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_rfc3042) = 1;
152 #define V_tcp_do_rfc3042 VNET(tcp_do_rfc3042)
153 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW,
154 &VNET_NAME(tcp_do_rfc3042), 0,
155 "Enable RFC 3042 (Limited Transmit)");
157 VNET_DEFINE(int, tcp_do_rfc3390) = 1;
158 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW,
159 &VNET_NAME(tcp_do_rfc3390), 0,
160 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
162 VNET_DEFINE(int, tcp_initcwnd_segments) = 10;
163 SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments,
164 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0,
165 "Slow-start flight size (initial congestion window) in number of segments");
167 VNET_DEFINE(int, tcp_do_rfc3465) = 1;
168 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW,
169 &VNET_NAME(tcp_do_rfc3465), 0,
170 "Enable RFC 3465 (Appropriate Byte Counting)");
172 VNET_DEFINE(int, tcp_abc_l_var) = 2;
173 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW,
174 &VNET_NAME(tcp_abc_l_var), 2,
175 "Cap the max cwnd increment during slow-start to this number of segments");
177 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, CTLFLAG_RW, 0, "TCP ECN");
179 VNET_DEFINE(int, tcp_do_ecn) = 0;
180 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW,
181 &VNET_NAME(tcp_do_ecn), 0,
184 VNET_DEFINE(int, tcp_ecn_maxretries) = 1;
185 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_VNET | CTLFLAG_RW,
186 &VNET_NAME(tcp_ecn_maxretries), 0,
187 "Max retries before giving up on ECN");
189 VNET_DEFINE(int, tcp_insecure_syn) = 0;
190 #define V_tcp_insecure_syn VNET(tcp_insecure_syn)
191 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW,
192 &VNET_NAME(tcp_insecure_syn), 0,
193 "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets");
195 VNET_DEFINE(int, tcp_insecure_rst) = 0;
196 #define V_tcp_insecure_rst VNET(tcp_insecure_rst)
197 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW,
198 &VNET_NAME(tcp_insecure_rst), 0,
199 "Follow RFC793 instead of RFC5961 criteria for accepting RST packets");
201 VNET_DEFINE(int, tcp_recvspace) = 1024*64;
202 #define V_tcp_recvspace VNET(tcp_recvspace)
203 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW,
204 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size");
206 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1;
207 #define V_tcp_do_autorcvbuf VNET(tcp_do_autorcvbuf)
208 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW,
209 &VNET_NAME(tcp_do_autorcvbuf), 0,
210 "Enable automatic receive buffer sizing");
212 VNET_DEFINE(int, tcp_autorcvbuf_inc) = 16*1024;
213 #define V_tcp_autorcvbuf_inc VNET(tcp_autorcvbuf_inc)
214 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_VNET | CTLFLAG_RW,
215 &VNET_NAME(tcp_autorcvbuf_inc), 0,
216 "Incrementor step size of automatic receive buffer");
218 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024;
219 #define V_tcp_autorcvbuf_max VNET(tcp_autorcvbuf_max)
220 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW,
221 &VNET_NAME(tcp_autorcvbuf_max), 0,
222 "Max size of automatic receive buffer");
224 VNET_DEFINE(struct inpcbhead, tcb);
225 #define tcb6 tcb /* for KAME src sync over BSD*'s */
226 VNET_DEFINE(struct inpcbinfo, tcbinfo);
228 static void tcp_dooptions(struct tcpopt *, u_char *, int, int);
229 static void tcp_do_segment(struct mbuf *, struct tcphdr *,
230 struct socket *, struct tcpcb *, int, int, uint8_t,
232 static void tcp_dropwithreset(struct mbuf *, struct tcphdr *,
233 struct tcpcb *, int, int);
234 static void tcp_pulloutofband(struct socket *,
235 struct tcphdr *, struct mbuf *, int);
236 static void tcp_xmit_timer(struct tcpcb *, int);
237 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *);
238 static void inline cc_ack_received(struct tcpcb *tp, struct tcphdr *th,
240 static void inline cc_conn_init(struct tcpcb *tp);
241 static void inline cc_post_recovery(struct tcpcb *tp, struct tcphdr *th);
242 static void inline hhook_run_tcp_est_in(struct tcpcb *tp,
243 struct tcphdr *th, struct tcpopt *to);
246 * TCP statistics are stored in an "array" of counter(9)s.
248 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat);
249 VNET_PCPUSTAT_SYSINIT(tcpstat);
250 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat,
251 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
254 VNET_PCPUSTAT_SYSUNINIT(tcpstat);
257 * Kernel module interface for updating tcpstat. The argument is an index
258 * into tcpstat treated as an array.
261 kmod_tcpstat_inc(int statnum)
264 counter_u64_add(VNET(tcpstat)[statnum], 1);
268 * Wrapper for the TCP established input helper hook.
271 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to)
273 struct tcp_hhook_data hhook_data;
275 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) {
280 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data,
286 * CC wrapper hook functions
289 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t type)
291 INP_WLOCK_ASSERT(tp->t_inpcb);
293 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th);
294 if (tp->snd_cwnd <= tp->snd_wnd)
295 tp->ccv->flags |= CCF_CWND_LIMITED;
297 tp->ccv->flags &= ~CCF_CWND_LIMITED;
299 if (type == CC_ACK) {
300 if (tp->snd_cwnd > tp->snd_ssthresh) {
301 tp->t_bytes_acked += min(tp->ccv->bytes_this_ack,
302 V_tcp_abc_l_var * tp->t_maxseg);
303 if (tp->t_bytes_acked >= tp->snd_cwnd) {
304 tp->t_bytes_acked -= tp->snd_cwnd;
305 tp->ccv->flags |= CCF_ABC_SENTAWND;
308 tp->ccv->flags &= ~CCF_ABC_SENTAWND;
309 tp->t_bytes_acked = 0;
313 if (CC_ALGO(tp)->ack_received != NULL) {
314 /* XXXLAS: Find a way to live without this */
315 tp->ccv->curack = th->th_ack;
316 CC_ALGO(tp)->ack_received(tp->ccv, type);
321 cc_conn_init(struct tcpcb *tp)
323 struct hc_metrics_lite metrics;
324 struct inpcb *inp = tp->t_inpcb;
327 INP_WLOCK_ASSERT(tp->t_inpcb);
329 tcp_hc_get(&inp->inp_inc, &metrics);
331 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) {
333 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
334 TCPSTAT_INC(tcps_usedrtt);
335 if (metrics.rmx_rttvar) {
336 tp->t_rttvar = metrics.rmx_rttvar;
337 TCPSTAT_INC(tcps_usedrttvar);
339 /* default variation is +- 1 rtt */
341 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
343 TCPT_RANGESET(tp->t_rxtcur,
344 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
345 tp->t_rttmin, TCPTV_REXMTMAX);
347 if (metrics.rmx_ssthresh) {
349 * There's some sort of gateway or interface
350 * buffer limit on the path. Use this to set
351 * the slow start threshhold, but set the
352 * threshold to no less than 2*mss.
354 tp->snd_ssthresh = max(2 * tp->t_maxseg, metrics.rmx_ssthresh);
355 TCPSTAT_INC(tcps_usedssthresh);
359 * Set the initial slow-start flight size.
361 * RFC5681 Section 3.1 specifies the default conservative values.
362 * RFC3390 specifies slightly more aggressive values.
363 * RFC6928 increases it to ten segments.
364 * Support for user specified value for initial flight size.
366 * If a SYN or SYN/ACK was lost and retransmitted, we have to
367 * reduce the initial CWND to one segment as congestion is likely
368 * requiring us to be cautious.
370 if (tp->snd_cwnd == 1)
371 tp->snd_cwnd = tp->t_maxseg; /* SYN(-ACK) lost */
372 else if (V_tcp_initcwnd_segments)
373 tp->snd_cwnd = min(V_tcp_initcwnd_segments * tp->t_maxseg,
374 max(2 * tp->t_maxseg, V_tcp_initcwnd_segments * 1460));
375 else if (V_tcp_do_rfc3390)
376 tp->snd_cwnd = min(4 * tp->t_maxseg,
377 max(2 * tp->t_maxseg, 4380));
379 /* Per RFC5681 Section 3.1 */
380 if (tp->t_maxseg > 2190)
381 tp->snd_cwnd = 2 * tp->t_maxseg;
382 else if (tp->t_maxseg > 1095)
383 tp->snd_cwnd = 3 * tp->t_maxseg;
385 tp->snd_cwnd = 4 * tp->t_maxseg;
388 if (CC_ALGO(tp)->conn_init != NULL)
389 CC_ALGO(tp)->conn_init(tp->ccv);
393 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type)
395 INP_WLOCK_ASSERT(tp->t_inpcb);
399 if (!IN_FASTRECOVERY(tp->t_flags)) {
400 tp->snd_recover = tp->snd_max;
401 if (tp->t_flags & TF_ECN_PERMIT)
402 tp->t_flags |= TF_ECN_SND_CWR;
406 if (!IN_CONGRECOVERY(tp->t_flags)) {
407 TCPSTAT_INC(tcps_ecn_rcwnd);
408 tp->snd_recover = tp->snd_max;
409 if (tp->t_flags & TF_ECN_PERMIT)
410 tp->t_flags |= TF_ECN_SND_CWR;
415 tp->t_bytes_acked = 0;
416 EXIT_RECOVERY(tp->t_flags);
417 tp->snd_ssthresh = max(2, min(tp->snd_wnd, tp->snd_cwnd) / 2 /
418 tp->t_maxseg) * tp->t_maxseg;
419 tp->snd_cwnd = tp->t_maxseg;
422 TCPSTAT_INC(tcps_sndrexmitbad);
423 /* RTO was unnecessary, so reset everything. */
424 tp->snd_cwnd = tp->snd_cwnd_prev;
425 tp->snd_ssthresh = tp->snd_ssthresh_prev;
426 tp->snd_recover = tp->snd_recover_prev;
427 if (tp->t_flags & TF_WASFRECOVERY)
428 ENTER_FASTRECOVERY(tp->t_flags);
429 if (tp->t_flags & TF_WASCRECOVERY)
430 ENTER_CONGRECOVERY(tp->t_flags);
431 tp->snd_nxt = tp->snd_max;
432 tp->t_flags &= ~TF_PREVVALID;
437 if (CC_ALGO(tp)->cong_signal != NULL) {
439 tp->ccv->curack = th->th_ack;
440 CC_ALGO(tp)->cong_signal(tp->ccv, type);
445 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th)
447 INP_WLOCK_ASSERT(tp->t_inpcb);
449 /* XXXLAS: KASSERT that we're in recovery? */
451 if (CC_ALGO(tp)->post_recovery != NULL) {
452 tp->ccv->curack = th->th_ack;
453 CC_ALGO(tp)->post_recovery(tp->ccv);
455 /* XXXLAS: EXIT_RECOVERY ? */
456 tp->t_bytes_acked = 0;
461 tcp_signature_verify_input(struct mbuf *m, int off0, int tlen, int optlen,
462 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag)
466 tcp_fields_to_net(th);
467 ret = tcp_signature_verify(m, off0, tlen, optlen, to, th, tcpbflag);
468 tcp_fields_to_host(th);
474 * Indicate whether this ack should be delayed. We can delay the ack if
475 * following conditions are met:
476 * - There is no delayed ack timer in progress.
477 * - Our last ack wasn't a 0-sized window. We never want to delay
478 * the ack that opens up a 0-sized window.
479 * - LRO wasn't used for this segment. We make sure by checking that the
480 * segment size is not larger than the MSS.
481 * - Delayed acks are enabled or this is a half-synchronized T/TCP
484 #define DELAY_ACK(tp, tlen) \
485 ((!tcp_timer_active(tp, TT_DELACK) && \
486 (tp->t_flags & TF_RXWIN0SENT) == 0) && \
487 (tlen <= tp->t_maxopd) && \
488 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
491 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos)
493 INP_WLOCK_ASSERT(tp->t_inpcb);
495 if (CC_ALGO(tp)->ecnpkt_handler != NULL) {
496 switch (iptos & IPTOS_ECN_MASK) {
498 tp->ccv->flags |= CCF_IPHDR_CE;
501 tp->ccv->flags &= ~CCF_IPHDR_CE;
504 tp->ccv->flags &= ~CCF_IPHDR_CE;
508 if (th->th_flags & TH_CWR)
509 tp->ccv->flags |= CCF_TCPHDR_CWR;
511 tp->ccv->flags &= ~CCF_TCPHDR_CWR;
513 if (tp->t_flags & TF_DELACK)
514 tp->ccv->flags |= CCF_DELACK;
516 tp->ccv->flags &= ~CCF_DELACK;
518 CC_ALGO(tp)->ecnpkt_handler(tp->ccv);
520 if (tp->ccv->flags & CCF_ACKNOW)
521 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
526 * TCP input handling is split into multiple parts:
527 * tcp6_input is a thin wrapper around tcp_input for the extended
528 * ip6_protox[] call format in ip6_input
529 * tcp_input handles primary segment validation, inpcb lookup and
530 * SYN processing on listen sockets
531 * tcp_do_segment processes the ACK and text of the segment for
532 * establishing, established and closing connections
536 tcp6_input(struct mbuf **mp, int *offp, int proto)
538 struct mbuf *m = *mp;
539 struct in6_ifaddr *ia6;
542 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
545 * draft-itojun-ipv6-tcp-to-anycast
546 * better place to put this in?
