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>
107 #include <netinet/tcp_syncache.h>
109 #include <netinet/tcp_debug.h>
110 #endif /* TCPDEBUG */
112 #include <netinet/tcp_offload.h>
116 #include <netipsec/ipsec.h>
117 #include <netipsec/ipsec6.h>
120 #include <machine/in_cksum.h>
122 #include <security/mac/mac_framework.h>
124 const int tcprexmtthresh = 3;
126 int tcp_log_in_vain = 0;
127 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
129 "Log all incoming TCP segments to closed ports");
131 VNET_DEFINE(int, blackhole) = 0;
132 #define V_blackhole VNET(blackhole)
133 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW,
134 &VNET_NAME(blackhole), 0,
135 "Do not send RST on segments to closed ports");
137 VNET_DEFINE(int, tcp_delack_enabled) = 1;
138 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW,
139 &VNET_NAME(tcp_delack_enabled), 0,
140 "Delay ACK to try and piggyback it onto a data packet");
142 VNET_DEFINE(int, drop_synfin) = 0;
143 #define V_drop_synfin VNET(drop_synfin)
144 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW,
145 &VNET_NAME(drop_synfin), 0,
146 "Drop TCP packets with SYN+FIN set");
148 VNET_DEFINE(int, tcp_do_rfc3042) = 1;
149 #define V_tcp_do_rfc3042 VNET(tcp_do_rfc3042)
150 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_RW,
151 &VNET_NAME(tcp_do_rfc3042), 0,
152 "Enable RFC 3042 (Limited Transmit)");
154 VNET_DEFINE(int, tcp_do_rfc3390) = 1;
155 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW,
156 &VNET_NAME(tcp_do_rfc3390), 0,
157 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
159 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, experimental, CTLFLAG_RW, 0,
160 "Experimental TCP extensions");
162 VNET_DEFINE(int, tcp_do_initcwnd10) = 1;
163 SYSCTL_VNET_INT(_net_inet_tcp_experimental, OID_AUTO, initcwnd10, CTLFLAG_RW,
164 &VNET_NAME(tcp_do_initcwnd10), 0,
165 "Enable RFC 6928 (Increasing initial CWND to 10)");
167 VNET_DEFINE(int, tcp_do_rfc3465) = 1;
168 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3465, 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_VNET_INT(_net_inet_tcp, OID_AUTO, abc_l_var, 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_VNET_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_RW,
181 &VNET_NAME(tcp_do_ecn), 0,
184 VNET_DEFINE(int, tcp_ecn_maxretries) = 1;
185 SYSCTL_VNET_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_RW,
186 &VNET_NAME(tcp_ecn_maxretries), 0,
187 "Max retries before giving up on ECN");
189 VNET_DEFINE(int, tcp_insecure_rst) = 0;
190 #define V_tcp_insecure_rst VNET(tcp_insecure_rst)
191 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_RW,
192 &VNET_NAME(tcp_insecure_rst), 0,
193 "Follow the old (insecure) criteria for accepting RST packets");
195 VNET_DEFINE(int, tcp_recvspace) = 1024*64;
196 #define V_tcp_recvspace VNET(tcp_recvspace)
197 SYSCTL_VNET_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
198 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size");
200 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1;
201 #define V_tcp_do_autorcvbuf VNET(tcp_do_autorcvbuf)
202 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_RW,
203 &VNET_NAME(tcp_do_autorcvbuf), 0,
204 "Enable automatic receive buffer sizing");
206 VNET_DEFINE(int, tcp_autorcvbuf_inc) = 16*1024;
207 #define V_tcp_autorcvbuf_inc VNET(tcp_autorcvbuf_inc)
208 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_RW,
209 &VNET_NAME(tcp_autorcvbuf_inc), 0,
210 "Incrementor step size of automatic receive buffer");
212 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024;
213 #define V_tcp_autorcvbuf_max VNET(tcp_autorcvbuf_max)
214 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_RW,
215 &VNET_NAME(tcp_autorcvbuf_max), 0,
216 "Max size of automatic receive buffer");
218 VNET_DEFINE(struct inpcbhead, tcb);
219 #define tcb6 tcb /* for KAME src sync over BSD*'s */
220 VNET_DEFINE(struct inpcbinfo, tcbinfo);
222 static void tcp_dooptions(struct tcpopt *, u_char *, int, int);
223 static void tcp_do_segment(struct mbuf *, struct tcphdr *,
224 struct socket *, struct tcpcb *, int, int, uint8_t,
226 static void tcp_dropwithreset(struct mbuf *, struct tcphdr *,
227 struct tcpcb *, int, int);
228 static void tcp_pulloutofband(struct socket *,
229 struct tcphdr *, struct mbuf *, int);
230 static void tcp_xmit_timer(struct tcpcb *, int);
231 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *);
232 static void inline tcp_fields_to_host(struct tcphdr *);
234 static void inline tcp_fields_to_net(struct tcphdr *);
235 static int inline tcp_signature_verify_input(struct mbuf *, int, int,
236 int, struct tcpopt *, struct tcphdr *, u_int);
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.
365 * If a SYN or SYN/ACK was lost and retransmitted, we have to
366 * reduce the initial CWND to one segment as congestion is likely
367 * requiring us to be cautious.
369 if (tp->snd_cwnd == 1)
370 tp->snd_cwnd = tp->t_maxseg; /* SYN(-ACK) lost */
371 else if (V_tcp_do_initcwnd10)
372 tp->snd_cwnd = min(10 * tp->t_maxseg,
373 max(2 * tp->t_maxseg, 14600));
374 else if (V_tcp_do_rfc3390)
375 tp->snd_cwnd = min(4 * tp->t_maxseg,
376 max(2 * tp->t_maxseg, 4380));
378 /* Per RFC5681 Section 3.1 */
379 if (tp->t_maxseg > 2190)
380 tp->snd_cwnd = 2 * tp->t_maxseg;
381 else if (tp->t_maxseg > 1095)
382 tp->snd_cwnd = 3 * tp->t_maxseg;
384 tp->snd_cwnd = 4 * tp->t_maxseg;
387 if (CC_ALGO(tp)->conn_init != NULL)
388 CC_ALGO(tp)->conn_init(tp->ccv);
392 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type)
394 INP_WLOCK_ASSERT(tp->t_inpcb);
398 if (!IN_FASTRECOVERY(tp->t_flags)) {
399 tp->snd_recover = tp->snd_max;
400 if (tp->t_flags & TF_ECN_PERMIT)
401 tp->t_flags |= TF_ECN_SND_CWR;
405 if (!IN_CONGRECOVERY(tp->t_flags)) {
406 TCPSTAT_INC(tcps_ecn_rcwnd);
407 tp->snd_recover = tp->snd_max;
408 if (tp->t_flags & TF_ECN_PERMIT)
409 tp->t_flags |= TF_ECN_SND_CWR;
414 tp->t_bytes_acked = 0;
415 EXIT_RECOVERY(tp->t_flags);
416 tp->snd_ssthresh = max(2, min(tp->snd_wnd, tp->snd_cwnd) / 2 /
417 tp->t_maxseg) * tp->t_maxseg;
418 tp->snd_cwnd = tp->t_maxseg;
421 TCPSTAT_INC(tcps_sndrexmitbad);
422 /* RTO was unnecessary, so reset everything. */
423 tp->snd_cwnd = tp->snd_cwnd_prev;
424 tp->snd_ssthresh = tp->snd_ssthresh_prev;
425 tp->snd_recover = tp->snd_recover_prev;
426 if (tp->t_flags & TF_WASFRECOVERY)
427 ENTER_FASTRECOVERY(tp->t_flags);
428 if (tp->t_flags & TF_WASCRECOVERY)
429 ENTER_CONGRECOVERY(tp->t_flags);
430 tp->snd_nxt = tp->snd_max;
431 tp->t_flags &= ~TF_PREVVALID;
436 if (CC_ALGO(tp)->cong_signal != NULL) {
438 tp->ccv->curack = th->th_ack;
439 CC_ALGO(tp)->cong_signal(tp->ccv, type);
444 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th)
446 INP_WLOCK_ASSERT(tp->t_inpcb);
448 /* XXXLAS: KASSERT that we're in recovery? */
450 if (CC_ALGO(tp)->post_recovery != NULL) {
451 tp->ccv->curack = th->th_ack;
452 CC_ALGO(tp)->post_recovery(tp->ccv);
454 /* XXXLAS: EXIT_RECOVERY ? */
455 tp->t_bytes_acked = 0;
459 tcp_fields_to_host(struct tcphdr *th)
462 th->th_seq = ntohl(th->th_seq);
463 th->th_ack = ntohl(th->th_ack);
464 th->th_win = ntohs(th->th_win);
465 th->th_urp = ntohs(th->th_urp);
470 tcp_fields_to_net(struct tcphdr *th)
473 th->th_seq = htonl(th->th_seq);
474 th->th_ack = htonl(th->th_ack);
475 th->th_win = htons(th->th_win);
476 th->th_urp = htons(th->th_urp);
480 tcp_signature_verify_input(struct mbuf *m, int off0, int tlen, int optlen,
481 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag)
485 tcp_fields_to_net(th);
486 ret = tcp_signature_verify(m, off0, tlen, optlen, to, th, tcpbflag);
487 tcp_fields_to_host(th);
492 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
494 #define ND6_HINT(tp) \
496 if ((tp) && (tp)->t_inpcb && \
497 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \
498 nd6_nud_hint(NULL, NULL, 0); \
505 * Indicate whether this ack should be delayed. We can delay the ack if
506 * following conditions are met:
507 * - There is no delayed ack timer in progress.
508 * - Our last ack wasn't a 0-sized window. We never want to delay
509 * the ack that opens up a 0-sized window.
510 * - LRO wasn't used for this segment. We make sure by checking that the
511 * segment size is not larger than the MSS.
512 * - Delayed acks are enabled or this is a half-synchronized T/TCP
515 #define DELAY_ACK(tp, tlen) \
516 ((!tcp_timer_active(tp, TT_DELACK) && \
517 (tp->t_flags & TF_RXWIN0SENT) == 0) && \
518 (tlen <= tp->t_maxopd) && \
519 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
522 * TCP input handling is split into multiple parts:
523 * tcp6_input is a thin wrapper around tcp_input for the extended
524 * ip6_protox[] call format in ip6_input
525 * tcp_input handles primary segment validation, inpcb lookup and
526 * SYN processing on listen sockets
527 * tcp_do_segment processes the ACK and text of the segment for
528 * establishing, established and closing connections
532 tcp6_input(struct mbuf **mp, int *offp, int proto)
534 struct mbuf *m = *mp;
535 struct in6_ifaddr *ia6;
537 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
540 * draft-itojun-ipv6-tcp-to-anycast
541 * better place to put this in?
543 ia6 = ip6_getdstifaddr(m);
544 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
547 ifa_free(&ia6->ia_ifa);
548 ip6 = mtod(m, struct ip6_hdr *);
549 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
550 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
554 ifa_free(&ia6->ia_ifa);
562 tcp_input(struct mbuf *m, int off0)
564 struct tcphdr *th = NULL;
565 struct ip *ip = NULL;
566 struct inpcb *inp = NULL;
567 struct tcpcb *tp = NULL;
568 struct socket *so = NULL;
577 int rstreason = 0; /* For badport_bandlim accounting purposes */
579 uint8_t sig_checked = 0;
582 struct m_tag *fwd_tag = NULL;
584 struct ip6_hdr *ip6 = NULL;
587 const void *ip6 = NULL;
589 struct tcpopt to; /* options in this segment */
590 char *s = NULL; /* address and port logging */
592 #define TI_UNLOCKED 1
597 * The size of tcp_saveipgen must be the size of the max ip header,
600 u_char tcp_saveipgen[IP6_HDR_LEN];
601 struct tcphdr tcp_savetcp;
606 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
610 TCPSTAT_INC(tcps_rcvtotal);
614 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */
616 if (m->m_len < (sizeof(*ip6) + sizeof(*th))) {
617 m = m_pullup(m, sizeof(*ip6) + sizeof(*th));
619 TCPSTAT_INC(tcps_rcvshort);
624 ip6 = mtod(m, struct ip6_hdr *);
625 th = (struct tcphdr *)((caddr_t)ip6 + off0);
626 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
627 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) {
628 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
629 th->th_sum = m->m_pkthdr.csum_data;
631 th->th_sum = in6_cksum_pseudo(ip6, tlen,
632 IPPROTO_TCP, m->m_pkthdr.csum_data);
633 th->th_sum ^= 0xffff;
635 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen);
637 TCPSTAT_INC(tcps_rcvbadsum);
642 * Be proactive about unspecified IPv6 address in source.
