte, tlen);
1824 if (so->so_options & SO_DEBUG)
1825 tcp_trace(TA_INPUT, ostate, tp,
1826 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1828 TCP_PROBE3(debug__input, tp, th, mtod(m, const char *));
1831 * Automatic sizing of receive socket buffer. Often the send
1832 * buffer size is not optimally adjusted to the actual network
1833 * conditions at hand (delay bandwidth product). Setting the
1834 * buffer size too small limits throughput on links with high
1835 * bandwidth and high delay (eg. trans-continental/oceanic links).
1837 * On the receive side the socket buffer memory is only rarely
1838 * used to any significant extent. This allows us to be much
1839 * more aggressive in scaling the receive socket buffer. For
1840 * the case that the buffer space is actually used to a large
1841 * extent and we run out of kernel memory we can simply drop
1842 * the new segments; TCP on the sender will just retransmit it
1843 * later. Setting the buffer size too big may only consume too
1844 * much kernel memory if the application doesn't read() from
1845 * the socket or packet loss or reordering makes use of the
1848 * The criteria to step up the receive buffer one notch are:
1849 * 1. Application has not set receive buffer size with
1850 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE.
1851 * 2. the number of bytes received during the time it takes
1852 * one timestamp to be reflected back to us (the RTT);
1853 * 3. received bytes per RTT is within seven eighth of the
1854 * current socket buffer size;
1855 * 4. receive buffer size has not hit maximal automatic size;
1857 * This algorithm does one step per RTT at most and only if
1858 * we receive a bulk stream w/o packet losses or reorderings.
1859 * Shrinking the buffer during idle times is not necessary as
1860 * it doesn't consume any memory when idle.
1862 * TODO: Only step up if the application is actually serving
1863 * the buffer to better manage the socket buffer resources.
1865 if (V_tcp_do_autorcvbuf &&
1866 (to.to_flags & TOF_TS) &&
1868 (so->so_rcv.sb_flags & SB_AUTOSIZE)) {
1869 if (TSTMP_GT(to.to_tsecr, tp->rfbuf_ts) &&
1870 to.to_tsecr - tp->rfbuf_ts < hz) {
1872 (so->so_rcv.sb_hiwat / 8 * 7) &&
1873 so->so_rcv.sb_hiwat <
1874 V_tcp_autorcvbuf_max) {
1876 min(so->so_rcv.sb_hiwat +
1877 V_tcp_autorcvbuf_inc,
1878 V_tcp_autorcvbuf_max);
1880 /* Start over with next RTT. */
1884 tp->rfbuf_cnt += tlen; /* add up */
1887 /* Add data to socket buffer. */
1888 SOCKBUF_LOCK(&so->so_rcv);
1889 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1893 * Set new socket buffer size.
1894 * Give up when limit is reached.
1897 if (!sbreserve_locked(&so->so_rcv,
1899 so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1900 m_adj(m, drop_hdrlen); /* delayed header drop */
1901 sbappendstream_locked(&so->so_rcv, m, 0);
1903 /* NB: sorwakeup_locked() does an implicit unlock. */
1904 sorwakeup_locked(so);
1905 if (DELAY_ACK(tp, tlen)) {
1906 tp->t_flags |= TF_DELACK;
1908 tp->t_flags |= TF_ACKNOW;
1916 * Calculate amount of space in receive window,
1917 * and then do TCP input processing.
1918 * Receive window is amount of space in rcv queue,
1919 * but not less than advertised window.
1921 win = sbspace(&so->so_rcv);
1924 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1926 /* Reset receive buffer auto scaling when not in bulk receive mode. */
1930 switch (tp->t_state) {
1933 * If the state is SYN_RECEIVED:
1934 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1936 case TCPS_SYN_RECEIVED:
1937 if ((thflags & TH_ACK) &&
1938 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1939 SEQ_GT(th->th_ack, tp->snd_max))) {
1940 rstreason = BANDLIM_RST_OPENPORT;
1946 * If the state is SYN_SENT:
1947 * if seg contains an ACK, but not for our SYN, drop the input.
1948 * if seg contains a RST, then drop the connection.
1949 * if seg does not contain SYN, then drop it.
1950 * Otherwise this is an acceptable SYN segment
1951 * initialize tp->rcv_nxt and tp->irs
1952 * if seg contains ack then advance tp->snd_una
1953 * if seg contains an ECE and ECN support is enabled, the stream
1955 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1956 * arrange for segment to be acked (eventually)
1957 * continue processing rest of data/controls, beginning with URG
1960 if ((thflags & TH_ACK) &&
1961 (SEQ_LEQ(th->th_ack, tp->iss) ||
1962 SEQ_GT(th->th_ack, tp->snd_max))) {
1963 rstreason = BANDLIM_UNLIMITED;
1966 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) {
1967 TCP_PROBE5(connect__refused, NULL, tp,
1968 mtod(m, const char *), tp, th);
1969 tp = tcp_drop(tp, ECONNREFUSED);
1971 if (thflags & TH_RST)
1973 if (!(thflags & TH_SYN))
1976 tp->irs = th->th_seq;
1978 if (thflags & TH_ACK) {
1979 TCPSTAT_INC(tcps_connects);
1982 mac_socketpeer_set_from_mbuf(m, so);
1984 /* Do window scaling on this connection? */
1985 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1986 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1987 tp->rcv_scale = tp->request_r_scale;
1989 tp->rcv_adv += imin(tp->rcv_wnd,
1990 TCP_MAXWIN << tp->rcv_scale);
1991 tp->snd_una++; /* SYN is acked */
1993 * If there's data, delay ACK; if there's also a FIN
1994 * ACKNOW will be turned on later.
1996 if (DELAY_ACK(tp, tlen) && tlen != 0)
1997 tcp_timer_activate(tp, TT_DELACK,
2000 tp->t_flags |= TF_ACKNOW;
2002 if ((thflags & TH_ECE) && V_tcp_do_ecn) {
2003 tp->t_flags |= TF_ECN_PERMIT;
2004 TCPSTAT_INC(tcps_ecn_shs);
2008 * Received <SYN,ACK> in SYN_SENT[*] state.
2010 * SYN_SENT --> ESTABLISHED
2011 * SYN_SENT* --> FIN_WAIT_1
2013 tp->t_starttime = ticks;
2014 if (tp->t_flags & TF_NEEDFIN) {
2015 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2016 tp->t_flags &= ~TF_NEEDFIN;
2019 tcp_state_change(tp, TCPS_ESTABLISHED);
2020 TCP_PROBE5(connect__established, NULL, tp,
2021 mtod(m, const char *), tp, th);
2023 tcp_timer_activate(tp, TT_KEEP,
2028 * Received initial SYN in SYN-SENT[*] state =>
2029 * simultaneous open.
2030 * If it succeeds, connection is * half-synchronized.
