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("$FreeBSD$");
54 #include "opt_inet6.h"
55 #include "opt_ipsec.h"
56 #include "opt_tcpdebug.h"
58 #include <sys/param.h>
59 #include <sys/kernel.h>
61 #include <sys/hhook.h>
63 #include <sys/malloc.h>
65 #include <sys/proc.h> /* for proc0 declaration */
66 #include <sys/protosw.h>
68 #include <sys/signalvar.h>
69 #include <sys/socket.h>
70 #include <sys/socketvar.h>
71 #include <sys/sysctl.h>
72 #include <sys/syslog.h>
73 #include <sys/systm.h>
75 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
80 #include <net/if_var.h>
81 #include <net/route.h>
84 #define TCPSTATES /* for logging */
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/ip.h>
91 #include <netinet/ip_icmp.h> /* required for icmp_var.h */
92 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */
93 #include <netinet/ip_var.h>
94 #include <netinet/ip_options.h>
95 #include <netinet/ip6.h>
96 #include <netinet/icmp6.h>
97 #include <netinet6/in6_pcb.h>
98 #include <netinet6/in6_var.h>
99 #include <netinet6/ip6_var.h>
100 #include <netinet6/nd6.h>
102 #include <netinet/tcp_fastopen.h>
104 #include <netinet/tcp.h>
105 #include <netinet/tcp_fsm.h>
106 #include <netinet/tcp_seq.h>
107 #include <netinet/tcp_timer.h>
108 #include <netinet/tcp_var.h>
109 #include <netinet6/tcp6_var.h>
110 #include <netinet/tcpip.h>
111 #include <netinet/cc/cc.h>
113 #include <netinet/tcp_pcap.h>
115 #include <netinet/tcp_syncache.h>
117 #include <netinet/tcp_debug.h>
118 #endif /* TCPDEBUG */
120 #include <netinet/tcp_offload.h>
124 #include <netipsec/ipsec.h>
125 #include <netipsec/ipsec6.h>
128 #include <machine/in_cksum.h>
130 #include <security/mac/mac_framework.h>
132 const int tcprexmtthresh = 3;
134 int tcp_log_in_vain = 0;
135 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
137 "Log all incoming TCP segments to closed ports");
139 VNET_DEFINE(int, blackhole) = 0;
140 #define V_blackhole VNET(blackhole)
141 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW,
142 &VNET_NAME(blackhole), 0,
143 "Do not send RST on segments to closed ports");
145 VNET_DEFINE(int, tcp_delack_enabled) = 1;
146 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW,
147 &VNET_NAME(tcp_delack_enabled), 0,
148 "Delay ACK to try and piggyback it onto a data packet");
150 VNET_DEFINE(int, drop_synfin) = 0;
151 #define V_drop_synfin VNET(drop_synfin)
152 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_VNET | CTLFLAG_RW,
153 &VNET_NAME(drop_synfin), 0,
154 "Drop TCP packets with SYN+FIN set");
156 VNET_DEFINE(int, tcp_do_rfc6675_pipe) = 0;
157 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc6675_pipe, CTLFLAG_VNET | CTLFLAG_RW,
158 &VNET_NAME(tcp_do_rfc6675_pipe), 0,
159 "Use calculated pipe/in-flight bytes per RFC 6675");
161 VNET_DEFINE(int, tcp_do_rfc3042) = 1;
162 #define V_tcp_do_rfc3042 VNET(tcp_do_rfc3042)
163 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW,
164 &VNET_NAME(tcp_do_rfc3042), 0,
165 "Enable RFC 3042 (Limited Transmit)");
167 VNET_DEFINE(int, tcp_do_rfc3390) = 1;
168 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW,
169 &VNET_NAME(tcp_do_rfc3390), 0,
170 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
172 VNET_DEFINE(int, tcp_initcwnd_segments) = 10;
173 SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments,
174 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0,
175 "Slow-start flight size (initial congestion window) in number of segments");
177 VNET_DEFINE(int, tcp_do_rfc3465) = 1;
178 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW,
179 &VNET_NAME(tcp_do_rfc3465), 0,
180 "Enable RFC 3465 (Appropriate Byte Counting)");
182 VNET_DEFINE(int, tcp_abc_l_var) = 2;
183 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW,
184 &VNET_NAME(tcp_abc_l_var), 2,
185 "Cap the max cwnd increment during slow-start to this number of segments");
187 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, CTLFLAG_RW, 0, "TCP ECN");
189 VNET_DEFINE(int, tcp_do_ecn) = 2;
190 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_VNET | CTLFLAG_RW,
191 &VNET_NAME(tcp_do_ecn), 0,
194 VNET_DEFINE(int, tcp_ecn_maxretries) = 1;
195 SYSCTL_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_VNET | CTLFLAG_RW,
196 &VNET_NAME(tcp_ecn_maxretries), 0,
197 "Max retries before giving up on ECN");
199 VNET_DEFINE(int, tcp_insecure_syn) = 0;
200 #define V_tcp_insecure_syn VNET(tcp_insecure_syn)
201 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW,
202 &VNET_NAME(tcp_insecure_syn), 0,
203 "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets");
205 VNET_DEFINE(int, tcp_insecure_rst) = 0;
206 #define V_tcp_insecure_rst VNET(tcp_insecure_rst)
207 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW,
208 &VNET_NAME(tcp_insecure_rst), 0,
209 "Follow RFC793 instead of RFC5961 criteria for accepting RST packets");
211 VNET_DEFINE(int, tcp_recvspace) = 1024*64;
212 #define V_tcp_recvspace VNET(tcp_recvspace)
213 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW,
214 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size");
216 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1;
217 #define V_tcp_do_autorcvbuf VNET(tcp_do_autorcvbuf)
218 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW,
219 &VNET_NAME(tcp_do_autorcvbuf), 0,
220 "Enable automatic receive buffer sizing");
222 VNET_DEFINE(int, tcp_autorcvbuf_inc) = 16*1024;
223 #define V_tcp_autorcvbuf_inc VNET(tcp_autorcvbuf_inc)
224 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_VNET | CTLFLAG_RW,
225 &VNET_NAME(tcp_autorcvbuf_inc), 0,
226 "Incrementor step size of automatic receive buffer");
228 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024;
229 #define V_tcp_autorcvbuf_max VNET(tcp_autorcvbuf_max)
230 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW,
231 &VNET_NAME(tcp_autorcvbuf_max), 0,
232 "Max size of automatic receive buffer");
234 VNET_DEFINE(struct inpcbhead, tcb);
235 #define tcb6 tcb /* for KAME src sync over BSD*'s */
236 VNET_DEFINE(struct inpcbinfo, tcbinfo);
239 * TCP statistics are stored in an array of counter(9)s, which size matches
240 * size of struct tcpstat. TCP running connection count is a regular array.
242 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat);
243 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat,
244 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
245 VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]);
246 SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD |
247 CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES,
248 "TCP connection counts by TCP state");
251 tcp_vnet_init(const void *unused)
254 COUNTER_ARRAY_ALLOC(V_tcps_states, TCP_NSTATES, M_WAITOK);
255 VNET_PCPUSTAT_ALLOC(tcpstat, M_WAITOK);
257 VNET_SYSINIT(tcp_vnet_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
258 tcp_vnet_init, NULL);
262 tcp_vnet_uninit(const void *unused)
265 COUNTER_ARRAY_FREE(V_tcps_states, TCP_NSTATES);
266 VNET_PCPUSTAT_FREE(tcpstat);
268 VNET_SYSUNINIT(tcp_vnet_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
269 tcp_vnet_uninit, NULL);
273 * Kernel module interface for updating tcpstat. The argument is an index
274 * into tcpstat treated as an array.
277 kmod_tcpstat_inc(int statnum)
280 counter_u64_add(VNET(tcpstat)[statnum], 1);
285 * Wrapper for the TCP established input helper hook.
288 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to)
290 struct tcp_hhook_data hhook_data;
292 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) {
297 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data,
304 * CC wrapper hook functions
307 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs,
310 INP_WLOCK_ASSERT(tp->t_inpcb);
312 tp->ccv->nsegs = nsegs;
313 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th);
314 if (tp->snd_cwnd <= tp->snd_wnd)
315 tp->ccv->flags |= CCF_CWND_LIMITED;
317 tp->ccv->flags &= ~CCF_CWND_LIMITED;
319 if (type == CC_ACK) {
320 if (tp->snd_cwnd > tp->snd_ssthresh) {
321 tp->t_bytes_acked += min(tp->ccv->bytes_this_ack,
322 nsegs * V_tcp_abc_l_var * tcp_maxseg(tp));
323 if (tp->t_bytes_acked >= tp->snd_cwnd) {
324 tp->t_bytes_acked -= tp->snd_cwnd;
325 tp->ccv->flags |= CCF_ABC_SENTAWND;
328 tp->ccv->flags &= ~CCF_ABC_SENTAWND;
329 tp->t_bytes_acked = 0;
333 if (CC_ALGO(tp)->ack_received != NULL) {
334 /* XXXLAS: Find a way to live without this */
335 tp->ccv->curack = th->th_ack;
336 CC_ALGO(tp)->ack_received(tp->ccv, type);
341 cc_conn_init(struct tcpcb *tp)
343 struct hc_metrics_lite metrics;
344 struct inpcb *inp = tp->t_inpcb;
348 INP_WLOCK_ASSERT(tp->t_inpcb);
350 tcp_hc_get(&inp->inp_inc, &metrics);
351 maxseg = tcp_maxseg(tp);
353 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) {
355 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
356 TCPSTAT_INC(tcps_usedrtt);
357 if (metrics.rmx_rttvar) {
358 tp->t_rttvar = metrics.rmx_rttvar;
359 TCPSTAT_INC(tcps_usedrttvar);
361 /* default variation is +- 1 rtt */
363 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
365 TCPT_RANGESET(tp->t_rxtcur,
366 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
367 tp->t_rttmin, TCPTV_REXMTMAX);
369 if (metrics.rmx_ssthresh) {
371 * There's some sort of gateway or interface
372 * buffer limit on the path. Use this to set
373 * the slow start threshold, but set the
374 * threshold to no less than 2*mss.
376 tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh);
377 TCPSTAT_INC(tcps_usedssthresh);
381 * Set the initial slow-start flight size.
383 * RFC5681 Section 3.1 specifies the default conservative values.
384 * RFC3390 specifies slightly more aggressive values.
385 * RFC6928 increases it to ten segments.
386 * Support for user specified value for initial flight size.
388 * If a SYN or SYN/ACK was lost and retransmitted, we have to
389 * reduce the initial CWND to one segment as congestion is likely
390 * requiring us to be cautious.
392 if (tp->snd_cwnd == 1)
393 tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */
394 else if (V_tcp_initcwnd_segments)
395 tp->snd_cwnd = min(V_tcp_initcwnd_segments * maxseg,
396 max(2 * maxseg, V_tcp_initcwnd_segments * 1460));
397 else if (V_tcp_do_rfc3390)
398 tp->snd_cwnd = min(4 * maxseg, max(2 * maxseg, 4380));
400 /* Per RFC5681 Section 3.1 */
402 tp->snd_cwnd = 2 * maxseg;
403 else if (maxseg > 1095)
404 tp->snd_cwnd = 3 * maxseg;
406 tp->snd_cwnd = 4 * maxseg;
409 if (CC_ALGO(tp)->conn_init != NULL)
410 CC_ALGO(tp)->conn_init(tp->ccv);
414 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type)
418 INP_WLOCK_ASSERT(tp->t_inpcb);
422 if (!IN_FASTRECOVERY(tp->t_flags)) {
423 tp->snd_recover = tp->snd_max;
424 if (tp->t_flags & TF_ECN_PERMIT)
425 tp->t_flags |= TF_ECN_SND_CWR;
429 if (!IN_CONGRECOVERY(tp->t_flags)) {
430 TCPSTAT_INC(tcps_ecn_rcwnd);
431 tp->snd_recover = tp->snd_max;
432 if (tp->t_flags & TF_ECN_PERMIT)
433 tp->t_flags |= TF_ECN_SND_CWR;
437 maxseg = tcp_maxseg(tp);
439 tp->t_bytes_acked = 0;
440 EXIT_RECOVERY(tp->t_flags);
441 if (CC_ALGO(tp)->cong_signal == NULL) {
443 * RFC5681 Section 3.1
444 * ssthresh = max (FlightSize / 2, 2*SMSS) eq (4)
447 max((tp->snd_max - tp->snd_una) / 2 / maxseg, 2)
449 tp->snd_cwnd = maxseg;
453 TCPSTAT_INC(tcps_sndrexmitbad);
454 /* RTO was unnecessary, so reset everything. */
455 tp->snd_cwnd = tp->snd_cwnd_prev;
456 tp->snd_ssthresh = tp->snd_ssthresh_prev;
457 tp->snd_recover = tp->snd_recover_prev;
458 if (tp->t_flags & TF_WASFRECOVERY)
459 ENTER_FASTRECOVERY(tp->t_flags);
460 if (tp->t_flags & TF_WASCRECOVERY)
461 ENTER_CONGRECOVERY(tp->t_flags);
462 tp->snd_nxt = tp->snd_max;
463 tp->t_flags &= ~TF_PREVVALID;
468 if (CC_ALGO(tp)->cong_signal != NULL) {
470 tp->ccv->curack = th->th_ack;
471 CC_ALGO(tp)->cong_signal(tp->ccv, type);
476 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th)
478 INP_WLOCK_ASSERT(tp->t_inpcb);
480 /* XXXLAS: KASSERT that we're in recovery? */
482 if (CC_ALGO(tp)->post_recovery != NULL) {
483 tp->ccv->curack = th->th_ack;
484 CC_ALGO(tp)->post_recovery(tp->ccv);
486 /* XXXLAS: EXIT_RECOVERY ? */
487 tp->t_bytes_acked = 0;
492 tcp_signature_verify_input(struct mbuf *m, int off0, int tlen, int optlen,
493 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag)
497 tcp_fields_to_net(th);
498 ret = tcp_signature_verify(m, off0, tlen, optlen, to, th, tcpbflag);
499 tcp_fields_to_host(th);
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.