548 ip6 = mtod(m, struct ip6_hdr *);
549 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */);
550 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
553 ifa_free(&ia6->ia_ifa);
554 ip6 = mtod(m, struct ip6_hdr *);
555 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
556 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
557 return (IPPROTO_DONE);
560 ifa_free(&ia6->ia_ifa);
562 return (tcp_input(mp, offp, proto));
567 tcp_input(struct mbuf **mp, int *offp, int proto)
569 struct mbuf *m = *mp;
570 struct tcphdr *th = NULL;
571 struct ip *ip = NULL;
572 struct inpcb *inp = NULL;
573 struct tcpcb *tp = NULL;
574 struct socket *so = NULL;
584 int rstreason = 0; /* For badport_bandlim accounting purposes */
586 uint8_t sig_checked = 0;
589 struct m_tag *fwd_tag = NULL;
591 struct ip6_hdr *ip6 = NULL;
594 const void *ip6 = NULL;
596 struct tcpopt to; /* options in this segment */
597 char *s = NULL; /* address and port logging */
599 #define TI_UNLOCKED 1
604 * The size of tcp_saveipgen must be the size of the max ip header,
607 u_char tcp_saveipgen[IP6_HDR_LEN];
608 struct tcphdr tcp_savetcp;
613 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
620 TCPSTAT_INC(tcps_rcvtotal);
624 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */
626 if (m->m_len < (sizeof(*ip6) + sizeof(*th))) {
627 m = m_pullup(m, sizeof(*ip6) + sizeof(*th));
629 TCPSTAT_INC(tcps_rcvshort);
630 return (IPPROTO_DONE);
634 ip6 = mtod(m, struct ip6_hdr *);
635 th = (struct tcphdr *)((caddr_t)ip6 + off0);
636 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
637 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) {
638 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
639 th->th_sum = m->m_pkthdr.csum_data;
641 th->th_sum = in6_cksum_pseudo(ip6, tlen,
642 IPPROTO_TCP, m->m_pkthdr.csum_data);
643 th->th_sum ^= 0xffff;
645 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen);
647 TCPSTAT_INC(tcps_rcvbadsum);
652 * Be proactive about unspecified IPv6 address in source.
653 * As we use all-zero to indicate unbounded/unconnected pcb,
654 * unspecified IPv6 address can be used to confuse us.
656 * Note that packets with unspecified IPv6 destination is
657 * already dropped in ip6_input.
659 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
665 #if defined(INET) && defined(INET6)
671 * Get IP and TCP header together in first mbuf.
672 * Note: IP leaves IP header in first mbuf.
674 if (off0 > sizeof (struct ip)) {
676 off0 = sizeof(struct ip);
678 if (m->m_len < sizeof (struct tcpiphdr)) {
679 if ((m = m_pullup(m, sizeof (struct tcpiphdr)))
681 TCPSTAT_INC(tcps_rcvshort);
682 return (IPPROTO_DONE);
685 ip = mtod(m, struct ip *);
686 th = (struct tcphdr *)((caddr_t)ip + off0);
687 tlen = ntohs(ip->ip_len) - off0;
689 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
690 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
691 th->th_sum = m->m_pkthdr.csum_data;
693 th->th_sum = in_pseudo(ip->ip_src.s_addr,
695 htonl(m->m_pkthdr.csum_data + tlen +
697 th->th_sum ^= 0xffff;
699 struct ipovly *ipov = (struct ipovly *)ip;
702 * Checksum extended TCP header and data.
705 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
706 ipov->ih_len = htons(tlen);
707 th->th_sum = in_cksum(m, len);
708 /* Reset length for SDT probes. */
709 ip->ip_len = htons(tlen + off0);
713 TCPSTAT_INC(tcps_rcvbadsum);
716 /* Re-initialization for later version check */
717 ip->ip_v = IPVERSION;
723 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
725 #if defined(INET) && defined(INET6)
733 * Check that TCP offset makes sense,
734 * pull out TCP options and adjust length. XXX
736 off = th->th_off << 2;
737 if (off < sizeof (struct tcphdr) || off > tlen) {
738 TCPSTAT_INC(tcps_rcvbadoff);
741 tlen -= off; /* tlen is used instead of ti->ti_len */
742 if (off > sizeof (struct tcphdr)) {
745 IP6_EXTHDR_CHECK(m, off0, off, IPPROTO_DONE);
746 ip6 = mtod(m, struct ip6_hdr *);
747 th = (struct tcphdr *)((caddr_t)ip6 + off0);
750 #if defined(INET) && defined(INET6)
755 if (m->m_len < sizeof(struct ip) + off) {
756 if ((m = m_pullup(m, sizeof (struct ip) + off))
758 TCPSTAT_INC(tcps_rcvshort);
759 return (IPPROTO_DONE);
761 ip = mtod(m, struct ip *);
762 th = (struct tcphdr *)((caddr_t)ip + off0);
766 optlen = off - sizeof (struct tcphdr);
767 optp = (u_char *)(th + 1);
769 thflags = th->th_flags;
772 * Convert TCP protocol specific fields to host format.
774 tcp_fields_to_host(th);
777 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options.
779 drop_hdrlen = off0 + off;
782 * Locate pcb for segment; if we're likely to add or remove a
783 * connection then first acquire pcbinfo lock. There are three cases
784 * where we might discover later we need a write lock despite the
785 * flags: ACKs moving a connection out of the syncache, ACKs for a
786 * connection in TIMEWAIT and SYNs not targeting a listening socket.
788 if ((thflags & (TH_FIN | TH_RST)) != 0) {
789 INP_INFO_RLOCK(&V_tcbinfo);
790 ti_locked = TI_RLOCKED;
792 ti_locked = TI_UNLOCKED;
795 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
799 (isipv6 && (m->m_flags & M_IP6_NEXTHOP))
801 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP))
804 #if defined(INET) && !defined(INET6)
805 (m->m_flags & M_IP_NEXTHOP)
808 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
812 if (ti_locked == TI_RLOCKED) {
813 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
815 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
819 if (isipv6 && fwd_tag != NULL) {
820 struct sockaddr_in6 *next_hop6;
822 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1);
824 * Transparently forwarded. Pretend to be the destination.
825 * Already got one like this?
827 inp = in6_pcblookup_mbuf(&V_tcbinfo,
828 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport,
829 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m);
832 * It's new. Try to find the ambushing socket.
833 * Because we've rewritten the destination address,
834 * any hardware-generated hash is ignored.
836 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src,
837 th->th_sport, &next_hop6->sin6_addr,
838 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) :
839 th->th_dport, INPLOOKUP_WILDCARD |
840 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
843 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src,
844 th->th_sport, &ip6->ip6_dst, th->th_dport,
845 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
846 m->m_pkthdr.rcvif, m);
849 #if defined(INET6) && defined(INET)
853 if (fwd_tag != NULL) {
854 struct sockaddr_in *next_hop;
856 next_hop = (struct sockaddr_in *)(fwd_tag+1);
858 * Transparently forwarded. Pretend to be the destination.
859 * already got one like this?
861 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport,
862 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB,
863 m->m_pkthdr.rcvif, m);
866 * It's new. Try to find the ambushing socket.
867 * Because we've rewritten the destination address,
868 * any hardware-generated hash is ignored.
870 inp = in_pcblookup(&V_tcbinfo, ip->ip_src,
871 th->th_sport, next_hop->sin_addr,
872 next_hop->sin_port ? ntohs(next_hop->sin_port) :
873 th->th_dport, INPLOOKUP_WILDCARD |
874 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
877 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src,
878 th->th_sport, ip->ip_dst, th->th_dport,
879 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
880 m->m_pkthdr.rcvif, m);
884 * If the INPCB does not exist then all data in the incoming
885 * segment is discarded and an appropriate RST is sent back.
886 * XXX MRT Send RST using which routing table?
890 * Log communication attempts to ports that are not
893 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) ||
894 tcp_log_in_vain == 2) {
895 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6)))
896 log(LOG_INFO, "%s; %s: Connection attempt "
897 "to closed port\n", s, __func__);
900 * When blackholing do not respond with a RST but
901 * completely ignore the segment and drop it.
903 if ((V_blackhole == 1 && (thflags & TH_SYN)) ||
907 rstreason = BANDLIM_RST_CLOSEDPORT;
910 INP_WLOCK_ASSERT(inp);
911 if ((inp->inp_flowtype == M_HASHTYPE_NONE) &&
912 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) &&
913 ((inp->inp_socket == NULL) ||
914 (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0)) {
915 inp->inp_flowid = m->m_pkthdr.flowid;
916 inp->inp_flowtype = M_HASHTYPE_GET(m);
920 if (isipv6 && ipsec6_in_reject(m, inp)) {
924 if (ipsec4_in_reject(m, inp) != 0) {
930 * Check the minimum TTL for socket.
932 if (inp->inp_ip_minttl != 0) {
934 if (isipv6 && inp->inp_ip_minttl > ip6->ip6_hlim)
938 if (inp->inp_ip_minttl > ip->ip_ttl)
943 * A previous connection in TIMEWAIT state is supposed to catch stray
944 * or duplicate segments arriving late. If this segment was a
945 * legitimate new connection attempt, the old INPCB gets removed and
946 * we can try again to find a listening socket.
948 * At this point, due to earlier optimism, we may hold only an inpcb
949 * lock, and not the inpcbinfo write lock. If so, we need to try to
950 * acquire it, or if that fails, acquire a reference on the inpcb,
951 * drop all locks, acquire a global write lock, and then re-acquire
952 * the inpcb lock. We may at that point discover that another thread
953 * has tried to free the inpcb, in which case we need to loop back
954 * and try to find a new inpcb to deliver to.
956 * XXXRW: It may be time to rethink timewait locking.
959 if (inp->inp_flags & INP_TIMEWAIT) {
960 if (ti_locked == TI_UNLOCKED) {
961 if (INP_INFO_TRY_RLOCK(&V_tcbinfo) == 0) {
964 INP_INFO_RLOCK(&V_tcbinfo);
965 ti_locked = TI_RLOCKED;
967 if (in_pcbrele_wlocked(inp)) {
972 ti_locked = TI_RLOCKED;
974 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
976 if (thflags & TH_SYN)
977 tcp_dooptions(&to, optp, optlen, TO_SYN);
979 * NB: tcp_twcheck unlocks the INP and frees the mbuf.
981 if (tcp_twcheck(inp, &to, th, m, tlen))
983 INP_INFO_RUNLOCK(&V_tcbinfo);
984 return (IPPROTO_DONE);
987 * The TCPCB may no longer exist if the connection is winding
988 * down or it is in the CLOSED state. Either way we drop the
989 * segment and send an appropriate response.
992 if (tp == NULL || tp->t_state == TCPS_CLOSED) {
993 rstreason = BANDLIM_RST_CLOSEDPORT;
998 if (tp->t_flags & TF_TOE) {
999 tcp_offload_input(tp, m);
1000 m = NULL; /* consumed by the TOE driver */
1006 * We've identified a valid inpcb, but it could be that we need an
1007 * inpcbinfo write lock but don't hold it. In this case, attempt to
1008 * acquire using the same strategy as the TIMEWAIT case above. If we
1009 * relock, we have to jump back to 'relocked' as the connection might
1010 * now be in TIMEWAIT.
1013 if ((thflags & (TH_FIN | TH_RST)) != 0)
1014 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1016 if (!((tp->t_state == TCPS_ESTABLISHED && (thflags & TH_SYN) == 0) ||
1017 (tp->t_state == TCPS_LISTEN && (thflags & TH_SYN)))) {
1018 if (ti_locked == TI_UNLOCKED) {
1019 if (INP_INFO_TRY_RLOCK(&V_tcbinfo) == 0) {
1022 INP_INFO_RLOCK(&V_tcbinfo);
1023 ti_locked = TI_RLOCKED;
1025 if (in_pcbrele_wlocked(inp)) {
1031 ti_locked = TI_RLOCKED;
1033 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1037 INP_WLOCK_ASSERT(inp);
1038 if (mac_inpcb_check_deliver(inp, m))
1041 so = inp->inp_socket;
1042 KASSERT(so != NULL, ("%s: so == NULL", __func__));
1044 if (so->so_options & SO_DEBUG) {
1045 ostate = tp->t_state;
1048 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
1051 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
1054 #endif /* TCPDEBUG */
1056 * When the socket is accepting connections (the INPCB is in LISTEN
1057 * state) we look into the SYN cache if this is a new connection
1058 * attempt or the completion of a previous one.