2031 * Otherwise, do 3-way handshake:
2032 * SYN-SENT -> SYN-RECEIVED
2033 * SYN-SENT* -> SYN-RECEIVED*
2035 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
2036 tcp_timer_activate(tp, TT_REXMT, 0);
2037 tcp_state_change(tp, TCPS_SYN_RECEIVED);
2040 KASSERT(ti_locked == TI_RLOCKED, ("%s: trimthenstep6: "
2041 "ti_locked %d", __func__, ti_locked));
2042 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2043 INP_WLOCK_ASSERT(tp->t_inpcb);
2046 * Advance th->th_seq to correspond to first data byte.
2047 * If data, trim to stay within window,
2048 * dropping FIN if necessary.
2051 if (tlen > tp->rcv_wnd) {
2052 todrop = tlen - tp->rcv_wnd;
2056 TCPSTAT_INC(tcps_rcvpackafterwin);
2057 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2059 tp->snd_wl1 = th->th_seq - 1;
2060 tp->rcv_up = th->th_seq;
2062 * Client side of transaction: already sent SYN and data.
2063 * If the remote host used T/TCP to validate the SYN,
2064 * our data will be ACK'd; if so, enter normal data segment
2065 * processing in the middle of step 5, ack processing.
2066 * Otherwise, goto step 6.
2068 if (thflags & TH_ACK)
2074 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
2075 * do normal processing.
2077 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later.
2081 break; /* continue normal processing */
2085 * States other than LISTEN or SYN_SENT.
2086 * First check the RST flag and sequence number since reset segments
2087 * are exempt from the timestamp and connection count tests. This
2088 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
2089 * below which allowed reset segments in half the sequence space
2090 * to fall though and be processed (which gives forged reset
2091 * segments with a random sequence number a 50 percent chance of
2092 * killing a connection).
2093 * Then check timestamp, if present.
2094 * Then check the connection count, if present.
2095 * Then check that at least some bytes of segment are within
2096 * receive window. If segment begins before rcv_nxt,
2097 * drop leading data (and SYN); if nothing left, just ack.
2099 if (thflags & TH_RST) {
2101 * RFC5961 Section 3.2
2103 * - RST drops connection only if SEG.SEQ == RCV.NXT.
2104 * - If RST is in window, we send challenge ACK.
2106 * Note: to take into account delayed ACKs, we should
2107 * test against last_ack_sent instead of rcv_nxt.
2108 * Note 2: we handle special case of closed window, not
2109 * covered by the RFC.
2111 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2112 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) ||
2113 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) {
2115 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2116 KASSERT(ti_locked == TI_RLOCKED,
2117 ("%s: TH_RST ti_locked %d, th %p tp %p",
2118 __func__, ti_locked, th, tp));
2119 KASSERT(tp->t_state != TCPS_SYN_SENT,
2120 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p",
2123 if (V_tcp_insecure_rst ||
2124 tp->last_ack_sent == th->th_seq) {
2125 TCPSTAT_INC(tcps_drops);
2126 /* Drop the connection. */
2127 switch (tp->t_state) {
2128 case TCPS_SYN_RECEIVED:
2129 so->so_error = ECONNREFUSED;
2131 case TCPS_ESTABLISHED:
2132 case TCPS_FIN_WAIT_1:
2133 case TCPS_FIN_WAIT_2:
2134 case TCPS_CLOSE_WAIT:
2135 so->so_error = ECONNRESET;
2137 tcp_state_change(tp, TCPS_CLOSED);
2143 TCPSTAT_INC(tcps_badrst);
2144 /* Send challenge ACK. */
2145 tcp_respond(tp, mtod(m, void *), th, m,
2146 tp->rcv_nxt, tp->snd_nxt, TH_ACK);
2147 tp->last_ack_sent = tp->rcv_nxt;
2155 * RFC5961 Section 4.2
2156 * Send challenge ACK for any SYN in synchronized state.
2158 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT) {
2159 KASSERT(ti_locked == TI_RLOCKED,
2160 ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked));
2161 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2163 TCPSTAT_INC(tcps_badsyn);
2164 if (V_tcp_insecure_syn &&
2165 SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2166 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
2167 tp = tcp_drop(tp, ECONNRESET);
2168 rstreason = BANDLIM_UNLIMITED;
2170 /* Send challenge ACK. */
2171 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt,
2172 tp->snd_nxt, TH_ACK);
2173 tp->last_ack_sent = tp->rcv_nxt;
2180 * RFC 1323 PAWS: If we have a timestamp reply on this segment
2181 * and it's less than ts_recent, drop it.
2183 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
2184 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
2186 /* Check to see if ts_recent is over 24 days old. */
2187 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) {
2189 * Invalidate ts_recent. If this segment updates
2190 * ts_recent, the age will be reset later and ts_recent
2191 * will get a valid value. If it does not, setting
2192 * ts_recent to zero will at least satisfy the
2193 * requirement that zero be placed in the timestamp
2194 * echo reply when ts_recent isn't valid. The
2195 * age isn't reset until we get a valid ts_recent
2196 * because we don't want out-of-order segments to be
2197 * dropped when ts_recent is old.
2201 TCPSTAT_INC(tcps_rcvduppack);
2202 TCPSTAT_ADD(tcps_rcvdupbyte, tlen);
2203 TCPSTAT_INC(tcps_pawsdrop);
2211 * In the SYN-RECEIVED state, validate that the packet belongs to
2212 * this connection before trimming the data to fit the receive
2213 * window. Check the sequence number versus IRS since we know
2214 * the sequence numbers haven't wrapped. This is a partial fix
2215 * for the "LAND" DoS attack.
2217 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
2218 rstreason = BANDLIM_RST_OPENPORT;
2222 todrop = tp->rcv_nxt - th->th_seq;
2224 if (thflags & TH_SYN) {
2234 * Following if statement from Stevens, vol. 2, p. 960.
2237 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
2239 * Any valid FIN must be to the left of the window.
2240 * At this point the FIN must be a duplicate or out
2241 * of sequence; drop it.
2246 * Send an ACK to resynchronize and drop any data.
2247 * But keep on processing for RST or ACK.
2249 tp->t_flags |= TF_ACKNOW;
2251 TCPSTAT_INC(tcps_rcvduppack);
2252 TCPSTAT_ADD(tcps_rcvdupbyte, todrop);
2254 TCPSTAT_INC(tcps_rcvpartduppack);
2255 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop);
2257 drop_hdrlen += todrop; /* drop from the top afterwards */
2258 th->th_seq += todrop;
2260 if (th->th_urp > todrop)
2261 th->th_urp -= todrop;
2269 * If new data are received on a connection after the
2270 * user processes are gone, then RST the other end.
2272 if ((so->so_state & SS_NOFDREF) &&
2273 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
2274 KASSERT(ti_locked == TI_RLOCKED, ("%s: SS_NOFDEREF && "
2275 "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked));
2276 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2278 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
2279 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data "
2280 "after socket was closed, "
2281 "sending RST and removing tcpcb\n",
2282 s, __func__, tcpstates[tp->t_state], tlen);
2286 TCPSTAT_INC(tcps_rcvafterclose);
2287 rstreason = BANDLIM_UNLIMITED;
2292 * If segment ends after window, drop trailing data
2293 * (and PUSH and FIN); if nothing left, just ACK.