643 * As we use all-zero to indicate unbounded/unconnected pcb,
644 * unspecified IPv6 address can be used to confuse us.
646 * Note that packets with unspecified IPv6 destination is
647 * already dropped in ip6_input.
649 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
655 #if defined(INET) && defined(INET6)
661 * Get IP and TCP header together in first mbuf.
662 * Note: IP leaves IP header in first mbuf.
664 if (off0 > sizeof (struct ip)) {
666 off0 = sizeof(struct ip);
668 if (m->m_len < sizeof (struct tcpiphdr)) {
669 if ((m = m_pullup(m, sizeof (struct tcpiphdr)))
671 TCPSTAT_INC(tcps_rcvshort);
675 ip = mtod(m, struct ip *);
676 th = (struct tcphdr *)((caddr_t)ip + off0);
677 tlen = ntohs(ip->ip_len) - off0;
679 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
680 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
681 th->th_sum = m->m_pkthdr.csum_data;
683 th->th_sum = in_pseudo(ip->ip_src.s_addr,
685 htonl(m->m_pkthdr.csum_data + tlen +
687 th->th_sum ^= 0xffff;
689 struct ipovly *ipov = (struct ipovly *)ip;
692 * Checksum extended TCP header and data.
695 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
696 ipov->ih_len = htons(tlen);
697 th->th_sum = in_cksum(m, len);
698 /* Reset length for SDT probes. */
699 ip->ip_len = htons(tlen + off0);
703 TCPSTAT_INC(tcps_rcvbadsum);
706 /* Re-initialization for later version check */
707 ip->ip_v = IPVERSION;
713 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
715 #if defined(INET) && defined(INET6)
723 * Check that TCP offset makes sense,
724 * pull out TCP options and adjust length. XXX
726 off = th->th_off << 2;
727 if (off < sizeof (struct tcphdr) || off > tlen) {
728 TCPSTAT_INC(tcps_rcvbadoff);
731 tlen -= off; /* tlen is used instead of ti->ti_len */
732 if (off > sizeof (struct tcphdr)) {
735 IP6_EXTHDR_CHECK(m, off0, off, );
736 ip6 = mtod(m, struct ip6_hdr *);
737 th = (struct tcphdr *)((caddr_t)ip6 + off0);
740 #if defined(INET) && defined(INET6)
745 if (m->m_len < sizeof(struct ip) + off) {
746 if ((m = m_pullup(m, sizeof (struct ip) + off))
748 TCPSTAT_INC(tcps_rcvshort);
751 ip = mtod(m, struct ip *);
752 th = (struct tcphdr *)((caddr_t)ip + off0);
756 optlen = off - sizeof (struct tcphdr);
757 optp = (u_char *)(th + 1);
759 thflags = th->th_flags;
762 * Convert TCP protocol specific fields to host format.
764 tcp_fields_to_host(th);
767 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options.
769 drop_hdrlen = off0 + off;
772 * Locate pcb for segment; if we're likely to add or remove a
773 * connection then first acquire pcbinfo lock. There are two cases
774 * where we might discover later we need a write lock despite the
775 * flags: ACKs moving a connection out of the syncache, and ACKs for
776 * a connection in TIMEWAIT.
778 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0) {
779 INP_INFO_WLOCK(&V_tcbinfo);
780 ti_locked = TI_WLOCKED;
782 ti_locked = TI_UNLOCKED;
785 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
789 (isipv6 && (m->m_flags & M_IP6_NEXTHOP))
791 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP))
794 #if defined(INET) && !defined(INET6)
795 (m->m_flags & M_IP_NEXTHOP)
798 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
802 if (ti_locked == TI_WLOCKED) {
803 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
805 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
809 if (isipv6 && fwd_tag != NULL) {
810 struct sockaddr_in6 *next_hop6;
812 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1);
814 * Transparently forwarded. Pretend to be the destination.
815 * Already got one like this?
817 inp = in6_pcblookup_mbuf(&V_tcbinfo,
818 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport,
819 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m);
822 * It's new. Try to find the ambushing socket.
823 * Because we've rewritten the destination address,
824 * any hardware-generated hash is ignored.
826 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src,
827 th->th_sport, &next_hop6->sin6_addr,
828 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) :
829 th->th_dport, INPLOOKUP_WILDCARD |
830 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
833 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src,
834 th->th_sport, &ip6->ip6_dst, th->th_dport,
835 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
836 m->m_pkthdr.rcvif, m);
839 #if defined(INET6) && defined(INET)
843 if (fwd_tag != NULL) {
844 struct sockaddr_in *next_hop;
846 next_hop = (struct sockaddr_in *)(fwd_tag+1);
848 * Transparently forwarded. Pretend to be the destination.
849 * already got one like this?
851 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport,
852 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB,
853 m->m_pkthdr.rcvif, m);
856 * It's new. Try to find the ambushing socket.
857 * Because we've rewritten the destination address,
858 * any hardware-generated hash is ignored.
860 inp = in_pcblookup(&V_tcbinfo, ip->ip_src,
861 th->th_sport, next_hop->sin_addr,
862 next_hop->sin_port ? ntohs(next_hop->sin_port) :
863 th->th_dport, INPLOOKUP_WILDCARD |
864 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
867 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src,
868 th->th_sport, ip->ip_dst, th->th_dport,
869 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
870 m->m_pkthdr.rcvif, m);
874 * If the INPCB does not exist then all data in the incoming
875 * segment is discarded and an appropriate RST is sent back.
876 * XXX MRT Send RST using which routing table?
880 * Log communication attempts to ports that are not
883 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) ||
884 tcp_log_in_vain == 2) {
885 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6)))
886 log(LOG_INFO, "%s; %s: Connection attempt "
887 "to closed port\n", s, __func__);
890 * When blackholing do not respond with a RST but
891 * completely ignore the segment and drop it.
893 if ((V_blackhole == 1 && (thflags & TH_SYN)) ||
897 rstreason = BANDLIM_RST_CLOSEDPORT;
900 INP_WLOCK_ASSERT(inp);
901 if (!(inp->inp_flags & INP_HW_FLOWID)
902 && (m->m_flags & M_FLOWID)
903 && ((inp->inp_socket == NULL)
904 || !(inp->inp_socket->so_options & SO_ACCEPTCONN))) {
905 inp->inp_flags |= INP_HW_FLOWID;
906 inp->inp_flags &= ~INP_SW_FLOWID;
907 inp->inp_flowid = m->m_pkthdr.flowid;
911 if (isipv6 && ipsec6_in_reject(m, inp)) {
912 IPSEC6STAT_INC(ips_in_polvio);
916 if (ipsec4_in_reject(m, inp) != 0) {
917 IPSECSTAT_INC(ips_in_polvio);
923 * Check the minimum TTL for socket.
925 if (inp->inp_ip_minttl != 0) {
927 if (isipv6 && inp->inp_ip_minttl > ip6->ip6_hlim)
931 if (inp->inp_ip_minttl > ip->ip_ttl)
936 * A previous connection in TIMEWAIT state is supposed to catch stray
937 * or duplicate segments arriving late. If this segment was a
938 * legitimate new connection attempt, the old INPCB gets removed and
939 * we can try again to find a listening socket.
941 * At this point, due to earlier optimism, we may hold only an inpcb
942 * lock, and not the inpcbinfo write lock. If so, we need to try to
943 * acquire it, or if that fails, acquire a reference on the inpcb,
944 * drop all locks, acquire a global write lock, and then re-acquire
945 * the inpcb lock. We may at that point discover that another thread
946 * has tried to free the inpcb, in which case we need to loop back
947 * and try to find a new inpcb to deliver to.
949 * XXXRW: It may be time to rethink timewait locking.
952 if (inp->inp_flags & INP_TIMEWAIT) {
953 if (ti_locked == TI_UNLOCKED) {
954 if (INP_INFO_TRY_WLOCK(&V_tcbinfo) == 0) {
957 INP_INFO_WLOCK(&V_tcbinfo);
958 ti_locked = TI_WLOCKED;
960 if (in_pcbrele_wlocked(inp)) {
965 ti_locked = TI_WLOCKED;
967 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
969 if (thflags & TH_SYN)
970 tcp_dooptions(&to, optp, optlen, TO_SYN);
972 * NB: tcp_twcheck unlocks the INP and frees the mbuf.
974 if (tcp_twcheck(inp, &to, th, m, tlen))
976 INP_INFO_WUNLOCK(&V_tcbinfo);
980 * The TCPCB may no longer exist if the connection is winding
981 * down or it is in the CLOSED state. Either way we drop the
982 * segment and send an appropriate response.
985 if (tp == NULL || tp->t_state == TCPS_CLOSED) {
986 rstreason = BANDLIM_RST_CLOSEDPORT;
991 if (tp->t_flags & TF_TOE) {
992 tcp_offload_input(tp, m);
993 m = NULL; /* consumed by the TOE driver */
999 * We've identified a valid inpcb, but it could be that we need an
1000 * inpcbinfo write lock but don't hold it. In this case, attempt to
1001 * acquire using the same strategy as the TIMEWAIT case above. If we
1002 * relock, we have to jump back to 'relocked' as the connection might
1003 * now be in TIMEWAIT.
1006 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0)
1007 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1009 if (tp->t_state != TCPS_ESTABLISHED) {
1010 if (ti_locked == TI_UNLOCKED) {
1011 if (INP_INFO_TRY_WLOCK(&V_tcbinfo) == 0) {
1014 INP_INFO_WLOCK(&V_tcbinfo);
1015 ti_locked = TI_WLOCKED;
1017 if (in_pcbrele_wlocked(inp)) {
1023 ti_locked = TI_WLOCKED;
1025 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1029 INP_WLOCK_ASSERT(inp);
1030 if (mac_inpcb_check_deliver(inp, m))
1033 so = inp->inp_socket;
1034 KASSERT(so != NULL, ("%s: so == NULL", __func__));
1036 if (so->so_options & SO_DEBUG) {
1037 ostate = tp->t_state;
1040 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
1043 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
1046 #endif /* TCPDEBUG */
1048 * When the socket is accepting connections (the INPCB is in LISTEN
1049 * state) we look into the SYN cache if this is a new connection
1050 * attempt or the completion of a previous one. Because listen
1051 * sockets are never in TCPS_ESTABLISHED, the V_tcbinfo lock will be
1052 * held in this case.
1054 if (so->so_options & SO_ACCEPTCONN) {
1055 struct in_conninfo inc;
1057 KASSERT(tp->t_state == TCPS_LISTEN, ("%s: so accepting but "
1058 "tp not listening", __func__));
1059 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1061 bzero(&inc, sizeof(inc));
1064 inc.inc_flags |= INC_ISIPV6;
1065 inc.inc6_faddr = ip6->ip6_src;
1066 inc.inc6_laddr = ip6->ip6_dst;
1070 inc.inc_faddr = ip->ip_src;
1071 inc.inc_laddr = ip->ip_dst;
1073 inc.inc_fport = th->th_sport;
1074 inc.inc_lport = th->th_dport;
1075 inc.inc_fibnum = so->so_fibnum;
1078 * Check for an existing connection attempt in syncache if
1079 * the flag is only ACK. A successful lookup creates a new
1080 * socket appended to the listen queue in SYN_RECEIVED state.
1082 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
1084 * Parse the TCP options here because
1085 * syncookies need access to the reflected
1088 tcp_dooptions(&to, optp, optlen, 0);
1090 * NB: syncache_expand() doesn't unlock
1091 * inp and tcpinfo locks.
1093 if (!syncache_expand(&inc, &to, th, &so, m)) {
1095 * No syncache entry or ACK was not
1096 * for our SYN/ACK. Send a RST.
1097 * NB: syncache did its own logging
1098 * of the failure cause.