1060 if (so->so_options & SO_ACCEPTCONN) {
1061 struct in_conninfo inc;
1063 KASSERT(tp->t_state == TCPS_LISTEN, ("%s: so accepting but "
1064 "tp not listening", __func__));
1065 bzero(&inc, sizeof(inc));
1068 inc.inc_flags |= INC_ISIPV6;
1069 inc.inc6_faddr = ip6->ip6_src;
1070 inc.inc6_laddr = ip6->ip6_dst;
1074 inc.inc_faddr = ip->ip_src;
1075 inc.inc_laddr = ip->ip_dst;
1077 inc.inc_fport = th->th_sport;
1078 inc.inc_lport = th->th_dport;
1079 inc.inc_fibnum = so->so_fibnum;
1082 * Check for an existing connection attempt in syncache if
1083 * the flag is only ACK. A successful lookup creates a new
1084 * socket appended to the listen queue in SYN_RECEIVED state.
1086 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
1088 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1090 * Parse the TCP options here because
1091 * syncookies need access to the reflected
1094 tcp_dooptions(&to, optp, optlen, 0);
1096 * NB: syncache_expand() doesn't unlock
1097 * inp and tcpinfo locks.
1099 if (!syncache_expand(&inc, &to, th, &so, m)) {
1101 * No syncache entry or ACK was not
1102 * for our SYN/ACK. Send a RST.
1103 * NB: syncache did its own logging
1104 * of the failure cause.
1106 rstreason = BANDLIM_RST_OPENPORT;
1111 * We completed the 3-way handshake
1112 * but could not allocate a socket
1113 * either due to memory shortage,
1114 * listen queue length limits or
1115 * global socket limits. Send RST
1116 * or wait and have the remote end
1117 * retransmit the ACK for another
1120 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1121 log(LOG_DEBUG, "%s; %s: Listen socket: "
1122 "Socket allocation failed due to "
1123 "limits or memory shortage, %s\n",
1125 V_tcp_sc_rst_sock_fail ?
1126 "sending RST" : "try again");
1127 if (V_tcp_sc_rst_sock_fail) {
1128 rstreason = BANDLIM_UNLIMITED;
1134 * Socket is created in state SYN_RECEIVED.
1135 * Unlock the listen socket, lock the newly
1136 * created socket and update the tp variable.
1138 INP_WUNLOCK(inp); /* listen socket */
1139 inp = sotoinpcb(so);
1141 * New connection inpcb is already locked by
1142 * syncache_expand().
1144 INP_WLOCK_ASSERT(inp);
1145 tp = intotcpcb(inp);
1146 KASSERT(tp->t_state == TCPS_SYN_RECEIVED,
1147 ("%s: ", __func__));
1148 #ifdef TCP_SIGNATURE
1149 if (sig_checked == 0) {
1150 tcp_dooptions(&to, optp, optlen,
1151 (thflags & TH_SYN) ? TO_SYN : 0);
1152 if (!tcp_signature_verify_input(m, off0, tlen,
1153 optlen, &to, th, tp->t_flags)) {
1156 * In SYN_SENT state if it receives an
1157 * RST, it is allowed for further
1160 if ((thflags & TH_RST) == 0 ||
1161 (tp->t_state == TCPS_SYN_SENT) == 0)
1169 * Process the segment and the data it
1170 * contains. tcp_do_segment() consumes
1171 * the mbuf chain and unlocks the inpcb.
1173 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen,
1175 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1176 return (IPPROTO_DONE);
1179 * Segment flag validation for new connection attempts:
1181 * Our (SYN|ACK) response was rejected.
1182 * Check with syncache and remove entry to prevent
1185 * NB: syncache_chkrst does its own logging of failure
1188 if (thflags & TH_RST) {
1189 syncache_chkrst(&inc, th);
1193 * We can't do anything without SYN.
1195 if ((thflags & TH_SYN) == 0) {
1196 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1197 log(LOG_DEBUG, "%s; %s: Listen socket: "
1198 "SYN is missing, segment ignored\n",
1200 TCPSTAT_INC(tcps_badsyn);
1204 * (SYN|ACK) is bogus on a listen socket.
1206 if (thflags & TH_ACK) {
1207 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1208 log(LOG_DEBUG, "%s; %s: Listen socket: "
1209 "SYN|ACK invalid, segment rejected\n",
1211 syncache_badack(&inc); /* XXX: Not needed! */
1212 TCPSTAT_INC(tcps_badsyn);
1213 rstreason = BANDLIM_RST_OPENPORT;
1217 * If the drop_synfin option is enabled, drop all
1218 * segments with both the SYN and FIN bits set.
1219 * This prevents e.g. nmap from identifying the
1221 * XXX: Poor reasoning. nmap has other methods
1222 * and is constantly refining its stack detection
1224 * XXX: This is a violation of the TCP specification
1225 * and was used by RFC1644.
1227 if ((thflags & TH_FIN) && V_drop_synfin) {
1228 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1229 log(LOG_DEBUG, "%s; %s: Listen socket: "
1230 "SYN|FIN segment ignored (based on "
1231 "sysctl setting)\n", s, __func__);
1232 TCPSTAT_INC(tcps_badsyn);
1236 * Segment's flags are (SYN) or (SYN|FIN).
1238 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored
1239 * as they do not affect the state of the TCP FSM.
1240 * The data pointed to by TH_URG and th_urp is ignored.
1242 KASSERT((thflags & (TH_RST|TH_ACK)) == 0,
1243 ("%s: Listen socket: TH_RST or TH_ACK set", __func__));
1244 KASSERT(thflags & (TH_SYN),
1245 ("%s: Listen socket: TH_SYN not set", __func__));
1248 * If deprecated address is forbidden,
1249 * we do not accept SYN to deprecated interface
1250 * address to prevent any new inbound connection from
1251 * getting established.
1252 * When we do not accept SYN, we send a TCP RST,
1253 * with deprecated source address (instead of dropping
1254 * it). We compromise it as it is much better for peer
1255 * to send a RST, and RST will be the final packet
1258 * If we do not forbid deprecated addresses, we accept
1259 * the SYN packet. RFC2462 does not suggest dropping
1261 * If we decipher RFC2462 5.5.4, it says like this:
1262 * 1. use of deprecated addr with existing
1263 * communication is okay - "SHOULD continue to be
1265 * 2. use of it with new communication:
1266 * (2a) "SHOULD NOT be used if alternate address
1267 * with sufficient scope is available"
1268 * (2b) nothing mentioned otherwise.
1269 * Here we fall into (2b) case as we have no choice in
1270 * our source address selection - we must obey the peer.
1272 * The wording in RFC2462 is confusing, and there are
1273 * multiple description text for deprecated address
1274 * handling - worse, they are not exactly the same.
1275 * I believe 5.5.4 is the best one, so we follow 5.5.4.
1277 if (isipv6 && !V_ip6_use_deprecated) {
1278 struct in6_ifaddr *ia6;
1280 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */);
1282 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1283 ifa_free(&ia6->ia_ifa);
1284 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1285 log(LOG_DEBUG, "%s; %s: Listen socket: "
1286 "Connection attempt to deprecated "
1287 "IPv6 address rejected\n",
1289 rstreason = BANDLIM_RST_OPENPORT;
1293 ifa_free(&ia6->ia_ifa);
1297 * Basic sanity checks on incoming SYN requests:
1298 * Don't respond if the destination is a link layer
1299 * broadcast according to RFC1122 4.2.3.10, p. 104.
1300 * If it is from this socket it must be forged.
1301 * Don't respond if the source or destination is a
1302 * global or subnet broad- or multicast address.
1303 * Note that it is quite possible to receive unicast
1304 * link-layer packets with a broadcast IP address. Use
1305 * in_broadcast() to find them.
1307 if (m->m_flags & (M_BCAST|M_MCAST)) {
1308 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1309 log(LOG_DEBUG, "%s; %s: Listen socket: "
1310 "Connection attempt from broad- or multicast "
1311 "link layer address ignored\n", s, __func__);
1316 if (th->th_dport == th->th_sport &&
1317 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) {
1318 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1319 log(LOG_DEBUG, "%s; %s: Listen socket: "
1320 "Connection attempt to/from self "
1321 "ignored\n", s, __func__);
1324 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1325 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) {
1326 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1327 log(LOG_DEBUG, "%s; %s: Listen socket: "
1328 "Connection attempt from/to multicast "
1329 "address ignored\n", s, __func__);
1334 #if defined(INET) && defined(INET6)
1339 if (th->th_dport == th->th_sport &&
1340 ip->ip_dst.s_addr == ip->ip_src.s_addr) {
1341 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1342 log(LOG_DEBUG, "%s; %s: Listen socket: "
1343 "Connection attempt from/to self "
1344 "ignored\n", s, __func__);
1347 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1348 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1349 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1350 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
1351 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1352 log(LOG_DEBUG, "%s; %s: Listen socket: "
1353 "Connection attempt from/to broad- "
1354 "or multicast address ignored\n",
1361 * SYN appears to be valid. Create compressed TCP state
1365 if (so->so_options & SO_DEBUG)
1366 tcp_trace(TA_INPUT, ostate, tp,
1367 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1369 TCP_PROBE3(debug__input, tp, th, mtod(m, const char *));
1370 tcp_dooptions(&to, optp, optlen, TO_SYN);
1371 syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL);
1373 * Entry added to syncache and mbuf consumed.
1374 * Only the listen socket is unlocked by syncache_add().
1376 if (ti_locked == TI_RLOCKED) {
1377 INP_INFO_RUNLOCK(&V_tcbinfo);
1378 ti_locked = TI_UNLOCKED;
1380 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1381 return (IPPROTO_DONE);
1382 } else if (tp->t_state == TCPS_LISTEN) {
1384 * When a listen socket is torn down the SO_ACCEPTCONN
1385 * flag is removed first while connections are drained
1386 * from the accept queue in a unlock/lock cycle of the
1387 * ACCEPT_LOCK, opening a race condition allowing a SYN
1388 * attempt go through unhandled.
1393 #ifdef TCP_SIGNATURE
1394 if (sig_checked == 0) {
1395 tcp_dooptions(&to, optp, optlen,
1396 (thflags & TH_SYN) ? TO_SYN : 0);
1397 if (!tcp_signature_verify_input(m, off0, tlen, optlen, &to,
1401 * In SYN_SENT state if it receives an RST, it is
1402 * allowed for further processing.
1404 if ((thflags & TH_RST) == 0 ||
1405 (tp->t_state == TCPS_SYN_SENT) == 0)
1412 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th);
1415 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later
1416 * state. tcp_do_segment() always consumes the mbuf chain, unlocks
1417 * the inpcb, and unlocks pcbinfo.
1419 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos, ti_locked);
1420 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1421 return (IPPROTO_DONE);
1424 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th);
1426 if (ti_locked == TI_RLOCKED) {
1427 INP_INFO_RUNLOCK(&V_tcbinfo);
1428 ti_locked = TI_UNLOCKED;
1432 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset "
1433 "ti_locked: %d", __func__, ti_locked));
1434 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1439 tcp_dropwithreset(m, th, tp, tlen, rstreason);
1442 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
1443 m = NULL; /* mbuf chain got consumed. */
1448 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th);
1450 if (ti_locked == TI_RLOCKED) {
1451 INP_INFO_RUNLOCK(&V_tcbinfo);
1452 ti_locked = TI_UNLOCKED;
1456 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock "
1457 "ti_locked: %d", __func__, ti_locked));
1458 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1466 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1471 return (IPPROTO_DONE);
1475 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so,
1476 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos,
1479 int thflags, acked, ourfinisacked, needoutput = 0;
1480 int rstreason, todrop, win;
1483 struct in_conninfo *inc;
1489 * The size of tcp_saveipgen must be the size of the max ip header,
1492 u_char tcp_saveipgen[IP6_HDR_LEN];
1493 struct tcphdr tcp_savetcp;
1496 thflags = th->th_flags;
1497 inc = &tp->t_inpcb->inp_inc;
1498 tp->sackhint.last_sack_ack = 0;
1501 * If this is either a state-changing packet or current state isn't
1502 * established, we require a write lock on tcbinfo. Otherwise, we
1503 * allow the tcbinfo to be in either alocked or unlocked, as the
1504 * caller may have unnecessarily acquired a write lock due to a race.