2295 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
2297 TCPSTAT_INC(tcps_rcvpackafterwin);
2298 if (todrop >= tlen) {
2299 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen);
2301 * If window is closed can only take segments at
2302 * window edge, and have to drop data and PUSH from
2303 * incoming segments. Continue processing, but
2304 * remember to ack. Otherwise, drop segment
2307 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2308 tp->t_flags |= TF_ACKNOW;
2309 TCPSTAT_INC(tcps_rcvwinprobe);
2313 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2316 thflags &= ~(TH_PUSH|TH_FIN);
2320 * If last ACK falls within this segment's sequence numbers,
2321 * record its timestamp.
2323 * 1) That the test incorporates suggestions from the latest
2324 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
2325 * 2) That updating only on newer timestamps interferes with
2326 * our earlier PAWS tests, so this check should be solely
2327 * predicated on the sequence space of this segment.
2328 * 3) That we modify the segment boundary check to be
2329 * Last.ACK.Sent <= SEG.SEQ + SEG.Len
2330 * instead of RFC1323's
2331 * Last.ACK.Sent < SEG.SEQ + SEG.Len,
2332 * This modified check allows us to overcome RFC1323's
2333 * limitations as described in Stevens TCP/IP Illustrated
2334 * Vol. 2 p.869. In such cases, we can still calculate the
2335 * RTT correctly when RCV.NXT == Last.ACK.Sent.
2337 if ((to.to_flags & TOF_TS) != 0 &&
2338 SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2339 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
2340 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
2341 tp->ts_recent_age = tcp_ts_getticks();
2342 tp->ts_recent = to.to_tsval;
2346 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
2347 * flag is on (half-synchronized state), then queue data for
2348 * later processing; else drop segment and return.
2350 if ((thflags & TH_ACK) == 0) {
2351 if (tp->t_state == TCPS_SYN_RECEIVED ||
2352 (tp->t_flags & TF_NEEDSYN))
2354 else if (tp->t_flags & TF_ACKNOW)
2363 switch (tp->t_state) {
2366 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
2367 * ESTABLISHED state and continue processing.
2368 * The ACK was checked above.
2370 case TCPS_SYN_RECEIVED:
2372 TCPSTAT_INC(tcps_connects);
2374 /* Do window scaling? */
2375 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2376 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2377 tp->rcv_scale = tp->request_r_scale;
2378 tp->snd_wnd = tiwin;
2382 * SYN-RECEIVED -> ESTABLISHED
2383 * SYN-RECEIVED* -> FIN-WAIT-1
2385 tp->t_starttime = ticks;
2386 if (tp->t_flags & TF_NEEDFIN) {
2387 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2388 tp->t_flags &= ~TF_NEEDFIN;
2390 tcp_state_change(tp, TCPS_ESTABLISHED);
2391 TCP_PROBE5(accept__established, NULL, tp,
2392 mtod(m, const char *), tp, th);
2394 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
2397 * If segment contains data or ACK, will call tcp_reass()
2398 * later; if not, do so now to pass queued data to user.
2400 if (tlen == 0 && (thflags & TH_FIN) == 0)
2401 (void) tcp_reass(tp, (struct tcphdr *)0, 0,
2403 tp->snd_wl1 = th->th_seq - 1;
2407 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2408 * ACKs. If the ack is in the range
2409 * tp->snd_una < th->th_ack <= tp->snd_max
2410 * then advance tp->snd_una to th->th_ack and drop
2411 * data from the retransmission queue. If this ACK reflects
2412 * more up to date window information we update our window information.
2414 case TCPS_ESTABLISHED:
2415 case TCPS_FIN_WAIT_1:
2416 case TCPS_FIN_WAIT_2:
2417 case TCPS_CLOSE_WAIT:
2420 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2421 TCPSTAT_INC(tcps_rcvacktoomuch);
2424 if ((tp->t_flags & TF_SACK_PERMIT) &&
2425 ((to.to_flags & TOF_SACK) ||
2426 !TAILQ_EMPTY(&tp->snd_holes)))
2427 tcp_sack_doack(tp, &to, th->th_ack);
2430 * Reset the value so that previous (valid) value
2431 * from the last ack with SACK doesn't get used.
2433 tp->sackhint.sacked_bytes = 0;
2435 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
2436 hhook_run_tcp_est_in(tp, th, &to);
2438 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2439 if (tlen == 0 && tiwin == tp->snd_wnd) {
2441 * If this is the first time we've seen a
2442 * FIN from the remote, this is not a
2443 * duplicate and it needs to be processed
2444 * normally. This happens during a
2445 * simultaneous close.
2447 if ((thflags & TH_FIN) &&
2448 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2452 TCPSTAT_INC(tcps_rcvdupack);
2454 * If we have outstanding data (other than
2455 * a window probe), this is a completely
2456 * duplicate ack (ie, window info didn't
2457 * change and FIN isn't set),
2458 * the ack is the biggest we've
2459 * seen and we've seen exactly our rexmt
2460 * threshhold of them, assume a packet
2461 * has been dropped and retransmit it.
2462 * Kludge snd_nxt & the congestion
2463 * window so we send only this one
2466 * We know we're losing at the current
2467 * window size so do congestion avoidance
2468 * (set ssthresh to half the current window
2469 * and pull our congestion window back to
2470 * the new ssthresh).
2472 * Dup acks mean that packets have left the
2473 * network (they're now cached at the receiver)
2474 * so bump cwnd by the amount in the receiver
2475 * to keep a constant cwnd packets in the
2478 * When using TCP ECN, notify the peer that
2479 * we reduced the cwnd.
2481 if (!tcp_timer_active(tp, TT_REXMT) ||
2482 th->th_ack != tp->snd_una)
2484 else if (++tp->t_dupacks > tcprexmtthresh ||
2485 IN_FASTRECOVERY(tp->t_flags)) {
2486 cc_ack_received(tp, th, CC_DUPACK);
2487 if ((tp->t_flags & TF_SACK_PERMIT) &&
2488 IN_FASTRECOVERY(tp->t_flags)) {
2492 * Compute the amount of data in flight first.
2493 * We can inject new data into the pipe iff
2494 * we have less than 1/2 the original window's
2495 * worth of data in flight.