1100 rstreason = BANDLIM_RST_OPENPORT;
1105 * We completed the 3-way handshake
1106 * but could not allocate a socket
1107 * either due to memory shortage,
1108 * listen queue length limits or
1109 * global socket limits. Send RST
1110 * or wait and have the remote end
1111 * retransmit the ACK for another
1114 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1115 log(LOG_DEBUG, "%s; %s: Listen socket: "
1116 "Socket allocation failed due to "
1117 "limits or memory shortage, %s\n",
1119 V_tcp_sc_rst_sock_fail ?
1120 "sending RST" : "try again");
1121 if (V_tcp_sc_rst_sock_fail) {
1122 rstreason = BANDLIM_UNLIMITED;
1128 * Socket is created in state SYN_RECEIVED.
1129 * Unlock the listen socket, lock the newly
1130 * created socket and update the tp variable.
1132 INP_WUNLOCK(inp); /* listen socket */
1133 inp = sotoinpcb(so);
1134 INP_WLOCK(inp); /* new connection */
1135 tp = intotcpcb(inp);
1136 KASSERT(tp->t_state == TCPS_SYN_RECEIVED,
1137 ("%s: ", __func__));
1138 #ifdef TCP_SIGNATURE
1139 if (sig_checked == 0) {
1140 tcp_dooptions(&to, optp, optlen,
1141 (thflags & TH_SYN) ? TO_SYN : 0);
1142 if (!tcp_signature_verify_input(m, off0, tlen,
1143 optlen, &to, th, tp->t_flags)) {
1146 * In SYN_SENT state if it receives an
1147 * RST, it is allowed for further
1150 if ((thflags & TH_RST) == 0 ||
1151 (tp->t_state == TCPS_SYN_SENT) == 0)
1159 * Process the segment and the data it
1160 * contains. tcp_do_segment() consumes
1161 * the mbuf chain and unlocks the inpcb.
1163 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen,
1165 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1169 * Segment flag validation for new connection attempts:
1171 * Our (SYN|ACK) response was rejected.
1172 * Check with syncache and remove entry to prevent
1175 * NB: syncache_chkrst does its own logging of failure
1178 if (thflags & TH_RST) {
1179 syncache_chkrst(&inc, th);
1183 * We can't do anything without SYN.
1185 if ((thflags & TH_SYN) == 0) {
1186 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1187 log(LOG_DEBUG, "%s; %s: Listen socket: "
1188 "SYN is missing, segment ignored\n",
1190 TCPSTAT_INC(tcps_badsyn);
1194 * (SYN|ACK) is bogus on a listen socket.
1196 if (thflags & TH_ACK) {
1197 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1198 log(LOG_DEBUG, "%s; %s: Listen socket: "
1199 "SYN|ACK invalid, segment rejected\n",
1201 syncache_badack(&inc); /* XXX: Not needed! */
1202 TCPSTAT_INC(tcps_badsyn);
1203 rstreason = BANDLIM_RST_OPENPORT;
1207 * If the drop_synfin option is enabled, drop all
1208 * segments with both the SYN and FIN bits set.
1209 * This prevents e.g. nmap from identifying the
1211 * XXX: Poor reasoning. nmap has other methods
1212 * and is constantly refining its stack detection
1214 * XXX: This is a violation of the TCP specification
1215 * and was used by RFC1644.
1217 if ((thflags & TH_FIN) && V_drop_synfin) {
1218 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1219 log(LOG_DEBUG, "%s; %s: Listen socket: "
1220 "SYN|FIN segment ignored (based on "
1221 "sysctl setting)\n", s, __func__);
1222 TCPSTAT_INC(tcps_badsyn);
1226 * Segment's flags are (SYN) or (SYN|FIN).
1228 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored
1229 * as they do not affect the state of the TCP FSM.
1230 * The data pointed to by TH_URG and th_urp is ignored.
1232 KASSERT((thflags & (TH_RST|TH_ACK)) == 0,
1233 ("%s: Listen socket: TH_RST or TH_ACK set", __func__));
1234 KASSERT(thflags & (TH_SYN),
1235 ("%s: Listen socket: TH_SYN not set", __func__));
1238 * If deprecated address is forbidden,
1239 * we do not accept SYN to deprecated interface
1240 * address to prevent any new inbound connection from
1241 * getting established.
1242 * When we do not accept SYN, we send a TCP RST,
1243 * with deprecated source address (instead of dropping
1244 * it). We compromise it as it is much better for peer
1245 * to send a RST, and RST will be the final packet
1248 * If we do not forbid deprecated addresses, we accept
1249 * the SYN packet. RFC2462 does not suggest dropping
1251 * If we decipher RFC2462 5.5.4, it says like this:
1252 * 1. use of deprecated addr with existing
1253 * communication is okay - "SHOULD continue to be
1255 * 2. use of it with new communication:
1256 * (2a) "SHOULD NOT be used if alternate address
1257 * with sufficient scope is available"
1258 * (2b) nothing mentioned otherwise.
1259 * Here we fall into (2b) case as we have no choice in
1260 * our source address selection - we must obey the peer.
1262 * The wording in RFC2462 is confusing, and there are
1263 * multiple description text for deprecated address
1264 * handling - worse, they are not exactly the same.
1265 * I believe 5.5.4 is the best one, so we follow 5.5.4.
1267 if (isipv6 && !V_ip6_use_deprecated) {
1268 struct in6_ifaddr *ia6;
1270 ia6 = ip6_getdstifaddr(m);
1272 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1273 ifa_free(&ia6->ia_ifa);
1274 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1275 log(LOG_DEBUG, "%s; %s: Listen socket: "
1276 "Connection attempt to deprecated "
1277 "IPv6 address rejected\n",
1279 rstreason = BANDLIM_RST_OPENPORT;
1283 ifa_free(&ia6->ia_ifa);
1287 * Basic sanity checks on incoming SYN requests:
1288 * Don't respond if the destination is a link layer
1289 * broadcast according to RFC1122 4.2.3.10, p. 104.
1290 * If it is from this socket it must be forged.
1291 * Don't respond if the source or destination is a
1292 * global or subnet broad- or multicast address.
1293 * Note that it is quite possible to receive unicast
1294 * link-layer packets with a broadcast IP address. Use
1295 * in_broadcast() to find them.
1297 if (m->m_flags & (M_BCAST|M_MCAST)) {
1298 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1299 log(LOG_DEBUG, "%s; %s: Listen socket: "
1300 "Connection attempt from broad- or multicast "
1301 "link layer address ignored\n", s, __func__);
1306 if (th->th_dport == th->th_sport &&
1307 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) {
1308 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1309 log(LOG_DEBUG, "%s; %s: Listen socket: "
1310 "Connection attempt to/from self "
1311 "ignored\n", s, __func__);
1314 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1315 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) {
1316 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1317 log(LOG_DEBUG, "%s; %s: Listen socket: "
1318 "Connection attempt from/to multicast "
1319 "address ignored\n", s, __func__);
1324 #if defined(INET) && defined(INET6)
1329 if (th->th_dport == th->th_sport &&
1330 ip->ip_dst.s_addr == ip->ip_src.s_addr) {
1331 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1332 log(LOG_DEBUG, "%s; %s: Listen socket: "
1333 "Connection attempt from/to self "
1334 "ignored\n", s, __func__);
1337 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1338 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1339 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1340 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
1341 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1342 log(LOG_DEBUG, "%s; %s: Listen socket: "
1343 "Connection attempt from/to broad- "
1344 "or multicast address ignored\n",
1351 * SYN appears to be valid. Create compressed TCP state
1355 if (so->so_options & SO_DEBUG)
1356 tcp_trace(TA_INPUT, ostate, tp,
1357 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1359 tcp_dooptions(&to, optp, optlen, TO_SYN);
1360 syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL);
1362 * Entry added to syncache and mbuf consumed.
1363 * Everything already unlocked by syncache_add().
1365 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1367 } else if (tp->t_state == TCPS_LISTEN) {
1369 * When a listen socket is torn down the SO_ACCEPTCONN
1370 * flag is removed first while connections are drained
1371 * from the accept queue in a unlock/lock cycle of the
1372 * ACCEPT_LOCK, opening a race condition allowing a SYN
1373 * attempt go through unhandled.
1378 #ifdef TCP_SIGNATURE
1379 if (sig_checked == 0) {
1380 tcp_dooptions(&to, optp, optlen,
1381 (thflags & TH_SYN) ? TO_SYN : 0);
1382 if (!tcp_signature_verify_input(m, off0, tlen, optlen, &to,
1386 * In SYN_SENT state if it receives an RST, it is
1387 * allowed for further processing.
1389 if ((thflags & TH_RST) == 0 ||
1390 (tp->t_state == TCPS_SYN_SENT) == 0)
1397 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th);
1400 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later
1401 * state. tcp_do_segment() always consumes the mbuf chain, unlocks
1402 * the inpcb, and unlocks pcbinfo.
1404 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos, ti_locked);
1405 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1409 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th);
1411 if (ti_locked == TI_WLOCKED) {
1412 INP_INFO_WUNLOCK(&V_tcbinfo);
1413 ti_locked = TI_UNLOCKED;
1417 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset "
1418 "ti_locked: %d", __func__, ti_locked));
1419 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1424 tcp_dropwithreset(m, th, tp, tlen, rstreason);
1427 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
1428 m = NULL; /* mbuf chain got consumed. */
1433 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th);
1435 if (ti_locked == TI_WLOCKED) {
1436 INP_INFO_WUNLOCK(&V_tcbinfo);
1437 ti_locked = TI_UNLOCKED;
1441 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock "
1442 "ti_locked: %d", __func__, ti_locked));
1443 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1451 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1459 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so,
1460 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos,
1463 int thflags, acked, ourfinisacked, needoutput = 0;
1464 int rstreason, todrop, win;
1467 struct in_conninfo *inc;
1473 * The size of tcp_saveipgen must be the size of the max ip header,
1476 u_char tcp_saveipgen[IP6_HDR_LEN];
1477 struct tcphdr tcp_savetcp;
1480 thflags = th->th_flags;
1481 inc = &tp->t_inpcb->inp_inc;
1482 tp->sackhint.last_sack_ack = 0;
1485 * If this is either a state-changing packet or current state isn't
1486 * established, we require a write lock on tcbinfo. Otherwise, we
1487 * allow the tcbinfo to be in either alocked or unlocked, as the
1488 * caller may have unnecessarily acquired a write lock due to a race.
1490 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 ||
1491 tp->t_state != TCPS_ESTABLISHED) {
1492 KASSERT(ti_locked == TI_WLOCKED, ("%s ti_locked %d for "
1493 "SYN/FIN/RST/!EST", __func__, ti_locked));
1494 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1497 if (ti_locked == TI_WLOCKED)
1498 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1500 KASSERT(ti_locked == TI_UNLOCKED, ("%s: EST "
1501 "ti_locked: %d", __func__, ti_locked));
1502 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1506 INP_WLOCK_ASSERT(tp->t_inpcb);
1507 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN",
1509 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT",
1513 * Segment received on connection.
1514 * Reset idle time and keep-alive timer.
1515 * XXX: This should be done after segment
1516 * validation to ignore broken/spoofed segs.
1518 tp->t_rcvtime = ticks;
1519 if (TCPS_HAVEESTABLISHED(tp->t_state))
1520 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
1523 * Unscale the window into a 32-bit value.
1524 * For the SYN_SENT state the scale is zero.
1526 tiwin = th->th_win << tp->snd_scale;
1529 * TCP ECN processing.
1531 if (tp->t_flags & TF_ECN_PERMIT) {
1532 if (thflags & TH_CWR)
1533 tp->t_flags &= ~TF_ECN_SND_ECE;
1534 switch (iptos & IPTOS_ECN_MASK) {
1536 tp->t_flags |= TF_ECN_SND_ECE;
1537 TCPSTAT_INC(tcps_ecn_ce);
1539 case IPTOS_ECN_ECT0:
1540 TCPSTAT_INC(tcps_ecn_ect0);
1542 case IPTOS_ECN_ECT1:
1543 TCPSTAT_INC(tcps_ecn_ect1);
1546 /* Congestion experienced. */
1547 if (thflags & TH_ECE) {
1548 cc_cong_signal(tp, th, CC_ECN);
1553 * Parse options on any incoming segment.
1555 tcp_dooptions(&to, (u_char *)(th + 1),
1556 (th->th_off << 2) - sizeof(struct tcphdr),
1557 (thflags & TH_SYN) ? TO_SYN : 0);
1560 * If echoed timestamp is later than the current time,
1561 * fall back to non RFC1323 RTT calculation. Normalize
1562 * timestamp if syncookies were used when this connection
1565 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1566 to.to_tsecr -= tp->ts_offset;
1567 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks()))
1571 * If timestamps were negotiated during SYN/ACK they should
1572 * appear on every segment during this session and vice versa.