1506 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 ||
1507 tp->t_state != TCPS_ESTABLISHED) {
1508 KASSERT(ti_locked == TI_RLOCKED, ("%s ti_locked %d for "
1509 "SYN/FIN/RST/!EST", __func__, ti_locked));
1510 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1513 if (ti_locked == TI_RLOCKED)
1514 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1516 KASSERT(ti_locked == TI_UNLOCKED, ("%s: EST "
1517 "ti_locked: %d", __func__, ti_locked));
1518 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1522 INP_WLOCK_ASSERT(tp->t_inpcb);
1523 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN",
1525 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT",
1529 /* Save segment, if requested. */
1530 tcp_pcap_add(th, m, &(tp->t_inpkts));
1534 * Segment received on connection.
1535 * Reset idle time and keep-alive timer.
1536 * XXX: This should be done after segment
1537 * validation to ignore broken/spoofed segs.
1539 tp->t_rcvtime = ticks;
1540 if (TCPS_HAVEESTABLISHED(tp->t_state))
1541 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
1544 * Scale up the window into a 32-bit value.
1545 * For the SYN_SENT state the scale is zero.
1547 tiwin = th->th_win << tp->snd_scale;
1550 * TCP ECN processing.
1552 if (tp->t_flags & TF_ECN_PERMIT) {
1553 if (thflags & TH_CWR)
1554 tp->t_flags &= ~TF_ECN_SND_ECE;
1555 switch (iptos & IPTOS_ECN_MASK) {
1557 tp->t_flags |= TF_ECN_SND_ECE;
1558 TCPSTAT_INC(tcps_ecn_ce);
1560 case IPTOS_ECN_ECT0:
1561 TCPSTAT_INC(tcps_ecn_ect0);
1563 case IPTOS_ECN_ECT1:
1564 TCPSTAT_INC(tcps_ecn_ect1);
1568 /* Process a packet differently from RFC3168. */
1569 cc_ecnpkt_handler(tp, th, iptos);
1571 /* Congestion experienced. */
1572 if (thflags & TH_ECE) {
1573 cc_cong_signal(tp, th, CC_ECN);
1578 * Parse options on any incoming segment.
1580 tcp_dooptions(&to, (u_char *)(th + 1),
1581 (th->th_off << 2) - sizeof(struct tcphdr),
1582 (thflags & TH_SYN) ? TO_SYN : 0);
1585 * If echoed timestamp is later than the current time,
1586 * fall back to non RFC1323 RTT calculation. Normalize
1587 * timestamp if syncookies were used when this connection
1590 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1591 to.to_tsecr -= tp->ts_offset;
1592 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks()))
1596 * If timestamps were negotiated during SYN/ACK they should
1597 * appear on every segment during this session and vice versa.
1599 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) {
1600 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1601 log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1602 "no action\n", s, __func__);
1606 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) {
1607 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1608 log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
1609 "no action\n", s, __func__);
1615 * Process options only when we get SYN/ACK back. The SYN case
1616 * for incoming connections is handled in tcp_syncache.
1617 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
1618 * or <SYN,ACK>) segment itself is never scaled.
1619 * XXX this is traditional behavior, may need to be cleaned up.
1621 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1622 if ((to.to_flags & TOF_SCALE) &&
1623 (tp->t_flags & TF_REQ_SCALE)) {
1624 tp->t_flags |= TF_RCVD_SCALE;
1625 tp->snd_scale = to.to_wscale;
1628 * Initial send window. It will be updated with
1629 * the next incoming segment to the scaled value.
1631 tp->snd_wnd = th->th_win;
1632 if (to.to_flags & TOF_TS) {
1633 tp->t_flags |= TF_RCVD_TSTMP;
1634 tp->ts_recent = to.to_tsval;
1635 tp->ts_recent_age = tcp_ts_getticks();
1637 if (to.to_flags & TOF_MSS)
1638 tcp_mss(tp, to.to_mss);
1639 if ((tp->t_flags & TF_SACK_PERMIT) &&
1640 (to.to_flags & TOF_SACKPERM) == 0)
1641 tp->t_flags &= ~TF_SACK_PERMIT;
1645 * Header prediction: check for the two common cases
1646 * of a uni-directional data xfer. If the packet has
1647 * no control flags, is in-sequence, the window didn't
1648 * change and we're not retransmitting, it's a
1649 * candidate. If the length is zero and the ack moved
1650 * forward, we're the sender side of the xfer. Just
1651 * free the data acked & wake any higher level process
1652 * that was blocked waiting for space. If the length
1653 * is non-zero and the ack didn't move, we're the
1654 * receiver side. If we're getting packets in-order
1655 * (the reassembly queue is empty), add the data to
1656 * the socket buffer and note that we need a delayed ack.
1657 * Make sure that the hidden state-flags are also off.
1658 * Since we check for TCPS_ESTABLISHED first, it can only
1661 if (tp->t_state == TCPS_ESTABLISHED &&
1662 th->th_seq == tp->rcv_nxt &&
1663 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1664 tp->snd_nxt == tp->snd_max &&
1665 tiwin && tiwin == tp->snd_wnd &&
1666 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1667 LIST_EMPTY(&tp->t_segq) &&
1668 ((to.to_flags & TOF_TS) == 0 ||
1669 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) {
1672 * If last ACK falls within this segment's sequence numbers,
1673 * record the timestamp.
1674 * NOTE that the test is modified according to the latest
1675 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1677 if ((to.to_flags & TOF_TS) != 0 &&
1678 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1679 tp->ts_recent_age = tcp_ts_getticks();
1680 tp->ts_recent = to.to_tsval;
1684 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1685 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1686 !IN_RECOVERY(tp->t_flags) &&
1687 (to.to_flags & TOF_SACK) == 0 &&
1688 TAILQ_EMPTY(&tp->snd_holes)) {
1690 * This is a pure ack for outstanding data.
1692 if (ti_locked == TI_RLOCKED)
1693 INP_INFO_RUNLOCK(&V_tcbinfo);
1694 ti_locked = TI_UNLOCKED;
1696 TCPSTAT_INC(tcps_predack);
1699 * "bad retransmit" recovery.
1701 if (tp->t_rxtshift == 1 &&
1702 tp->t_flags & TF_PREVVALID &&
1703 (int)(ticks - tp->t_badrxtwin) < 0) {
1704 cc_cong_signal(tp, th, CC_RTO_ERR);
1708 * Recalculate the transmit timer / rtt.
1710 * Some boxes send broken timestamp replies
1711 * during the SYN+ACK phase, ignore
1712 * timestamps of 0 or we could calculate a
1713 * huge RTT and blow up the retransmit timer.
1715 if ((to.to_flags & TOF_TS) != 0 &&
1719 t = tcp_ts_getticks() - to.to_tsecr;
1720 if (!tp->t_rttlow || tp->t_rttlow > t)
1723 TCP_TS_TO_TICKS(t) + 1);
1724 } else if (tp->t_rtttime &&
1725 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1726 if (!tp->t_rttlow ||
1727 tp->t_rttlow > ticks - tp->t_rtttime)
1728 tp->t_rttlow = ticks - tp->t_rtttime;
1730 ticks - tp->t_rtttime);
1732 acked = BYTES_THIS_ACK(tp, th);
1734 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
1735 hhook_run_tcp_est_in(tp, th, &to);
1737 TCPSTAT_INC(tcps_rcvackpack);
1738 TCPSTAT_ADD(tcps_rcvackbyte, acked);
1739 sbdrop(&so->so_snd, acked);
1740 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1741 SEQ_LEQ(th->th_ack, tp->snd_recover))
1742 tp->snd_recover = th->th_ack - 1;
1745 * Let the congestion control algorithm update
1746 * congestion control related information. This
1747 * typically means increasing the congestion
1750 cc_ack_received(tp, th, CC_ACK);
1752 tp->snd_una = th->th_ack;
1754 * Pull snd_wl2 up to prevent seq wrap relative
1757 tp->snd_wl2 = th->th_ack;
1762 * If all outstanding data are acked, stop
1763 * retransmit timer, otherwise restart timer
1764 * using current (possibly backed-off) value.
1765 * If process is waiting for space,
1766 * wakeup/selwakeup/signal. If data
1767 * are ready to send, let tcp_output
1768 * decide between more output or persist.
1771 if (so->so_options & SO_DEBUG)
1772 tcp_trace(TA_INPUT, ostate, tp,
1773 (void *)tcp_saveipgen,
1776 TCP_PROBE3(debug__input, tp, th,
1777 mtod(m, const char *));
1778 if (tp->snd_una == tp->snd_max)
1779 tcp_timer_activate(tp, TT_REXMT, 0);
1780 else if (!tcp_timer_active(tp, TT_PERSIST))
1781 tcp_timer_activate(tp, TT_REXMT,
1784 if (sbavail(&so->so_snd))
1785 (void) tcp_output(tp);
1788 } else if (th->th_ack == tp->snd_una &&
1789 tlen <= sbspace(&so->so_rcv)) {
1790 int newsize = 0; /* automatic sockbuf scaling */
1793 * This is a pure, in-sequence data packet with
1794 * nothing on the reassembly queue and we have enough
1795 * buffer space to take it.
1797 if (ti_locked == TI_RLOCKED)
1798 INP_INFO_RUNLOCK(&V_tcbinfo);
1799 ti_locked = TI_UNLOCKED;
1801 /* Clean receiver SACK report if present */
1802 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks)
1803 tcp_clean_sackreport(tp);
1804 TCPSTAT_INC(tcps_preddat);
1805 tp->rcv_nxt += tlen;
1807 * Pull snd_wl1 up to prevent seq wrap relative to
1810 tp->snd_wl1 = th->th_seq;
1812 * Pull rcv_up up to prevent seq wrap relative to
1815 tp->rcv_up = tp->rcv_nxt;
1816 TCPSTAT_INC(tcps_rcvpack);
1817 TCPSTAT_ADD(tcps_rcvbyte, tlen);
1819 if (so->so_options & SO_DEBUG)
1820 tcp_trace(TA_INPUT, ostate, tp,
1821 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1823 TCP_PROBE3(debug__input, tp, th, mtod(m, const char *));
1826 * Automatic sizing of receive socket buffer. Often the send
1827 * buffer size is not optimally adjusted to the actual network
1828 * conditions at hand (delay bandwidth product). Setting the
1829 * buffer size too small limits throughput on links with high
1830 * bandwidth and high delay (eg. trans-continental/oceanic links).
1832 * On the receive side the socket buffer memory is only rarely
1833 * used to any significant extent. This allows us to be much
1834 * more aggressive in scaling the receive socket buffer. For
1835 * the case that the buffer space is actually used to a large
1836 * extent and we run out of kernel memory we can simply drop
1837 * the new segments; TCP on the sender will just retransmit it
1838 * later. Setting the buffer size too big may only consume too
1839 * much kernel memory if the application doesn't read() from
1840 * the socket or packet loss or reordering makes use of the
1843 * The criteria to step up the receive buffer one notch are:
1844 * 1. Application has not set receive buffer size with
1845 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE.
1846 * 2. the number of bytes received during the time it takes
1847 * one timestamp to be reflected back to us (the RTT);
1848 * 3. received bytes per RTT is within seven eighth of the
1849 * current socket buffer size;
1850 * 4. receive buffer size has not hit maximal automatic size;
1852 * This algorithm does one step per RTT at most and only if
1853 * we receive a bulk stream w/o packet losses or reorderings.
1854 * Shrinking the buffer during idle times is not necessary as
1855 * it doesn't consume any memory when idle.
1857 * TODO: Only step up if the application is actually serving
1858 * the buffer to better manage the socket buffer resources.
1860 if (V_tcp_do_autorcvbuf &&
1861 (to.to_flags & TOF_TS) &&
1863 (so->so_rcv.sb_flags & SB_AUTOSIZE)) {
1864 if (TSTMP_GT(to.to_tsecr, tp->rfbuf_ts) &&
1865 to.to_tsecr - tp->rfbuf_ts < hz) {
1867 (so->so_rcv.sb_hiwat / 8 * 7) &&
1868 so->so_rcv.sb_hiwat <
1869 V_tcp_autorcvbuf_max) {
1871 min(so->so_rcv.sb_hiwat +
1872 V_tcp_autorcvbuf_inc,
1873 V_tcp_autorcvbuf_max);
1875 /* Start over with next RTT. */
1879 tp->rfbuf_cnt += tlen; /* add up */
1882 /* Add data to socket buffer. */
1883 SOCKBUF_LOCK(&so->so_rcv);
1884 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1888 * Set new socket buffer size.
1889 * Give up when limit is reached.
1892 if (!sbreserve_locked(&so->so_rcv,
1894 so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1895 m_adj(m, drop_hdrlen); /* delayed header drop */
1896 sbappendstream_locked(&so->so_rcv, m, 0);
1898 /* NB: sorwakeup_locked() does an implicit unlock. */
1899 sorwakeup_locked(so);
1900 if (DELAY_ACK(tp, tlen)) {
1901 tp->t_flags |= TF_DELACK;
1903 tp->t_flags |= TF_ACKNOW;
1911 * Calculate amount of space in receive window,
1912 * and then do TCP input processing.