2497 if (V_tcp_do_rfc6675_pipe)
2498 awnd = tcp_compute_pipe(tp);
2500 awnd = (tp->snd_nxt - tp->snd_fack) +
2501 tp->sackhint.sack_bytes_rexmit;
2503 if (awnd < tp->snd_ssthresh) {
2504 tp->snd_cwnd += tp->t_maxseg;
2505 if (tp->snd_cwnd > tp->snd_ssthresh)
2506 tp->snd_cwnd = tp->snd_ssthresh;
2509 tp->snd_cwnd += tp->t_maxseg;
2510 (void) tcp_output(tp);
2512 } else if (tp->t_dupacks == tcprexmtthresh) {
2513 tcp_seq onxt = tp->snd_nxt;
2516 * If we're doing sack, check to
2517 * see if we're already in sack
2518 * recovery. If we're not doing sack,
2519 * check to see if we're in newreno
2522 if (tp->t_flags & TF_SACK_PERMIT) {
2523 if (IN_FASTRECOVERY(tp->t_flags)) {
2528 if (SEQ_LEQ(th->th_ack,
2534 /* Congestion signal before ack. */
2535 cc_cong_signal(tp, th, CC_NDUPACK);
2536 cc_ack_received(tp, th, CC_DUPACK);
2537 tcp_timer_activate(tp, TT_REXMT, 0);
2539 if (tp->t_flags & TF_SACK_PERMIT) {
2541 tcps_sack_recovery_episode);
2542 tp->sack_newdata = tp->snd_nxt;
2543 tp->snd_cwnd = tp->t_maxseg;
2544 (void) tcp_output(tp);
2547 tp->snd_nxt = th->th_ack;
2548 tp->snd_cwnd = tp->t_maxseg;
2549 (void) tcp_output(tp);
2550 KASSERT(tp->snd_limited <= 2,
2551 ("%s: tp->snd_limited too big",
2553 tp->snd_cwnd = tp->snd_ssthresh +
2555 (tp->t_dupacks - tp->snd_limited);
2556 if (SEQ_GT(onxt, tp->snd_nxt))
2559 } else if (V_tcp_do_rfc3042) {
2561 * Process first and second duplicate
2562 * ACKs. Each indicates a segment
2563 * leaving the network, creating room
2564 * for more. Make sure we can send a
2565 * packet on reception of each duplicate
2566 * ACK by increasing snd_cwnd by one
2567 * segment. Restore the original
2568 * snd_cwnd after packet transmission.
2570 cc_ack_received(tp, th, CC_DUPACK);
2571 u_long oldcwnd = tp->snd_cwnd;
2572 tcp_seq oldsndmax = tp->snd_max;
2576 KASSERT(tp->t_dupacks == 1 ||
2578 ("%s: dupacks not 1 or 2",
2580 if (tp->t_dupacks == 1)
2581 tp->snd_limited = 0;
2583 (tp->snd_nxt - tp->snd_una) +
2584 (tp->t_dupacks - tp->snd_limited) *
2587 * Only call tcp_output when there
2588 * is new data available to be sent.
2589 * Otherwise we would send pure ACKs.
2591 SOCKBUF_LOCK(&so->so_snd);
2592 avail = sbavail(&so->so_snd) -
2593 (tp->snd_nxt - tp->snd_una);
2594 SOCKBUF_UNLOCK(&so->so_snd);
2596 (void) tcp_output(tp);
2597 sent = tp->snd_max - oldsndmax;
2598 if (sent > tp->t_maxseg) {
2599 KASSERT((tp->t_dupacks == 2 &&
2600 tp->snd_limited == 0) ||
2601 (sent == tp->t_maxseg + 1 &&
2602 tp->t_flags & TF_SENTFIN),
2603 ("%s: sent too much",
2605 tp->snd_limited = 2;
2606 } else if (sent > 0)
2608 tp->snd_cwnd = oldcwnd;
2616 KASSERT(SEQ_GT(th->th_ack, tp->snd_una),
2617 ("%s: th_ack <= snd_una", __func__));
2620 * If the congestion window was inflated to account
2621 * for the other side's cached packets, retract it.
2623 if (IN_FASTRECOVERY(tp->t_flags)) {
2624 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2625 if (tp->t_flags & TF_SACK_PERMIT)
2626 tcp_sack_partialack(tp, th);
2628 tcp_newreno_partial_ack(tp, th);
2630 cc_post_recovery(tp, th);
2634 * If we reach this point, ACK is not a duplicate,
2635 * i.e., it ACKs something we sent.
2637 if (tp->t_flags & TF_NEEDSYN) {
2639 * T/TCP: Connection was half-synchronized, and our
2640 * SYN has been ACK'd (so connection is now fully
2641 * synchronized). Go to non-starred state,
2642 * increment snd_una for ACK of SYN, and check if
2643 * we can do window scaling.
2645 tp->t_flags &= ~TF_NEEDSYN;
2647 /* Do window scaling? */
2648 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2649 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2650 tp->rcv_scale = tp->request_r_scale;
2651 /* Send window already scaled. */
2656 INP_WLOCK_ASSERT(tp->t_inpcb);
2658 acked = BYTES_THIS_ACK(tp, th);
2659 TCPSTAT_INC(tcps_rcvackpack);
2660 TCPSTAT_ADD(tcps_rcvackbyte, acked);
2663 * If we just performed our first retransmit, and the ACK
2664 * arrives within our recovery window, then it was a mistake
2665 * to do the retransmit in the first place. Recover our
2666 * original cwnd and ssthresh, and proceed to transmit where
2669 if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID &&
2670 (int)(ticks - tp->t_badrxtwin) < 0)
2671 cc_cong_signal(tp, th, CC_RTO_ERR);
2674 * If we have a timestamp reply, update smoothed
2675 * round trip time. If no timestamp is present but
2676 * transmit timer is running and timed sequence
2677 * number was acked, update smoothed round trip time.
2678 * Since we now have an rtt measurement, cancel the
2679 * timer backoff (cf., Phil Karn's retransmit alg.).
2680 * Recompute the initial retransmit timer.
2682 * Some boxes send broken timestamp replies
2683 * during the SYN+ACK phase, ignore
2684 * timestamps of 0 or we could calculate a
2685 * huge RTT and blow up the retransmit timer.
2687 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) {
2690 t = tcp_ts_getticks() - to.to_tsecr;
2691 if (!tp->t_rttlow || tp->t_rttlow > t)
2693 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1);
2694 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2695 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2696 tp->t_rttlow = ticks - tp->t_rtttime;
2697 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2701 * If all outstanding data is acked, stop retransmit
2702 * timer and remember to restart (more output or persist).
2703 * If there is more data to be acked, restart retransmit
2704 * timer, using current (possibly backed-off) value.
2706 if (th->th_ack == tp->snd_max) {
2707 tcp_timer_activate(tp, TT_REXMT, 0);
2709 } else if (!tcp_timer_active(tp, TT_PERSIST))
2710 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
2713 * If no data (only SYN) was ACK'd,
2714 * skip rest of ACK processing.
2720 * Let the congestion control algorithm update congestion
2721 * control related information. This typically means increasing
2722 * the congestion window.