513 #define DELAY_ACK(tp, tlen) \
514 ((!tcp_timer_active(tp, TT_DELACK) && \
515 (tp->t_flags & TF_RXWIN0SENT) == 0) && \
516 (tlen <= tp->t_maxseg) && \
517 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
520 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos)
522 INP_WLOCK_ASSERT(tp->t_inpcb);
524 if (CC_ALGO(tp)->ecnpkt_handler != NULL) {
525 switch (iptos & IPTOS_ECN_MASK) {
527 tp->ccv->flags |= CCF_IPHDR_CE;
530 tp->ccv->flags &= ~CCF_IPHDR_CE;
533 tp->ccv->flags &= ~CCF_IPHDR_CE;
537 if (th->th_flags & TH_CWR)
538 tp->ccv->flags |= CCF_TCPHDR_CWR;
540 tp->ccv->flags &= ~CCF_TCPHDR_CWR;
542 if (tp->t_flags & TF_DELACK)
543 tp->ccv->flags |= CCF_DELACK;
545 tp->ccv->flags &= ~CCF_DELACK;
547 CC_ALGO(tp)->ecnpkt_handler(tp->ccv);
549 if (tp->ccv->flags & CCF_ACKNOW)
550 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
555 * TCP input handling is split into multiple parts:
556 * tcp6_input is a thin wrapper around tcp_input for the extended
557 * ip6_protox[] call format in ip6_input
558 * tcp_input handles primary segment validation, inpcb lookup and
559 * SYN processing on listen sockets
560 * tcp_do_segment processes the ACK and text of the segment for
561 * establishing, established and closing connections
565 tcp6_input(struct mbuf **mp, int *offp, int proto)
567 struct mbuf *m = *mp;
568 struct in6_ifaddr *ia6;
571 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
574 * draft-itojun-ipv6-tcp-to-anycast
575 * better place to put this in?
577 ip6 = mtod(m, struct ip6_hdr *);
578 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */);
579 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
582 ifa_free(&ia6->ia_ifa);
583 ip6 = mtod(m, struct ip6_hdr *);
584 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
585 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
586 return (IPPROTO_DONE);
589 ifa_free(&ia6->ia_ifa);
591 return (tcp_input(mp, offp, proto));
596 tcp_input(struct mbuf **mp, int *offp, int proto)
598 struct mbuf *m = *mp;
599 struct tcphdr *th = NULL;
600 struct ip *ip = NULL;
601 struct inpcb *inp = NULL;
602 struct tcpcb *tp = NULL;
603 struct socket *so = NULL;
613 int rstreason = 0; /* For badport_bandlim accounting purposes */
615 uint8_t sig_checked = 0;
618 struct m_tag *fwd_tag = NULL;
620 struct ip6_hdr *ip6 = NULL;
623 const void *ip6 = NULL;
625 struct tcpopt to; /* options in this segment */
626 char *s = NULL; /* address and port logging */
630 * The size of tcp_saveipgen must be the size of the max ip header,
633 u_char tcp_saveipgen[IP6_HDR_LEN];
634 struct tcphdr tcp_savetcp;
639 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
646 TCPSTAT_INC(tcps_rcvtotal);
650 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */
652 if (m->m_len < (sizeof(*ip6) + sizeof(*th))) {
653 m = m_pullup(m, sizeof(*ip6) + sizeof(*th));
655 TCPSTAT_INC(tcps_rcvshort);
656 return (IPPROTO_DONE);
660 ip6 = mtod(m, struct ip6_hdr *);
661 th = (struct tcphdr *)((caddr_t)ip6 + off0);
662 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
663 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) {
664 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
665 th->th_sum = m->m_pkthdr.csum_data;
667 th->th_sum = in6_cksum_pseudo(ip6, tlen,
668 IPPROTO_TCP, m->m_pkthdr.csum_data);
669 th->th_sum ^= 0xffff;
671 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen);
673 TCPSTAT_INC(tcps_rcvbadsum);
678 * Be proactive about unspecified IPv6 address in source.
679 * As we use all-zero to indicate unbounded/unconnected pcb,
680 * unspecified IPv6 address can be used to confuse us.
682 * Note that packets with unspecified IPv6 destination is
683 * already dropped in ip6_input.
685 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
689 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
692 #if defined(INET) && defined(INET6)
698 * Get IP and TCP header together in first mbuf.
699 * Note: IP leaves IP header in first mbuf.
701 if (off0 > sizeof (struct ip)) {
703 off0 = sizeof(struct ip);
705 if (m->m_len < sizeof (struct tcpiphdr)) {
706 if ((m = m_pullup(m, sizeof (struct tcpiphdr)))
708 TCPSTAT_INC(tcps_rcvshort);
709 return (IPPROTO_DONE);
712 ip = mtod(m, struct ip *);
713 th = (struct tcphdr *)((caddr_t)ip + off0);
714 tlen = ntohs(ip->ip_len) - off0;
717 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
718 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
719 th->th_sum = m->m_pkthdr.csum_data;
721 th->th_sum = in_pseudo(ip->ip_src.s_addr,
723 htonl(m->m_pkthdr.csum_data + tlen +
725 th->th_sum ^= 0xffff;
727 struct ipovly *ipov = (struct ipovly *)ip;
730 * Checksum extended TCP header and data.
733 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
734 ipov->ih_len = htons(tlen);
735 th->th_sum = in_cksum(m, len);
736 /* Reset length for SDT probes. */
737 ip->ip_len = htons(len);
740 /* Re-initialization for later version check */
741 ip->ip_v = IPVERSION;
745 TCPSTAT_INC(tcps_rcvbadsum);
752 * Check that TCP offset makes sense,
753 * pull out TCP options and adjust length. XXX
755 off = th->th_off << 2;
756 if (off < sizeof (struct tcphdr) || off > tlen) {
757 TCPSTAT_INC(tcps_rcvbadoff);
760 tlen -= off; /* tlen is used instead of ti->ti_len */
761 if (off > sizeof (struct tcphdr)) {
764 IP6_EXTHDR_CHECK(m, off0, off, IPPROTO_DONE);
765 ip6 = mtod(m, struct ip6_hdr *);
766 th = (struct tcphdr *)((caddr_t)ip6 + off0);
769 #if defined(INET) && defined(INET6)
774 if (m->m_len < sizeof(struct ip) + off) {
775 if ((m = m_pullup(m, sizeof (struct ip) + off))
777 TCPSTAT_INC(tcps_rcvshort);
778 return (IPPROTO_DONE);
780 ip = mtod(m, struct ip *);
781 th = (struct tcphdr *)((caddr_t)ip + off0);
785 optlen = off - sizeof (struct tcphdr);
786 optp = (u_char *)(th + 1);
788 thflags = th->th_flags;
791 * Convert TCP protocol specific fields to host format.
793 tcp_fields_to_host(th);
796 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options.
798 drop_hdrlen = off0 + off;
801 * Locate pcb for segment; if we're likely to add or remove a
802 * connection then first acquire pcbinfo lock. There are three cases
803 * where we might discover later we need a write lock despite the
804 * flags: ACKs moving a connection out of the syncache, ACKs for a
805 * connection in TIMEWAIT and SYNs not targeting a listening socket.
807 if ((thflags & (TH_FIN | TH_RST)) != 0) {
808 INP_INFO_RLOCK(&V_tcbinfo);
809 ti_locked = TI_RLOCKED;
811 ti_locked = TI_UNLOCKED;
814 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
818 (isipv6 && (m->m_flags & M_IP6_NEXTHOP))
820 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP))
823 #if defined(INET) && !defined(INET6)
824 (m->m_flags & M_IP_NEXTHOP)
827 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
831 if (ti_locked == TI_RLOCKED) {
832 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
834 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
838 if (isipv6 && fwd_tag != NULL) {
839 struct sockaddr_in6 *next_hop6;
841 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1);
843 * Transparently forwarded. Pretend to be the destination.
844 * Already got one like this?
846 inp = in6_pcblookup_mbuf(&V_tcbinfo,
847 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport,
848 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m);
851 * It's new. Try to find the ambushing socket.
852 * Because we've rewritten the destination address,
853 * any hardware-generated hash is ignored.
855 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src,
856 th->th_sport, &next_hop6->sin6_addr,
857 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) :
858 th->th_dport, INPLOOKUP_WILDCARD |
859 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
862 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src,
863 th->th_sport, &ip6->ip6_dst, th->th_dport,
864 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
865 m->m_pkthdr.rcvif, m);
868 #if defined(INET6) && defined(INET)
872 if (fwd_tag != NULL) {
873 struct sockaddr_in *next_hop;
875 next_hop = (struct sockaddr_in *)(fwd_tag+1);
877 * Transparently forwarded. Pretend to be the destination.
878 * already got one like this?
880 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport,
881 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB,
882 m->m_pkthdr.rcvif, m);
885 * It's new. Try to find the ambushing socket.
886 * Because we've rewritten the destination address,
887 * any hardware-generated hash is ignored.
889 inp = in_pcblookup(&V_tcbinfo, ip->ip_src,
890 th->th_sport, next_hop->sin_addr,
891 next_hop->sin_port ? ntohs(next_hop->sin_port) :
892 th->th_dport, INPLOOKUP_WILDCARD |
893 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
896 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src,
897 th->th_sport, ip->ip_dst, th->th_dport,
898 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
899 m->m_pkthdr.rcvif, m);
903 * If the INPCB does not exist then all data in the incoming
904 * segment is discarded and an appropriate RST is sent back.
905 * XXX MRT Send RST using which routing table?
909 * Log communication attempts to ports that are not
912 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) ||
913 tcp_log_in_vain == 2) {
914 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6)))
915 log(LOG_INFO, "%s; %s: Connection attempt "
916 "to closed port\n", s, __func__);
919 * When blackholing do not respond with a RST but
920 * completely ignore the segment and drop it.
922 if ((V_blackhole == 1 && (thflags & TH_SYN)) ||
926 rstreason = BANDLIM_RST_CLOSEDPORT;
929 INP_WLOCK_ASSERT(inp);
931 * While waiting for inp lock during the lookup, another thread
932 * can have dropped the inpcb, in which case we need to loop back
933 * and try to find a new inpcb to deliver to.
935 if (inp->inp_flags & INP_DROPPED) {
940 if ((inp->inp_flowtype == M_HASHTYPE_NONE) &&
941 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) &&
942 ((inp->inp_socket == NULL) ||
943 (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0)) {
944 inp->inp_flowid = m->m_pkthdr.flowid;
945 inp->inp_flowtype = M_HASHTYPE_GET(m);
949 if (isipv6 && ipsec6_in_reject(m, inp)) {
953 if (ipsec4_in_reject(m, inp) != 0) {
959 * Check the minimum TTL for socket.
961 if (inp->inp_ip_minttl != 0) {
964 if (inp->inp_ip_minttl > ip6->ip6_hlim)
968 if (inp->inp_ip_minttl > ip->ip_ttl)
973 * A previous connection in TIMEWAIT state is supposed to catch stray
974 * or duplicate segments arriving late. If this segment was a
975 * legitimate new connection attempt, the old INPCB gets removed and
976 * we can try again to find a listening socket.
978 * At this point, due to earlier optimism, we may hold only an inpcb
979 * lock, and not the inpcbinfo write lock. If so, we need to try to
980 * acquire it, or if that fails, acquire a reference on the inpcb,
981 * drop all locks, acquire a global write lock, and then re-acquire
982 * the inpcb lock. We may at that point discover that another thread
983 * has tried to free the inpcb, in which case we need to loop back
984 * and try to find a new inpcb to deliver to.
986 * XXXRW: It may be time to rethink timewait locking.
989 if (inp->inp_flags & INP_TIMEWAIT) {
990 if (ti_locked == TI_UNLOCKED) {
991 if (INP_INFO_TRY_RLOCK(&V_tcbinfo) == 0) {
994 INP_INFO_RLOCK(&V_tcbinfo);
995 ti_locked = TI_RLOCKED;
997 if (in_pcbrele_wlocked(inp)) {
1000 } else if (inp->inp_flags & INP_DROPPED) {
1006 ti_locked = TI_RLOCKED;
1008 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1010 if (thflags & TH_SYN)
1011 tcp_dooptions(&to, optp, optlen, TO_SYN);
1013 * NB: tcp_twcheck unlocks the INP and frees the mbuf.
1015 if (tcp_twcheck(inp, &to, th, m, tlen))
1017 INP_INFO_RUNLOCK(&V_tcbinfo);
1018 return (IPPROTO_DONE);
1021 * The TCPCB may no longer exist if the connection is winding
1022 * down or it is in the CLOSED state. Either way we drop the
1023 * segment and send an appropriate response.
1025 tp = intotcpcb(inp);
1026 if (tp == NULL || tp->t_state == TCPS_CLOSED) {
1027 rstreason = BANDLIM_RST_CLOSEDPORT;
1032 if (tp->t_flags & TF_TOE) {
1033 tcp_offload_input(tp, m);
1034 m = NULL; /* consumed by the TOE driver */
1040 * We've identified a valid inpcb, but it could be that we need an
1041 * inpcbinfo write lock but don't hold it. In this case, attempt to
1042 * acquire using the same strategy as the TIMEWAIT case above. If we
1043 * relock, we have to jump back to 'relocked' as the connection might
1044 * now be in TIMEWAIT.