1574 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) {
1575 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1576 log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1577 "no action\n", s, __func__);
1581 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) {
1582 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1583 log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
1584 "no action\n", s, __func__);
1590 * Process options only when we get SYN/ACK back. The SYN case
1591 * for incoming connections is handled in tcp_syncache.
1592 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
1593 * or <SYN,ACK>) segment itself is never scaled.
1594 * XXX this is traditional behavior, may need to be cleaned up.
1596 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1597 if ((to.to_flags & TOF_SCALE) &&
1598 (tp->t_flags & TF_REQ_SCALE)) {
1599 tp->t_flags |= TF_RCVD_SCALE;
1600 tp->snd_scale = to.to_wscale;
1603 * Initial send window. It will be updated with
1604 * the next incoming segment to the scaled value.
1606 tp->snd_wnd = th->th_win;
1607 if (to.to_flags & TOF_TS) {
1608 tp->t_flags |= TF_RCVD_TSTMP;
1609 tp->ts_recent = to.to_tsval;
1610 tp->ts_recent_age = tcp_ts_getticks();
1612 if (to.to_flags & TOF_MSS)
1613 tcp_mss(tp, to.to_mss);
1614 if ((tp->t_flags & TF_SACK_PERMIT) &&
1615 (to.to_flags & TOF_SACKPERM) == 0)
1616 tp->t_flags &= ~TF_SACK_PERMIT;
1620 * Header prediction: check for the two common cases
1621 * of a uni-directional data xfer. If the packet has
1622 * no control flags, is in-sequence, the window didn't
1623 * change and we're not retransmitting, it's a
1624 * candidate. If the length is zero and the ack moved
1625 * forward, we're the sender side of the xfer. Just
1626 * free the data acked & wake any higher level process
1627 * that was blocked waiting for space. If the length
1628 * is non-zero and the ack didn't move, we're the
1629 * receiver side. If we're getting packets in-order
1630 * (the reassembly queue is empty), add the data to
1631 * the socket buffer and note that we need a delayed ack.
1632 * Make sure that the hidden state-flags are also off.
1633 * Since we check for TCPS_ESTABLISHED first, it can only
1636 if (tp->t_state == TCPS_ESTABLISHED &&
1637 th->th_seq == tp->rcv_nxt &&
1638 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1639 tp->snd_nxt == tp->snd_max &&
1640 tiwin && tiwin == tp->snd_wnd &&
1641 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1642 tp->t_segq == NULL && ((to.to_flags & TOF_TS) == 0 ||
1643 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) {
1646 * If last ACK falls within this segment's sequence numbers,
1647 * record the timestamp.
1648 * NOTE that the test is modified according to the latest
1649 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1651 if ((to.to_flags & TOF_TS) != 0 &&
1652 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1653 tp->ts_recent_age = tcp_ts_getticks();
1654 tp->ts_recent = to.to_tsval;
1658 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1659 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1660 !IN_RECOVERY(tp->t_flags) &&
1661 (to.to_flags & TOF_SACK) == 0 &&
1662 TAILQ_EMPTY(&tp->snd_holes)) {
1664 * This is a pure ack for outstanding data.
1666 if (ti_locked == TI_WLOCKED)
1667 INP_INFO_WUNLOCK(&V_tcbinfo);
1668 ti_locked = TI_UNLOCKED;
1670 TCPSTAT_INC(tcps_predack);
1673 * "bad retransmit" recovery.
1675 if (tp->t_rxtshift == 1 &&
1676 tp->t_flags & TF_PREVVALID &&
1677 (int)(ticks - tp->t_badrxtwin) < 0) {
1678 cc_cong_signal(tp, th, CC_RTO_ERR);
1682 * Recalculate the transmit timer / rtt.
1684 * Some boxes send broken timestamp replies
1685 * during the SYN+ACK phase, ignore
1686 * timestamps of 0 or we could calculate a
1687 * huge RTT and blow up the retransmit timer.
1689 if ((to.to_flags & TOF_TS) != 0 &&
1693 t = tcp_ts_getticks() - to.to_tsecr;
1694 if (!tp->t_rttlow || tp->t_rttlow > t)
1697 TCP_TS_TO_TICKS(t) + 1);
1698 } else if (tp->t_rtttime &&
1699 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1700 if (!tp->t_rttlow ||
1701 tp->t_rttlow > ticks - tp->t_rtttime)
1702 tp->t_rttlow = ticks - tp->t_rtttime;
1704 ticks - tp->t_rtttime);
1706 acked = BYTES_THIS_ACK(tp, th);
1708 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
1709 hhook_run_tcp_est_in(tp, th, &to);
1711 TCPSTAT_INC(tcps_rcvackpack);
1712 TCPSTAT_ADD(tcps_rcvackbyte, acked);
1713 sbdrop(&so->so_snd, acked);
1714 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1715 SEQ_LEQ(th->th_ack, tp->snd_recover))
1716 tp->snd_recover = th->th_ack - 1;
1719 * Let the congestion control algorithm update
1720 * congestion control related information. This
1721 * typically means increasing the congestion
1724 cc_ack_received(tp, th, CC_ACK);
1726 tp->snd_una = th->th_ack;
1728 * Pull snd_wl2 up to prevent seq wrap relative
1731 tp->snd_wl2 = th->th_ack;
1734 ND6_HINT(tp); /* Some progress has been made. */
1737 * If all outstanding data are acked, stop
1738 * retransmit timer, otherwise restart timer
1739 * using current (possibly backed-off) value.
1740 * If process is waiting for space,
1741 * wakeup/selwakeup/signal. If data
1742 * are ready to send, let tcp_output
1743 * decide between more output or persist.
1746 if (so->so_options & SO_DEBUG)
1747 tcp_trace(TA_INPUT, ostate, tp,
1748 (void *)tcp_saveipgen,
1751 if (tp->snd_una == tp->snd_max)
1752 tcp_timer_activate(tp, TT_REXMT, 0);
1753 else if (!tcp_timer_active(tp, TT_PERSIST))
1754 tcp_timer_activate(tp, TT_REXMT,
1757 if (so->so_snd.sb_cc)
1758 (void) tcp_output(tp);
1761 } else if (th->th_ack == tp->snd_una &&
1762 tlen <= sbspace(&so->so_rcv)) {
1763 int newsize = 0; /* automatic sockbuf scaling */
1766 * This is a pure, in-sequence data packet with
1767 * nothing on the reassembly queue and we have enough
1768 * buffer space to take it.
1770 if (ti_locked == TI_WLOCKED)
1771 INP_INFO_WUNLOCK(&V_tcbinfo);
1772 ti_locked = TI_UNLOCKED;
1774 /* Clean receiver SACK report if present */
1775 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks)
1776 tcp_clean_sackreport(tp);
1777 TCPSTAT_INC(tcps_preddat);
1778 tp->rcv_nxt += tlen;
1780 * Pull snd_wl1 up to prevent seq wrap relative to
1783 tp->snd_wl1 = th->th_seq;
1785 * Pull rcv_up up to prevent seq wrap relative to
1788 tp->rcv_up = tp->rcv_nxt;
1789 TCPSTAT_INC(tcps_rcvpack);
1790 TCPSTAT_ADD(tcps_rcvbyte, tlen);
1791 ND6_HINT(tp); /* Some progress has been made */
1793 if (so->so_options & SO_DEBUG)
1794 tcp_trace(TA_INPUT, ostate, tp,
1795 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1798 * Automatic sizing of receive socket buffer. Often the send
1799 * buffer size is not optimally adjusted to the actual network
1800 * conditions at hand (delay bandwidth product). Setting the
1801 * buffer size too small limits throughput on links with high
1802 * bandwidth and high delay (eg. trans-continental/oceanic links).
1804 * On the receive side the socket buffer memory is only rarely
1805 * used to any significant extent. This allows us to be much
1806 * more aggressive in scaling the receive socket buffer. For
1807 * the case that the buffer space is actually used to a large
1808 * extent and we run out of kernel memory we can simply drop
1809 * the new segments; TCP on the sender will just retransmit it
1810 * later. Setting the buffer size too big may only consume too
1811 * much kernel memory if the application doesn't read() from
1812 * the socket or packet loss or reordering makes use of the
1815 * The criteria to step up the receive buffer one notch are:
1816 * 1. the number of bytes received during the time it takes
1817 * one timestamp to be reflected back to us (the RTT);
1818 * 2. received bytes per RTT is within seven eighth of the
1819 * current socket buffer size;
1820 * 3. receive buffer size has not hit maximal automatic size;
1822 * This algorithm does one step per RTT at most and only if
1823 * we receive a bulk stream w/o packet losses or reorderings.
1824 * Shrinking the buffer during idle times is not necessary as
1825 * it doesn't consume any memory when idle.
1827 * TODO: Only step up if the application is actually serving
1828 * the buffer to better manage the socket buffer resources.
1830 if (V_tcp_do_autorcvbuf &&
1832 (so->so_rcv.sb_flags & SB_AUTOSIZE)) {
1833 if (TSTMP_GT(to.to_tsecr, tp->rfbuf_ts) &&
1834 to.to_tsecr - tp->rfbuf_ts < hz) {
1836 (so->so_rcv.sb_hiwat / 8 * 7) &&
1837 so->so_rcv.sb_hiwat <
1838 V_tcp_autorcvbuf_max) {
1840 min(so->so_rcv.sb_hiwat +
1841 V_tcp_autorcvbuf_inc,
1842 V_tcp_autorcvbuf_max);
1844 /* Start over with next RTT. */
1848 tp->rfbuf_cnt += tlen; /* add up */
1851 /* Add data to socket buffer. */
1852 SOCKBUF_LOCK(&so->so_rcv);
1853 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1857 * Set new socket buffer size.
1858 * Give up when limit is reached.
1861 if (!sbreserve_locked(&so->so_rcv,
1863 so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1864 m_adj(m, drop_hdrlen); /* delayed header drop */
1865 sbappendstream_locked(&so->so_rcv, m);
1867 /* NB: sorwakeup_locked() does an implicit unlock. */
1868 sorwakeup_locked(so);
1869 if (DELAY_ACK(tp, tlen)) {
1870 tp->t_flags |= TF_DELACK;
1872 tp->t_flags |= TF_ACKNOW;
1880 * Calculate amount of space in receive window,
1881 * and then do TCP input processing.
1882 * Receive window is amount of space in rcv queue,
1883 * but not less than advertised window.
1885 win = sbspace(&so->so_rcv);
1888 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1890 /* Reset receive buffer auto scaling when not in bulk receive mode. */
1894 switch (tp->t_state) {
1897 * If the state is SYN_RECEIVED:
1898 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1900 case TCPS_SYN_RECEIVED:
1901 if ((thflags & TH_ACK) &&
1902 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1903 SEQ_GT(th->th_ack, tp->snd_max))) {
1904 rstreason = BANDLIM_RST_OPENPORT;
1910 * If the state is SYN_SENT:
1911 * if seg contains an ACK, but not for our SYN, drop the input.
1912 * if seg contains a RST, then drop the connection.
1913 * if seg does not contain SYN, then drop it.
1914 * Otherwise this is an acceptable SYN segment
1915 * initialize tp->rcv_nxt and tp->irs
1916 * if seg contains ack then advance tp->snd_una
1917 * if seg contains an ECE and ECN support is enabled, the stream
1919 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1920 * arrange for segment to be acked (eventually)
1921 * continue processing rest of data/controls, beginning with URG
1924 if ((thflags & TH_ACK) &&
1925 (SEQ_LEQ(th->th_ack, tp->iss) ||
1926 SEQ_GT(th->th_ack, tp->snd_max))) {
1927 rstreason = BANDLIM_UNLIMITED;
1930 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) {
1931 TCP_PROBE5(connect__refused, NULL, tp,
1932 mtod(m, const char *), tp, th);
1933 tp = tcp_drop(tp, ECONNREFUSED);
1935 if (thflags & TH_RST)
1937 if (!(thflags & TH_SYN))
1940 tp->irs = th->th_seq;
1942 if (thflags & TH_ACK) {
1943 TCPSTAT_INC(tcps_connects);
1946 mac_socketpeer_set_from_mbuf(m, so);
1948 /* Do window scaling on this connection? */
1949 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1950 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1951 tp->rcv_scale = tp->request_r_scale;
1953 tp->rcv_adv += imin(tp->rcv_wnd,
1954 TCP_MAXWIN << tp->rcv_scale);
1955 tp->snd_una++; /* SYN is acked */
1957 * If there's data, delay ACK; if there's also a FIN
1958 * ACKNOW will be turned on later.