2724 cc_ack_received(tp, th, CC_ACK);
2726 SOCKBUF_LOCK(&so->so_snd);
2727 if (acked > sbavail(&so->so_snd)) {
2728 tp->snd_wnd -= sbavail(&so->so_snd);
2729 mfree = sbcut_locked(&so->so_snd,
2730 (int)sbavail(&so->so_snd));
2733 mfree = sbcut_locked(&so->so_snd, acked);
2734 tp->snd_wnd -= acked;
2737 /* NB: sowwakeup_locked() does an implicit unlock. */
2738 sowwakeup_locked(so);
2740 /* Detect una wraparound. */
2741 if (!IN_RECOVERY(tp->t_flags) &&
2742 SEQ_GT(tp->snd_una, tp->snd_recover) &&
2743 SEQ_LEQ(th->th_ack, tp->snd_recover))
2744 tp->snd_recover = th->th_ack - 1;
2745 /* XXXLAS: Can this be moved up into cc_post_recovery? */
2746 if (IN_RECOVERY(tp->t_flags) &&
2747 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2748 EXIT_RECOVERY(tp->t_flags);
2750 tp->snd_una = th->th_ack;
2751 if (tp->t_flags & TF_SACK_PERMIT) {
2752 if (SEQ_GT(tp->snd_una, tp->snd_recover))
2753 tp->snd_recover = tp->snd_una;
2755 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2756 tp->snd_nxt = tp->snd_una;
2758 switch (tp->t_state) {
2761 * In FIN_WAIT_1 STATE in addition to the processing
2762 * for the ESTABLISHED state if our FIN is now acknowledged
2763 * then enter FIN_WAIT_2.
2765 case TCPS_FIN_WAIT_1:
2766 if (ourfinisacked) {
2768 * If we can't receive any more
2769 * data, then closing user can proceed.
2770 * Starting the timer is contrary to the
2771 * specification, but if we don't get a FIN
2772 * we'll hang forever.
2775 * we should release the tp also, and use a
2778 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2779 soisdisconnected(so);
2780 tcp_timer_activate(tp, TT_2MSL,
2781 (tcp_fast_finwait2_recycle ?
2782 tcp_finwait2_timeout :
2785 tcp_state_change(tp, TCPS_FIN_WAIT_2);
2790 * In CLOSING STATE in addition to the processing for
2791 * the ESTABLISHED state if the ACK acknowledges our FIN
2792 * then enter the TIME-WAIT state, otherwise ignore
2796 if (ourfinisacked) {
2797 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2799 INP_INFO_RUNLOCK(&V_tcbinfo);
2806 * In LAST_ACK, we may still be waiting for data to drain
2807 * and/or to be acked, as well as for the ack of our FIN.
2808 * If our FIN is now acknowledged, delete the TCB,
2809 * enter the closed state and return.
2812 if (ourfinisacked) {
2813 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2822 INP_WLOCK_ASSERT(tp->t_inpcb);
2825 * Update window information.
2826 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2828 if ((thflags & TH_ACK) &&
2829 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2830 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2831 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2832 /* keep track of pure window updates */
2834 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2835 TCPSTAT_INC(tcps_rcvwinupd);
2836 tp->snd_wnd = tiwin;
2837 tp->snd_wl1 = th->th_seq;
2838 tp->snd_wl2 = th->th_ack;
2839 if (tp->snd_wnd > tp->max_sndwnd)
2840 tp->max_sndwnd = tp->snd_wnd;
2845 * Process segments with URG.
2847 if ((thflags & TH_URG) && th->th_urp &&
2848 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2850 * This is a kludge, but if we receive and accept
2851 * random urgent pointers, we'll crash in
2852 * soreceive. It's hard to imagine someone
2853 * actually wanting to send this much urgent data.
2855 SOCKBUF_LOCK(&so->so_rcv);
2856 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) {
2857 th->th_urp = 0; /* XXX */
2858 thflags &= ~TH_URG; /* XXX */
2859 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */
2860 goto dodata; /* XXX */
2863 * If this segment advances the known urgent pointer,
2864 * then mark the data stream. This should not happen
2865 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2866 * a FIN has been received from the remote side.
2867 * In these states we ignore the URG.
2869 * According to RFC961 (Assigned Protocols),
2870 * the urgent pointer points to the last octet
2871 * of urgent data. We continue, however,
2872 * to consider it to indicate the first octet
2873 * of data past the urgent section as the original
2874 * spec states (in one of two places).
2876 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2877 tp->rcv_up = th->th_seq + th->th_urp;
2878 so->so_oobmark = sbavail(&so->so_rcv) +
2879 (tp->rcv_up - tp->rcv_nxt) - 1;
2880 if (so->so_oobmark == 0)
2881 so->so_rcv.sb_state |= SBS_RCVATMARK;
2883 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2885 SOCKBUF_UNLOCK(&so->so_rcv);
2887 * Remove out of band data so doesn't get presented to user.
2888 * This can happen independent of advancing the URG pointer,
2889 * but if two URG's are pending at once, some out-of-band
2890 * data may creep in... ick.
2892 if (th->th_urp <= (u_long)tlen &&
2893 !(so->so_options & SO_OOBINLINE)) {
2894 /* hdr drop is delayed */
2895 tcp_pulloutofband(so, th, m, drop_hdrlen);
2899 * If no out of band data is expected,
2900 * pull receive urgent pointer along
2901 * with the receive window.
2903 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2904 tp->rcv_up = tp->rcv_nxt;
2907 INP_WLOCK_ASSERT(tp->t_inpcb);
2910 * Process the segment text, merging it into the TCP sequencing queue,
2911 * and arranging for acknowledgment of receipt if necessary.
2912 * This process logically involves adjusting tp->rcv_wnd as data
2913 * is presented to the user (this happens in tcp_usrreq.c,
2914 * case PRU_RCVD). If a FIN has already been received on this
2915 * connection then we just ignore the text.
2917 if ((tlen || (thflags & TH_FIN)) &&
2918 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2919 tcp_seq save_start = th->th_seq;
2920 m_adj(m, drop_hdrlen); /* delayed header drop */
2922 * Insert segment which includes th into TCP reassembly queue
2923 * with control block tp. Set thflags to whether reassembly now
2924 * includes a segment with FIN. This handles the common case
2925 * inline (segment is the next to be received on an established
2926 * connection, and the queue is empty), avoiding linkage into
2927 * and removal from the queue and repetition of various
2929 * Set DELACK for segments received in order, but ack
2930 * immediately when segments are out of order (so
2931 * fast retransmit can work).
2933 if (th->th_seq == tp->rcv_nxt &&
2934 LIST_EMPTY(&tp->t_segq) &&
2935 TCPS_HAVEESTABLISHED(tp->t_state)) {
2936 if (DELAY_ACK(tp, tlen))
2937 tp->t_flags |= TF_DELACK;
2939 tp->t_flags |= TF_ACKNOW;
2940 tp->rcv_nxt += tlen;
2941 thflags = th->th_flags & TH_FIN;
2942 TCPSTAT_INC(tcps_rcvpack);
2943 TCPSTAT_ADD(tcps_rcvbyte, tlen);
2944 SOCKBUF_LOCK(&so->so_rcv);
2945 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
2948 sbappendstream_locked(&so->so_rcv, m, 0);
2949 /* NB: sorwakeup_locked() does an implicit unlock. */
2950 sorwakeup_locked(so);
2953 * XXX: Due to the header drop above "th" is
2954 * theoretically invalid by now. Fortunately
2955 * m_adj() doesn't actually frees any mbufs
2956 * when trimming from the head.