1913 * Receive window is amount of space in rcv queue,
1914 * but not less than advertised window.
1916 win = sbspace(&so->so_rcv);
1919 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1921 /* Reset receive buffer auto scaling when not in bulk receive mode. */
1925 switch (tp->t_state) {
1928 * If the state is SYN_RECEIVED:
1929 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1931 case TCPS_SYN_RECEIVED:
1932 if ((thflags & TH_ACK) &&
1933 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1934 SEQ_GT(th->th_ack, tp->snd_max))) {
1935 rstreason = BANDLIM_RST_OPENPORT;
1941 * If the state is SYN_SENT:
1942 * if seg contains an ACK, but not for our SYN, drop the input.
1943 * if seg contains a RST, then drop the connection.
1944 * if seg does not contain SYN, then drop it.
1945 * Otherwise this is an acceptable SYN segment
1946 * initialize tp->rcv_nxt and tp->irs
1947 * if seg contains ack then advance tp->snd_una
1948 * if seg contains an ECE and ECN support is enabled, the stream
1950 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1951 * arrange for segment to be acked (eventually)
1952 * continue processing rest of data/controls, beginning with URG
1955 if ((thflags & TH_ACK) &&
1956 (SEQ_LEQ(th->th_ack, tp->iss) ||
1957 SEQ_GT(th->th_ack, tp->snd_max))) {
1958 rstreason = BANDLIM_UNLIMITED;
1961 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) {
1962 TCP_PROBE5(connect__refused, NULL, tp,
1963 mtod(m, const char *), tp, th);
1964 tp = tcp_drop(tp, ECONNREFUSED);
1966 if (thflags & TH_RST)
1968 if (!(thflags & TH_SYN))
1971 tp->irs = th->th_seq;
1973 if (thflags & TH_ACK) {
1974 TCPSTAT_INC(tcps_connects);
1977 mac_socketpeer_set_from_mbuf(m, so);
1979 /* Do window scaling on this connection? */
1980 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1981 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1982 tp->rcv_scale = tp->request_r_scale;
1984 tp->rcv_adv += imin(tp->rcv_wnd,
1985 TCP_MAXWIN << tp->rcv_scale);
1986 tp->snd_una++; /* SYN is acked */
1988 * If there's data, delay ACK; if there's also a FIN
1989 * ACKNOW will be turned on later.
1991 if (DELAY_ACK(tp, tlen) && tlen != 0)
1992 tcp_timer_activate(tp, TT_DELACK,
1995 tp->t_flags |= TF_ACKNOW;
1997 if ((thflags & TH_ECE) && V_tcp_do_ecn) {
1998 tp->t_flags |= TF_ECN_PERMIT;
1999 TCPSTAT_INC(tcps_ecn_shs);
2003 * Received <SYN,ACK> in SYN_SENT[*] state.
2005 * SYN_SENT --> ESTABLISHED
2006 * SYN_SENT* --> FIN_WAIT_1
2008 tp->t_starttime = ticks;
2009 if (tp->t_flags & TF_NEEDFIN) {
2010 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2011 tp->t_flags &= ~TF_NEEDFIN;
2014 tcp_state_change(tp, TCPS_ESTABLISHED);
2015 TCP_PROBE5(connect__established, NULL, tp,
2016 mtod(m, const char *), tp, th);
2018 tcp_timer_activate(tp, TT_KEEP,
2023 * Received initial SYN in SYN-SENT[*] state =>
2024 * simultaneous open.
2025 * If it succeeds, connection is * half-synchronized.
2026 * Otherwise, do 3-way handshake:
2027 * SYN-SENT -> SYN-RECEIVED
2028 * SYN-SENT* -> SYN-RECEIVED*
2030 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
2031 tcp_timer_activate(tp, TT_REXMT, 0);
2032 tcp_state_change(tp, TCPS_SYN_RECEIVED);
2035 KASSERT(ti_locked == TI_RLOCKED, ("%s: trimthenstep6: "
2036 "ti_locked %d", __func__, ti_locked));
2037 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2038 INP_WLOCK_ASSERT(tp->t_inpcb);
2041 * Advance th->th_seq to correspond to first data byte.
2042 * If data, trim to stay within window,
2043 * dropping FIN if necessary.
2046 if (tlen > tp->rcv_wnd) {
2047 todrop = tlen - tp->rcv_wnd;
2051 TCPSTAT_INC(tcps_rcvpackafterwin);
2052 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2054 tp->snd_wl1 = th->th_seq - 1;
2055 tp->rcv_up = th->th_seq;
2057 * Client side of transaction: already sent SYN and data.
2058 * If the remote host used T/TCP to validate the SYN,
2059 * our data will be ACK'd; if so, enter normal data segment
2060 * processing in the middle of step 5, ack processing.
2061 * Otherwise, goto step 6.
2063 if (thflags & TH_ACK)
2069 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
2070 * do normal processing.
2072 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later.
2076 break; /* continue normal processing */
2080 * States other than LISTEN or SYN_SENT.
2081 * First check the RST flag and sequence number since reset segments
2082 * are exempt from the timestamp and connection count tests. This
2083 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
2084 * below which allowed reset segments in half the sequence space
2085 * to fall though and be processed (which gives forged reset
2086 * segments with a random sequence number a 50 percent chance of
2087 * killing a connection).
2088 * Then check timestamp, if present.
2089 * Then check the connection count, if present.
2090 * Then check that at least some bytes of segment are within
2091 * receive window. If segment begins before rcv_nxt,
2092 * drop leading data (and SYN); if nothing left, just ack.
2094 if (thflags & TH_RST) {
2096 * RFC5961 Section 3.2
2098 * - RST drops connection only if SEG.SEQ == RCV.NXT.
2099 * - If RST is in window, we send challenge ACK.
2101 * Note: to take into account delayed ACKs, we should
2102 * test against last_ack_sent instead of rcv_nxt.
2103 * Note 2: we handle special case of closed window, not
2104 * covered by the RFC.
2106 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2107 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) ||
2108 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) {
2110 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2111 KASSERT(ti_locked == TI_RLOCKED,
2112 ("%s: TH_RST ti_locked %d, th %p tp %p",
2113 __func__, ti_locked, th, tp));
2114 KASSERT(tp->t_state != TCPS_SYN_SENT,
2115 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p",
2118 if (V_tcp_insecure_rst ||
2119 tp->last_ack_sent == th->th_seq) {
2120 TCPSTAT_INC(tcps_drops);
2121 /* Drop the connection. */
2122 switch (tp->t_state) {
2123 case TCPS_SYN_RECEIVED:
2124 so->so_error = ECONNREFUSED;
2126 case TCPS_ESTABLISHED:
2127 case TCPS_FIN_WAIT_1:
2128 case TCPS_FIN_WAIT_2:
2129 case TCPS_CLOSE_WAIT:
2130 so->so_error = ECONNRESET;
2132 tcp_state_change(tp, TCPS_CLOSED);
2138 TCPSTAT_INC(tcps_badrst);
2139 /* Send challenge ACK. */
2140 tcp_respond(tp, mtod(m, void *), th, m,
2141 tp->rcv_nxt, tp->snd_nxt, TH_ACK);
2142 tp->last_ack_sent = tp->rcv_nxt;
2150 * RFC5961 Section 4.2
2151 * Send challenge ACK for any SYN in synchronized state.
2153 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT) {
2154 KASSERT(ti_locked == TI_RLOCKED,
2155 ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked));
2156 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2158 TCPSTAT_INC(tcps_badsyn);
2159 if (V_tcp_insecure_syn &&
2160 SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2161 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
2162 tp = tcp_drop(tp, ECONNRESET);
2163 rstreason = BANDLIM_UNLIMITED;
2165 /* Send challenge ACK. */
2166 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt,
2167 tp->snd_nxt, TH_ACK);
2168 tp->last_ack_sent = tp->rcv_nxt;
2175 * RFC 1323 PAWS: If we have a timestamp reply on this segment
2176 * and it's less than ts_recent, drop it.
2178 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
2179 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
2181 /* Check to see if ts_recent is over 24 days old. */
2182 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) {
2184 * Invalidate ts_recent. If this segment updates
2185 * ts_recent, the age will be reset later and ts_recent
2186 * will get a valid value. If it does not, setting
2187 * ts_recent to zero will at least satisfy the
2188 * requirement that zero be placed in the timestamp
2189 * echo reply when ts_recent isn't valid. The
2190 * age isn't reset until we get a valid ts_recent
2191 * because we don't want out-of-order segments to be
2192 * dropped when ts_recent is old.
2196 TCPSTAT_INC(tcps_rcvduppack);
2197 TCPSTAT_ADD(tcps_rcvdupbyte, tlen);
2198 TCPSTAT_INC(tcps_pawsdrop);
2206 * In the SYN-RECEIVED state, validate that the packet belongs to
2207 * this connection before trimming the data to fit the receive
2208 * window. Check the sequence number versus IRS since we know
2209 * the sequence numbers haven't wrapped. This is a partial fix
2210 * for the "LAND" DoS attack.
2212 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
2213 rstreason = BANDLIM_RST_OPENPORT;
2217 todrop = tp->rcv_nxt - th->th_seq;
2219 if (thflags & TH_SYN) {
2229 * Following if statement from Stevens, vol. 2, p. 960.
2232 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
2234 * Any valid FIN must be to the left of the window.
2235 * At this point the FIN must be a duplicate or out
2236 * of sequence; drop it.
2241 * Send an ACK to resynchronize and drop any data.
2242 * But keep on processing for RST or ACK.
2244 tp->t_flags |= TF_ACKNOW;
2246 TCPSTAT_INC(tcps_rcvduppack);
2247 TCPSTAT_ADD(tcps_rcvdupbyte, todrop);
2249 TCPSTAT_INC(tcps_rcvpartduppack);
2250 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop);
2252 drop_hdrlen += todrop; /* drop from the top afterwards */
2253 th->th_seq += todrop;
2255 if (th->th_urp > todrop)
2256 th->th_urp -= todrop;
2264 * If new data are received on a connection after the
2265 * user processes are gone, then RST the other end.
2267 if ((so->so_state & SS_NOFDREF) &&
2268 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
2269 KASSERT(ti_locked == TI_RLOCKED, ("%s: SS_NOFDEREF && "
2270 "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked));
2271 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2273 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
2274 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data "
2275 "after socket was closed, "
2276 "sending RST and removing tcpcb\n",
2277 s, __func__, tcpstates[tp->t_state], tlen);
2281 TCPSTAT_INC(tcps_rcvafterclose);
2282 rstreason = BANDLIM_UNLIMITED;
2287 * If segment ends after window, drop trailing data
2288 * (and PUSH and FIN); if nothing left, just ACK.
2290 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
2292 TCPSTAT_INC(tcps_rcvpackafterwin);
2293 if (todrop >= tlen) {
2294 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen);
2296 * If window is closed can only take segments at
2297 * window edge, and have to drop data and PUSH from
2298 * incoming segments. Continue processing, but
2299 * remember to ack. Otherwise, drop segment
2302 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2303 tp->t_flags |= TF_ACKNOW;
2304 TCPSTAT_INC(tcps_rcvwinprobe);
2308 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2311 thflags &= ~(TH_PUSH|TH_FIN);
2315 * If last ACK falls within this segment's sequence numbers,
2316 * record its timestamp.
2318 * 1) That the test incorporates suggestions from the latest
2319 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
2320 * 2) That updating only on newer timestamps interferes with
2321 * our earlier PAWS tests, so this check should be solely
2322 * predicated on the sequence space of this segment.
2323 * 3) That we modify the segment boundary check to be
2324 * Last.ACK.Sent <= SEG.SEQ + SEG.Len
2325 * instead of RFC1323's
2326 * Last.ACK.Sent < SEG.SEQ + SEG.Len,
2327 * This modified check allows us to overcome RFC1323's
2328 * limitations as described in Stevens TCP/IP Illustrated
2329 * Vol. 2 p.869. In such cases, we can still calculate the
2330 * RTT correctly when RCV.NXT == Last.ACK.Sent.
2332 if ((to.to_flags & TOF_TS) != 0 &&
2333 SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2334 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
2335 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
2336 tp->ts_recent_age = tcp_ts_getticks();
2337 tp->ts_recent = to.to_tsval;
2341 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
2342 * flag is on (half-synchronized state), then queue data for
2343 * later processing; else drop segment and return.
2345 if ((thflags & TH_ACK) == 0) {
2346 if (tp->t_state == TCPS_SYN_RECEIVED ||
2347 (tp->t_flags & TF_NEEDSYN))
2349 else if (tp->t_flags & TF_ACKNOW)
2358 switch (tp->t_state) {
2361 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
2362 * ESTABLISHED state and continue processing.