1047 if ((thflags & (TH_FIN | TH_RST)) != 0)
1048 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1050 if (!((tp->t_state == TCPS_ESTABLISHED && (thflags & TH_SYN) == 0) ||
1051 (tp->t_state == TCPS_LISTEN && (thflags & TH_SYN) &&
1052 !IS_FASTOPEN(tp->t_flags)))) {
1053 if (ti_locked == TI_UNLOCKED) {
1054 if (INP_INFO_TRY_RLOCK(&V_tcbinfo) == 0) {
1057 INP_INFO_RLOCK(&V_tcbinfo);
1058 ti_locked = TI_RLOCKED;
1060 if (in_pcbrele_wlocked(inp)) {
1063 } else if (inp->inp_flags & INP_DROPPED) {
1070 ti_locked = TI_RLOCKED;
1072 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1076 INP_WLOCK_ASSERT(inp);
1077 if (mac_inpcb_check_deliver(inp, m))
1080 so = inp->inp_socket;
1081 KASSERT(so != NULL, ("%s: so == NULL", __func__));
1083 if (so->so_options & SO_DEBUG) {
1084 ostate = tp->t_state;
1087 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
1090 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
1093 #endif /* TCPDEBUG */
1095 * When the socket is accepting connections (the INPCB is in LISTEN
1096 * state) we look into the SYN cache if this is a new connection
1097 * attempt or the completion of a previous one.
1099 KASSERT(tp->t_state == TCPS_LISTEN || !(so->so_options & SO_ACCEPTCONN),
1100 ("%s: so accepting but tp %p not listening", __func__, tp));
1101 if (tp->t_state == TCPS_LISTEN && (so->so_options & SO_ACCEPTCONN)) {
1102 struct in_conninfo inc;
1104 bzero(&inc, sizeof(inc));
1107 inc.inc_flags |= INC_ISIPV6;
1108 inc.inc6_faddr = ip6->ip6_src;
1109 inc.inc6_laddr = ip6->ip6_dst;
1113 inc.inc_faddr = ip->ip_src;
1114 inc.inc_laddr = ip->ip_dst;
1116 inc.inc_fport = th->th_sport;
1117 inc.inc_lport = th->th_dport;
1118 inc.inc_fibnum = so->so_fibnum;
1121 * Check for an existing connection attempt in syncache if
1122 * the flag is only ACK. A successful lookup creates a new
1123 * socket appended to the listen queue in SYN_RECEIVED state.
1125 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
1127 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1129 * Parse the TCP options here because
1130 * syncookies need access to the reflected
1133 tcp_dooptions(&to, optp, optlen, 0);
1135 * NB: syncache_expand() doesn't unlock
1136 * inp and tcpinfo locks.
1138 if (!syncache_expand(&inc, &to, th, &so, m)) {
1140 * No syncache entry or ACK was not
1141 * for our SYN/ACK. Send a RST.
1142 * NB: syncache did its own logging
1143 * of the failure cause.
1145 rstreason = BANDLIM_RST_OPENPORT;
1153 * We completed the 3-way handshake
1154 * but could not allocate a socket
1155 * either due to memory shortage,
1156 * listen queue length limits or
1157 * global socket limits. Send RST
1158 * or wait and have the remote end
1159 * retransmit the ACK for another
1162 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1163 log(LOG_DEBUG, "%s; %s: Listen socket: "
1164 "Socket allocation failed due to "
1165 "limits or memory shortage, %s\n",
1167 V_tcp_sc_rst_sock_fail ?
1168 "sending RST" : "try again");
1169 if (V_tcp_sc_rst_sock_fail) {
1170 rstreason = BANDLIM_UNLIMITED;
1176 * Socket is created in state SYN_RECEIVED.
1177 * Unlock the listen socket, lock the newly
1178 * created socket and update the tp variable.
1180 INP_WUNLOCK(inp); /* listen socket */
1181 inp = sotoinpcb(so);
1183 * New connection inpcb is already locked by
1184 * syncache_expand().
1186 INP_WLOCK_ASSERT(inp);
1187 tp = intotcpcb(inp);
1188 KASSERT(tp->t_state == TCPS_SYN_RECEIVED,
1189 ("%s: ", __func__));
1190 #ifdef TCP_SIGNATURE
1191 if (sig_checked == 0) {
1192 tcp_dooptions(&to, optp, optlen,
1193 (thflags & TH_SYN) ? TO_SYN : 0);
1194 if (!tcp_signature_verify_input(m, off0, tlen,
1195 optlen, &to, th, tp->t_flags)) {
1198 * In SYN_SENT state if it receives an
1199 * RST, it is allowed for further
1202 if ((thflags & TH_RST) == 0 ||
1203 (tp->t_state == TCPS_SYN_SENT) == 0)
1211 * Process the segment and the data it
1212 * contains. tcp_do_segment() consumes
1213 * the mbuf chain and unlocks the inpcb.
1215 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen,
1217 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1218 return (IPPROTO_DONE);
1221 * Segment flag validation for new connection attempts:
1223 * Our (SYN|ACK) response was rejected.
1224 * Check with syncache and remove entry to prevent
1227 * NB: syncache_chkrst does its own logging of failure
1230 if (thflags & TH_RST) {
1231 syncache_chkrst(&inc, th);
1235 * We can't do anything without SYN.
1237 if ((thflags & TH_SYN) == 0) {
1238 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1239 log(LOG_DEBUG, "%s; %s: Listen socket: "
1240 "SYN is missing, segment ignored\n",
1242 TCPSTAT_INC(tcps_badsyn);
1246 * (SYN|ACK) is bogus on a listen socket.
1248 if (thflags & TH_ACK) {
1249 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1250 log(LOG_DEBUG, "%s; %s: Listen socket: "
1251 "SYN|ACK invalid, segment rejected\n",
1253 syncache_badack(&inc); /* XXX: Not needed! */
1254 TCPSTAT_INC(tcps_badsyn);
1255 rstreason = BANDLIM_RST_OPENPORT;
1259 * If the drop_synfin option is enabled, drop all
1260 * segments with both the SYN and FIN bits set.
1261 * This prevents e.g. nmap from identifying the
1263 * XXX: Poor reasoning. nmap has other methods
1264 * and is constantly refining its stack detection
1266 * XXX: This is a violation of the TCP specification
1267 * and was used by RFC1644.
1269 if ((thflags & TH_FIN) && V_drop_synfin) {
1270 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1271 log(LOG_DEBUG, "%s; %s: Listen socket: "
1272 "SYN|FIN segment ignored (based on "
1273 "sysctl setting)\n", s, __func__);
1274 TCPSTAT_INC(tcps_badsyn);
1278 * Segment's flags are (SYN) or (SYN|FIN).
1280 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored
1281 * as they do not affect the state of the TCP FSM.
1282 * The data pointed to by TH_URG and th_urp is ignored.
1284 KASSERT((thflags & (TH_RST|TH_ACK)) == 0,
1285 ("%s: Listen socket: TH_RST or TH_ACK set", __func__));
1286 KASSERT(thflags & (TH_SYN),
1287 ("%s: Listen socket: TH_SYN not set", __func__));
1290 * If deprecated address is forbidden,
1291 * we do not accept SYN to deprecated interface
1292 * address to prevent any new inbound connection from
1293 * getting established.
1294 * When we do not accept SYN, we send a TCP RST,
1295 * with deprecated source address (instead of dropping
1296 * it). We compromise it as it is much better for peer
1297 * to send a RST, and RST will be the final packet
1300 * If we do not forbid deprecated addresses, we accept
1301 * the SYN packet. RFC2462 does not suggest dropping
1303 * If we decipher RFC2462 5.5.4, it says like this:
1304 * 1. use of deprecated addr with existing
1305 * communication is okay - "SHOULD continue to be
1307 * 2. use of it with new communication:
1308 * (2a) "SHOULD NOT be used if alternate address
1309 * with sufficient scope is available"
1310 * (2b) nothing mentioned otherwise.
1311 * Here we fall into (2b) case as we have no choice in
1312 * our source address selection - we must obey the peer.
1314 * The wording in RFC2462 is confusing, and there are
1315 * multiple description text for deprecated address
1316 * handling - worse, they are not exactly the same.
1317 * I believe 5.5.4 is the best one, so we follow 5.5.4.
1319 if (isipv6 && !V_ip6_use_deprecated) {
1320 struct in6_ifaddr *ia6;
1322 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */);
1324 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1325 ifa_free(&ia6->ia_ifa);
1326 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1327 log(LOG_DEBUG, "%s; %s: Listen socket: "
1328 "Connection attempt to deprecated "
1329 "IPv6 address rejected\n",
1331 rstreason = BANDLIM_RST_OPENPORT;
1335 ifa_free(&ia6->ia_ifa);
1339 * Basic sanity checks on incoming SYN requests:
1340 * Don't respond if the destination is a link layer
1341 * broadcast according to RFC1122 4.2.3.10, p. 104.
1342 * If it is from this socket it must be forged.
1343 * Don't respond if the source or destination is a
1344 * global or subnet broad- or multicast address.
1345 * Note that it is quite possible to receive unicast
1346 * link-layer packets with a broadcast IP address. Use
1347 * in_broadcast() to find them.
1349 if (m->m_flags & (M_BCAST|M_MCAST)) {
1350 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1351 log(LOG_DEBUG, "%s; %s: Listen socket: "
1352 "Connection attempt from broad- or multicast "
1353 "link layer address ignored\n", s, __func__);
1358 if (th->th_dport == th->th_sport &&
1359 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) {
1360 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1361 log(LOG_DEBUG, "%s; %s: Listen socket: "
1362 "Connection attempt to/from self "
1363 "ignored\n", s, __func__);
1366 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1367 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) {
1368 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1369 log(LOG_DEBUG, "%s; %s: Listen socket: "
1370 "Connection attempt from/to multicast "
1371 "address ignored\n", s, __func__);
1376 #if defined(INET) && defined(INET6)
1381 if (th->th_dport == th->th_sport &&
1382 ip->ip_dst.s_addr == ip->ip_src.s_addr) {
1383 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1384 log(LOG_DEBUG, "%s; %s: Listen socket: "
1385 "Connection attempt from/to self "
1386 "ignored\n", s, __func__);
1389 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1390 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1391 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1392 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
1393 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1394 log(LOG_DEBUG, "%s; %s: Listen socket: "
1395 "Connection attempt from/to broad- "
1396 "or multicast address ignored\n",
1403 * SYN appears to be valid. Create compressed TCP state
1407 if (so->so_options & SO_DEBUG)
1408 tcp_trace(TA_INPUT, ostate, tp,
1409 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1411 TCP_PROBE3(debug__input, tp, th, mtod(m, const char *));
1412 tcp_dooptions(&to, optp, optlen, TO_SYN);
1414 if (syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL))
1415 goto tfo_socket_result;
1417 syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL);
1420 * Entry added to syncache and mbuf consumed.
1421 * Only the listen socket is unlocked by syncache_add().
1423 if (ti_locked == TI_RLOCKED) {
1424 INP_INFO_RUNLOCK(&V_tcbinfo);
1425 ti_locked = TI_UNLOCKED;
1427 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1428 return (IPPROTO_DONE);
1429 } else if (tp->t_state == TCPS_LISTEN) {
1431 * When a listen socket is torn down the SO_ACCEPTCONN
1432 * flag is removed first while connections are drained
1433 * from the accept queue in a unlock/lock cycle of the
1434 * ACCEPT_LOCK, opening a race condition allowing a SYN
1435 * attempt go through unhandled.
1440 #ifdef TCP_SIGNATURE
1441 if (sig_checked == 0) {
1442 tcp_dooptions(&to, optp, optlen,
1443 (thflags & TH_SYN) ? TO_SYN : 0);
1444 if (!tcp_signature_verify_input(m, off0, tlen, optlen, &to,
1448 * In SYN_SENT state if it receives an RST, it is
1449 * allowed for further processing.
1451 if ((thflags & TH_RST) == 0 ||
1452 (tp->t_state == TCPS_SYN_SENT) == 0)
1459 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th);
1462 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later
1463 * state. tcp_do_segment() always consumes the mbuf chain, unlocks
1464 * the inpcb, and unlocks pcbinfo.
1466 tp->t_fb->tfb_tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos, ti_locked);
1467 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1468 return (IPPROTO_DONE);
1471 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th);
1473 if (ti_locked == TI_RLOCKED) {
1474 INP_INFO_RUNLOCK(&V_tcbinfo);
1475 ti_locked = TI_UNLOCKED;
1479 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset "
1480 "ti_locked: %d", __func__, ti_locked));
1481 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1486 tcp_dropwithreset(m, th, tp, tlen, rstreason);
1489 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
1490 m = NULL; /* mbuf chain got consumed. */
1495 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th);
1497 if (ti_locked == TI_RLOCKED) {
1498 INP_INFO_RUNLOCK(&V_tcbinfo);
1499 ti_locked = TI_UNLOCKED;
1503 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock "
1504 "ti_locked: %d", __func__, ti_locked));
1505 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1513 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1518 return (IPPROTO_DONE);
1522 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so,
1523 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos,
1526 int thflags, acked, ourfinisacked, needoutput = 0, sack_changed;
1527 int rstreason, todrop, win;
1531 struct in_conninfo *inc;
1540 * The size of tcp_saveipgen must be the size of the max ip header,
1543 u_char tcp_saveipgen[IP6_HDR_LEN];
1544 struct tcphdr tcp_savetcp;
1547 thflags = th->th_flags;
1548 inc = &tp->t_inpcb->inp_inc;
1549 tp->sackhint.last_sack_ack = 0;
1551 nsegs = max(1, m->m_pkthdr.lro_nsegs);
1554 * If this is either a state-changing packet or current state isn't
1555 * established, we require a write lock on tcbinfo. Otherwise, we
1556 * allow the tcbinfo to be in either alocked or unlocked, as the
1557 * caller may have unnecessarily acquired a write lock due to a race.