1960 if (DELAY_ACK(tp, tlen) && tlen != 0)
1961 tcp_timer_activate(tp, TT_DELACK,
1964 tp->t_flags |= TF_ACKNOW;
1966 if ((thflags & TH_ECE) && V_tcp_do_ecn) {
1967 tp->t_flags |= TF_ECN_PERMIT;
1968 TCPSTAT_INC(tcps_ecn_shs);
1972 * Received <SYN,ACK> in SYN_SENT[*] state.
1974 * SYN_SENT --> ESTABLISHED
1975 * SYN_SENT* --> FIN_WAIT_1
1977 tp->t_starttime = ticks;
1978 if (tp->t_flags & TF_NEEDFIN) {
1979 tcp_state_change(tp, TCPS_FIN_WAIT_1);
1980 tp->t_flags &= ~TF_NEEDFIN;
1983 tcp_state_change(tp, TCPS_ESTABLISHED);
1984 TCP_PROBE5(connect__established, NULL, tp,
1985 mtod(m, const char *), tp, th);
1987 tcp_timer_activate(tp, TT_KEEP,
1992 * Received initial SYN in SYN-SENT[*] state =>
1993 * simultaneous open.
1994 * If it succeeds, connection is * half-synchronized.
1995 * Otherwise, do 3-way handshake:
1996 * SYN-SENT -> SYN-RECEIVED
1997 * SYN-SENT* -> SYN-RECEIVED*
1999 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
2000 tcp_timer_activate(tp, TT_REXMT, 0);
2001 tcp_state_change(tp, TCPS_SYN_RECEIVED);
2004 KASSERT(ti_locked == TI_WLOCKED, ("%s: trimthenstep6: "
2005 "ti_locked %d", __func__, ti_locked));
2006 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2007 INP_WLOCK_ASSERT(tp->t_inpcb);
2010 * Advance th->th_seq to correspond to first data byte.
2011 * If data, trim to stay within window,
2012 * dropping FIN if necessary.
2015 if (tlen > tp->rcv_wnd) {
2016 todrop = tlen - tp->rcv_wnd;
2020 TCPSTAT_INC(tcps_rcvpackafterwin);
2021 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2023 tp->snd_wl1 = th->th_seq - 1;
2024 tp->rcv_up = th->th_seq;
2026 * Client side of transaction: already sent SYN and data.
2027 * If the remote host used T/TCP to validate the SYN,
2028 * our data will be ACK'd; if so, enter normal data segment
2029 * processing in the middle of step 5, ack processing.
2030 * Otherwise, goto step 6.
2032 if (thflags & TH_ACK)
2038 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
2039 * do normal processing.
2041 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later.
2045 break; /* continue normal processing */
2049 * States other than LISTEN or SYN_SENT.
2050 * First check the RST flag and sequence number since reset segments
2051 * are exempt from the timestamp and connection count tests. This
2052 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
2053 * below which allowed reset segments in half the sequence space
2054 * to fall though and be processed (which gives forged reset
2055 * segments with a random sequence number a 50 percent chance of
2056 * killing a connection).
2057 * Then check timestamp, if present.
2058 * Then check the connection count, if present.
2059 * Then check that at least some bytes of segment are within
2060 * receive window. If segment begins before rcv_nxt,
2061 * drop leading data (and SYN); if nothing left, just ack.
2064 * If the RST bit is set, check the sequence number to see
2065 * if this is a valid reset segment.
2067 * In all states except SYN-SENT, all reset (RST) segments
2068 * are validated by checking their SEQ-fields. A reset is
2069 * valid if its sequence number is in the window.
2070 * Note: this does not take into account delayed ACKs, so
2071 * we should test against last_ack_sent instead of rcv_nxt.
2072 * The sequence number in the reset segment is normally an
2073 * echo of our outgoing acknowlegement numbers, but some hosts
2074 * send a reset with the sequence number at the rightmost edge
2075 * of our receive window, and we have to handle this case.
2076 * Note 2: Paul Watson's paper "Slipping in the Window" has shown
2077 * that brute force RST attacks are possible. To combat this,
2078 * we use a much stricter check while in the ESTABLISHED state,
2079 * only accepting RSTs where the sequence number is equal to
2080 * last_ack_sent. In all other states (the states in which a
2081 * RST is more likely), the more permissive check is used.
2082 * If we have multiple segments in flight, the initial reset
2083 * segment sequence numbers will be to the left of last_ack_sent,
2084 * but they will eventually catch up.
2085 * In any case, it never made sense to trim reset segments to
2086 * fit the receive window since RFC 1122 says:
2087 * 4.2.2.12 RST Segment: RFC-793 Section 3.4
2089 * A TCP SHOULD allow a received RST segment to include data.
2092 * It has been suggested that a RST segment could contain
2093 * ASCII text that encoded and explained the cause of the
2094 * RST. No standard has yet been established for such
2097 * If the reset segment passes the sequence number test examine
2099 * SYN_RECEIVED STATE:
2100 * If passive open, return to LISTEN state.
2101 * If active open, inform user that connection was refused.
2102 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
2103 * Inform user that connection was reset, and close tcb.
2104 * CLOSING, LAST_ACK STATES:
2107 * Drop the segment - see Stevens, vol. 2, p. 964 and
2110 if (thflags & TH_RST) {
2111 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) &&
2112 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
2113 switch (tp->t_state) {
2115 case TCPS_SYN_RECEIVED:
2116 so->so_error = ECONNREFUSED;
2119 case TCPS_ESTABLISHED:
2120 if (V_tcp_insecure_rst == 0 &&
2121 !(SEQ_GEQ(th->th_seq, tp->rcv_nxt - 1) &&
2122 SEQ_LEQ(th->th_seq, tp->rcv_nxt + 1)) &&
2123 !(SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) &&
2124 SEQ_LEQ(th->th_seq, tp->last_ack_sent + 1))) {
2125 TCPSTAT_INC(tcps_badrst);
2129 case TCPS_FIN_WAIT_1:
2130 case TCPS_FIN_WAIT_2:
2131 case TCPS_CLOSE_WAIT:
2132 so->so_error = ECONNRESET;
2134 KASSERT(ti_locked == TI_WLOCKED,
2135 ("tcp_do_segment: TH_RST 1 ti_locked %d",
2137 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2139 tcp_state_change(tp, TCPS_CLOSED);
2140 TCPSTAT_INC(tcps_drops);
2146 KASSERT(ti_locked == TI_WLOCKED,
2147 ("tcp_do_segment: TH_RST 2 ti_locked %d",
2149 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2159 * RFC 1323 PAWS: If we have a timestamp reply on this segment
2160 * and it's less than ts_recent, drop it.
2162 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
2163 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
2165 /* Check to see if ts_recent is over 24 days old. */
2166 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) {
2168 * Invalidate ts_recent. If this segment updates
2169 * ts_recent, the age will be reset later and ts_recent
2170 * will get a valid value. If it does not, setting
2171 * ts_recent to zero will at least satisfy the
2172 * requirement that zero be placed in the timestamp
2173 * echo reply when ts_recent isn't valid. The
2174 * age isn't reset until we get a valid ts_recent
2175 * because we don't want out-of-order segments to be
2176 * dropped when ts_recent is old.
2180 TCPSTAT_INC(tcps_rcvduppack);
2181 TCPSTAT_ADD(tcps_rcvdupbyte, tlen);
2182 TCPSTAT_INC(tcps_pawsdrop);
2190 * In the SYN-RECEIVED state, validate that the packet belongs to
2191 * this connection before trimming the data to fit the receive
2192 * window. Check the sequence number versus IRS since we know
2193 * the sequence numbers haven't wrapped. This is a partial fix
2194 * for the "LAND" DoS attack.
2196 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
2197 rstreason = BANDLIM_RST_OPENPORT;
2201 todrop = tp->rcv_nxt - th->th_seq;
2204 * If this is a duplicate SYN for our current connection,
2205 * advance over it and pretend and it's not a SYN.
2207 if (thflags & TH_SYN && th->th_seq == tp->irs) {
2217 * Following if statement from Stevens, vol. 2, p. 960.
2220 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
2222 * Any valid FIN must be to the left of the window.
2223 * At this point the FIN must be a duplicate or out
2224 * of sequence; drop it.
2229 * Send an ACK to resynchronize and drop any data.
2230 * But keep on processing for RST or ACK.
2232 tp->t_flags |= TF_ACKNOW;
2234 TCPSTAT_INC(tcps_rcvduppack);
2235 TCPSTAT_ADD(tcps_rcvdupbyte, todrop);
2237 TCPSTAT_INC(tcps_rcvpartduppack);
2238 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop);
2240 drop_hdrlen += todrop; /* drop from the top afterwards */
2241 th->th_seq += todrop;
2243 if (th->th_urp > todrop)
2244 th->th_urp -= todrop;
2252 * If new data are received on a connection after the
2253 * user processes are gone, then RST the other end.
2255 if ((so->so_state & SS_NOFDREF) &&
2256 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
2257 KASSERT(ti_locked == TI_WLOCKED, ("%s: SS_NOFDEREF && "
2258 "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked));
2259 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2261 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
2262 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data "
2263 "after socket was closed, "
2264 "sending RST and removing tcpcb\n",
2265 s, __func__, tcpstates[tp->t_state], tlen);
2269 TCPSTAT_INC(tcps_rcvafterclose);
2270 rstreason = BANDLIM_UNLIMITED;
2275 * If segment ends after window, drop trailing data
2276 * (and PUSH and FIN); if nothing left, just ACK.
2278 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
2280 TCPSTAT_INC(tcps_rcvpackafterwin);
2281 if (todrop >= tlen) {
2282 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen);
2284 * If window is closed can only take segments at
2285 * window edge, and have to drop data and PUSH from
2286 * incoming segments. Continue processing, but
2287 * remember to ack. Otherwise, drop segment
2290 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2291 tp->t_flags |= TF_ACKNOW;
2292 TCPSTAT_INC(tcps_rcvwinprobe);
2296 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2299 thflags &= ~(TH_PUSH|TH_FIN);
2303 * If last ACK falls within this segment's sequence numbers,
2304 * record its timestamp.
2306 * 1) That the test incorporates suggestions from the latest
2307 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
2308 * 2) That updating only on newer timestamps interferes with
2309 * our earlier PAWS tests, so this check should be solely
2310 * predicated on the sequence space of this segment.
2311 * 3) That we modify the segment boundary check to be
2312 * Last.ACK.Sent <= SEG.SEQ + SEG.Len
2313 * instead of RFC1323's
2314 * Last.ACK.Sent < SEG.SEQ + SEG.Len,
2315 * This modified check allows us to overcome RFC1323's
2316 * limitations as described in Stevens TCP/IP Illustrated
2317 * Vol. 2 p.869. In such cases, we can still calculate the
2318 * RTT correctly when RCV.NXT == Last.ACK.Sent.
2320 if ((to.to_flags & TOF_TS) != 0 &&
2321 SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2322 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
2323 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
2324 tp->ts_recent_age = tcp_ts_getticks();
2325 tp->ts_recent = to.to_tsval;
2329 * If a SYN is in the window, then this is an
2330 * error and we send an RST and drop the connection.
2332 if (thflags & TH_SYN) {
2333 KASSERT(ti_locked == TI_WLOCKED,
2334 ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked));
2335 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2337 tp = tcp_drop(tp, ECONNRESET);
2338 rstreason = BANDLIM_UNLIMITED;
2343 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
2344 * flag is on (half-synchronized state), then queue data for
2345 * later processing; else drop segment and return.
2347 if ((thflags & TH_ACK) == 0) {
2348 if (tp->t_state == TCPS_SYN_RECEIVED ||
2349 (tp->t_flags & TF_NEEDSYN))
2351 else if (tp->t_flags & TF_ACKNOW)
2360 switch (tp->t_state) {
2363 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
2364 * ESTABLISHED state and continue processing.
2365 * The ACK was checked above.