2958 thflags = tcp_reass(tp, th, &tlen, m);
2959 tp->t_flags |= TF_ACKNOW;
2961 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT))
2962 tcp_update_sack_list(tp, save_start, save_start + tlen);
2965 * Note the amount of data that peer has sent into
2966 * our window, in order to estimate the sender's
2970 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
2971 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2973 len = so->so_rcv.sb_hiwat;
2981 * If FIN is received ACK the FIN and let the user know
2982 * that the connection is closing.
2984 if (thflags & TH_FIN) {
2985 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2988 * If connection is half-synchronized
2989 * (ie NEEDSYN flag on) then delay ACK,
2990 * so it may be piggybacked when SYN is sent.
2991 * Otherwise, since we received a FIN then no
2992 * more input can be expected, send ACK now.
2994 if (tp->t_flags & TF_NEEDSYN)
2995 tp->t_flags |= TF_DELACK;
2997 tp->t_flags |= TF_ACKNOW;
3000 switch (tp->t_state) {
3003 * In SYN_RECEIVED and ESTABLISHED STATES
3004 * enter the CLOSE_WAIT state.
3006 case TCPS_SYN_RECEIVED:
3007 tp->t_starttime = ticks;
3009 case TCPS_ESTABLISHED:
3010 tcp_state_change(tp, TCPS_CLOSE_WAIT);
3014 * If still in FIN_WAIT_1 STATE FIN has not been acked so
3015 * enter the CLOSING state.
3017 case TCPS_FIN_WAIT_1:
3018 tcp_state_change(tp, TCPS_CLOSING);
3022 * In FIN_WAIT_2 state enter the TIME_WAIT state,
3023 * starting the time-wait timer, turning off the other
3026 case TCPS_FIN_WAIT_2:
3027 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
3028 KASSERT(ti_locked == TI_RLOCKED, ("%s: dodata "
3029 "TCP_FIN_WAIT_2 ti_locked: %d", __func__,
3033 INP_INFO_RUNLOCK(&V_tcbinfo);
3037 if (ti_locked == TI_RLOCKED)
3038 INP_INFO_RUNLOCK(&V_tcbinfo);
3039 ti_locked = TI_UNLOCKED;
3042 if (so->so_options & SO_DEBUG)
3043 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
3046 TCP_PROBE3(debug__input, tp, th, mtod(m, const char *));
3049 * Return any desired output.
3051 if (needoutput || (tp->t_flags & TF_ACKNOW))
3052 (void) tcp_output(tp);
3055 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d",
3056 __func__, ti_locked));
3057 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
3058 INP_WLOCK_ASSERT(tp->t_inpcb);
3060 if (tp->t_flags & TF_DELACK) {
3061 tp->t_flags &= ~TF_DELACK;
3062 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
3064 INP_WUNLOCK(tp->t_inpcb);
3069 * Generate an ACK dropping incoming segment if it occupies
3070 * sequence space, where the ACK reflects our state.
3072 * We can now skip the test for the RST flag since all
3073 * paths to this code happen after packets containing
3074 * RST have been dropped.
3076 * In the SYN-RECEIVED state, don't send an ACK unless the
3077 * segment we received passes the SYN-RECEIVED ACK test.
3078 * If it fails send a RST. This breaks the loop in the
3079 * "LAND" DoS attack, and also prevents an ACK storm
3080 * between two listening ports that have been sent forged
3081 * SYN segments, each with the source address of the other.
3083 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
3084 (SEQ_GT(tp->snd_una, th->th_ack) ||
3085 SEQ_GT(th->th_ack, tp->snd_max)) ) {
3086 rstreason = BANDLIM_RST_OPENPORT;
3090 if (so->so_options & SO_DEBUG)
3091 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3094 TCP_PROBE3(debug__input, tp, th, mtod(m, const char *));
3095 if (ti_locked == TI_RLOCKED)
3096 INP_INFO_RUNLOCK(&V_tcbinfo);
3097 ti_locked = TI_UNLOCKED;
3099 tp->t_flags |= TF_ACKNOW;
3100 (void) tcp_output(tp);
3101 INP_WUNLOCK(tp->t_inpcb);
3106 if (ti_locked == TI_RLOCKED)
3107 INP_INFO_RUNLOCK(&V_tcbinfo);
3108 ti_locked = TI_UNLOCKED;
3111 tcp_dropwithreset(m, th, tp, tlen, rstreason);
3112 INP_WUNLOCK(tp->t_inpcb);
3114 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
3118 if (ti_locked == TI_RLOCKED) {
3119 INP_INFO_RUNLOCK(&V_tcbinfo);
3120 ti_locked = TI_UNLOCKED;
3124 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
3128 * Drop space held by incoming segment and return.
3131 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
3132 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3135 TCP_PROBE3(debug__input, tp, th, mtod(m, const char *));
3137 INP_WUNLOCK(tp->t_inpcb);
3142 * Issue RST and make ACK acceptable to originator of segment.
3143 * The mbuf must still include the original packet header.
3147 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
3148 int tlen, int rstreason)
3154 struct ip6_hdr *ip6;
3158 INP_WLOCK_ASSERT(tp->t_inpcb);
3161 /* Don't bother if destination was broadcast/multicast. */
3162 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
3165 if (mtod(m, struct ip *)->ip_v == 6) {
3166 ip6 = mtod(m, struct ip6_hdr *);
3167 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3168 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3170 /* IPv6 anycast check is done at tcp6_input() */
3173 #if defined(INET) && defined(INET6)
3178 ip = mtod(m, struct ip *);
3179 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3180 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3181 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3182 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3187 /* Perform bandwidth limiting. */
3188 if (badport_bandlim(rstreason) < 0)
3191 /* tcp_respond consumes the mbuf chain. */
3192 if (th->th_flags & TH_ACK) {
3193 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0,
3194 th->th_ack, TH_RST);
3196 if (th->th_flags & TH_SYN)
3198 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
3199 (tcp_seq)0, TH_RST|TH_ACK);
3207 * Parse TCP options and place in tcpopt.