1559 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 ||
1560 tp->t_state != TCPS_ESTABLISHED) {
1561 KASSERT(ti_locked == TI_RLOCKED, ("%s ti_locked %d for "
1562 "SYN/FIN/RST/!EST", __func__, ti_locked));
1563 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1566 if (ti_locked == TI_RLOCKED)
1567 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1569 KASSERT(ti_locked == TI_UNLOCKED, ("%s: EST "
1570 "ti_locked: %d", __func__, ti_locked));
1571 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1575 INP_WLOCK_ASSERT(tp->t_inpcb);
1576 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN",
1578 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT",
1582 /* Save segment, if requested. */
1583 tcp_pcap_add(th, m, &(tp->t_inpkts));
1587 * Segment received on connection.
1588 * Reset idle time and keep-alive timer.
1589 * XXX: This should be done after segment
1590 * validation to ignore broken/spoofed segs.
1592 tp->t_rcvtime = ticks;
1595 * Scale up the window into a 32-bit value.
1596 * For the SYN_SENT state the scale is zero.
1598 tiwin = th->th_win << tp->snd_scale;
1601 * TCP ECN processing.
1603 if (tp->t_flags & TF_ECN_PERMIT) {
1604 if (thflags & TH_CWR)
1605 tp->t_flags &= ~TF_ECN_SND_ECE;
1606 switch (iptos & IPTOS_ECN_MASK) {
1608 tp->t_flags |= TF_ECN_SND_ECE;
1609 TCPSTAT_INC(tcps_ecn_ce);
1611 case IPTOS_ECN_ECT0:
1612 TCPSTAT_INC(tcps_ecn_ect0);
1614 case IPTOS_ECN_ECT1:
1615 TCPSTAT_INC(tcps_ecn_ect1);
1619 /* Process a packet differently from RFC3168. */
1620 cc_ecnpkt_handler(tp, th, iptos);
1622 /* Congestion experienced. */
1623 if (thflags & TH_ECE) {
1624 cc_cong_signal(tp, th, CC_ECN);
1629 * Parse options on any incoming segment.
1631 tcp_dooptions(&to, (u_char *)(th + 1),
1632 (th->th_off << 2) - sizeof(struct tcphdr),
1633 (thflags & TH_SYN) ? TO_SYN : 0);
1636 * If echoed timestamp is later than the current time,
1637 * fall back to non RFC1323 RTT calculation. Normalize
1638 * timestamp if syncookies were used when this connection
1641 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1642 to.to_tsecr -= tp->ts_offset;
1643 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks()))
1647 * If timestamps were negotiated during SYN/ACK they should
1648 * appear on every segment during this session and vice versa.
1650 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) {
1651 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1652 log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1653 "no action\n", s, __func__);
1657 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) {
1658 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1659 log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
1660 "no action\n", s, __func__);
1666 * Process options only when we get SYN/ACK back. The SYN case
1667 * for incoming connections is handled in tcp_syncache.
1668 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
1669 * or <SYN,ACK>) segment itself is never scaled.
1670 * XXX this is traditional behavior, may need to be cleaned up.
1672 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1673 if ((to.to_flags & TOF_SCALE) &&
1674 (tp->t_flags & TF_REQ_SCALE)) {
1675 tp->t_flags |= TF_RCVD_SCALE;
1676 tp->snd_scale = to.to_wscale;
1679 * Initial send window. It will be updated with
1680 * the next incoming segment to the scaled value.
1682 tp->snd_wnd = th->th_win;
1683 if (to.to_flags & TOF_TS) {
1684 tp->t_flags |= TF_RCVD_TSTMP;
1685 tp->ts_recent = to.to_tsval;
1686 tp->ts_recent_age = tcp_ts_getticks();
1688 if (to.to_flags & TOF_MSS)
1689 tcp_mss(tp, to.to_mss);
1690 if ((tp->t_flags & TF_SACK_PERMIT) &&
1691 (to.to_flags & TOF_SACKPERM) == 0)
1692 tp->t_flags &= ~TF_SACK_PERMIT;
1696 * Header prediction: check for the two common cases
1697 * of a uni-directional data xfer. If the packet has
1698 * no control flags, is in-sequence, the window didn't
1699 * change and we're not retransmitting, it's a
1700 * candidate. If the length is zero and the ack moved
1701 * forward, we're the sender side of the xfer. Just
1702 * free the data acked & wake any higher level process
1703 * that was blocked waiting for space. If the length
1704 * is non-zero and the ack didn't move, we're the
1705 * receiver side. If we're getting packets in-order
1706 * (the reassembly queue is empty), add the data to
1707 * the socket buffer and note that we need a delayed ack.
1708 * Make sure that the hidden state-flags are also off.
1709 * Since we check for TCPS_ESTABLISHED first, it can only
1712 if (tp->t_state == TCPS_ESTABLISHED &&
1713 th->th_seq == tp->rcv_nxt &&
1714 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1715 tp->snd_nxt == tp->snd_max &&
1716 tiwin && tiwin == tp->snd_wnd &&
1717 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1718 LIST_EMPTY(&tp->t_segq) &&
1719 ((to.to_flags & TOF_TS) == 0 ||
1720 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) {
1723 * If last ACK falls within this segment's sequence numbers,
1724 * record the timestamp.
1725 * NOTE that the test is modified according to the latest
1726 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1728 if ((to.to_flags & TOF_TS) != 0 &&
1729 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1730 tp->ts_recent_age = tcp_ts_getticks();
1731 tp->ts_recent = to.to_tsval;
1735 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1736 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1737 !IN_RECOVERY(tp->t_flags) &&
1738 (to.to_flags & TOF_SACK) == 0 &&
1739 TAILQ_EMPTY(&tp->snd_holes)) {
1741 * This is a pure ack for outstanding data.
1743 if (ti_locked == TI_RLOCKED)
1744 INP_INFO_RUNLOCK(&V_tcbinfo);
1745 ti_locked = TI_UNLOCKED;
1747 TCPSTAT_INC(tcps_predack);
1750 * "bad retransmit" recovery.
1752 if (tp->t_rxtshift == 1 &&
1753 tp->t_flags & TF_PREVVALID &&
1754 (int)(ticks - tp->t_badrxtwin) < 0) {
1755 cc_cong_signal(tp, th, CC_RTO_ERR);
1759 * Recalculate the transmit timer / rtt.
1761 * Some boxes send broken timestamp replies
1762 * during the SYN+ACK phase, ignore
1763 * timestamps of 0 or we could calculate a
1764 * huge RTT and blow up the retransmit timer.
1766 if ((to.to_flags & TOF_TS) != 0 &&
1770 t = tcp_ts_getticks() - to.to_tsecr;
1771 if (!tp->t_rttlow || tp->t_rttlow > t)
1774 TCP_TS_TO_TICKS(t) + 1);
1775 } else if (tp->t_rtttime &&
1776 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1777 if (!tp->t_rttlow ||
1778 tp->t_rttlow > ticks - tp->t_rtttime)
1779 tp->t_rttlow = ticks - tp->t_rtttime;
1781 ticks - tp->t_rtttime);
1783 acked = BYTES_THIS_ACK(tp, th);
1786 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
1787 hhook_run_tcp_est_in(tp, th, &to);
1790 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
1791 TCPSTAT_ADD(tcps_rcvackbyte, acked);
1792 sbdrop(&so->so_snd, acked);
1793 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1794 SEQ_LEQ(th->th_ack, tp->snd_recover))
1795 tp->snd_recover = th->th_ack - 1;
1798 * Let the congestion control algorithm update
1799 * congestion control related information. This
1800 * typically means increasing the congestion
1803 cc_ack_received(tp, th, nsegs, CC_ACK);
1805 tp->snd_una = th->th_ack;
1807 * Pull snd_wl2 up to prevent seq wrap relative
1810 tp->snd_wl2 = th->th_ack;
1815 * If all outstanding data are acked, stop
1816 * retransmit timer, otherwise restart timer
1817 * using current (possibly backed-off) value.
1818 * If process is waiting for space,
1819 * wakeup/selwakeup/signal. If data
1820 * are ready to send, let tcp_output
1821 * decide between more output or persist.
1824 if (so->so_options & SO_DEBUG)
1825 tcp_trace(TA_INPUT, ostate, tp,
1826 (void *)tcp_saveipgen,
1829 TCP_PROBE3(debug__input, tp, th,
1830 mtod(m, const char *));
1831 if (tp->snd_una == tp->snd_max)
1832 tcp_timer_activate(tp, TT_REXMT, 0);
1833 else if (!tcp_timer_active(tp, TT_PERSIST))
1834 tcp_timer_activate(tp, TT_REXMT,
1837 if (sbavail(&so->so_snd))
1838 (void) tp->t_fb->tfb_tcp_output(tp);
1841 } else if (th->th_ack == tp->snd_una &&
1842 tlen <= sbspace(&so->so_rcv)) {
1843 int newsize = 0; /* automatic sockbuf scaling */
1846 * This is a pure, in-sequence data packet with
1847 * nothing on the reassembly queue and we have enough
1848 * buffer space to take it.
1850 if (ti_locked == TI_RLOCKED)
1851 INP_INFO_RUNLOCK(&V_tcbinfo);
1852 ti_locked = TI_UNLOCKED;
1854 /* Clean receiver SACK report if present */
1855 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks)
1856 tcp_clean_sackreport(tp);
1857 TCPSTAT_INC(tcps_preddat);
1858 tp->rcv_nxt += tlen;
1860 * Pull snd_wl1 up to prevent seq wrap relative to
1863 tp->snd_wl1 = th->th_seq;
1865 * Pull rcv_up up to prevent seq wrap relative to
1868 tp->rcv_up = tp->rcv_nxt;
1869 TCPSTAT_ADD(tcps_rcvpack, nsegs);
1870 TCPSTAT_ADD(tcps_rcvbyte, tlen);
1872 if (so->so_options & SO_DEBUG)
1873 tcp_trace(TA_INPUT, ostate, tp,
1874 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1876 TCP_PROBE3(debug__input, tp, th, mtod(m, const char *));
1879 * Automatic sizing of receive socket buffer. Often the send
1880 * buffer size is not optimally adjusted to the actual network
1881 * conditions at hand (delay bandwidth product). Setting the
1882 * buffer size too small limits throughput on links with high
1883 * bandwidth and high delay (eg. trans-continental/oceanic links).
1885 * On the receive side the socket buffer memory is only rarely
1886 * used to any significant extent. This allows us to be much
1887 * more aggressive in scaling the receive socket buffer. For
1888 * the case that the buffer space is actually used to a large
1889 * extent and we run out of kernel memory we can simply drop
1890 * the new segments; TCP on the sender will just retransmit it
1891 * later. Setting the buffer size too big may only consume too
1892 * much kernel memory if the application doesn't read() from
1893 * the socket or packet loss or reordering makes use of the
1896 * The criteria to step up the receive buffer one notch are:
1897 * 1. Application has not set receive buffer size with
1898 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE.
1899 * 2. the number of bytes received during the time it takes
1900 * one timestamp to be reflected back to us (the RTT);
1901 * 3. received bytes per RTT is within seven eighth of the
1902 * current socket buffer size;
1903 * 4. receive buffer size has not hit maximal automatic size;
1905 * This algorithm does one step per RTT at most and only if
1906 * we receive a bulk stream w/o packet losses or reorderings.
1907 * Shrinking the buffer during idle times is not necessary as
1908 * it doesn't consume any memory when idle.
1910 * TODO: Only step up if the application is actually serving
1911 * the buffer to better manage the socket buffer resources.
1913 if (V_tcp_do_autorcvbuf &&
1914 (to.to_flags & TOF_TS) &&
1916 (so->so_rcv.sb_flags & SB_AUTOSIZE)) {
1917 if (TSTMP_GT(to.to_tsecr, tp->rfbuf_ts) &&
1918 to.to_tsecr - tp->rfbuf_ts < hz) {
1920 (so->so_rcv.sb_hiwat / 8 * 7) &&
1921 so->so_rcv.sb_hiwat <
1922 V_tcp_autorcvbuf_max) {
1924 min(so->so_rcv.sb_hiwat +
1925 V_tcp_autorcvbuf_inc,
1926 V_tcp_autorcvbuf_max);
1928 /* Start over with next RTT. */
1932 tp->rfbuf_cnt += tlen; /* add up */
1935 /* Add data to socket buffer. */
1936 SOCKBUF_LOCK(&so->so_rcv);
1937 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1941 * Set new socket buffer size.
1942 * Give up when limit is reached.
1945 if (!sbreserve_locked(&so->so_rcv,
1947 so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1948 m_adj(m, drop_hdrlen); /* delayed header drop */
1949 sbappendstream_locked(&so->so_rcv, m, 0);
1951 /* NB: sorwakeup_locked() does an implicit unlock. */
1952 sorwakeup_locked(so);
1953 if (DELAY_ACK(tp, tlen)) {
1954 tp->t_flags |= TF_DELACK;
1956 tp->t_flags |= TF_ACKNOW;
1957 tp->t_fb->tfb_tcp_output(tp);
1964 * Calculate amount of space in receive window,
1965 * and then do TCP input processing.
1966 * Receive window is amount of space in rcv queue,
1967 * but not less than advertised window.
