2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
3 * The Regents of the University of California. All rights reserved.
4 * Copyright (c) 2007-2008,2010
5 * Swinburne University of Technology, Melbourne, Australia.
6 * Copyright (c) 2009-2010 Lawrence Stewart <lstewart@freebsd.org>
7 * Copyright (c) 2010 The FreeBSD Foundation
8 * Copyright (c) 2010-2011 Juniper Networks, Inc.
11 * Portions of this software were developed at the Centre for Advanced Internet
12 * Architectures, Swinburne University of Technology, by Lawrence Stewart,
13 * James Healy and David Hayes, made possible in part by a grant from the Cisco
14 * University Research Program Fund at Community Foundation Silicon Valley.
16 * Portions of this software were developed at the Centre for Advanced
17 * Internet Architectures, Swinburne University of Technology, Melbourne,
18 * Australia by David Hayes under sponsorship from the FreeBSD Foundation.
20 * Portions of this software were developed by Robert N. M. Watson under
21 * contract to Juniper Networks, Inc.
23 * Redistribution and use in source and binary forms, with or without
24 * modification, are permitted provided that the following conditions
26 * 1. Redistributions of source code must retain the above copyright
27 * notice, this list of conditions and the following disclaimer.
28 * 2. Redistributions in binary form must reproduce the above copyright
29 * notice, this list of conditions and the following disclaimer in the
30 * documentation and/or other materials provided with the distribution.
31 * 4. Neither the name of the University nor the names of its contributors
32 * may be used to endorse or promote products derived from this software
33 * without specific prior written permission.
35 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
36 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
37 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
38 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
39 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
40 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
41 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
42 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
43 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
44 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
47 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95
50 #include <sys/cdefs.h>
51 __FBSDID("$FreeBSD$");
53 #include "opt_ipfw.h" /* for ipfw_fwd */
55 #include "opt_inet6.h"
56 #include "opt_ipsec.h"
57 #include "opt_kdtrace.h"
58 #include "opt_tcpdebug.h"
60 #include <sys/param.h>
61 #include <sys/kernel.h>
62 #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/route.h>
83 #define TCPSTATES /* for logging */
85 #include <netinet/cc.h>
86 #include <netinet/in.h>
87 #include <netinet/in_kdtrace.h>
88 #include <netinet/in_pcb.h>
89 #include <netinet/in_systm.h>
90 #include <netinet/in_var.h>
91 #include <netinet/ip.h>
92 #include <netinet/ip_icmp.h> /* required for icmp_var.h */
93 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */
94 #include <netinet/ip_var.h>
95 #include <netinet/ip_options.h>
96 #include <netinet/ip6.h>
97 #include <netinet/icmp6.h>
98 #include <netinet6/in6_pcb.h>
99 #include <netinet6/ip6_var.h>
100 #include <netinet6/nd6.h>
101 #include <netinet/tcp_fsm.h>
102 #include <netinet/tcp_seq.h>
103 #include <netinet/tcp_timer.h>
104 #include <netinet/tcp_var.h>
105 #include <netinet6/tcp6_var.h>
106 #include <netinet/tcpip.h>
107 #include <netinet/tcp_syncache.h>
109 #include <netinet/tcp_debug.h>
110 #endif /* TCPDEBUG */
112 #include <netinet/tcp_offload.h>
116 #include <netipsec/ipsec.h>
117 #include <netipsec/ipsec6.h>
120 #include <machine/in_cksum.h>
122 #include <security/mac/mac_framework.h>
124 const int tcprexmtthresh = 3;
126 int tcp_log_in_vain = 0;
127 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
129 "Log all incoming TCP segments to closed ports");
131 VNET_DEFINE(int, blackhole) = 0;
132 #define V_blackhole VNET(blackhole)
133 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW,
134 &VNET_NAME(blackhole), 0,
135 "Do not send RST on segments to closed ports");
137 VNET_DEFINE(int, tcp_delack_enabled) = 1;
138 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW,
139 &VNET_NAME(tcp_delack_enabled), 0,
140 "Delay ACK to try and piggyback it onto a data packet");
142 VNET_DEFINE(int, drop_synfin) = 0;
143 #define V_drop_synfin VNET(drop_synfin)
144 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW,