3210 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags)
3215 for (; cnt > 0; cnt -= optlen, cp += optlen) {
3217 if (opt == TCPOPT_EOL)
3219 if (opt == TCPOPT_NOP)
3225 if (optlen < 2 || optlen > cnt)
3230 if (optlen != TCPOLEN_MAXSEG)
3232 if (!(flags & TO_SYN))
3234 to->to_flags |= TOF_MSS;
3235 bcopy((char *)cp + 2,
3236 (char *)&to->to_mss, sizeof(to->to_mss));
3237 to->to_mss = ntohs(to->to_mss);
3240 if (optlen != TCPOLEN_WINDOW)
3242 if (!(flags & TO_SYN))
3244 to->to_flags |= TOF_SCALE;
3245 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT);
3247 case TCPOPT_TIMESTAMP:
3248 if (optlen != TCPOLEN_TIMESTAMP)
3250 to->to_flags |= TOF_TS;
3251 bcopy((char *)cp + 2,
3252 (char *)&to->to_tsval, sizeof(to->to_tsval));
3253 to->to_tsval = ntohl(to->to_tsval);
3254 bcopy((char *)cp + 6,
3255 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
3256 to->to_tsecr = ntohl(to->to_tsecr);
3258 #ifdef TCP_SIGNATURE
3260 * XXX In order to reply to a host which has set the
3261 * TCP_SIGNATURE option in its initial SYN, we have to
3262 * record the fact that the option was observed here
3263 * for the syncache code to perform the correct response.
3265 case TCPOPT_SIGNATURE:
3266 if (optlen != TCPOLEN_SIGNATURE)
3268 to->to_flags |= TOF_SIGNATURE;
3269 to->to_signature = cp + 2;
3272 case TCPOPT_SACK_PERMITTED:
3273 if (optlen != TCPOLEN_SACK_PERMITTED)
3275 if (!(flags & TO_SYN))
3279 to->to_flags |= TOF_SACKPERM;
3282 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
3286 to->to_flags |= TOF_SACK;
3287 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
3288 to->to_sacks = cp + 2;
3289 TCPSTAT_INC(tcps_sack_rcv_blocks);
3298 * Pull out of band byte out of a segment so
3299 * it doesn't appear in the user's data queue.
3300 * It is still reflected in the segment length for
3301 * sequencing purposes.
3304 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m,
3307 int cnt = off + th->th_urp - 1;
3310 if (m->m_len > cnt) {
3311 char *cp = mtod(m, caddr_t) + cnt;
3312 struct tcpcb *tp = sototcpcb(so);
3314 INP_WLOCK_ASSERT(tp->t_inpcb);
3317 tp->t_oobflags |= TCPOOB_HAVEDATA;
3318 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3320 if (m->m_flags & M_PKTHDR)
3329 panic("tcp_pulloutofband");
3333 * Collect new round-trip time estimate
3334 * and update averages and current timeout.
3337 tcp_xmit_timer(struct tcpcb *tp, int rtt)
3341 INP_WLOCK_ASSERT(tp->t_inpcb);
3343 TCPSTAT_INC(tcps_rttupdated);
3345 if (tp->t_srtt != 0) {
3347 * srtt is stored as fixed point with 5 bits after the
3348 * binary point (i.e., scaled by 8). The following magic
3349 * is equivalent to the smoothing algorithm in rfc793 with
3350 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3351 * point). Adjust rtt to origin 0.
3353 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
3354 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
3356 if ((tp->t_srtt += delta) <= 0)
3360 * We accumulate a smoothed rtt variance (actually, a
3361 * smoothed mean difference), then set the retransmit
3362 * timer to smoothed rtt + 4 times the smoothed variance.
3363 * rttvar is stored as fixed point with 4 bits after the
3364 * binary point (scaled by 16). The following is
3365 * equivalent to rfc793 smoothing with an alpha of .75
3366 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
3367 * rfc793's wired-in beta.
3371 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
3372 if ((tp->t_rttvar += delta) <= 0)
3374 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
3375 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3378 * No rtt measurement yet - use the unsmoothed rtt.
3379 * Set the variance to half the rtt (so our first
3380 * retransmit happens at 3*rtt).
3382 tp->t_srtt = rtt << TCP_RTT_SHIFT;
3383 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
3384 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3390 * the retransmit should happen at rtt + 4 * rttvar.
3391 * Because of the way we do the smoothing, srtt and rttvar
3392 * will each average +1/2 tick of bias. When we compute
3393 * the retransmit timer, we want 1/2 tick of rounding and
3394 * 1 extra tick because of +-1/2 tick uncertainty in the
3395 * firing of the timer. The bias will give us exactly the
3396 * 1.5 tick we need. But, because the bias is
3397 * statistical, we have to test that we don't drop below
3398 * the minimum feasible timer (which is 2 ticks).
3400 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3401 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3404 * We received an ack for a packet that wasn't retransmitted;
3405 * it is probably safe to discard any error indications we've
3406 * received recently. This isn't quite right, but close enough
3407 * for now (a route might have failed after we sent a segment,
3408 * and the return path might not be symmetrical).
3410 tp->t_softerror = 0;
3414 * Determine a reasonable value for maxseg size.
3415 * If the route is known, check route for mtu.
3416 * If none, use an mss that can be handled on the outgoing interface
3417 * without forcing IP to fragment. If no route is found, route has no mtu,
3418 * or the destination isn't local, use a default, hopefully conservative
3419 * size (usually 512 or the default IP max size, but no more than the mtu
3420 * of the interface), as we can't discover anything about intervening
3421 * gateways or networks. We also initialize the congestion/slow start
3422 * window to be a single segment if the destination isn't local.
3423 * While looking at the routing entry, we also initialize other path-dependent
3424 * parameters from pre-set or cached values in the routing entry.
3426 * Also take into account the space needed for options that we
3427 * send regularly. Make maxseg shorter by that amount to assure
3428 * that we can send maxseg amount of data even when the options
3429 * are present. Store the upper limit of the length of options plus
3432 * NOTE that this routine is only called when we process an incoming
3433 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS
3434 * settings are handled in tcp_mssopt().
3437 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer,
3438 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap)
3442 struct inpcb *inp = tp->t_inpcb;
3443 struct hc_metrics_lite metrics;
3446 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
3447 size_t min_protoh = isipv6 ?
3448 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
3449 sizeof (struct tcpiphdr);
3451 const size_t min_protoh = sizeof(struct tcpiphdr);
3454 INP_WLOCK_ASSERT(tp->t_inpcb);
3456 if (mtuoffer != -1) {
3457 KASSERT(offer == -1, ("%s: conflict", __func__));
3458 offer = mtuoffer - min_protoh;
3465 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap);
3466 tp->t_maxopd = tp->t_maxseg = V_tcp_v6mssdflt;
3469 #if defined(INET) && defined(INET6)
3474 maxmtu = tcp_maxmtu(&inp->inp_inc, cap);
3475 tp->t_maxopd = tp->t_maxseg = V_tcp_mssdflt;
3480 * No route to sender, stay with default mss and return.
3484 * In case we return early we need to initialize metrics
3485 * to a defined state as tcp_hc_get() would do for us
3486 * if there was no cache hit.