1969 win = sbspace(&so->so_rcv);
1972 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1974 /* Reset receive buffer auto scaling when not in bulk receive mode. */
1978 switch (tp->t_state) {
1981 * If the state is SYN_RECEIVED:
1982 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1984 case TCPS_SYN_RECEIVED:
1985 if ((thflags & TH_ACK) &&
1986 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1987 SEQ_GT(th->th_ack, tp->snd_max))) {
1988 rstreason = BANDLIM_RST_OPENPORT;
1992 if (IS_FASTOPEN(tp->t_flags)) {
1994 * When a TFO connection is in SYN_RECEIVED, the
1995 * only valid packets are the initial SYN, a
1996 * retransmit/copy of the initial SYN (possibly with
1997 * a subset of the original data), a valid ACK, a
2000 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) {
2001 rstreason = BANDLIM_RST_OPENPORT;
2003 } else if (thflags & TH_SYN) {
2004 /* non-initial SYN is ignored */
2005 if ((tcp_timer_active(tp, TT_DELACK) ||
2006 tcp_timer_active(tp, TT_REXMT)))
2008 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) {
2016 * If the state is SYN_SENT:
2017 * if seg contains an ACK, but not for our SYN, drop the input.
2018 * if seg contains a RST, then drop the connection.
2019 * if seg does not contain SYN, then drop it.
2020 * Otherwise this is an acceptable SYN segment
2021 * initialize tp->rcv_nxt and tp->irs
2022 * if seg contains ack then advance tp->snd_una
2023 * if seg contains an ECE and ECN support is enabled, the stream
2025 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
2026 * arrange for segment to be acked (eventually)
2027 * continue processing rest of data/controls, beginning with URG
2030 if ((thflags & TH_ACK) &&
2031 (SEQ_LEQ(th->th_ack, tp->iss) ||
2032 SEQ_GT(th->th_ack, tp->snd_max))) {
2033 rstreason = BANDLIM_UNLIMITED;
2036 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) {
2037 TCP_PROBE5(connect__refused, NULL, tp,
2038 mtod(m, const char *), tp, th);
2039 tp = tcp_drop(tp, ECONNREFUSED);
2041 if (thflags & TH_RST)
2043 if (!(thflags & TH_SYN))
2046 tp->irs = th->th_seq;
2048 if (thflags & TH_ACK) {
2049 TCPSTAT_INC(tcps_connects);
2052 mac_socketpeer_set_from_mbuf(m, so);
2054 /* Do window scaling on this connection? */
2055 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2056 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2057 tp->rcv_scale = tp->request_r_scale;
2059 tp->rcv_adv += min(tp->rcv_wnd,
2060 TCP_MAXWIN << tp->rcv_scale);
2061 tp->snd_una++; /* SYN is acked */
2063 * If there's data, delay ACK; if there's also a FIN
2064 * ACKNOW will be turned on later.
2066 if (DELAY_ACK(tp, tlen) && tlen != 0)
2067 tcp_timer_activate(tp, TT_DELACK,
2070 tp->t_flags |= TF_ACKNOW;
2072 if ((thflags & TH_ECE) && V_tcp_do_ecn) {
2073 tp->t_flags |= TF_ECN_PERMIT;
2074 TCPSTAT_INC(tcps_ecn_shs);
2078 * Received <SYN,ACK> in SYN_SENT[*] state.
2080 * SYN_SENT --> ESTABLISHED
2081 * SYN_SENT* --> FIN_WAIT_1
2083 tp->t_starttime = ticks;
2084 if (tp->t_flags & TF_NEEDFIN) {
2085 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2086 tp->t_flags &= ~TF_NEEDFIN;
2089 tcp_state_change(tp, TCPS_ESTABLISHED);
2090 TCP_PROBE5(connect__established, NULL, tp,
2091 mtod(m, const char *), tp, th);
2093 tcp_timer_activate(tp, TT_KEEP,
2098 * Received initial SYN in SYN-SENT[*] state =>
2099 * simultaneous open.
2100 * If it succeeds, connection is * half-synchronized.
2101 * Otherwise, do 3-way handshake:
2102 * SYN-SENT -> SYN-RECEIVED
2103 * SYN-SENT* -> SYN-RECEIVED*
2105 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
2106 tcp_timer_activate(tp, TT_REXMT, 0);
2107 tcp_state_change(tp, TCPS_SYN_RECEIVED);
2110 KASSERT(ti_locked == TI_RLOCKED, ("%s: trimthenstep6: "
2111 "ti_locked %d", __func__, ti_locked));
2112 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2113 INP_WLOCK_ASSERT(tp->t_inpcb);
2116 * Advance th->th_seq to correspond to first data byte.
2117 * If data, trim to stay within window,
2118 * dropping FIN if necessary.
2121 if (tlen > tp->rcv_wnd) {
2122 todrop = tlen - tp->rcv_wnd;
2126 TCPSTAT_INC(tcps_rcvpackafterwin);
2127 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2129 tp->snd_wl1 = th->th_seq - 1;
2130 tp->rcv_up = th->th_seq;
2132 * Client side of transaction: already sent SYN and data.
2133 * If the remote host used T/TCP to validate the SYN,
2134 * our data will be ACK'd; if so, enter normal data segment
2135 * processing in the middle of step 5, ack processing.
2136 * Otherwise, goto step 6.
2138 if (thflags & TH_ACK)
2144 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
2145 * do normal processing.
2147 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later.
2151 break; /* continue normal processing */
2155 * States other than LISTEN or SYN_SENT.
2156 * First check the RST flag and sequence number since reset segments
2157 * are exempt from the timestamp and connection count tests. This
2158 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
2159 * below which allowed reset segments in half the sequence space
2160 * to fall though and be processed (which gives forged reset
2161 * segments with a random sequence number a 50 percent chance of
2162 * killing a connection).
2163 * Then check timestamp, if present.
2164 * Then check the connection count, if present.
2165 * Then check that at least some bytes of segment are within
2166 * receive window. If segment begins before rcv_nxt,
2167 * drop leading data (and SYN); if nothing left, just ack.
2169 if (thflags & TH_RST) {
2171 * RFC5961 Section 3.2
2173 * - RST drops connection only if SEG.SEQ == RCV.NXT.
2174 * - If RST is in window, we send challenge ACK.
2176 * Note: to take into account delayed ACKs, we should
2177 * test against last_ack_sent instead of rcv_nxt.
2178 * Note 2: we handle special case of closed window, not
2179 * covered by the RFC.
2181 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2182 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) ||
2183 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) {
2185 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2186 KASSERT(ti_locked == TI_RLOCKED,
2187 ("%s: TH_RST ti_locked %d, th %p tp %p",
2188 __func__, ti_locked, th, tp));
2189 KASSERT(tp->t_state != TCPS_SYN_SENT,
2190 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p",
2193 if (V_tcp_insecure_rst ||
2194 tp->last_ack_sent == th->th_seq) {
2195 TCPSTAT_INC(tcps_drops);
2196 /* Drop the connection. */
2197 switch (tp->t_state) {
2198 case TCPS_SYN_RECEIVED:
2199 so->so_error = ECONNREFUSED;
2201 case TCPS_ESTABLISHED:
2202 case TCPS_FIN_WAIT_1:
2203 case TCPS_FIN_WAIT_2:
2204 case TCPS_CLOSE_WAIT:
2207 so->so_error = ECONNRESET;
2214 TCPSTAT_INC(tcps_badrst);
2215 /* Send challenge ACK. */
2216 tcp_respond(tp, mtod(m, void *), th, m,
2217 tp->rcv_nxt, tp->snd_nxt, TH_ACK);
2218 tp->last_ack_sent = tp->rcv_nxt;
2226 * RFC5961 Section 4.2
2227 * Send challenge ACK for any SYN in synchronized state.
2229 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT &&
2230 tp->t_state != TCPS_SYN_RECEIVED) {
2231 KASSERT(ti_locked == TI_RLOCKED,
2232 ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked));
2233 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2235 TCPSTAT_INC(tcps_badsyn);
2236 if (V_tcp_insecure_syn &&
2237 SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2238 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
2239 tp = tcp_drop(tp, ECONNRESET);
2240 rstreason = BANDLIM_UNLIMITED;
2242 /* Send challenge ACK. */
2243 tcp_respond(tp, mtod(m, void *), th, m, tp->rcv_nxt,
2244 tp->snd_nxt, TH_ACK);
2245 tp->last_ack_sent = tp->rcv_nxt;
2252 * RFC 1323 PAWS: If we have a timestamp reply on this segment
2253 * and it's less than ts_recent, drop it.
2255 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
2256 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
2258 /* Check to see if ts_recent is over 24 days old. */
2259 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) {
2261 * Invalidate ts_recent. If this segment updates
2262 * ts_recent, the age will be reset later and ts_recent
2263 * will get a valid value. If it does not, setting
2264 * ts_recent to zero will at least satisfy the
2265 * requirement that zero be placed in the timestamp
2266 * echo reply when ts_recent isn't valid. The
2267 * age isn't reset until we get a valid ts_recent
2268 * because we don't want out-of-order segments to be
2269 * dropped when ts_recent is old.
2273 TCPSTAT_INC(tcps_rcvduppack);
2274 TCPSTAT_ADD(tcps_rcvdupbyte, tlen);
2275 TCPSTAT_INC(tcps_pawsdrop);
2283 * In the SYN-RECEIVED state, validate that the packet belongs to
2284 * this connection before trimming the data to fit the receive
2285 * window. Check the sequence number versus IRS since we know
2286 * the sequence numbers haven't wrapped. This is a partial fix
2287 * for the "LAND" DoS attack.
2289 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
2290 rstreason = BANDLIM_RST_OPENPORT;
2294 todrop = tp->rcv_nxt - th->th_seq;
2296 if (thflags & TH_SYN) {
2306 * Following if statement from Stevens, vol. 2, p. 960.
2309 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
2311 * Any valid FIN must be to the left of the window.
2312 * At this point the FIN must be a duplicate or out
2313 * of sequence; drop it.
2318 * Send an ACK to resynchronize and drop any data.
2319 * But keep on processing for RST or ACK.
2321 tp->t_flags |= TF_ACKNOW;
2323 TCPSTAT_INC(tcps_rcvduppack);
2324 TCPSTAT_ADD(tcps_rcvdupbyte, todrop);
2326 TCPSTAT_INC(tcps_rcvpartduppack);
2327 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop);
2329 drop_hdrlen += todrop; /* drop from the top afterwards */
2330 th->th_seq += todrop;
2332 if (th->th_urp > todrop)
2333 th->th_urp -= todrop;
2341 * If new data are received on a connection after the
2342 * user processes are gone, then RST the other end.
2344 if ((so->so_state & SS_NOFDREF) &&
2345 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
2346 KASSERT(ti_locked == TI_RLOCKED, ("%s: SS_NOFDEREF && "
2347 "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked));
2348 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2350 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
2351 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data "
2352 "after socket was closed, "
2353 "sending RST and removing tcpcb\n",
2354 s, __func__, tcpstates[tp->t_state], tlen);
2358 TCPSTAT_INC(tcps_rcvafterclose);
2359 rstreason = BANDLIM_UNLIMITED;
2364 * If segment ends after window, drop trailing data
2365 * (and PUSH and FIN); if nothing left, just ACK.
2367 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
2369 TCPSTAT_INC(tcps_rcvpackafterwin);
2370 if (todrop >= tlen) {
2371 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen);
2373 * If window is closed can only take segments at
2374 * window edge, and have to drop data and PUSH from
2375 * incoming segments. Continue processing, but
2376 * remember to ack. Otherwise, drop segment
2379 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2380 tp->t_flags |= TF_ACKNOW;
2381 TCPSTAT_INC(tcps_rcvwinprobe);
2385 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2388 thflags &= ~(TH_PUSH|TH_FIN);
2392 * If last ACK falls within this segment's sequence numbers,
2393 * record its timestamp.
2395 * 1) That the test incorporates suggestions from the latest
2396 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
2397 * 2) That updating only on newer timestamps interferes with
2398 * our earlier PAWS tests, so this check should be solely
2399 * predicated on the sequence space of this segment.
2400 * 3) That we modify the segment boundary check to be
2401 * Last.ACK.Sent <= SEG.SEQ + SEG.Len
2402 * instead of RFC1323's
2403 * Last.ACK.Sent < SEG.SEQ + SEG.Len,
2404 * This modified check allows us to overcome RFC1323's
2405 * limitations as described in Stevens TCP/IP Illustrated
2406 * Vol. 2 p.869. In such cases, we can still calculate the
2407 * RTT correctly when RCV.NXT == Last.ACK.Sent.
2409 if ((to.to_flags & TOF_TS) != 0 &&
2410 SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2411 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
2412 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
2413 tp->ts_recent_age = tcp_ts_getticks();
2414 tp->ts_recent = to.to_tsval;
2418 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
2419 * flag is on (half-synchronized state), then queue data for
2420 * later processing; else drop segment and return.
2422 if ((thflags & TH_ACK) == 0) {
2423 if (tp->t_state == TCPS_SYN_RECEIVED ||
2424 (tp->t_flags & TF_NEEDSYN)) {
2426 if (tp->t_state == TCPS_SYN_RECEIVED &&
2427 IS_FASTOPEN(tp->t_flags)) {
2428 tp->snd_wnd = tiwin;
2433 } else if (tp->t_flags & TF_ACKNOW)
2442 switch (tp->t_state) {
2445 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
2446 * ESTABLISHED state and continue processing.
2447 * The ACK was checked above.