2363 * The ACK was checked above.
2365 case TCPS_SYN_RECEIVED:
2367 TCPSTAT_INC(tcps_connects);
2369 /* Do window scaling? */
2370 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2371 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2372 tp->rcv_scale = tp->request_r_scale;
2373 tp->snd_wnd = tiwin;
2377 * SYN-RECEIVED -> ESTABLISHED
2378 * SYN-RECEIVED* -> FIN-WAIT-1
2380 tp->t_starttime = ticks;
2381 if (tp->t_flags & TF_NEEDFIN) {
2382 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2383 tp->t_flags &= ~TF_NEEDFIN;
2385 tcp_state_change(tp, TCPS_ESTABLISHED);
2386 TCP_PROBE5(accept__established, NULL, tp,
2387 mtod(m, const char *), tp, th);
2389 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
2392 * If segment contains data or ACK, will call tcp_reass()
2393 * later; if not, do so now to pass queued data to user.
2395 if (tlen == 0 && (thflags & TH_FIN) == 0)
2396 (void) tcp_reass(tp, (struct tcphdr *)0, 0,
2398 tp->snd_wl1 = th->th_seq - 1;
2402 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2403 * ACKs. If the ack is in the range
2404 * tp->snd_una < th->th_ack <= tp->snd_max
2405 * then advance tp->snd_una to th->th_ack and drop
2406 * data from the retransmission queue. If this ACK reflects
2407 * more up to date window information we update our window information.
2409 case TCPS_ESTABLISHED:
2410 case TCPS_FIN_WAIT_1:
2411 case TCPS_FIN_WAIT_2:
2412 case TCPS_CLOSE_WAIT:
2415 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2416 TCPSTAT_INC(tcps_rcvacktoomuch);
2419 if ((tp->t_flags & TF_SACK_PERMIT) &&
2420 ((to.to_flags & TOF_SACK) ||
2421 !TAILQ_EMPTY(&tp->snd_holes)))
2422 tcp_sack_doack(tp, &to, th->th_ack);
2424 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
2425 hhook_run_tcp_est_in(tp, th, &to);
2427 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2428 if (tlen == 0 && tiwin == tp->snd_wnd) {
2430 * If this is the first time we've seen a
2431 * FIN from the remote, this is not a
2432 * duplicate and it needs to be processed
2433 * normally. This happens during a
2434 * simultaneous close.
2436 if ((thflags & TH_FIN) &&
2437 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2441 TCPSTAT_INC(tcps_rcvdupack);
2443 * If we have outstanding data (other than
2444 * a window probe), this is a completely
2445 * duplicate ack (ie, window info didn't
2446 * change and FIN isn't set),
2447 * the ack is the biggest we've
2448 * seen and we've seen exactly our rexmt
2449 * threshhold of them, assume a packet
2450 * has been dropped and retransmit it.
2451 * Kludge snd_nxt & the congestion
2452 * window so we send only this one
2455 * We know we're losing at the current
2456 * window size so do congestion avoidance
2457 * (set ssthresh to half the current window
2458 * and pull our congestion window back to
2459 * the new ssthresh).
2461 * Dup acks mean that packets have left the
2462 * network (they're now cached at the receiver)
2463 * so bump cwnd by the amount in the receiver
2464 * to keep a constant cwnd packets in the
2467 * When using TCP ECN, notify the peer that
2468 * we reduced the cwnd.
2470 if (!tcp_timer_active(tp, TT_REXMT) ||
2471 th->th_ack != tp->snd_una)
2473 else if (++tp->t_dupacks > tcprexmtthresh ||
2474 IN_FASTRECOVERY(tp->t_flags)) {
2475 cc_ack_received(tp, th, CC_DUPACK);
2476 if ((tp->t_flags & TF_SACK_PERMIT) &&
2477 IN_FASTRECOVERY(tp->t_flags)) {
2481 * Compute the amount of data in flight first.
2482 * We can inject new data into the pipe iff
2483 * we have less than 1/2 the original window's
2484 * worth of data in flight.
2486 awnd = (tp->snd_nxt - tp->snd_fack) +
2487 tp->sackhint.sack_bytes_rexmit;
2488 if (awnd < tp->snd_ssthresh) {
2489 tp->snd_cwnd += tp->t_maxseg;
2490 if (tp->snd_cwnd > tp->snd_ssthresh)
2491 tp->snd_cwnd = tp->snd_ssthresh;
2494 tp->snd_cwnd += tp->t_maxseg;
2495 (void) tcp_output(tp);
2497 } else if (tp->t_dupacks == tcprexmtthresh) {
2498 tcp_seq onxt = tp->snd_nxt;
2501 * If we're doing sack, check to
2502 * see if we're already in sack
2503 * recovery. If we're not doing sack,
2504 * check to see if we're in newreno
2507 if (tp->t_flags & TF_SACK_PERMIT) {
2508 if (IN_FASTRECOVERY(tp->t_flags)) {
2513 if (SEQ_LEQ(th->th_ack,
2519 /* Congestion signal before ack. */
2520 cc_cong_signal(tp, th, CC_NDUPACK);
2521 cc_ack_received(tp, th, CC_DUPACK);
2522 tcp_timer_activate(tp, TT_REXMT, 0);
2524 if (tp->t_flags & TF_SACK_PERMIT) {
2526 tcps_sack_recovery_episode);
2527 tp->sack_newdata = tp->snd_nxt;
2528 tp->snd_cwnd = tp->t_maxseg;
2529 (void) tcp_output(tp);
2532 tp->snd_nxt = th->th_ack;
2533 tp->snd_cwnd = tp->t_maxseg;
2534 (void) tcp_output(tp);
2535 KASSERT(tp->snd_limited <= 2,
2536 ("%s: tp->snd_limited too big",
2538 tp->snd_cwnd = tp->snd_ssthresh +
2540 (tp->t_dupacks - tp->snd_limited);
2541 if (SEQ_GT(onxt, tp->snd_nxt))
2544 } else if (V_tcp_do_rfc3042) {
2546 * Process first and second duplicate
2547 * ACKs. Each indicates a segment
2548 * leaving the network, creating room
2549 * for more. Make sure we can send a
2550 * packet on reception of each duplicate
2551 * ACK by increasing snd_cwnd by one
2552 * segment. Restore the original
2553 * snd_cwnd after packet transmission.
2555 cc_ack_received(tp, th, CC_DUPACK);
2556 u_long oldcwnd = tp->snd_cwnd;
2557 tcp_seq oldsndmax = tp->snd_max;
2561 KASSERT(tp->t_dupacks == 1 ||
2563 ("%s: dupacks not 1 or 2",
2565 if (tp->t_dupacks == 1)
2566 tp->snd_limited = 0;
2568 (tp->snd_nxt - tp->snd_una) +
2569 (tp->t_dupacks - tp->snd_limited) *
2572 * Only call tcp_output when there
2573 * is new data available to be sent.
2574 * Otherwise we would send pure ACKs.
2576 SOCKBUF_LOCK(&so->so_snd);
2577 avail = sbavail(&so->so_snd) -
2578 (tp->snd_nxt - tp->snd_una);
2579 SOCKBUF_UNLOCK(&so->so_snd);
2581 (void) tcp_output(tp);
2582 sent = tp->snd_max - oldsndmax;
2583 if (sent > tp->t_maxseg) {
2584 KASSERT((tp->t_dupacks == 2 &&
2585 tp->snd_limited == 0) ||
2586 (sent == tp->t_maxseg + 1 &&
2587 tp->t_flags & TF_SENTFIN),
2588 ("%s: sent too much",
2590 tp->snd_limited = 2;
2591 } else if (sent > 0)
2593 tp->snd_cwnd = oldcwnd;
2601 KASSERT(SEQ_GT(th->th_ack, tp->snd_una),
2602 ("%s: th_ack <= snd_una", __func__));
2605 * If the congestion window was inflated to account
2606 * for the other side's cached packets, retract it.
2608 if (IN_FASTRECOVERY(tp->t_flags)) {
2609 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2610 if (tp->t_flags & TF_SACK_PERMIT)
2611 tcp_sack_partialack(tp, th);
2613 tcp_newreno_partial_ack(tp, th);
2615 cc_post_recovery(tp, th);
2619 * If we reach this point, ACK is not a duplicate,
2620 * i.e., it ACKs something we sent.
2622 if (tp->t_flags & TF_NEEDSYN) {
2624 * T/TCP: Connection was half-synchronized, and our
2625 * SYN has been ACK'd (so connection is now fully
2626 * synchronized). Go to non-starred state,
2627 * increment snd_una for ACK of SYN, and check if
2628 * we can do window scaling.
2630 tp->t_flags &= ~TF_NEEDSYN;
2632 /* Do window scaling? */
2633 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2634 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2635 tp->rcv_scale = tp->request_r_scale;
2636 /* Send window already scaled. */
2641 INP_WLOCK_ASSERT(tp->t_inpcb);
2643 acked = BYTES_THIS_ACK(tp, th);
2644 TCPSTAT_INC(tcps_rcvackpack);
2645 TCPSTAT_ADD(tcps_rcvackbyte, acked);
2648 * If we just performed our first retransmit, and the ACK
2649 * arrives within our recovery window, then it was a mistake
2650 * to do the retransmit in the first place. Recover our
2651 * original cwnd and ssthresh, and proceed to transmit where
2654 if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID &&
2655 (int)(ticks - tp->t_badrxtwin) < 0)
2656 cc_cong_signal(tp, th, CC_RTO_ERR);
2659 * If we have a timestamp reply, update smoothed
2660 * round trip time. If no timestamp is present but
2661 * transmit timer is running and timed sequence
2662 * number was acked, update smoothed round trip time.
2663 * Since we now have an rtt measurement, cancel the
2664 * timer backoff (cf., Phil Karn's retransmit alg.).
2665 * Recompute the initial retransmit timer.
2667 * Some boxes send broken timestamp replies
2668 * during the SYN+ACK phase, ignore
2669 * timestamps of 0 or we could calculate a
2670 * huge RTT and blow up the retransmit timer.
2672 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) {
2675 t = tcp_ts_getticks() - to.to_tsecr;
2676 if (!tp->t_rttlow || tp->t_rttlow > t)
2678 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1);
2679 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2680 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2681 tp->t_rttlow = ticks - tp->t_rtttime;
2682 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2686 * If all outstanding data is acked, stop retransmit
2687 * timer and remember to restart (more output or persist).
2688 * If there is more data to be acked, restart retransmit
2689 * timer, using current (possibly backed-off) value.
2691 if (th->th_ack == tp->snd_max) {
2692 tcp_timer_activate(tp, TT_REXMT, 0);
2694 } else if (!tcp_timer_active(tp, TT_PERSIST))
2695 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
2698 * If no data (only SYN) was ACK'd,
2699 * skip rest of ACK processing.
2705 * Let the congestion control algorithm update congestion
2706 * control related information. This typically means increasing
2707 * the congestion window.
2709 cc_ack_received(tp, th, CC_ACK);
2711 SOCKBUF_LOCK(&so->so_snd);
2712 if (acked > sbavail(&so->so_snd)) {
2713 tp->snd_wnd -= sbavail(&so->so_snd);
2714 mfree = sbcut_locked(&so->so_snd,
2715 (int)sbavail(&so->so_snd));
2718 mfree = sbcut_locked(&so->so_snd, acked);
2719 tp->snd_wnd -= acked;
2722 /* NB: sowwakeup_locked() does an implicit unlock. */
2723 sowwakeup_locked(so);
2725 /* Detect una wraparound. */
2726 if (!IN_RECOVERY(tp->t_flags) &&
2727 SEQ_GT(tp->snd_una, tp->snd_recover) &&
2728 SEQ_LEQ(th->th_ack, tp->snd_recover))
2729 tp->snd_recover = th->th_ack - 1;
2730 /* XXXLAS: Can this be moved up into cc_post_recovery? */
2731 if (IN_RECOVERY(tp->t_flags) &&
2732 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2733 EXIT_RECOVERY(tp->t_flags);
2735 tp->snd_una = th->th_ack;
2736 if (tp->t_flags & TF_SACK_PERMIT) {
2737 if (SEQ_GT(tp->snd_una, tp->snd_recover))
2738 tp->snd_recover = tp->snd_una;
2740 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2741 tp->snd_nxt = tp->snd_una;
2743 switch (tp->t_state) {
2746 * In FIN_WAIT_1 STATE in addition to the processing
2747 * for the ESTABLISHED state if our FIN is now acknowledged
2748 * then enter FIN_WAIT_2.
2750 case TCPS_FIN_WAIT_1:
2751 if (ourfinisacked) {
2753 * If we can't receive any more
2754 * data, then closing user can proceed.