145 &VNET_NAME(drop_synfin), 0,
146 "Drop TCP packets with SYN+FIN set");
148 VNET_DEFINE(int, tcp_do_rfc3042) = 1;
149 #define V_tcp_do_rfc3042 VNET(tcp_do_rfc3042)
150 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_RW,
151 &VNET_NAME(tcp_do_rfc3042), 0,
152 "Enable RFC 3042 (Limited Transmit)");
154 VNET_DEFINE(int, tcp_do_rfc3390) = 1;
155 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW,
156 &VNET_NAME(tcp_do_rfc3390), 0,
157 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
159 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, experimental, CTLFLAG_RW, 0,
160 "Experimental TCP extensions");
162 VNET_DEFINE(int, tcp_do_initcwnd10) = 1;
163 SYSCTL_VNET_INT(_net_inet_tcp_experimental, OID_AUTO, initcwnd10, CTLFLAG_RW,
164 &VNET_NAME(tcp_do_initcwnd10), 0,
165 "Enable RFC 6928 (Increasing initial CWND to 10)");
167 VNET_DEFINE(int, tcp_do_rfc3465) = 1;
168 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_RW,
169 &VNET_NAME(tcp_do_rfc3465), 0,
170 "Enable RFC 3465 (Appropriate Byte Counting)");
172 VNET_DEFINE(int, tcp_abc_l_var) = 2;
173 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_RW,
174 &VNET_NAME(tcp_abc_l_var), 2,
175 "Cap the max cwnd increment during slow-start to this number of segments");
177 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, CTLFLAG_RW, 0, "TCP ECN");
179 VNET_DEFINE(int, tcp_do_ecn) = 0;
180 SYSCTL_VNET_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_RW,
181 &VNET_NAME(tcp_do_ecn), 0,
184 VNET_DEFINE(int, tcp_ecn_maxretries) = 1;
185 SYSCTL_VNET_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_RW,
186 &VNET_NAME(tcp_ecn_maxretries), 0,
187 "Max retries before giving up on ECN");
189 VNET_DEFINE(int, tcp_insecure_rst) = 0;
190 #define V_tcp_insecure_rst VNET(tcp_insecure_rst)
191 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_RW,
192 &VNET_NAME(tcp_insecure_rst), 0,
193 "Follow the old (insecure) criteria for accepting RST packets");
195 VNET_DEFINE(int, tcp_recvspace) = 1024*64;
196 #define V_tcp_recvspace VNET(tcp_recvspace)
197 SYSCTL_VNET_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
198 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size");
200 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1;
201 #define V_tcp_do_autorcvbuf VNET(tcp_do_autorcvbuf)
202 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_RW,
203 &VNET_NAME(tcp_do_autorcvbuf), 0,
204 "Enable automatic receive buffer sizing");
206 VNET_DEFINE(int, tcp_autorcvbuf_inc) = 16*1024;
207 #define V_tcp_autorcvbuf_inc VNET(tcp_autorcvbuf_inc)
208 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_RW,
209 &VNET_NAME(tcp_autorcvbuf_inc), 0,
210 "Incrementor step size of automatic receive buffer");
212 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024;
213 #define V_tcp_autorcvbuf_max VNET(tcp_autorcvbuf_max)
214 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_RW,
215 &VNET_NAME(tcp_autorcvbuf_max), 0,
216 "Max size of automatic receive buffer");
218 VNET_DEFINE(struct inpcbhead, tcb);
219 #define tcb6 tcb /* for KAME src sync over BSD*'s */
220 VNET_DEFINE(struct inpcbinfo, tcbinfo);
222 static void tcp_dooptions(struct tcpopt *, u_char *, int, int);
223 static void tcp_do_segment(struct mbuf *, struct tcphdr *,
224 struct socket *, struct tcpcb *, int, int, uint8_t,
226 static void tcp_dropwithreset(struct mbuf *, struct tcphdr *,
227 struct tcpcb *, int, int);
228 static void tcp_pulloutofband(struct socket *,
229 struct tcphdr *, struct mbuf *, int);
230 static void tcp_xmit_timer(struct tcpcb *, int);
231 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *);
232 static void inline cc_ack_received(struct tcpcb *tp, struct tcphdr *th,
234 static void inline cc_conn_init(struct tcpcb *tp);
235 static void inline cc_post_recovery(struct tcpcb *tp, struct tcphdr *th);
236 static void inline hhook_run_tcp_est_in(struct tcpcb *tp,
237 struct tcphdr *th, struct tcpopt *to);
240 * TCP statistics are stored in an "array" of counter(9)s.
242 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat);
243 VNET_PCPUSTAT_SYSINIT(tcpstat);
244 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat,
245 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
248 VNET_PCPUSTAT_SYSUNINIT(tcpstat);
251 * Kernel module interface for updating tcpstat. The argument is an index
252 * into tcpstat treated as an array.
255 kmod_tcpstat_inc(int statnum)
258 counter_u64_add(VNET(tcpstat)[statnum], 1);
262 * Wrapper for the TCP established input helper hook.