2367 case TCPS_SYN_RECEIVED:
2369 TCPSTAT_INC(tcps_connects);
2371 /* Do window scaling? */
2372 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2373 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2374 tp->rcv_scale = tp->request_r_scale;
2375 tp->snd_wnd = tiwin;
2379 * SYN-RECEIVED -> ESTABLISHED
2380 * SYN-RECEIVED* -> FIN-WAIT-1
2382 tp->t_starttime = ticks;
2383 if (tp->t_flags & TF_NEEDFIN) {
2384 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2385 tp->t_flags &= ~TF_NEEDFIN;
2387 tcp_state_change(tp, TCPS_ESTABLISHED);
2388 TCP_PROBE5(accept__established, NULL, tp,
2389 mtod(m, const char *), tp, th);
2391 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
2394 * If segment contains data or ACK, will call tcp_reass()
2395 * later; if not, do so now to pass queued data to user.
2397 if (tlen == 0 && (thflags & TH_FIN) == 0)
2398 (void) tcp_reass(tp, (struct tcphdr *)0, 0,
2400 tp->snd_wl1 = th->th_seq - 1;
2404 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2405 * ACKs. If the ack is in the range
2406 * tp->snd_una < th->th_ack <= tp->snd_max
2407 * then advance tp->snd_una to th->th_ack and drop
2408 * data from the retransmission queue. If this ACK reflects
2409 * more up to date window information we update our window information.
2411 case TCPS_ESTABLISHED:
2412 case TCPS_FIN_WAIT_1:
2413 case TCPS_FIN_WAIT_2:
2414 case TCPS_CLOSE_WAIT:
2417 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2418 TCPSTAT_INC(tcps_rcvacktoomuch);
2421 if ((tp->t_flags & TF_SACK_PERMIT) &&
2422 ((to.to_flags & TOF_SACK) ||
2423 !TAILQ_EMPTY(&tp->snd_holes)))
2424 tcp_sack_doack(tp, &to, th->th_ack);
2426 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
2427 hhook_run_tcp_est_in(tp, th, &to);
2429 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2430 if (tlen == 0 && tiwin == tp->snd_wnd) {
2432 * If this is the first time we've seen a
2433 * FIN from the remote, this is not a
2434 * duplicate and it needs to be processed
2435 * normally. This happens during a
2436 * simultaneous close.
2438 if ((thflags & TH_FIN) &&
2439 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2443 TCPSTAT_INC(tcps_rcvdupack);
2445 * If we have outstanding data (other than
2446 * a window probe), this is a completely
2447 * duplicate ack (ie, window info didn't
2448 * change and FIN isn't set),
2449 * the ack is the biggest we've
2450 * seen and we've seen exactly our rexmt
2451 * threshhold of them, assume a packet
2452 * has been dropped and retransmit it.
2453 * Kludge snd_nxt & the congestion
2454 * window so we send only this one
2457 * We know we're losing at the current
2458 * window size so do congestion avoidance
2459 * (set ssthresh to half the current window
2460 * and pull our congestion window back to
2461 * the new ssthresh).
2463 * Dup acks mean that packets have left the
2464 * network (they're now cached at the receiver)
2465 * so bump cwnd by the amount in the receiver
2466 * to keep a constant cwnd packets in the
2469 * When using TCP ECN, notify the peer that
2470 * we reduced the cwnd.
2472 if (!tcp_timer_active(tp, TT_REXMT) ||
2473 th->th_ack != tp->snd_una)
2475 else if (++tp->t_dupacks > tcprexmtthresh ||
2476 IN_FASTRECOVERY(tp->t_flags)) {
2477 cc_ack_received(tp, th, CC_DUPACK);
2478 if ((tp->t_flags & TF_SACK_PERMIT) &&
2479 IN_FASTRECOVERY(tp->t_flags)) {
2483 * Compute the amount of data in flight first.
2484 * We can inject new data into the pipe iff
2485 * we have less than 1/2 the original window's
2486 * worth of data in flight.
2488 awnd = (tp->snd_nxt - tp->snd_fack) +
2489 tp->sackhint.sack_bytes_rexmit;
2490 if (awnd < tp->snd_ssthresh) {
2491 tp->snd_cwnd += tp->t_maxseg;
2492 if (tp->snd_cwnd > tp->snd_ssthresh)
2493 tp->snd_cwnd = tp->snd_ssthresh;
2496 tp->snd_cwnd += tp->t_maxseg;
2497 (void) tcp_output(tp);
2499 } else if (tp->t_dupacks == tcprexmtthresh) {
2500 tcp_seq onxt = tp->snd_nxt;
2503 * If we're doing sack, check to
2504 * see if we're already in sack
2505 * recovery. If we're not doing sack,
2506 * check to see if we're in newreno
2509 if (tp->t_flags & TF_SACK_PERMIT) {
2510 if (IN_FASTRECOVERY(tp->t_flags)) {
2515 if (SEQ_LEQ(th->th_ack,
2521 /* Congestion signal before ack. */
2522 cc_cong_signal(tp, th, CC_NDUPACK);
2523 cc_ack_received(tp, th, CC_DUPACK);
2524 tcp_timer_activate(tp, TT_REXMT, 0);
2526 if (tp->t_flags & TF_SACK_PERMIT) {
2528 tcps_sack_recovery_episode);
2529 tp->sack_newdata = tp->snd_nxt;
2530 tp->snd_cwnd = tp->t_maxseg;
2531 (void) tcp_output(tp);
2534 tp->snd_nxt = th->th_ack;
2535 tp->snd_cwnd = tp->t_maxseg;
2536 (void) tcp_output(tp);
2537 KASSERT(tp->snd_limited <= 2,
2538 ("%s: tp->snd_limited too big",
2540 tp->snd_cwnd = tp->snd_ssthresh +
2542 (tp->t_dupacks - tp->snd_limited);
2543 if (SEQ_GT(onxt, tp->snd_nxt))
2546 } else if (V_tcp_do_rfc3042) {
2547 cc_ack_received(tp, th, CC_DUPACK);
2548 u_long oldcwnd = tp->snd_cwnd;
2549 tcp_seq oldsndmax = tp->snd_max;
2553 KASSERT(tp->t_dupacks == 1 ||
2555 ("%s: dupacks not 1 or 2",
2557 if (tp->t_dupacks == 1)
2558 tp->snd_limited = 0;
2560 (tp->snd_nxt - tp->snd_una) +
2561 (tp->t_dupacks - tp->snd_limited) *
2564 * Only call tcp_output when there
2565 * is new data available to be sent.
2566 * Otherwise we would send pure ACKs.
2568 SOCKBUF_LOCK(&so->so_snd);
2569 avail = so->so_snd.sb_cc -
2570 (tp->snd_nxt - tp->snd_una);
2571 SOCKBUF_UNLOCK(&so->so_snd);
2573 (void) tcp_output(tp);
2574 sent = tp->snd_max - oldsndmax;
2575 if (sent > tp->t_maxseg) {
2576 KASSERT((tp->t_dupacks == 2 &&
2577 tp->snd_limited == 0) ||
2578 (sent == tp->t_maxseg + 1 &&
2579 tp->t_flags & TF_SENTFIN),
2580 ("%s: sent too much",
2582 tp->snd_limited = 2;
2583 } else if (sent > 0)
2585 tp->snd_cwnd = oldcwnd;
2593 KASSERT(SEQ_GT(th->th_ack, tp->snd_una),
2594 ("%s: th_ack <= snd_una", __func__));
2597 * If the congestion window was inflated to account
2598 * for the other side's cached packets, retract it.
2600 if (IN_FASTRECOVERY(tp->t_flags)) {
2601 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2602 if (tp->t_flags & TF_SACK_PERMIT)
2603 tcp_sack_partialack(tp, th);
2605 tcp_newreno_partial_ack(tp, th);
2607 cc_post_recovery(tp, th);
2611 * If we reach this point, ACK is not a duplicate,
2612 * i.e., it ACKs something we sent.
2614 if (tp->t_flags & TF_NEEDSYN) {
2616 * T/TCP: Connection was half-synchronized, and our
2617 * SYN has been ACK'd (so connection is now fully
2618 * synchronized). Go to non-starred state,
2619 * increment snd_una for ACK of SYN, and check if
2620 * we can do window scaling.
2622 tp->t_flags &= ~TF_NEEDSYN;
2624 /* Do window scaling? */
2625 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2626 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2627 tp->rcv_scale = tp->request_r_scale;
2628 /* Send window already scaled. */
2633 INP_WLOCK_ASSERT(tp->t_inpcb);
2635 acked = BYTES_THIS_ACK(tp, th);
2636 TCPSTAT_INC(tcps_rcvackpack);
2637 TCPSTAT_ADD(tcps_rcvackbyte, acked);
2640 * If we just performed our first retransmit, and the ACK
2641 * arrives within our recovery window, then it was a mistake
2642 * to do the retransmit in the first place. Recover our
2643 * original cwnd and ssthresh, and proceed to transmit where
2646 if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID &&
2647 (int)(ticks - tp->t_badrxtwin) < 0)
2648 cc_cong_signal(tp, th, CC_RTO_ERR);
2651 * If we have a timestamp reply, update smoothed
2652 * round trip time. If no timestamp is present but
2653 * transmit timer is running and timed sequence
2654 * number was acked, update smoothed round trip time.
2655 * Since we now have an rtt measurement, cancel the
2656 * timer backoff (cf., Phil Karn's retransmit alg.).
2657 * Recompute the initial retransmit timer.
2659 * Some boxes send broken timestamp replies
2660 * during the SYN+ACK phase, ignore
2661 * timestamps of 0 or we could calculate a
2662 * huge RTT and blow up the retransmit timer.
2664 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) {
2667 t = tcp_ts_getticks() - to.to_tsecr;
2668 if (!tp->t_rttlow || tp->t_rttlow > t)
2670 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1);
2671 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2672 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2673 tp->t_rttlow = ticks - tp->t_rtttime;
2674 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2678 * If all outstanding data is acked, stop retransmit
2679 * timer and remember to restart (more output or persist).
2680 * If there is more data to be acked, restart retransmit
2681 * timer, using current (possibly backed-off) value.
2683 if (th->th_ack == tp->snd_max) {
2684 tcp_timer_activate(tp, TT_REXMT, 0);
2686 } else if (!tcp_timer_active(tp, TT_PERSIST))
2687 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
2690 * If no data (only SYN) was ACK'd,
2691 * skip rest of ACK processing.
2697 * Let the congestion control algorithm update congestion
2698 * control related information. This typically means increasing
2699 * the congestion window.
2701 cc_ack_received(tp, th, CC_ACK);
2703 SOCKBUF_LOCK(&so->so_snd);
2704 if (acked > so->so_snd.sb_cc) {
2705 tp->snd_wnd -= so->so_snd.sb_cc;
2706 mfree = sbcut_locked(&so->so_snd,
2707 (int)so->so_snd.sb_cc);
2710 mfree = sbcut_locked(&so->so_snd, acked);
2711 tp->snd_wnd -= acked;
2714 /* NB: sowwakeup_locked() does an implicit unlock. */
2715 sowwakeup_locked(so);
2717 /* Detect una wraparound. */
2718 if (!IN_RECOVERY(tp->t_flags) &&
2719 SEQ_GT(tp->snd_una, tp->snd_recover) &&
2720 SEQ_LEQ(th->th_ack, tp->snd_recover))
2721 tp->snd_recover = th->th_ack - 1;
2722 /* XXXLAS: Can this be moved up into cc_post_recovery? */
2723 if (IN_RECOVERY(tp->t_flags) &&
2724 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2725 EXIT_RECOVERY(tp->t_flags);
2727 tp->snd_una = th->th_ack;
2728 if (tp->t_flags & TF_SACK_PERMIT) {
2729 if (SEQ_GT(tp->snd_una, tp->snd_recover))
2730 tp->snd_recover = tp->snd_una;
2732 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2733 tp->snd_nxt = tp->snd_una;
2735 switch (tp->t_state) {
2738 * In FIN_WAIT_1 STATE in addition to the processing
2739 * for the ESTABLISHED state if our FIN is now acknowledged
2740 * then enter FIN_WAIT_2.
2742 case TCPS_FIN_WAIT_1:
2743 if (ourfinisacked) {
2745 * If we can't receive any more
2746 * data, then closing user can proceed.
2747 * Starting the timer is contrary to the
2748 * specification, but if we don't get a FIN
2749 * we'll hang forever.
2752 * we should release the tp also, and use a
2755 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2756 soisdisconnected(so);
2757 tcp_timer_activate(tp, TT_2MSL,
2758 (tcp_fast_finwait2_recycle ?