2755 * Starting the timer is contrary to the
2756 * specification, but if we don't get a FIN
2757 * we'll hang forever.
2760 * we should release the tp also, and use a
2763 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2764 soisdisconnected(so);
2765 tcp_timer_activate(tp, TT_2MSL,
2766 (tcp_fast_finwait2_recycle ?
2767 tcp_finwait2_timeout :
2770 tcp_state_change(tp, TCPS_FIN_WAIT_2);
2775 * In CLOSING STATE in addition to the processing for
2776 * the ESTABLISHED state if the ACK acknowledges our FIN
2777 * then enter the TIME-WAIT state, otherwise ignore
2781 if (ourfinisacked) {
2782 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2784 INP_INFO_RUNLOCK(&V_tcbinfo);
2791 * In LAST_ACK, we may still be waiting for data to drain
2792 * and/or to be acked, as well as for the ack of our FIN.
2793 * If our FIN is now acknowledged, delete the TCB,
2794 * enter the closed state and return.
2797 if (ourfinisacked) {
2798 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2807 INP_WLOCK_ASSERT(tp->t_inpcb);
2810 * Update window information.
2811 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2813 if ((thflags & TH_ACK) &&
2814 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2815 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2816 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2817 /* keep track of pure window updates */
2819 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2820 TCPSTAT_INC(tcps_rcvwinupd);
2821 tp->snd_wnd = tiwin;
2822 tp->snd_wl1 = th->th_seq;
2823 tp->snd_wl2 = th->th_ack;
2824 if (tp->snd_wnd > tp->max_sndwnd)
2825 tp->max_sndwnd = tp->snd_wnd;
2830 * Process segments with URG.
2832 if ((thflags & TH_URG) && th->th_urp &&
2833 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2835 * This is a kludge, but if we receive and accept
2836 * random urgent pointers, we'll crash in
2837 * soreceive. It's hard to imagine someone
2838 * actually wanting to send this much urgent data.
2840 SOCKBUF_LOCK(&so->so_rcv);
2841 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) {
2842 th->th_urp = 0; /* XXX */
2843 thflags &= ~TH_URG; /* XXX */
2844 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */
2845 goto dodata; /* XXX */
2848 * If this segment advances the known urgent pointer,
2849 * then mark the data stream. This should not happen
2850 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2851 * a FIN has been received from the remote side.
2852 * In these states we ignore the URG.
2854 * According to RFC961 (Assigned Protocols),
2855 * the urgent pointer points to the last octet
2856 * of urgent data. We continue, however,
2857 * to consider it to indicate the first octet
2858 * of data past the urgent section as the original
2859 * spec states (in one of two places).
2861 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2862 tp->rcv_up = th->th_seq + th->th_urp;
2863 so->so_oobmark = sbavail(&so->so_rcv) +
2864 (tp->rcv_up - tp->rcv_nxt) - 1;
2865 if (so->so_oobmark == 0)
2866 so->so_rcv.sb_state |= SBS_RCVATMARK;
2868 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2870 SOCKBUF_UNLOCK(&so->so_rcv);
2872 * Remove out of band data so doesn't get presented to user.
2873 * This can happen independent of advancing the URG pointer,
2874 * but if two URG's are pending at once, some out-of-band
2875 * data may creep in... ick.
2877 if (th->th_urp <= (u_long)tlen &&
2878 !(so->so_options & SO_OOBINLINE)) {
2879 /* hdr drop is delayed */
2880 tcp_pulloutofband(so, th, m, drop_hdrlen);
2884 * If no out of band data is expected,
2885 * pull receive urgent pointer along
2886 * with the receive window.
2888 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2889 tp->rcv_up = tp->rcv_nxt;
2892 INP_WLOCK_ASSERT(tp->t_inpcb);
2895 * Process the segment text, merging it into the TCP sequencing queue,
2896 * and arranging for acknowledgment of receipt if necessary.
2897 * This process logically involves adjusting tp->rcv_wnd as data
2898 * is presented to the user (this happens in tcp_usrreq.c,
2899 * case PRU_RCVD). If a FIN has already been received on this
2900 * connection then we just ignore the text.
2902 if ((tlen || (thflags & TH_FIN)) &&
2903 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2904 tcp_seq save_start = th->th_seq;
2905 m_adj(m, drop_hdrlen); /* delayed header drop */
2907 * Insert segment which includes th into TCP reassembly queue
2908 * with control block tp. Set thflags to whether reassembly now
2909 * includes a segment with FIN. This handles the common case
2910 * inline (segment is the next to be received on an established
2911 * connection, and the queue is empty), avoiding linkage into
2912 * and removal from the queue and repetition of various
2914 * Set DELACK for segments received in order, but ack
2915 * immediately when segments are out of order (so
2916 * fast retransmit can work).
2918 if (th->th_seq == tp->rcv_nxt &&
2919 LIST_EMPTY(&tp->t_segq) &&
2920 TCPS_HAVEESTABLISHED(tp->t_state)) {
2921 if (DELAY_ACK(tp, tlen))
2922 tp->t_flags |= TF_DELACK;
2924 tp->t_flags |= TF_ACKNOW;
2925 tp->rcv_nxt += tlen;
2926 thflags = th->th_flags & TH_FIN;
2927 TCPSTAT_INC(tcps_rcvpack);
2928 TCPSTAT_ADD(tcps_rcvbyte, tlen);
2929 SOCKBUF_LOCK(&so->so_rcv);
2930 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
2933 sbappendstream_locked(&so->so_rcv, m, 0);
2934 /* NB: sorwakeup_locked() does an implicit unlock. */
2935 sorwakeup_locked(so);
2938 * XXX: Due to the header drop above "th" is
2939 * theoretically invalid by now. Fortunately
2940 * m_adj() doesn't actually frees any mbufs
2941 * when trimming from the head.
2943 thflags = tcp_reass(tp, th, &tlen, m);
2944 tp->t_flags |= TF_ACKNOW;
2946 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT))
2947 tcp_update_sack_list(tp, save_start, save_start + tlen);
2950 * Note the amount of data that peer has sent into
2951 * our window, in order to estimate the sender's
2955 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
2956 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2958 len = so->so_rcv.sb_hiwat;
2966 * If FIN is received ACK the FIN and let the user know
2967 * that the connection is closing.
2969 if (thflags & TH_FIN) {
2970 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2973 * If connection is half-synchronized
2974 * (ie NEEDSYN flag on) then delay ACK,
2975 * so it may be piggybacked when SYN is sent.
2976 * Otherwise, since we received a FIN then no
2977 * more input can be expected, send ACK now.
2979 if (tp->t_flags & TF_NEEDSYN)
2980 tp->t_flags |= TF_DELACK;
2982 tp->t_flags |= TF_ACKNOW;
2985 switch (tp->t_state) {
2988 * In SYN_RECEIVED and ESTABLISHED STATES
2989 * enter the CLOSE_WAIT state.
2991 case TCPS_SYN_RECEIVED:
2992 tp->t_starttime = ticks;
2994 case TCPS_ESTABLISHED:
2995 tcp_state_change(tp, TCPS_CLOSE_WAIT);
2999 * If still in FIN_WAIT_1 STATE FIN has not been acked so
3000 * enter the CLOSING state.
3002 case TCPS_FIN_WAIT_1:
3003 tcp_state_change(tp, TCPS_CLOSING);
3007 * In FIN_WAIT_2 state enter the TIME_WAIT state,
3008 * starting the time-wait timer, turning off the other
3011 case TCPS_FIN_WAIT_2:
3012 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
3013 KASSERT(ti_locked == TI_RLOCKED, ("%s: dodata "
3014 "TCP_FIN_WAIT_2 ti_locked: %d", __func__,
3018 INP_INFO_RUNLOCK(&V_tcbinfo);
3022 if (ti_locked == TI_RLOCKED)
3023 INP_INFO_RUNLOCK(&V_tcbinfo);
3024 ti_locked = TI_UNLOCKED;
3027 if (so->so_options & SO_DEBUG)
3028 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
3031 TCP_PROBE3(debug__input, tp, th, mtod(m, const char *));
3034 * Return any desired output.
3036 if (needoutput || (tp->t_flags & TF_ACKNOW))
3037 (void) tcp_output(tp);
3040 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d",
3041 __func__, ti_locked));
3042 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
3043 INP_WLOCK_ASSERT(tp->t_inpcb);
3045 if (tp->t_flags & TF_DELACK) {
3046 tp->t_flags &= ~TF_DELACK;
3047 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
3049 INP_WUNLOCK(tp->t_inpcb);
3054 * Generate an ACK dropping incoming segment if it occupies
3055 * sequence space, where the ACK reflects our state.
3057 * We can now skip the test for the RST flag since all
3058 * paths to this code happen after packets containing
3059 * RST have been dropped.
3061 * In the SYN-RECEIVED state, don't send an ACK unless the
3062 * segment we received passes the SYN-RECEIVED ACK test.
3063 * If it fails send a RST. This breaks the loop in the
3064 * "LAND" DoS attack, and also prevents an ACK storm
3065 * between two listening ports that have been sent forged
3066 * SYN segments, each with the source address of the other.
3068 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
3069 (SEQ_GT(tp->snd_una, th->th_ack) ||
3070 SEQ_GT(th->th_ack, tp->snd_max)) ) {
3071 rstreason = BANDLIM_RST_OPENPORT;
3075 if (so->so_options & SO_DEBUG)
3076 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3079 TCP_PROBE3(debug__input, tp, th, mtod(m, const char *));
3080 if (ti_locked == TI_RLOCKED)
3081 INP_INFO_RUNLOCK(&V_tcbinfo);
3082 ti_locked = TI_UNLOCKED;
3084 tp->t_flags |= TF_ACKNOW;
3085 (void) tcp_output(tp);
3086 INP_WUNLOCK(tp->t_inpcb);
3091 if (ti_locked == TI_RLOCKED)
3092 INP_INFO_RUNLOCK(&V_tcbinfo);
3093 ti_locked = TI_UNLOCKED;
3096 tcp_dropwithreset(m, th, tp, tlen, rstreason);
3097 INP_WUNLOCK(tp->t_inpcb);
3099 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
3103 if (ti_locked == TI_RLOCKED) {
3104 INP_INFO_RUNLOCK(&V_tcbinfo);
3105 ti_locked = TI_UNLOCKED;
3109 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
3113 * Drop space held by incoming segment and return.
3116 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
3117 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3120 TCP_PROBE3(debug__input, tp, th, mtod(m, const char *));
3122 INP_WUNLOCK(tp->t_inpcb);
3127 * Issue RST and make ACK acceptable to originator of segment.
3128 * The mbuf must still include the original packet header.
3132 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
3133 int tlen, int rstreason)
3139 struct ip6_hdr *ip6;
3143 INP_WLOCK_ASSERT(tp->t_inpcb);
3146 /* Don't bother if destination was broadcast/multicast. */
3147 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
3150 if (mtod(m, struct ip *)->ip_v == 6) {
3151 ip6 = mtod(m, struct ip6_hdr *);
3152 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3153 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3155 /* IPv6 anycast check is done at tcp6_input() */
3158 #if defined(INET) && defined(INET6)
3163 ip = mtod(m, struct ip *);
3164 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3165 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3166 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3167 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3172 /* Perform bandwidth limiting. */
3173 if (badport_bandlim(rstreason) < 0)
3176 /* tcp_respond consumes the mbuf chain. */
3177 if (th->th_flags & TH_ACK) {
3178 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0,
3179 th->th_ack, TH_RST);
3181 if (th->th_flags & TH_SYN)
3183 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
3184 (tcp_seq)0, TH_RST|TH_ACK);
3192 * Parse TCP options and place in tcpopt.
3195 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags)
3200 for (; cnt > 0; cnt -= optlen, cp += optlen) {
3202 if (opt == TCPOPT_EOL)
3204 if (opt == TCPOPT_NOP)
3210 if (optlen < 2 || optlen > cnt)
3215 if (optlen != TCPOLEN_MAXSEG)
3217 if (!(flags & TO_SYN))
3219 to->to_flags |= TOF_MSS;
3220 bcopy((char *)cp + 2,
3221 (char *)&to->to_mss, sizeof(to->to_mss));
3222 to->to_mss = ntohs(to->to_mss);
3225 if (optlen != TCPOLEN_WINDOW)
3227 if (!(flags & TO_SYN))
3229 to->to_flags |= TOF_SCALE;
3230 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT);
3232 case TCPOPT_TIMESTAMP:
3233 if (optlen != TCPOLEN_TIMESTAMP)
3235 to->to_flags |= TOF_TS;
3236 bcopy((char *)cp + 2,
3237 (char *)&to->to_tsval, sizeof(to->to_tsval));
3238 to->to_tsval = ntohl(to->to_tsval);
3239 bcopy((char *)cp + 6,
3240 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
3241 to->to_tsecr = ntohl(to->to_tsecr);
3243 #ifdef TCP_SIGNATURE
3245 * XXX In order to reply to a host which has set the
3246 * TCP_SIGNATURE option in its initial SYN, we have to
3247 * record the fact that the option was observed here
3248 * for the syncache code to perform the correct response.