3488 if (metricptr != NULL)
3489 bzero(metricptr, sizeof(struct hc_metrics_lite));
3493 /* What have we got? */
3497 * Offer == 0 means that there was no MSS on the SYN
3498 * segment, in this case we use tcp_mssdflt as
3499 * already assigned to t_maxopd above.
3501 offer = tp->t_maxopd;
3506 * Offer == -1 means that we didn't receive SYN yet.
3512 * Prevent DoS attack with too small MSS. Round up
3513 * to at least minmss.
3515 offer = max(offer, V_tcp_minmss);
3519 * rmx information is now retrieved from tcp_hostcache.
3521 tcp_hc_get(&inp->inp_inc, &metrics);
3522 if (metricptr != NULL)
3523 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite));
3526 * If there's a discovered mtu in tcp hostcache, use it.
3527 * Else, use the link mtu.
3529 if (metrics.rmx_mtu)
3530 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
3534 mss = maxmtu - min_protoh;
3535 if (!V_path_mtu_discovery &&
3536 !in6_localaddr(&inp->in6p_faddr))
3537 mss = min(mss, V_tcp_v6mssdflt);
3540 #if defined(INET) && defined(INET6)
3545 mss = maxmtu - min_protoh;
3546 if (!V_path_mtu_discovery &&
3547 !in_localaddr(inp->inp_faddr))
3548 mss = min(mss, V_tcp_mssdflt);
3552 * XXX - The above conditional (mss = maxmtu - min_protoh)
3553 * probably violates the TCP spec.
3554 * The problem is that, since we don't know the
3555 * other end's MSS, we are supposed to use a conservative
3556 * default. But, if we do that, then MTU discovery will
3557 * never actually take place, because the conservative
3558 * default is much less than the MTUs typically seen
3559 * on the Internet today. For the moment, we'll sweep
3560 * this under the carpet.
3562 * The conservative default might not actually be a problem
3563 * if the only case this occurs is when sending an initial
3564 * SYN with options and data to a host we've never talked
3565 * to before. Then, they will reply with an MSS value which
3566 * will get recorded and the new parameters should get
3567 * recomputed. For Further Study.
3570 mss = min(mss, offer);
3573 * Sanity check: make sure that maxopd will be large
3574 * enough to allow some data on segments even if the
3575 * all the option space is used (40bytes). Otherwise
3576 * funny things may happen in tcp_output.
3581 * maxopd stores the maximum length of data AND options
3582 * in a segment; maxseg is the amount of data in a normal
3583 * segment. We need to store this value (maxopd) apart
3584 * from maxseg, because now every segment carries options
3585 * and thus we normally have somewhat less data in segments.
3590 * origoffer==-1 indicates that no segments were received yet.
3591 * In this case we just guess.
3593 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
3595 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
3596 mss -= TCPOLEN_TSTAMP_APPA;
3602 tcp_mss(struct tcpcb *tp, int offer)
3608 struct hc_metrics_lite metrics;
3609 struct tcp_ifcap cap;
3611 KASSERT(tp != NULL, ("%s: tp == NULL", __func__));
3613 bzero(&cap, sizeof(cap));
3614 tcp_mss_update(tp, offer, -1, &metrics, &cap);
3620 * If there's a pipesize, change the socket buffer to that size,
3621 * don't change if sb_hiwat is different than default (then it
3622 * has been changed on purpose with setsockopt).
3623 * Make the socket buffers an integral number of mss units;
3624 * if the mss is larger than the socket buffer, decrease the mss.
3626 so = inp->inp_socket;
3627 SOCKBUF_LOCK(&so->so_snd);
3628 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe)
3629 bufsize = metrics.rmx_sendpipe;
3631 bufsize = so->so_snd.sb_hiwat;
3635 bufsize = roundup(bufsize, mss);
3636 if (bufsize > sb_max)
3638 if (bufsize > so->so_snd.sb_hiwat)
3639 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
3641 SOCKBUF_UNLOCK(&so->so_snd);
3644 SOCKBUF_LOCK(&so->so_rcv);
3645 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe)
3646 bufsize = metrics.rmx_recvpipe;
3648 bufsize = so->so_rcv.sb_hiwat;
3649 if (bufsize > mss) {
3650 bufsize = roundup(bufsize, mss);
3651 if (bufsize > sb_max)
3653 if (bufsize > so->so_rcv.sb_hiwat)
3654 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
3656 SOCKBUF_UNLOCK(&so->so_rcv);
3658 /* Check the interface for TSO capabilities. */
3659 if (cap.ifcap & CSUM_TSO) {
3660 tp->t_flags |= TF_TSO;
3661 tp->t_tsomax = cap.tsomax;
3662 tp->t_tsomaxsegcount = cap.tsomaxsegcount;
3663 tp->t_tsomaxsegsize = cap.tsomaxsegsize;
3668 * Determine the MSS option to send on an outgoing SYN.
3671 tcp_mssopt(struct in_conninfo *inc)
3678 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3681 if (inc->inc_flags & INC_ISIPV6) {
3682 mss = V_tcp_v6mssdflt;
3683 maxmtu = tcp_maxmtu6(inc, NULL);
3684 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3687 #if defined(INET) && defined(INET6)
3692 mss = V_tcp_mssdflt;
3693 maxmtu = tcp_maxmtu(inc, NULL);
3694 min_protoh = sizeof(struct tcpiphdr);
3697 #if defined(INET6) || defined(INET)
3698 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3701 if (maxmtu && thcmtu)
3702 mss = min(maxmtu, thcmtu) - min_protoh;
3703 else if (maxmtu || thcmtu)
3704 mss = max(maxmtu, thcmtu) - min_protoh;
3711 * On a partial ack arrives, force the retransmission of the
3712 * next unacknowledged segment. Do not clear tp->t_dupacks.
3713 * By setting snd_nxt to ti_ack, this forces retransmission timer to
3717 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th)
3719 tcp_seq onxt = tp->snd_nxt;
3720 u_long ocwnd = tp->snd_cwnd;
3722 INP_WLOCK_ASSERT(tp->t_inpcb);
3724 tcp_timer_activate(tp, TT_REXMT, 0);
3726 tp->snd_nxt = th->th_ack;
3728 * Set snd_cwnd to one segment beyond acknowledged offset.
3729 * (tp->snd_una has not yet been updated when this function is called.)
3731 tp->snd_cwnd = tp->t_maxseg + BYTES_THIS_ACK(tp, th);
3732 tp->t_flags |= TF_ACKNOW;
3733 (void) tcp_output(tp);
3734 tp->snd_cwnd = ocwnd;
3735 if (SEQ_GT(onxt, tp->snd_nxt))
3738 * Partial window deflation. Relies on fact that tp->snd_una
3741 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th))
3742 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th);
3745 tp->snd_cwnd += tp->t_maxseg;
3749 tcp_compute_pipe(struct tcpcb *tp)
3751 return (tp->snd_max - tp->snd_una +
3752 tp->sackhint.sack_bytes_rexmit -
3753 tp->sackhint.sacked_bytes);