2449 case TCPS_SYN_RECEIVED:
2451 TCPSTAT_INC(tcps_connects);
2453 /* Do window scaling? */
2454 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2455 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2456 tp->rcv_scale = tp->request_r_scale;
2457 tp->snd_wnd = tiwin;
2461 * SYN-RECEIVED -> ESTABLISHED
2462 * SYN-RECEIVED* -> FIN-WAIT-1
2464 tp->t_starttime = ticks;
2465 if (tp->t_flags & TF_NEEDFIN) {
2466 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2467 tp->t_flags &= ~TF_NEEDFIN;
2469 tcp_state_change(tp, TCPS_ESTABLISHED);
2470 TCP_PROBE5(accept__established, NULL, tp,
2471 mtod(m, const char *), tp, th);
2473 if (tp->t_tfo_pending) {
2474 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2475 tp->t_tfo_pending = NULL;
2478 * Account for the ACK of our SYN prior to
2479 * regular ACK processing below.
2484 * TFO connections call cc_conn_init() during SYN
2485 * processing. Calling it again here for such
2486 * connections is not harmless as it would undo the
2487 * snd_cwnd reduction that occurs when a TFO SYN|ACK
2490 if (!IS_FASTOPEN(tp->t_flags))
2493 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
2496 * If segment contains data or ACK, will call tcp_reass()
2497 * later; if not, do so now to pass queued data to user.
2499 if (tlen == 0 && (thflags & TH_FIN) == 0)
2500 (void) tcp_reass(tp, (struct tcphdr *)0, 0,
2502 tp->snd_wl1 = th->th_seq - 1;
2506 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2507 * ACKs. If the ack is in the range
2508 * tp->snd_una < th->th_ack <= tp->snd_max
2509 * then advance tp->snd_una to th->th_ack and drop
2510 * data from the retransmission queue. If this ACK reflects
2511 * more up to date window information we update our window information.
2513 case TCPS_ESTABLISHED:
2514 case TCPS_FIN_WAIT_1:
2515 case TCPS_FIN_WAIT_2:
2516 case TCPS_CLOSE_WAIT:
2519 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2520 TCPSTAT_INC(tcps_rcvacktoomuch);
2523 if ((tp->t_flags & TF_SACK_PERMIT) &&
2524 ((to.to_flags & TOF_SACK) ||
2525 !TAILQ_EMPTY(&tp->snd_holes)))
2526 sack_changed = tcp_sack_doack(tp, &to, th->th_ack);
2529 * Reset the value so that previous (valid) value
2530 * from the last ack with SACK doesn't get used.
2532 tp->sackhint.sacked_bytes = 0;
2535 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
2536 hhook_run_tcp_est_in(tp, th, &to);
2539 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2542 maxseg = tcp_maxseg(tp);
2544 (tiwin == tp->snd_wnd ||
2545 (tp->t_flags & TF_SACK_PERMIT))) {
2547 * If this is the first time we've seen a
2548 * FIN from the remote, this is not a
2549 * duplicate and it needs to be processed
2550 * normally. This happens during a
2551 * simultaneous close.
2553 if ((thflags & TH_FIN) &&
2554 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2558 TCPSTAT_INC(tcps_rcvdupack);
2560 * If we have outstanding data (other than
2561 * a window probe), this is a completely
2562 * duplicate ack (ie, window info didn't
2563 * change and FIN isn't set),
2564 * the ack is the biggest we've
2565 * seen and we've seen exactly our rexmt
2566 * threshold of them, assume a packet
2567 * has been dropped and retransmit it.
2568 * Kludge snd_nxt & the congestion
2569 * window so we send only this one
2572 * We know we're losing at the current
2573 * window size so do congestion avoidance
2574 * (set ssthresh to half the current window
2575 * and pull our congestion window back to
2576 * the new ssthresh).
2578 * Dup acks mean that packets have left the
2579 * network (they're now cached at the receiver)
2580 * so bump cwnd by the amount in the receiver
2581 * to keep a constant cwnd packets in the
2584 * When using TCP ECN, notify the peer that
2585 * we reduced the cwnd.
2588 * Following 2 kinds of acks should not affect
2591 * 2) Acks with SACK but without any new SACK
2592 * information in them. These could result from
2593 * any anomaly in the network like a switch
2594 * duplicating packets or a possible DoS attack.
2596 if (th->th_ack != tp->snd_una ||
2597 ((tp->t_flags & TF_SACK_PERMIT) &&
2600 else if (!tcp_timer_active(tp, TT_REXMT))
2602 else if (++tp->t_dupacks > tcprexmtthresh ||
2603 IN_FASTRECOVERY(tp->t_flags)) {
2604 cc_ack_received(tp, th, nsegs,
2606 if ((tp->t_flags & TF_SACK_PERMIT) &&
2607 IN_FASTRECOVERY(tp->t_flags)) {
2611 * Compute the amount of data in flight first.
2612 * We can inject new data into the pipe iff
2613 * we have less than 1/2 the original window's
2614 * worth of data in flight.
2616 if (V_tcp_do_rfc6675_pipe)
2617 awnd = tcp_compute_pipe(tp);
2619 awnd = (tp->snd_nxt - tp->snd_fack) +
2620 tp->sackhint.sack_bytes_rexmit;
2622 if (awnd < tp->snd_ssthresh) {
2623 tp->snd_cwnd += maxseg;
2625 * RFC5681 Section 3.2 talks about cwnd
2626 * inflation on additional dupacks and
2627 * deflation on recovering from loss.
2629 * We keep cwnd into check so that
2630 * we don't have to 'deflate' it when we
2631 * get out of recovery.
2633 if (tp->snd_cwnd > tp->snd_ssthresh)
2634 tp->snd_cwnd = tp->snd_ssthresh;
2637 tp->snd_cwnd += maxseg;
2638 (void) tp->t_fb->tfb_tcp_output(tp);
2640 } else if (tp->t_dupacks == tcprexmtthresh) {
2641 tcp_seq onxt = tp->snd_nxt;
2644 * If we're doing sack, check to
2645 * see if we're already in sack
2646 * recovery. If we're not doing sack,
2647 * check to see if we're in newreno
2650 if (tp->t_flags & TF_SACK_PERMIT) {
2651 if (IN_FASTRECOVERY(tp->t_flags)) {
2656 if (SEQ_LEQ(th->th_ack,
2662 /* Congestion signal before ack. */
2663 cc_cong_signal(tp, th, CC_NDUPACK);
2664 cc_ack_received(tp, th, nsegs,
2666 tcp_timer_activate(tp, TT_REXMT, 0);
2668 if (tp->t_flags & TF_SACK_PERMIT) {
2670 tcps_sack_recovery_episode);
2671 tp->sack_newdata = tp->snd_nxt;
2672 if (CC_ALGO(tp)->cong_signal == NULL)
2673 tp->snd_cwnd = maxseg;
2674 (void) tp->t_fb->tfb_tcp_output(tp);
2677 tp->snd_nxt = th->th_ack;
2678 if (CC_ALGO(tp)->cong_signal == NULL)
2679 tp->snd_cwnd = maxseg;
2680 (void) tp->t_fb->tfb_tcp_output(tp);
2681 KASSERT(tp->snd_limited <= 2,
2682 ("%s: tp->snd_limited too big",
2684 if (CC_ALGO(tp)->cong_signal == NULL)
2685 tp->snd_cwnd = tp->snd_ssthresh +
2687 (tp->t_dupacks - tp->snd_limited);
2688 if (SEQ_GT(onxt, tp->snd_nxt))
2691 } else if (V_tcp_do_rfc3042) {
2693 * Process first and second duplicate
2694 * ACKs. Each indicates a segment
2695 * leaving the network, creating room
2696 * for more. Make sure we can send a
2697 * packet on reception of each duplicate
2698 * ACK by increasing snd_cwnd by one
2699 * segment. Restore the original
2700 * snd_cwnd after packet transmission.
2702 cc_ack_received(tp, th, nsegs,
2704 uint32_t oldcwnd = tp->snd_cwnd;
2705 tcp_seq oldsndmax = tp->snd_max;
2709 KASSERT(tp->t_dupacks == 1 ||
2711 ("%s: dupacks not 1 or 2",
2713 if (tp->t_dupacks == 1)
2714 tp->snd_limited = 0;
2716 (tp->snd_nxt - tp->snd_una) +
2717 (tp->t_dupacks - tp->snd_limited) *
2720 * Only call tcp_output when there
2721 * is new data available to be sent.
2722 * Otherwise we would send pure ACKs.
2724 SOCKBUF_LOCK(&so->so_snd);
2725 avail = sbavail(&so->so_snd) -
2726 (tp->snd_nxt - tp->snd_una);
2727 SOCKBUF_UNLOCK(&so->so_snd);
2729 (void) tp->t_fb->tfb_tcp_output(tp);
2730 sent = tp->snd_max - oldsndmax;
2731 if (sent > maxseg) {
2732 KASSERT((tp->t_dupacks == 2 &&
2733 tp->snd_limited == 0) ||
2734 (sent == maxseg + 1 &&
2735 tp->t_flags & TF_SENTFIN),
2736 ("%s: sent too much",
2738 tp->snd_limited = 2;
2739 } else if (sent > 0)
2741 tp->snd_cwnd = oldcwnd;
2748 * This ack is advancing the left edge, reset the
2753 * If this ack also has new SACK info, increment the
2754 * counter as per rfc6675.
2756 if ((tp->t_flags & TF_SACK_PERMIT) && sack_changed)
2760 KASSERT(SEQ_GT(th->th_ack, tp->snd_una),
2761 ("%s: th_ack <= snd_una", __func__));
2764 * If the congestion window was inflated to account
2765 * for the other side's cached packets, retract it.
2767 if (IN_FASTRECOVERY(tp->t_flags)) {
2768 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2769 if (tp->t_flags & TF_SACK_PERMIT)
2770 tcp_sack_partialack(tp, th);
2772 tcp_newreno_partial_ack(tp, th);
2774 cc_post_recovery(tp, th);
2777 * If we reach this point, ACK is not a duplicate,
2778 * i.e., it ACKs something we sent.
2780 if (tp->t_flags & TF_NEEDSYN) {
2782 * T/TCP: Connection was half-synchronized, and our
2783 * SYN has been ACK'd (so connection is now fully
2784 * synchronized). Go to non-starred state,
2785 * increment snd_una for ACK of SYN, and check if
2786 * we can do window scaling.
2788 tp->t_flags &= ~TF_NEEDSYN;
2790 /* Do window scaling? */
2791 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2792 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2793 tp->rcv_scale = tp->request_r_scale;
2794 /* Send window already scaled. */
2799 INP_WLOCK_ASSERT(tp->t_inpcb);
2801 acked = BYTES_THIS_ACK(tp, th);
2802 KASSERT(acked >= 0, ("%s: acked unexepectedly negative "
2803 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__,
2804 tp->snd_una, th->th_ack, tp, m));
2805 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
2806 TCPSTAT_ADD(tcps_rcvackbyte, acked);
2809 * If we just performed our first retransmit, and the ACK
2810 * arrives within our recovery window, then it was a mistake
2811 * to do the retransmit in the first place. Recover our
2812 * original cwnd and ssthresh, and proceed to transmit where
2815 if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID &&
2816 (int)(ticks - tp->t_badrxtwin) < 0)
2817 cc_cong_signal(tp, th, CC_RTO_ERR);
2820 * If we have a timestamp reply, update smoothed
2821 * round trip time. If no timestamp is present but
2822 * transmit timer is running and timed sequence
2823 * number was acked, update smoothed round trip time.
2824 * Since we now have an rtt measurement, cancel the
2825 * timer backoff (cf., Phil Karn's retransmit alg.).
2826 * Recompute the initial retransmit timer.
2828 * Some boxes send broken timestamp replies
2829 * during the SYN+ACK phase, ignore
2830 * timestamps of 0 or we could calculate a
2831 * huge RTT and blow up the retransmit timer.
2833 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) {
2836 t = tcp_ts_getticks() - to.to_tsecr;
2837 if (!tp->t_rttlow || tp->t_rttlow > t)
2839 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1);
2840 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2841 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2842 tp->t_rttlow = ticks - tp->t_rtttime;
2843 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2847 * If all outstanding data is acked, stop retransmit
2848 * timer and remember to restart (more output or persist).
2849 * If there is more data to be acked, restart retransmit
2850 * timer, using current (possibly backed-off) value.
2852 if (th->th_ack == tp->snd_max) {
2853 tcp_timer_activate(tp, TT_REXMT, 0);
2855 } else if (!tcp_timer_active(tp, TT_PERSIST))
2856 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
2859 * If no data (only SYN) was ACK'd,
2860 * skip rest of ACK processing.
2866 * Let the congestion control algorithm update congestion
2867 * control related information. This typically means increasing
2868 * the congestion window.
2870 cc_ack_received(tp, th, nsegs, CC_ACK);
2872 SOCKBUF_LOCK(&so->so_snd);
2873 if (acked > sbavail(&so->so_snd)) {
2874 if (tp->snd_wnd >= sbavail(&so->so_snd))
2875 tp->snd_wnd -= sbavail(&so->so_snd);
2878 mfree = sbcut_locked(&so->so_snd,
2879 (int)sbavail(&so->so_snd));
2882 mfree = sbcut_locked(&so->so_snd, acked);
2883 if (tp->snd_wnd >= (uint32_t) acked)
2884 tp->snd_wnd -= acked;
2889 /* NB: sowwakeup_locked() does an implicit unlock. */
2890 sowwakeup_locked(so);
2892 /* Detect una wraparound. */
2893 if (!IN_RECOVERY(tp->t_flags) &&
2894 SEQ_GT(tp->snd_una, tp->snd_recover) &&
2895 SEQ_LEQ(th->th_ack, tp->snd_recover))
2896 tp->snd_recover = th->th_ack - 1;
2897 /* XXXLAS: Can this be moved up into cc_post_recovery? */
2898 if (IN_RECOVERY(tp->t_flags) &&
2899 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2900 EXIT_RECOVERY(tp->t_flags);
2902 tp->snd_una = th->th_ack;
2903 if (tp->t_flags & TF_SACK_PERMIT) {
2904 if (SEQ_GT(tp->snd_una, tp->snd_recover))
2905 tp->snd_recover = tp->snd_una;
2907 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2908 tp->snd_nxt = tp->snd_una;
2910 switch (tp->t_state) {
2913 * In FIN_WAIT_1 STATE in addition to the processing
2914 * for the ESTABLISHED state if our FIN is now acknowledged
2915 * then enter FIN_WAIT_2.