3250 case TCPOPT_SIGNATURE:
3251 if (optlen != TCPOLEN_SIGNATURE)
3253 to->to_flags |= TOF_SIGNATURE;
3254 to->to_signature = cp + 2;
3257 case TCPOPT_SACK_PERMITTED:
3258 if (optlen != TCPOLEN_SACK_PERMITTED)
3260 if (!(flags & TO_SYN))
3264 to->to_flags |= TOF_SACKPERM;
3267 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
3271 to->to_flags |= TOF_SACK;
3272 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
3273 to->to_sacks = cp + 2;
3274 TCPSTAT_INC(tcps_sack_rcv_blocks);
3283 * Pull out of band byte out of a segment so
3284 * it doesn't appear in the user's data queue.
3285 * It is still reflected in the segment length for
3286 * sequencing purposes.
3289 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m,
3292 int cnt = off + th->th_urp - 1;
3295 if (m->m_len > cnt) {
3296 char *cp = mtod(m, caddr_t) + cnt;
3297 struct tcpcb *tp = sototcpcb(so);
3299 INP_WLOCK_ASSERT(tp->t_inpcb);
3302 tp->t_oobflags |= TCPOOB_HAVEDATA;
3303 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3305 if (m->m_flags & M_PKTHDR)
3314 panic("tcp_pulloutofband");
3318 * Collect new round-trip time estimate
3319 * and update averages and current timeout.
3322 tcp_xmit_timer(struct tcpcb *tp, int rtt)
3326 INP_WLOCK_ASSERT(tp->t_inpcb);
3328 TCPSTAT_INC(tcps_rttupdated);
3330 if (tp->t_srtt != 0) {
3332 * srtt is stored as fixed point with 5 bits after the
3333 * binary point (i.e., scaled by 8). The following magic
3334 * is equivalent to the smoothing algorithm in rfc793 with
3335 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3336 * point). Adjust rtt to origin 0.
3338 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
3339 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
3341 if ((tp->t_srtt += delta) <= 0)
3345 * We accumulate a smoothed rtt variance (actually, a
3346 * smoothed mean difference), then set the retransmit
3347 * timer to smoothed rtt + 4 times the smoothed variance.
3348 * rttvar is stored as fixed point with 4 bits after the
3349 * binary point (scaled by 16). The following is
3350 * equivalent to rfc793 smoothing with an alpha of .75
3351 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
3352 * rfc793's wired-in beta.
3356 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
3357 if ((tp->t_rttvar += delta) <= 0)
3359 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
3360 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3363 * No rtt measurement yet - use the unsmoothed rtt.
3364 * Set the variance to half the rtt (so our first
3365 * retransmit happens at 3*rtt).
3367 tp->t_srtt = rtt << TCP_RTT_SHIFT;
3368 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
3369 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3375 * the retransmit should happen at rtt + 4 * rttvar.
3376 * Because of the way we do the smoothing, srtt and rttvar
3377 * will each average +1/2 tick of bias. When we compute
3378 * the retransmit timer, we want 1/2 tick of rounding and
3379 * 1 extra tick because of +-1/2 tick uncertainty in the
3380 * firing of the timer. The bias will give us exactly the
3381 * 1.5 tick we need. But, because the bias is
3382 * statistical, we have to test that we don't drop below
3383 * the minimum feasible timer (which is 2 ticks).
3385 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3386 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3389 * We received an ack for a packet that wasn't retransmitted;
3390 * it is probably safe to discard any error indications we've
3391 * received recently. This isn't quite right, but close enough
3392 * for now (a route might have failed after we sent a segment,
3393 * and the return path might not be symmetrical).
3395 tp->t_softerror = 0;
3399 * Determine a reasonable value for maxseg size.
3400 * If the route is known, check route for mtu.
3401 * If none, use an mss that can be handled on the outgoing interface
3402 * without forcing IP to fragment. If no route is found, route has no mtu,
3403 * or the destination isn't local, use a default, hopefully conservative
3404 * size (usually 512 or the default IP max size, but no more than the mtu
3405 * of the interface), as we can't discover anything about intervening
3406 * gateways or networks. We also initialize the congestion/slow start
3407 * window to be a single segment if the destination isn't local.
3408 * While looking at the routing entry, we also initialize other path-dependent
3409 * parameters from pre-set or cached values in the routing entry.
3411 * Also take into account the space needed for options that we
3412 * send regularly. Make maxseg shorter by that amount to assure
3413 * that we can send maxseg amount of data even when the options
3414 * are present. Store the upper limit of the length of options plus
3417 * NOTE that this routine is only called when we process an incoming
3418 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS
3419 * settings are handled in tcp_mssopt().
3422 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer,
3423 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap)
3427 struct inpcb *inp = tp->t_inpcb;
3428 struct hc_metrics_lite metrics;
3431 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
3432 size_t min_protoh = isipv6 ?
3433 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
3434 sizeof (struct tcpiphdr);
3436 const size_t min_protoh = sizeof(struct tcpiphdr);
3439 INP_WLOCK_ASSERT(tp->t_inpcb);
3441 if (mtuoffer != -1) {
3442 KASSERT(offer == -1, ("%s: conflict", __func__));
3443 offer = mtuoffer - min_protoh;
3450 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap);
3451 tp->t_maxopd = tp->t_maxseg = V_tcp_v6mssdflt;
3454 #if defined(INET) && defined(INET6)
3459 maxmtu = tcp_maxmtu(&inp->inp_inc, cap);
3460 tp->t_maxopd = tp->t_maxseg = V_tcp_mssdflt;
3465 * No route to sender, stay with default mss and return.
3469 * In case we return early we need to initialize metrics
3470 * to a defined state as tcp_hc_get() would do for us
3471 * if there was no cache hit.
3473 if (metricptr != NULL)
3474 bzero(metricptr, sizeof(struct hc_metrics_lite));
3478 /* What have we got? */
3482 * Offer == 0 means that there was no MSS on the SYN
3483 * segment, in this case we use tcp_mssdflt as
3484 * already assigned to t_maxopd above.
3486 offer = tp->t_maxopd;
3491 * Offer == -1 means that we didn't receive SYN yet.
3497 * Prevent DoS attack with too small MSS. Round up
3498 * to at least minmss.
3500 offer = max(offer, V_tcp_minmss);
3504 * rmx information is now retrieved from tcp_hostcache.
3506 tcp_hc_get(&inp->inp_inc, &metrics);
3507 if (metricptr != NULL)
3508 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite));
3511 * If there's a discovered mtu in tcp hostcache, use it.
3512 * Else, use the link mtu.
3514 if (metrics.rmx_mtu)
3515 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
3519 mss = maxmtu - min_protoh;
3520 if (!V_path_mtu_discovery &&
3521 !in6_localaddr(&inp->in6p_faddr))
3522 mss = min(mss, V_tcp_v6mssdflt);
3525 #if defined(INET) && defined(INET6)
3530 mss = maxmtu - min_protoh;
3531 if (!V_path_mtu_discovery &&
3532 !in_localaddr(inp->inp_faddr))
3533 mss = min(mss, V_tcp_mssdflt);
3537 * XXX - The above conditional (mss = maxmtu - min_protoh)
3538 * probably violates the TCP spec.
3539 * The problem is that, since we don't know the
3540 * other end's MSS, we are supposed to use a conservative
3541 * default. But, if we do that, then MTU discovery will
3542 * never actually take place, because the conservative
3543 * default is much less than the MTUs typically seen
3544 * on the Internet today. For the moment, we'll sweep
3545 * this under the carpet.
3547 * The conservative default might not actually be a problem
3548 * if the only case this occurs is when sending an initial
3549 * SYN with options and data to a host we've never talked
3550 * to before. Then, they will reply with an MSS value which
3551 * will get recorded and the new parameters should get
3552 * recomputed. For Further Study.
3555 mss = min(mss, offer);
3558 * Sanity check: make sure that maxopd will be large
3559 * enough to allow some data on segments even if the
3560 * all the option space is used (40bytes). Otherwise
3561 * funny things may happen in tcp_output.
3566 * maxopd stores the maximum length of data AND options
3567 * in a segment; maxseg is the amount of data in a normal
3568 * segment. We need to store this value (maxopd) apart
3569 * from maxseg, because now every segment carries options
3570 * and thus we normally have somewhat less data in segments.
3575 * origoffer==-1 indicates that no segments were received yet.
3576 * In this case we just guess.
3578 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
3580 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
3581 mss -= TCPOLEN_TSTAMP_APPA;
3587 tcp_mss(struct tcpcb *tp, int offer)
3593 struct hc_metrics_lite metrics;
3594 struct tcp_ifcap cap;
3596 KASSERT(tp != NULL, ("%s: tp == NULL", __func__));
3598 bzero(&cap, sizeof(cap));
3599 tcp_mss_update(tp, offer, -1, &metrics, &cap);
3605 * If there's a pipesize, change the socket buffer to that size,
3606 * don't change if sb_hiwat is different than default (then it
3607 * has been changed on purpose with setsockopt).
3608 * Make the socket buffers an integral number of mss units;
3609 * if the mss is larger than the socket buffer, decrease the mss.
3611 so = inp->inp_socket;
3612 SOCKBUF_LOCK(&so->so_snd);
3613 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe)
3614 bufsize = metrics.rmx_sendpipe;
3616 bufsize = so->so_snd.sb_hiwat;
3620 bufsize = roundup(bufsize, mss);
3621 if (bufsize > sb_max)
3623 if (bufsize > so->so_snd.sb_hiwat)
3624 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
3626 SOCKBUF_UNLOCK(&so->so_snd);
3629 SOCKBUF_LOCK(&so->so_rcv);
3630 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe)
3631 bufsize = metrics.rmx_recvpipe;
3633 bufsize = so->so_rcv.sb_hiwat;
3634 if (bufsize > mss) {
3635 bufsize = roundup(bufsize, mss);
3636 if (bufsize > sb_max)
3638 if (bufsize > so->so_rcv.sb_hiwat)
3639 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
3641 SOCKBUF_UNLOCK(&so->so_rcv);
3643 /* Check the interface for TSO capabilities. */
3644 if (cap.ifcap & CSUM_TSO) {
3645 tp->t_flags |= TF_TSO;
3646 tp->t_tsomax = cap.tsomax;
3647 tp->t_tsomaxsegcount = cap.tsomaxsegcount;
3648 tp->t_tsomaxsegsize = cap.tsomaxsegsize;
3653 * Determine the MSS option to send on an outgoing SYN.
3656 tcp_mssopt(struct in_conninfo *inc)
3663 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3666 if (inc->inc_flags & INC_ISIPV6) {
3667 mss = V_tcp_v6mssdflt;
3668 maxmtu = tcp_maxmtu6(inc, NULL);
3669 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3672 #if defined(INET) && defined(INET6)
3677 mss = V_tcp_mssdflt;
3678 maxmtu = tcp_maxmtu(inc, NULL);
3679 min_protoh = sizeof(struct tcpiphdr);
3682 #if defined(INET6) || defined(INET)
3683 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3686 if (maxmtu && thcmtu)
3687 mss = min(maxmtu, thcmtu) - min_protoh;
3688 else if (maxmtu || thcmtu)
3689 mss = max(maxmtu, thcmtu) - min_protoh;
3696 * On a partial ack arrives, force the retransmission of the
3697 * next unacknowledged segment. Do not clear tp->t_dupacks.
3698 * By setting snd_nxt to ti_ack, this forces retransmission timer to
3702 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th)
3704 tcp_seq onxt = tp->snd_nxt;
3705 u_long ocwnd = tp->snd_cwnd;
3707 INP_WLOCK_ASSERT(tp->t_inpcb);
3709 tcp_timer_activate(tp, TT_REXMT, 0);
3711 tp->snd_nxt = th->th_ack;
3713 * Set snd_cwnd to one segment beyond acknowledged offset.
3714 * (tp->snd_una has not yet been updated when this function is called.)
3716 tp->snd_cwnd = tp->t_maxseg + BYTES_THIS_ACK(tp, th);
3717 tp->t_flags |= TF_ACKNOW;
3718 (void) tcp_output(tp);
3719 tp->snd_cwnd = ocwnd;
3720 if (SEQ_GT(onxt, tp->snd_nxt))
3723 * Partial window deflation. Relies on fact that tp->snd_una
3726 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th))
3727 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th);
3730 tp->snd_cwnd += tp->t_maxseg;