2759 tcp_finwait2_timeout :
2762 tcp_state_change(tp, TCPS_FIN_WAIT_2);
2767 * In CLOSING STATE in addition to the processing for
2768 * the ESTABLISHED state if the ACK acknowledges our FIN
2769 * then enter the TIME-WAIT state, otherwise ignore
2773 if (ourfinisacked) {
2774 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2776 INP_INFO_WUNLOCK(&V_tcbinfo);
2783 * In LAST_ACK, we may still be waiting for data to drain
2784 * and/or to be acked, as well as for the ack of our FIN.
2785 * If our FIN is now acknowledged, delete the TCB,
2786 * enter the closed state and return.
2789 if (ourfinisacked) {
2790 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2799 INP_WLOCK_ASSERT(tp->t_inpcb);
2802 * Update window information.
2803 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2805 if ((thflags & TH_ACK) &&
2806 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2807 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2808 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2809 /* keep track of pure window updates */
2811 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2812 TCPSTAT_INC(tcps_rcvwinupd);
2813 tp->snd_wnd = tiwin;
2814 tp->snd_wl1 = th->th_seq;
2815 tp->snd_wl2 = th->th_ack;
2816 if (tp->snd_wnd > tp->max_sndwnd)
2817 tp->max_sndwnd = tp->snd_wnd;
2822 * Process segments with URG.
2824 if ((thflags & TH_URG) && th->th_urp &&
2825 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2827 * This is a kludge, but if we receive and accept
2828 * random urgent pointers, we'll crash in
2829 * soreceive. It's hard to imagine someone
2830 * actually wanting to send this much urgent data.
2832 SOCKBUF_LOCK(&so->so_rcv);
2833 if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2834 th->th_urp = 0; /* XXX */
2835 thflags &= ~TH_URG; /* XXX */
2836 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */
2837 goto dodata; /* XXX */
2840 * If this segment advances the known urgent pointer,
2841 * then mark the data stream. This should not happen
2842 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2843 * a FIN has been received from the remote side.
2844 * In these states we ignore the URG.
2846 * According to RFC961 (Assigned Protocols),
2847 * the urgent pointer points to the last octet
2848 * of urgent data. We continue, however,
2849 * to consider it to indicate the first octet
2850 * of data past the urgent section as the original
2851 * spec states (in one of two places).
2853 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2854 tp->rcv_up = th->th_seq + th->th_urp;
2855 so->so_oobmark = so->so_rcv.sb_cc +
2856 (tp->rcv_up - tp->rcv_nxt) - 1;
2857 if (so->so_oobmark == 0)
2858 so->so_rcv.sb_state |= SBS_RCVATMARK;
2860 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2862 SOCKBUF_UNLOCK(&so->so_rcv);
2864 * Remove out of band data so doesn't get presented to user.
2865 * This can happen independent of advancing the URG pointer,
2866 * but if two URG's are pending at once, some out-of-band
2867 * data may creep in... ick.
2869 if (th->th_urp <= (u_long)tlen &&
2870 !(so->so_options & SO_OOBINLINE)) {
2871 /* hdr drop is delayed */
2872 tcp_pulloutofband(so, th, m, drop_hdrlen);
2876 * If no out of band data is expected,
2877 * pull receive urgent pointer along
2878 * with the receive window.
2880 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2881 tp->rcv_up = tp->rcv_nxt;
2884 INP_WLOCK_ASSERT(tp->t_inpcb);
2887 * Process the segment text, merging it into the TCP sequencing queue,
2888 * and arranging for acknowledgment of receipt if necessary.
2889 * This process logically involves adjusting tp->rcv_wnd as data
2890 * is presented to the user (this happens in tcp_usrreq.c,
2891 * case PRU_RCVD). If a FIN has already been received on this
2892 * connection then we just ignore the text.
2894 if ((tlen || (thflags & TH_FIN)) &&
2895 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2896 tcp_seq save_start = th->th_seq;
2897 m_adj(m, drop_hdrlen); /* delayed header drop */
2899 * Insert segment which includes th into TCP reassembly queue
2900 * with control block tp. Set thflags to whether reassembly now
2901 * includes a segment with FIN. This handles the common case
2902 * inline (segment is the next to be received on an established
2903 * connection, and the queue is empty), avoiding linkage into
2904 * and removal from the queue and repetition of various
2906 * Set DELACK for segments received in order, but ack
2907 * immediately when segments are out of order (so
2908 * fast retransmit can work).
2910 if (th->th_seq == tp->rcv_nxt && tp->t_segq == NULL &&
2911 TCPS_HAVEESTABLISHED(tp->t_state)) {
2912 if (DELAY_ACK(tp, tlen))
2913 tp->t_flags |= TF_DELACK;
2915 tp->t_flags |= TF_ACKNOW;
2916 tp->rcv_nxt += tlen;
2917 thflags = th->th_flags & TH_FIN;
2918 TCPSTAT_INC(tcps_rcvpack);
2919 TCPSTAT_ADD(tcps_rcvbyte, tlen);
2921 SOCKBUF_LOCK(&so->so_rcv);
2922 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
2925 sbappendstream_locked(&so->so_rcv, m);
2926 /* NB: sorwakeup_locked() does an implicit unlock. */
2927 sorwakeup_locked(so);
2930 * XXX: Due to the header drop above "th" is
2931 * theoretically invalid by now. Fortunately
2932 * m_adj() doesn't actually frees any mbufs
2933 * when trimming from the head.
2935 thflags = tcp_reass(tp, th, &tlen, m);
2936 tp->t_flags |= TF_ACKNOW;
2938 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT))
2939 tcp_update_sack_list(tp, save_start, save_start + tlen);
2942 * Note the amount of data that peer has sent into
2943 * our window, in order to estimate the sender's
2947 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
2948 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2950 len = so->so_rcv.sb_hiwat;
2958 * If FIN is received ACK the FIN and let the user know
2959 * that the connection is closing.
2961 if (thflags & TH_FIN) {
2962 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2965 * If connection is half-synchronized
2966 * (ie NEEDSYN flag on) then delay ACK,
2967 * so it may be piggybacked when SYN is sent.
2968 * Otherwise, since we received a FIN then no
2969 * more input can be expected, send ACK now.
2971 if (tp->t_flags & TF_NEEDSYN)
2972 tp->t_flags |= TF_DELACK;
2974 tp->t_flags |= TF_ACKNOW;
2977 switch (tp->t_state) {
2980 * In SYN_RECEIVED and ESTABLISHED STATES
2981 * enter the CLOSE_WAIT state.
2983 case TCPS_SYN_RECEIVED:
2984 tp->t_starttime = ticks;
2986 case TCPS_ESTABLISHED:
2987 tcp_state_change(tp, TCPS_CLOSE_WAIT);
2991 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2992 * enter the CLOSING state.
2994 case TCPS_FIN_WAIT_1:
2995 tcp_state_change(tp, TCPS_CLOSING);
2999 * In FIN_WAIT_2 state enter the TIME_WAIT state,
3000 * starting the time-wait timer, turning off the other
3003 case TCPS_FIN_WAIT_2:
3004 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
3005 KASSERT(ti_locked == TI_WLOCKED, ("%s: dodata "
3006 "TCP_FIN_WAIT_2 ti_locked: %d", __func__,
3010 INP_INFO_WUNLOCK(&V_tcbinfo);
3014 if (ti_locked == TI_WLOCKED)
3015 INP_INFO_WUNLOCK(&V_tcbinfo);
3016 ti_locked = TI_UNLOCKED;
3019 if (so->so_options & SO_DEBUG)
3020 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
3025 * Return any desired output.
3027 if (needoutput || (tp->t_flags & TF_ACKNOW))
3028 (void) tcp_output(tp);
3031 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d",
3032 __func__, ti_locked));
3033 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
3034 INP_WLOCK_ASSERT(tp->t_inpcb);
3036 if (tp->t_flags & TF_DELACK) {
3037 tp->t_flags &= ~TF_DELACK;
3038 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
3040 INP_WUNLOCK(tp->t_inpcb);
3045 * Generate an ACK dropping incoming segment if it occupies
3046 * sequence space, where the ACK reflects our state.
3048 * We can now skip the test for the RST flag since all
3049 * paths to this code happen after packets containing
3050 * RST have been dropped.
3052 * In the SYN-RECEIVED state, don't send an ACK unless the
3053 * segment we received passes the SYN-RECEIVED ACK test.
3054 * If it fails send a RST. This breaks the loop in the
3055 * "LAND" DoS attack, and also prevents an ACK storm
3056 * between two listening ports that have been sent forged
3057 * SYN segments, each with the source address of the other.
3059 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
3060 (SEQ_GT(tp->snd_una, th->th_ack) ||
3061 SEQ_GT(th->th_ack, tp->snd_max)) ) {
3062 rstreason = BANDLIM_RST_OPENPORT;
3066 if (so->so_options & SO_DEBUG)
3067 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3070 if (ti_locked == TI_WLOCKED)
3071 INP_INFO_WUNLOCK(&V_tcbinfo);
3072 ti_locked = TI_UNLOCKED;
3074 tp->t_flags |= TF_ACKNOW;
3075 (void) tcp_output(tp);
3076 INP_WUNLOCK(tp->t_inpcb);
3081 if (ti_locked == TI_WLOCKED)
3082 INP_INFO_WUNLOCK(&V_tcbinfo);
3083 ti_locked = TI_UNLOCKED;
3086 tcp_dropwithreset(m, th, tp, tlen, rstreason);
3087 INP_WUNLOCK(tp->t_inpcb);
3089 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
3093 if (ti_locked == TI_WLOCKED) {
3094 INP_INFO_WUNLOCK(&V_tcbinfo);
3095 ti_locked = TI_UNLOCKED;
3099 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
3103 * Drop space held by incoming segment and return.
3106 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
3107 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3111 INP_WUNLOCK(tp->t_inpcb);
3116 * Issue RST and make ACK acceptable to originator of segment.
3117 * The mbuf must still include the original packet header.
3121 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
3122 int tlen, int rstreason)
3128 struct ip6_hdr *ip6;
3132 INP_WLOCK_ASSERT(tp->t_inpcb);
3135 /* Don't bother if destination was broadcast/multicast. */
3136 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
3139 if (mtod(m, struct ip *)->ip_v == 6) {
3140 ip6 = mtod(m, struct ip6_hdr *);
3141 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3142 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3144 /* IPv6 anycast check is done at tcp6_input() */
3147 #if defined(INET) && defined(INET6)
3152 ip = mtod(m, struct ip *);
3153 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3154 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3155 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3156 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3161 /* Perform bandwidth limiting. */
3162 if (badport_bandlim(rstreason) < 0)
3165 /* tcp_respond consumes the mbuf chain. */
3166 if (th->th_flags & TH_ACK) {
3167 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0,
3168 th->th_ack, TH_RST);
3170 if (th->th_flags & TH_SYN)
3172 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
3173 (tcp_seq)0, TH_RST|TH_ACK);
3181 * Parse TCP options and place in tcpopt.
3184 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags)
3189 for (; cnt > 0; cnt -= optlen, cp += optlen) {
3191 if (opt == TCPOPT_EOL)
3193 if (opt == TCPOPT_NOP)
3199 if (optlen < 2 || optlen > cnt)
3204 if (optlen != TCPOLEN_MAXSEG)
3206 if (!(flags & TO_SYN))
3208 to->to_flags |= TOF_MSS;
3209 bcopy((char *)cp + 2,
3210 (char *)&to->to_mss, sizeof(to->to_mss));
3211 to->to_mss = ntohs(to->to_mss);
3214 if (optlen != TCPOLEN_WINDOW)
3216 if (!(flags & TO_SYN))
3218 to->to_flags |= TOF_SCALE;
3219 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT);
3221 case TCPOPT_TIMESTAMP:
3222 if (optlen != TCPOLEN_TIMESTAMP)
3224 to->to_flags |= TOF_TS;
3225 bcopy((char *)cp + 2,
3226 (char *)&to->to_tsval, sizeof(to->to_tsval));
3227 to->to_tsval = ntohl(to->to_tsval);
3228 bcopy((char *)cp + 6,
3229 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
3230 to->to_tsecr = ntohl(to->to_tsecr);
3232 #ifdef TCP_SIGNATURE
3234 * XXX In order to reply to a host which has set the
3235 * TCP_SIGNATURE option in its initial SYN, we have to
3236 * record the fact that the option was observed here
3237 * for the syncache code to perform the correct response.