2917 case TCPS_FIN_WAIT_1:
2918 if (ourfinisacked) {
2920 * If we can't receive any more
2921 * data, then closing user can proceed.
2922 * Starting the timer is contrary to the
2923 * specification, but if we don't get a FIN
2924 * we'll hang forever.
2927 * we should release the tp also, and use a
2930 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2931 soisdisconnected(so);
2932 tcp_timer_activate(tp, TT_2MSL,
2933 (tcp_fast_finwait2_recycle ?
2934 tcp_finwait2_timeout :
2937 tcp_state_change(tp, TCPS_FIN_WAIT_2);
2942 * In CLOSING STATE in addition to the processing for
2943 * the ESTABLISHED state if the ACK acknowledges our FIN
2944 * then enter the TIME-WAIT state, otherwise ignore
2948 if (ourfinisacked) {
2949 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2951 INP_INFO_RUNLOCK(&V_tcbinfo);
2958 * In LAST_ACK, we may still be waiting for data to drain
2959 * and/or to be acked, as well as for the ack of our FIN.
2960 * If our FIN is now acknowledged, delete the TCB,
2961 * enter the closed state and return.
2964 if (ourfinisacked) {
2965 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
2974 INP_WLOCK_ASSERT(tp->t_inpcb);
2977 * Update window information.
2978 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2980 if ((thflags & TH_ACK) &&
2981 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2982 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2983 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2984 /* keep track of pure window updates */
2986 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2987 TCPSTAT_INC(tcps_rcvwinupd);
2988 tp->snd_wnd = tiwin;
2989 tp->snd_wl1 = th->th_seq;
2990 tp->snd_wl2 = th->th_ack;
2991 if (tp->snd_wnd > tp->max_sndwnd)
2992 tp->max_sndwnd = tp->snd_wnd;
2997 * Process segments with URG.
2999 if ((thflags & TH_URG) && th->th_urp &&
3000 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3002 * This is a kludge, but if we receive and accept
3003 * random urgent pointers, we'll crash in
3004 * soreceive. It's hard to imagine someone
3005 * actually wanting to send this much urgent data.
3007 SOCKBUF_LOCK(&so->so_rcv);
3008 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) {
3009 th->th_urp = 0; /* XXX */
3010 thflags &= ~TH_URG; /* XXX */
3011 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */
3012 goto dodata; /* XXX */
3015 * If this segment advances the known urgent pointer,
3016 * then mark the data stream. This should not happen
3017 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
3018 * a FIN has been received from the remote side.
3019 * In these states we ignore the URG.
3021 * According to RFC961 (Assigned Protocols),
3022 * the urgent pointer points to the last octet
3023 * of urgent data. We continue, however,
3024 * to consider it to indicate the first octet
3025 * of data past the urgent section as the original
3026 * spec states (in one of two places).
3028 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
3029 tp->rcv_up = th->th_seq + th->th_urp;
3030 so->so_oobmark = sbavail(&so->so_rcv) +
3031 (tp->rcv_up - tp->rcv_nxt) - 1;
3032 if (so->so_oobmark == 0)
3033 so->so_rcv.sb_state |= SBS_RCVATMARK;
3035 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
3037 SOCKBUF_UNLOCK(&so->so_rcv);
3039 * Remove out of band data so doesn't get presented to user.
3040 * This can happen independent of advancing the URG pointer,
3041 * but if two URG's are pending at once, some out-of-band
3042 * data may creep in... ick.
3044 if (th->th_urp <= (uint32_t)tlen &&
3045 !(so->so_options & SO_OOBINLINE)) {
3046 /* hdr drop is delayed */
3047 tcp_pulloutofband(so, th, m, drop_hdrlen);
3051 * If no out of band data is expected,
3052 * pull receive urgent pointer along
3053 * with the receive window.
3055 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
3056 tp->rcv_up = tp->rcv_nxt;
3059 INP_WLOCK_ASSERT(tp->t_inpcb);
3062 * Process the segment text, merging it into the TCP sequencing queue,
3063 * and arranging for acknowledgment of receipt if necessary.
3064 * This process logically involves adjusting tp->rcv_wnd as data
3065 * is presented to the user (this happens in tcp_usrreq.c,
3066 * case PRU_RCVD). If a FIN has already been received on this
3067 * connection then we just ignore the text.
3070 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) &&
3071 IS_FASTOPEN(tp->t_flags));
3073 #define tfo_syn (false)
3075 if ((tlen || (thflags & TH_FIN) || tfo_syn) &&
3076 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3077 tcp_seq save_start = th->th_seq;
3078 m_adj(m, drop_hdrlen); /* delayed header drop */
3080 * Insert segment which includes th into TCP reassembly queue
3081 * with control block tp. Set thflags to whether reassembly now
3082 * includes a segment with FIN. This handles the common case
3083 * inline (segment is the next to be received on an established
3084 * connection, and the queue is empty), avoiding linkage into
3085 * and removal from the queue and repetition of various
3087 * Set DELACK for segments received in order, but ack
3088 * immediately when segments are out of order (so
3089 * fast retransmit can work).
3091 if (th->th_seq == tp->rcv_nxt &&
3092 LIST_EMPTY(&tp->t_segq) &&
3093 (TCPS_HAVEESTABLISHED(tp->t_state) ||
3095 if (DELAY_ACK(tp, tlen) || tfo_syn)
3096 tp->t_flags |= TF_DELACK;
3098 tp->t_flags |= TF_ACKNOW;
3099 tp->rcv_nxt += tlen;
3100 thflags = th->th_flags & TH_FIN;
3101 TCPSTAT_INC(tcps_rcvpack);
3102 TCPSTAT_ADD(tcps_rcvbyte, tlen);
3103 SOCKBUF_LOCK(&so->so_rcv);
3104 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
3107 sbappendstream_locked(&so->so_rcv, m, 0);
3108 /* NB: sorwakeup_locked() does an implicit unlock. */
3109 sorwakeup_locked(so);
3112 * XXX: Due to the header drop above "th" is
3113 * theoretically invalid by now. Fortunately
3114 * m_adj() doesn't actually frees any mbufs
3115 * when trimming from the head.
3117 thflags = tcp_reass(tp, th, &tlen, m);
3118 tp->t_flags |= TF_ACKNOW;
3120 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT))
3121 tcp_update_sack_list(tp, save_start, save_start + tlen);
3124 * Note the amount of data that peer has sent into
3125 * our window, in order to estimate the sender's
3129 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
3130 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
3132 len = so->so_rcv.sb_hiwat;
3140 * If FIN is received ACK the FIN and let the user know
3141 * that the connection is closing.
3143 if (thflags & TH_FIN) {
3144 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3147 * If connection is half-synchronized
3148 * (ie NEEDSYN flag on) then delay ACK,
3149 * so it may be piggybacked when SYN is sent.
3150 * Otherwise, since we received a FIN then no
3151 * more input can be expected, send ACK now.
3153 if (tp->t_flags & TF_NEEDSYN)
3154 tp->t_flags |= TF_DELACK;
3156 tp->t_flags |= TF_ACKNOW;
3159 switch (tp->t_state) {
3162 * In SYN_RECEIVED and ESTABLISHED STATES
3163 * enter the CLOSE_WAIT state.
3165 case TCPS_SYN_RECEIVED:
3166 tp->t_starttime = ticks;
3168 case TCPS_ESTABLISHED:
3169 tcp_state_change(tp, TCPS_CLOSE_WAIT);
3173 * If still in FIN_WAIT_1 STATE FIN has not been acked so
3174 * enter the CLOSING state.
3176 case TCPS_FIN_WAIT_1:
3177 tcp_state_change(tp, TCPS_CLOSING);
3181 * In FIN_WAIT_2 state enter the TIME_WAIT state,
3182 * starting the time-wait timer, turning off the other
3185 case TCPS_FIN_WAIT_2:
3186 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
3187 KASSERT(ti_locked == TI_RLOCKED, ("%s: dodata "
3188 "TCP_FIN_WAIT_2 ti_locked: %d", __func__,
3192 INP_INFO_RUNLOCK(&V_tcbinfo);
3196 if (ti_locked == TI_RLOCKED)
3197 INP_INFO_RUNLOCK(&V_tcbinfo);
3198 ti_locked = TI_UNLOCKED;
3201 if (so->so_options & SO_DEBUG)
3202 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
3205 TCP_PROBE3(debug__input, tp, th, mtod(m, const char *));
3208 * Return any desired output.
3210 if (needoutput || (tp->t_flags & TF_ACKNOW))
3211 (void) tp->t_fb->tfb_tcp_output(tp);
3214 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d",
3215 __func__, ti_locked));
3216 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
3217 INP_WLOCK_ASSERT(tp->t_inpcb);
3219 if (tp->t_flags & TF_DELACK) {
3220 tp->t_flags &= ~TF_DELACK;
3221 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
3223 INP_WUNLOCK(tp->t_inpcb);
3228 * Generate an ACK dropping incoming segment if it occupies
3229 * sequence space, where the ACK reflects our state.
3231 * We can now skip the test for the RST flag since all
3232 * paths to this code happen after packets containing
3233 * RST have been dropped.
3235 * In the SYN-RECEIVED state, don't send an ACK unless the
3236 * segment we received passes the SYN-RECEIVED ACK test.
3237 * If it fails send a RST. This breaks the loop in the
3238 * "LAND" DoS attack, and also prevents an ACK storm
3239 * between two listening ports that have been sent forged
3240 * SYN segments, each with the source address of the other.
3242 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
3243 (SEQ_GT(tp->snd_una, th->th_ack) ||
3244 SEQ_GT(th->th_ack, tp->snd_max)) ) {
3245 rstreason = BANDLIM_RST_OPENPORT;
3249 if (so->so_options & SO_DEBUG)
3250 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3253 TCP_PROBE3(debug__input, tp, th, mtod(m, const char *));
3254 if (ti_locked == TI_RLOCKED)
3255 INP_INFO_RUNLOCK(&V_tcbinfo);
3256 ti_locked = TI_UNLOCKED;
3258 tp->t_flags |= TF_ACKNOW;
3259 (void) tp->t_fb->tfb_tcp_output(tp);
3260 INP_WUNLOCK(tp->t_inpcb);
3265 if (ti_locked == TI_RLOCKED)
3266 INP_INFO_RUNLOCK(&V_tcbinfo);
3267 ti_locked = TI_UNLOCKED;
3270 tcp_dropwithreset(m, th, tp, tlen, rstreason);
3271 INP_WUNLOCK(tp->t_inpcb);
3273 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
3277 if (ti_locked == TI_RLOCKED) {
3278 INP_INFO_RUNLOCK(&V_tcbinfo);
3279 ti_locked = TI_UNLOCKED;
3283 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
3287 * Drop space held by incoming segment and return.
3290 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
3291 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3294 TCP_PROBE3(debug__input, tp, th, mtod(m, const char *));
3296 INP_WUNLOCK(tp->t_inpcb);
3304 * Issue RST and make ACK acceptable to originator of segment.
3305 * The mbuf must still include the original packet header.
3309 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
3310 int tlen, int rstreason)
3316 struct ip6_hdr *ip6;
3320 INP_WLOCK_ASSERT(tp->t_inpcb);
3323 /* Don't bother if destination was broadcast/multicast. */
3324 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
3327 if (mtod(m, struct ip *)->ip_v == 6) {
3328 ip6 = mtod(m, struct ip6_hdr *);
3329 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3330 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3332 /* IPv6 anycast check is done at tcp6_input() */
3335 #if defined(INET) && defined(INET6)
3340 ip = mtod(m, struct ip *);
3341 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3342 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3343 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3344 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3349 /* Perform bandwidth limiting. */
3350 if (badport_bandlim(rstreason) < 0)
3353 /* tcp_respond consumes the mbuf chain. */
3354 if (th->th_flags & TH_ACK) {
3355 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0,
3356 th->th_ack, TH_RST);
3358 if (th->th_flags & TH_SYN)
3360 if (th->th_flags & TH_FIN)
3362 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
3363 (tcp_seq)0, TH_RST|TH_ACK);
3371 * Parse TCP options and place in tcpopt.
3374 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags)
3379 for (; cnt > 0; cnt -= optlen, cp += optlen) {
3381 if (opt == TCPOPT_EOL)
3383 if (opt == TCPOPT_NOP)
3389 if (optlen < 2 || optlen > cnt)
3394 if (optlen != TCPOLEN_MAXSEG)
3396 if (!(flags & TO_SYN))
3398 to->to_flags |= TOF_MSS;
3399 bcopy((char *)cp + 2,
3400 (char *)&to->to_mss, sizeof(to->to_mss));
3401 to->to_mss = ntohs(to->to_mss);
3404 if (optlen != TCPOLEN_WINDOW)
3406 if (!(flags & TO_SYN))
3408 to->to_flags |= TOF_SCALE;
3409 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT);
3411 case TCPOPT_TIMESTAMP:
3412 if (optlen != TCPOLEN_TIMESTAMP)
3414 to->to_flags |= TOF_TS;
3415 bcopy((char *)cp + 2,
3416 (char *)&to->to_tsval, sizeof(to->to_tsval));
3417 to->to_tsval = ntohl(to->to_tsval);
3418 bcopy((char *)cp + 6,
3419 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
3420 to->to_tsecr = ntohl(to->to_tsecr);
3422 #ifdef TCP_SIGNATURE
3424 * XXX In order to reply to a host which has set the
3425 * TCP_SIGNATURE option in its initial SYN, we have to
3426 * record the fact that the option was observed here
3427 * for the syncache code to perform the correct response.