3239 case TCPOPT_SIGNATURE:
3240 if (optlen != TCPOLEN_SIGNATURE)
3242 to->to_flags |= TOF_SIGNATURE;
3243 to->to_signature = cp + 2;
3246 case TCPOPT_SACK_PERMITTED:
3247 if (optlen != TCPOLEN_SACK_PERMITTED)
3249 if (!(flags & TO_SYN))
3253 to->to_flags |= TOF_SACKPERM;
3256 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
3260 to->to_flags |= TOF_SACK;
3261 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
3262 to->to_sacks = cp + 2;
3263 TCPSTAT_INC(tcps_sack_rcv_blocks);
3272 * Pull out of band byte out of a segment so
3273 * it doesn't appear in the user's data queue.
3274 * It is still reflected in the segment length for
3275 * sequencing purposes.
3278 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m,
3281 int cnt = off + th->th_urp - 1;
3284 if (m->m_len > cnt) {
3285 char *cp = mtod(m, caddr_t) + cnt;
3286 struct tcpcb *tp = sototcpcb(so);
3288 INP_WLOCK_ASSERT(tp->t_inpcb);
3291 tp->t_oobflags |= TCPOOB_HAVEDATA;
3292 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3294 if (m->m_flags & M_PKTHDR)
3303 panic("tcp_pulloutofband");
3307 * Collect new round-trip time estimate
3308 * and update averages and current timeout.
3311 tcp_xmit_timer(struct tcpcb *tp, int rtt)
3315 INP_WLOCK_ASSERT(tp->t_inpcb);
3317 TCPSTAT_INC(tcps_rttupdated);
3319 if (tp->t_srtt != 0) {
3321 * srtt is stored as fixed point with 5 bits after the
3322 * binary point (i.e., scaled by 8). The following magic
3323 * is equivalent to the smoothing algorithm in rfc793 with
3324 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3325 * point). Adjust rtt to origin 0.
3327 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
3328 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
3330 if ((tp->t_srtt += delta) <= 0)
3334 * We accumulate a smoothed rtt variance (actually, a
3335 * smoothed mean difference), then set the retransmit
3336 * timer to smoothed rtt + 4 times the smoothed variance.
3337 * rttvar is stored as fixed point with 4 bits after the
3338 * binary point (scaled by 16). The following is
3339 * equivalent to rfc793 smoothing with an alpha of .75
3340 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
3341 * rfc793's wired-in beta.
3345 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
3346 if ((tp->t_rttvar += delta) <= 0)
3348 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
3349 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3352 * No rtt measurement yet - use the unsmoothed rtt.
3353 * Set the variance to half the rtt (so our first
3354 * retransmit happens at 3*rtt).
3356 tp->t_srtt = rtt << TCP_RTT_SHIFT;
3357 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
3358 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3364 * the retransmit should happen at rtt + 4 * rttvar.
3365 * Because of the way we do the smoothing, srtt and rttvar
3366 * will each average +1/2 tick of bias. When we compute
3367 * the retransmit timer, we want 1/2 tick of rounding and
3368 * 1 extra tick because of +-1/2 tick uncertainty in the
3369 * firing of the timer. The bias will give us exactly the
3370 * 1.5 tick we need. But, because the bias is
3371 * statistical, we have to test that we don't drop below
3372 * the minimum feasible timer (which is 2 ticks).
3374 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3375 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3378 * We received an ack for a packet that wasn't retransmitted;
3379 * it is probably safe to discard any error indications we've
3380 * received recently. This isn't quite right, but close enough
3381 * for now (a route might have failed after we sent a segment,
3382 * and the return path might not be symmetrical).
3384 tp->t_softerror = 0;
3388 * Determine a reasonable value for maxseg size.
3389 * If the route is known, check route for mtu.
3390 * If none, use an mss that can be handled on the outgoing interface
3391 * without forcing IP to fragment. If no route is found, route has no mtu,
3392 * or the destination isn't local, use a default, hopefully conservative
3393 * size (usually 512 or the default IP max size, but no more than the mtu
3394 * of the interface), as we can't discover anything about intervening
3395 * gateways or networks. We also initialize the congestion/slow start
3396 * window to be a single segment if the destination isn't local.
3397 * While looking at the routing entry, we also initialize other path-dependent
3398 * parameters from pre-set or cached values in the routing entry.
3400 * Also take into account the space needed for options that we
3401 * send regularly. Make maxseg shorter by that amount to assure
3402 * that we can send maxseg amount of data even when the options
3403 * are present. Store the upper limit of the length of options plus
3406 * NOTE that this routine is only called when we process an incoming
3407 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS
3408 * settings are handled in tcp_mssopt().
3411 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer,
3412 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap)
3416 struct inpcb *inp = tp->t_inpcb;
3417 struct hc_metrics_lite metrics;
3420 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
3421 size_t min_protoh = isipv6 ?
3422 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
3423 sizeof (struct tcpiphdr);
3425 const size_t min_protoh = sizeof(struct tcpiphdr);
3428 INP_WLOCK_ASSERT(tp->t_inpcb);
3430 if (mtuoffer != -1) {
3431 KASSERT(offer == -1, ("%s: conflict", __func__));
3432 offer = mtuoffer - min_protoh;
3439 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap);
3440 tp->t_maxopd = tp->t_maxseg = V_tcp_v6mssdflt;
3443 #if defined(INET) && defined(INET6)
3448 maxmtu = tcp_maxmtu(&inp->inp_inc, cap);
3449 tp->t_maxopd = tp->t_maxseg = V_tcp_mssdflt;
3454 * No route to sender, stay with default mss and return.
3458 * In case we return early we need to initialize metrics
3459 * to a defined state as tcp_hc_get() would do for us
3460 * if there was no cache hit.
3462 if (metricptr != NULL)
3463 bzero(metricptr, sizeof(struct hc_metrics_lite));
3467 /* What have we got? */
3471 * Offer == 0 means that there was no MSS on the SYN
3472 * segment, in this case we use tcp_mssdflt as
3473 * already assigned to t_maxopd above.
3475 offer = tp->t_maxopd;
3480 * Offer == -1 means that we didn't receive SYN yet.
3486 * Prevent DoS attack with too small MSS. Round up
3487 * to at least minmss.
3489 offer = max(offer, V_tcp_minmss);
3493 * rmx information is now retrieved from tcp_hostcache.
3495 tcp_hc_get(&inp->inp_inc, &metrics);
3496 if (metricptr != NULL)
3497 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite));
3500 * If there's a discovered mtu int tcp hostcache, use it
3501 * else, use the link mtu.
3503 if (metrics.rmx_mtu)
3504 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
3508 mss = maxmtu - min_protoh;
3509 if (!V_path_mtu_discovery &&
3510 !in6_localaddr(&inp->in6p_faddr))
3511 mss = min(mss, V_tcp_v6mssdflt);
3514 #if defined(INET) && defined(INET6)
3519 mss = maxmtu - min_protoh;
3520 if (!V_path_mtu_discovery &&
3521 !in_localaddr(inp->inp_faddr))
3522 mss = min(mss, V_tcp_mssdflt);
3526 * XXX - The above conditional (mss = maxmtu - min_protoh)
3527 * probably violates the TCP spec.
3528 * The problem is that, since we don't know the
3529 * other end's MSS, we are supposed to use a conservative
3530 * default. But, if we do that, then MTU discovery will
3531 * never actually take place, because the conservative
3532 * default is much less than the MTUs typically seen
3533 * on the Internet today. For the moment, we'll sweep
3534 * this under the carpet.
3536 * The conservative default might not actually be a problem
3537 * if the only case this occurs is when sending an initial
3538 * SYN with options and data to a host we've never talked
3539 * to before. Then, they will reply with an MSS value which
3540 * will get recorded and the new parameters should get
3541 * recomputed. For Further Study.
3544 mss = min(mss, offer);
3547 * Sanity check: make sure that maxopd will be large
3548 * enough to allow some data on segments even if the
3549 * all the option space is used (40bytes). Otherwise
3550 * funny things may happen in tcp_output.
3555 * maxopd stores the maximum length of data AND options
3556 * in a segment; maxseg is the amount of data in a normal
3557 * segment. We need to store this value (maxopd) apart
3558 * from maxseg, because now every segment carries options
3559 * and thus we normally have somewhat less data in segments.
3564 * origoffer==-1 indicates that no segments were received yet.
3565 * In this case we just guess.
3567 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
3569 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
3570 mss -= TCPOLEN_TSTAMP_APPA;
3576 tcp_mss(struct tcpcb *tp, int offer)
3582 struct hc_metrics_lite metrics;
3583 struct tcp_ifcap cap;
3585 KASSERT(tp != NULL, ("%s: tp == NULL", __func__));
3587 bzero(&cap, sizeof(cap));
3588 tcp_mss_update(tp, offer, -1, &metrics, &cap);
3594 * If there's a pipesize, change the socket buffer to that size,
3595 * don't change if sb_hiwat is different than default (then it
3596 * has been changed on purpose with setsockopt).
3597 * Make the socket buffers an integral number of mss units;
3598 * if the mss is larger than the socket buffer, decrease the mss.
3600 so = inp->inp_socket;
3601 SOCKBUF_LOCK(&so->so_snd);
3602 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe)
3603 bufsize = metrics.rmx_sendpipe;
3605 bufsize = so->so_snd.sb_hiwat;
3609 bufsize = roundup(bufsize, mss);
3610 if (bufsize > sb_max)
3612 if (bufsize > so->so_snd.sb_hiwat)
3613 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
3615 SOCKBUF_UNLOCK(&so->so_snd);
3618 SOCKBUF_LOCK(&so->so_rcv);
3619 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe)
3620 bufsize = metrics.rmx_recvpipe;
3622 bufsize = so->so_rcv.sb_hiwat;
3623 if (bufsize > mss) {
3624 bufsize = roundup(bufsize, mss);
3625 if (bufsize > sb_max)
3627 if (bufsize > so->so_rcv.sb_hiwat)
3628 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
3630 SOCKBUF_UNLOCK(&so->so_rcv);
3632 /* Check the interface for TSO capabilities. */
3633 if (cap.ifcap & CSUM_TSO) {
3634 tp->t_flags |= TF_TSO;
3635 tp->t_tsomax = cap.tsomax;
3640 * Determine the MSS option to send on an outgoing SYN.
3643 tcp_mssopt(struct in_conninfo *inc)
3650 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3653 if (inc->inc_flags & INC_ISIPV6) {
3654 mss = V_tcp_v6mssdflt;
3655 maxmtu = tcp_maxmtu6(inc, NULL);
3656 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3659 #if defined(INET) && defined(INET6)
3664 mss = V_tcp_mssdflt;
3665 maxmtu = tcp_maxmtu(inc, NULL);
3666 min_protoh = sizeof(struct tcpiphdr);
3669 #if defined(INET6) || defined(INET)
3670 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3673 if (maxmtu && thcmtu)
3674 mss = min(maxmtu, thcmtu) - min_protoh;
3675 else if (maxmtu || thcmtu)
3676 mss = max(maxmtu, thcmtu) - min_protoh;
3683 * On a partial ack arrives, force the retransmission of the
3684 * next unacknowledged segment. Do not clear tp->t_dupacks.
3685 * By setting snd_nxt to ti_ack, this forces retransmission timer to
3689 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th)
3691 tcp_seq onxt = tp->snd_nxt;
3692 u_long ocwnd = tp->snd_cwnd;
3694 INP_WLOCK_ASSERT(tp->t_inpcb);
3696 tcp_timer_activate(tp, TT_REXMT, 0);
3698 tp->snd_nxt = th->th_ack;
3700 * Set snd_cwnd to one segment beyond acknowledged offset.
3701 * (tp->snd_una has not yet been updated when this function is called.)
3703 tp->snd_cwnd = tp->t_maxseg + BYTES_THIS_ACK(tp, th);
3704 tp->t_flags |= TF_ACKNOW;
3705 (void) tcp_output(tp);
3706 tp->snd_cwnd = ocwnd;
3707 if (SEQ_GT(onxt, tp->snd_nxt))
3710 * Partial window deflation. Relies on fact that tp->snd_una
3713 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th))
3714 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th);
3717 tp->snd_cwnd += tp->t_maxseg;