3429 case TCPOPT_SIGNATURE:
3430 if (optlen != TCPOLEN_SIGNATURE)
3432 to->to_flags |= TOF_SIGNATURE;
3433 to->to_signature = cp + 2;
3436 case TCPOPT_SACK_PERMITTED:
3437 if (optlen != TCPOLEN_SACK_PERMITTED)
3439 if (!(flags & TO_SYN))
3443 to->to_flags |= TOF_SACKPERM;
3446 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
3450 to->to_flags |= TOF_SACK;
3451 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
3452 to->to_sacks = cp + 2;
3453 TCPSTAT_INC(tcps_sack_rcv_blocks);
3456 case TCPOPT_FAST_OPEN:
3457 if ((optlen != TCPOLEN_FAST_OPEN_EMPTY) &&
3458 (optlen < TCPOLEN_FAST_OPEN_MIN) &&
3459 (optlen > TCPOLEN_FAST_OPEN_MAX))
3461 if (!(flags & TO_SYN))
3463 if (!V_tcp_fastopen_enabled)
3465 to->to_flags |= TOF_FASTOPEN;
3466 to->to_tfo_len = optlen - 2;
3467 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL;
3477 * Pull out of band byte out of a segment so
3478 * it doesn't appear in the user's data queue.
3479 * It is still reflected in the segment length for
3480 * sequencing purposes.
3483 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m,
3486 int cnt = off + th->th_urp - 1;
3489 if (m->m_len > cnt) {
3490 char *cp = mtod(m, caddr_t) + cnt;
3491 struct tcpcb *tp = sototcpcb(so);
3493 INP_WLOCK_ASSERT(tp->t_inpcb);
3496 tp->t_oobflags |= TCPOOB_HAVEDATA;
3497 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3499 if (m->m_flags & M_PKTHDR)
3508 panic("tcp_pulloutofband");
3512 * Collect new round-trip time estimate
3513 * and update averages and current timeout.
3516 tcp_xmit_timer(struct tcpcb *tp, int rtt)
3520 INP_WLOCK_ASSERT(tp->t_inpcb);
3522 TCPSTAT_INC(tcps_rttupdated);
3524 if (tp->t_srtt != 0) {
3526 * srtt is stored as fixed point with 5 bits after the
3527 * binary point (i.e., scaled by 8). The following magic
3528 * is equivalent to the smoothing algorithm in rfc793 with
3529 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3530 * point). Adjust rtt to origin 0.
3532 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
3533 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
3535 if ((tp->t_srtt += delta) <= 0)
3539 * We accumulate a smoothed rtt variance (actually, a
3540 * smoothed mean difference), then set the retransmit
3541 * timer to smoothed rtt + 4 times the smoothed variance.
3542 * rttvar is stored as fixed point with 4 bits after the
3543 * binary point (scaled by 16). The following is
3544 * equivalent to rfc793 smoothing with an alpha of .75
3545 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
3546 * rfc793's wired-in beta.
3550 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
3551 if ((tp->t_rttvar += delta) <= 0)
3553 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
3554 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3557 * No rtt measurement yet - use the unsmoothed rtt.
3558 * Set the variance to half the rtt (so our first
3559 * retransmit happens at 3*rtt).
3561 tp->t_srtt = rtt << TCP_RTT_SHIFT;
3562 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
3563 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3569 * the retransmit should happen at rtt + 4 * rttvar.
3570 * Because of the way we do the smoothing, srtt and rttvar
3571 * will each average +1/2 tick of bias. When we compute
3572 * the retransmit timer, we want 1/2 tick of rounding and
3573 * 1 extra tick because of +-1/2 tick uncertainty in the
3574 * firing of the timer. The bias will give us exactly the
3575 * 1.5 tick we need. But, because the bias is
3576 * statistical, we have to test that we don't drop below
3577 * the minimum feasible timer (which is 2 ticks).
3579 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3580 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3583 * We received an ack for a packet that wasn't retransmitted;
3584 * it is probably safe to discard any error indications we've
3585 * received recently. This isn't quite right, but close enough
3586 * for now (a route might have failed after we sent a segment,
3587 * and the return path might not be symmetrical).
3589 tp->t_softerror = 0;
3593 * Determine a reasonable value for maxseg size.
3594 * If the route is known, check route for mtu.
3595 * If none, use an mss that can be handled on the outgoing interface
3596 * without forcing IP to fragment. If no route is found, route has no mtu,
3597 * or the destination isn't local, use a default, hopefully conservative
3598 * size (usually 512 or the default IP max size, but no more than the mtu
3599 * of the interface), as we can't discover anything about intervening
3600 * gateways or networks. We also initialize the congestion/slow start
3601 * window to be a single segment if the destination isn't local.
3602 * While looking at the routing entry, we also initialize other path-dependent
3603 * parameters from pre-set or cached values in the routing entry.
3605 * NOTE that resulting t_maxseg doesn't include space for TCP options or
3606 * IP options, e.g. IPSEC data, since length of this data may vary, and
3607 * thus it is calculated for every segment separately in tcp_output().
3609 * NOTE that this routine is only called when we process an incoming
3610 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS
3611 * settings are handled in tcp_mssopt().
3614 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer,
3615 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap)
3618 uint32_t maxmtu = 0;
3619 struct inpcb *inp = tp->t_inpcb;
3620 struct hc_metrics_lite metrics;
3622 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
3623 size_t min_protoh = isipv6 ?
3624 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
3625 sizeof (struct tcpiphdr);
3627 const size_t min_protoh = sizeof(struct tcpiphdr);
3630 INP_WLOCK_ASSERT(tp->t_inpcb);
3632 if (mtuoffer != -1) {
3633 KASSERT(offer == -1, ("%s: conflict", __func__));
3634 offer = mtuoffer - min_protoh;
3640 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap);
3641 tp->t_maxseg = V_tcp_v6mssdflt;
3644 #if defined(INET) && defined(INET6)
3649 maxmtu = tcp_maxmtu(&inp->inp_inc, cap);
3650 tp->t_maxseg = V_tcp_mssdflt;
3655 * No route to sender, stay with default mss and return.
3659 * In case we return early we need to initialize metrics
3660 * to a defined state as tcp_hc_get() would do for us
3661 * if there was no cache hit.
3663 if (metricptr != NULL)
3664 bzero(metricptr, sizeof(struct hc_metrics_lite));
3668 /* What have we got? */
3672 * Offer == 0 means that there was no MSS on the SYN
3673 * segment, in this case we use tcp_mssdflt as
3674 * already assigned to t_maxseg above.
3676 offer = tp->t_maxseg;
3681 * Offer == -1 means that we didn't receive SYN yet.
3687 * Prevent DoS attack with too small MSS. Round up
3688 * to at least minmss.
3690 offer = max(offer, V_tcp_minmss);
3694 * rmx information is now retrieved from tcp_hostcache.
3696 tcp_hc_get(&inp->inp_inc, &metrics);
3697 if (metricptr != NULL)
3698 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite));
3701 * If there's a discovered mtu in tcp hostcache, use it.
3702 * Else, use the link mtu.
3704 if (metrics.rmx_mtu)
3705 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
3709 mss = maxmtu - min_protoh;
3710 if (!V_path_mtu_discovery &&
3711 !in6_localaddr(&inp->in6p_faddr))
3712 mss = min(mss, V_tcp_v6mssdflt);
3715 #if defined(INET) && defined(INET6)
3720 mss = maxmtu - min_protoh;
3721 if (!V_path_mtu_discovery &&
3722 !in_localaddr(inp->inp_faddr))
3723 mss = min(mss, V_tcp_mssdflt);
3727 * XXX - The above conditional (mss = maxmtu - min_protoh)
3728 * probably violates the TCP spec.
3729 * The problem is that, since we don't know the
3730 * other end's MSS, we are supposed to use a conservative
3731 * default. But, if we do that, then MTU discovery will
3732 * never actually take place, because the conservative
3733 * default is much less than the MTUs typically seen
3734 * on the Internet today. For the moment, we'll sweep
3735 * this under the carpet.
3737 * The conservative default might not actually be a problem
3738 * if the only case this occurs is when sending an initial
3739 * SYN with options and data to a host we've never talked
3740 * to before. Then, they will reply with an MSS value which
3741 * will get recorded and the new parameters should get
3742 * recomputed. For Further Study.
3745 mss = min(mss, offer);
3748 * Sanity check: make sure that maxseg will be large
3749 * enough to allow some data on segments even if the
3750 * all the option space is used (40bytes). Otherwise
3751 * funny things may happen in tcp_output.
3753 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3761 tcp_mss(struct tcpcb *tp, int offer)
3767 struct hc_metrics_lite metrics;
3768 struct tcp_ifcap cap;
3770 KASSERT(tp != NULL, ("%s: tp == NULL", __func__));
3772 bzero(&cap, sizeof(cap));
3773 tcp_mss_update(tp, offer, -1, &metrics, &cap);
3779 * If there's a pipesize, change the socket buffer to that size,
3780 * don't change if sb_hiwat is different than default (then it
3781 * has been changed on purpose with setsockopt).
3782 * Make the socket buffers an integral number of mss units;
3783 * if the mss is larger than the socket buffer, decrease the mss.
3785 so = inp->inp_socket;
3786 SOCKBUF_LOCK(&so->so_snd);
3787 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe)
3788 bufsize = metrics.rmx_sendpipe;
3790 bufsize = so->so_snd.sb_hiwat;
3794 bufsize = roundup(bufsize, mss);
3795 if (bufsize > sb_max)
3797 if (bufsize > so->so_snd.sb_hiwat)
3798 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
3800 SOCKBUF_UNLOCK(&so->so_snd);
3802 * Sanity check: make sure that maxseg will be large
3803 * enough to allow some data on segments even if the
3804 * all the option space is used (40bytes). Otherwise
3805 * funny things may happen in tcp_output.
3807 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3809 tp->t_maxseg = max(mss, 64);
3811 SOCKBUF_LOCK(&so->so_rcv);
3812 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe)
3813 bufsize = metrics.rmx_recvpipe;
3815 bufsize = so->so_rcv.sb_hiwat;
3816 if (bufsize > mss) {
3817 bufsize = roundup(bufsize, mss);
3818 if (bufsize > sb_max)
3820 if (bufsize > so->so_rcv.sb_hiwat)
3821 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
3823 SOCKBUF_UNLOCK(&so->so_rcv);
3825 /* Check the interface for TSO capabilities. */
3826 if (cap.ifcap & CSUM_TSO) {
3827 tp->t_flags |= TF_TSO;
3828 tp->t_tsomax = cap.tsomax;
3829 tp->t_tsomaxsegcount = cap.tsomaxsegcount;
3830 tp->t_tsomaxsegsize = cap.tsomaxsegsize;
3835 * Determine the MSS option to send on an outgoing SYN.
3838 tcp_mssopt(struct in_conninfo *inc)
3841 uint32_t thcmtu = 0;
3842 uint32_t maxmtu = 0;
3845 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3848 if (inc->inc_flags & INC_ISIPV6) {
3849 mss = V_tcp_v6mssdflt;
3850 maxmtu = tcp_maxmtu6(inc, NULL);
3851 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3854 #if defined(INET) && defined(INET6)
3859 mss = V_tcp_mssdflt;
3860 maxmtu = tcp_maxmtu(inc, NULL);
3861 min_protoh = sizeof(struct tcpiphdr);
3864 #if defined(INET6) || defined(INET)
3865 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3868 if (maxmtu && thcmtu)
3869 mss = min(maxmtu, thcmtu) - min_protoh;
3870 else if (maxmtu || thcmtu)
3871 mss = max(maxmtu, thcmtu) - min_protoh;
3878 * On a partial ack arrives, force the retransmission of the
3879 * next unacknowledged segment. Do not clear tp->t_dupacks.
3880 * By setting snd_nxt to ti_ack, this forces retransmission timer to
3884 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th)
3886 tcp_seq onxt = tp->snd_nxt;
3887 uint32_t ocwnd = tp->snd_cwnd;
3888 u_int maxseg = tcp_maxseg(tp);
3890 INP_WLOCK_ASSERT(tp->t_inpcb);
3892 tcp_timer_activate(tp, TT_REXMT, 0);
3894 tp->snd_nxt = th->th_ack;
3896 * Set snd_cwnd to one segment beyond acknowledged offset.
3897 * (tp->snd_una has not yet been updated when this function is called.)
3899 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th);
3900 tp->t_flags |= TF_ACKNOW;
3901 (void) tp->t_fb->tfb_tcp_output(tp);
3902 tp->snd_cwnd = ocwnd;
3903 if (SEQ_GT(onxt, tp->snd_nxt))
3906 * Partial window deflation. Relies on fact that tp->snd_una
3909 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th))
3910 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th);
3913 tp->snd_cwnd += maxseg;
3917 tcp_compute_pipe(struct tcpcb *tp)
3919 return (tp->snd_max - tp->snd_una +
3920 tp->sackhint.sack_bytes_rexmit -
3921 tp->sackhint.sacked_bytes);