265 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to)
267 struct tcp_hhook_data hhook_data;
269 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) {
274 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data,
280 * CC wrapper hook functions
283 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t type)
285 INP_WLOCK_ASSERT(tp->t_inpcb);
287 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th);
288 if (tp->snd_cwnd <= tp->snd_wnd)
289 tp->ccv->flags |= CCF_CWND_LIMITED;
291 tp->ccv->flags &= ~CCF_CWND_LIMITED;
293 if (type == CC_ACK) {
294 if (tp->snd_cwnd > tp->snd_ssthresh) {
295 tp->t_bytes_acked += min(tp->ccv->bytes_this_ack,
296 V_tcp_abc_l_var * tp->t_maxseg);
297 if (tp->t_bytes_acked >= tp->snd_cwnd) {
298 tp->t_bytes_acked -= tp->snd_cwnd;
299 tp->ccv->flags |= CCF_ABC_SENTAWND;
302 tp->ccv->flags &= ~CCF_ABC_SENTAWND;
303 tp->t_bytes_acked = 0;
307 if (CC_ALGO(tp)->ack_received != NULL) {
308 /* XXXLAS: Find a way to live without this */
309 tp->ccv->curack = th->th_ack;
310 CC_ALGO(tp)->ack_received(tp->ccv, type);
315 cc_conn_init(struct tcpcb *tp)
317 struct hc_metrics_lite metrics;
318 struct inpcb *inp = tp->t_inpcb;
321 INP_WLOCK_ASSERT(tp->t_inpcb);
323 tcp_hc_get(&inp->inp_inc, &metrics);
325 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) {
327 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
328 TCPSTAT_INC(tcps_usedrtt);
329 if (metrics.rmx_rttvar) {
330 tp->t_rttvar = metrics.rmx_rttvar;
331 TCPSTAT_INC(tcps_usedrttvar);
333 /* default variation is +- 1 rtt */
335 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
337 TCPT_RANGESET(tp->t_rxtcur,
338 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
339 tp->t_rttmin, TCPTV_REXMTMAX);
341 if (metrics.rmx_ssthresh) {
343 * There's some sort of gateway or interface
344 * buffer limit on the path. Use this to set
345 * the slow start threshhold, but set the
346 * threshold to no less than 2*mss.
348 tp->snd_ssthresh = max(2 * tp->t_maxseg, metrics.rmx_ssthresh);
349 TCPSTAT_INC(tcps_usedssthresh);
353 * Set the initial slow-start flight size.
355 * RFC5681 Section 3.1 specifies the default conservative values.
356 * RFC3390 specifies slightly more aggressive values.
357 * RFC6928 increases it to ten segments.
359 * If a SYN or SYN/ACK was lost and retransmitted, we have to
360 * reduce the initial CWND to one segment as congestion is likely
361 * requiring us to be cautious.
363 if (tp->snd_cwnd == 1)
364 tp->snd_cwnd = tp->t_maxseg; /* SYN(-ACK) lost */
365 else if (V_tcp_do_initcwnd10)
366 tp->snd_cwnd = min(10 * tp->t_maxseg,
367 max(2 * tp->t_maxseg, 14600));
368 else if (V_tcp_do_rfc3390)
369 tp->snd_cwnd = min(4 * tp->t_maxseg,
370 max(2 * tp->t_maxseg, 4380));
372 /* Per RFC5681 Section 3.1 */
373 if (tp->t_maxseg > 2190)
374 tp->snd_cwnd = 2 * tp->t_maxseg;
375 else if (tp->t_maxseg > 1095)
376 tp->snd_cwnd = 3 * tp->t_maxseg;
378 tp->snd_cwnd = 4 * tp->t_maxseg;
381 if (CC_ALGO(tp)->conn_init != NULL)
382 CC_ALGO(tp)->conn_init(tp->ccv);
386 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type)
388 INP_WLOCK_ASSERT(tp->t_inpcb);
392 if (!IN_FASTRECOVERY(tp->t_flags)) {
393 tp->snd_recover = tp->snd_max;
394 if (tp->t_flags & TF_ECN_PERMIT)
395 tp->t_flags |= TF_ECN_SND_CWR;
399 if (!IN_CONGRECOVERY(tp->t_flags)) {
400 TCPSTAT_INC(tcps_ecn_rcwnd);
401 tp->snd_recover = tp->snd_max;
402 if (tp->t_flags & TF_ECN_PERMIT)
403 tp->t_flags |= TF_ECN_SND_CWR;
408 tp->t_bytes_acked = 0;
409 EXIT_RECOVERY(tp->t_flags);
410 tp->snd_ssthresh = max(2, min(tp->snd_wnd, tp->snd_cwnd) / 2 /
411 tp->t_maxseg) * tp->t_maxseg;
412 tp->snd_cwnd = tp->t_maxseg;
415 TCPSTAT_INC(tcps_sndrexmitbad);
416 /* RTO was unnecessary, so reset everything. */
417 tp->snd_cwnd = tp->snd_cwnd_prev;
418 tp->snd_ssthresh = tp->snd_ssthresh_prev;
419 tp->snd_recover = tp->snd_recover_prev;
420 if (tp->t_flags & TF_WASFRECOVERY)
421 ENTER_FASTRECOVERY(tp->t_flags);
422 if (tp->t_flags & TF_WASCRECOVERY)
423 ENTER_CONGRECOVERY(tp->t_flags);
424 tp->snd_nxt = tp->snd_max;
425 tp->t_flags &= ~TF_PREVVALID;
430 if (CC_ALGO(tp)->cong_signal != NULL) {
432 tp->ccv->curack = th->th_ack;
433 CC_ALGO(tp)->cong_signal(tp->ccv, type);
438 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th)
440 INP_WLOCK_ASSERT(tp->t_inpcb);
442 /* XXXLAS: KASSERT that we're in recovery? */
444 if (CC_ALGO(tp)->post_recovery != NULL) {
445 tp->ccv->curack = th->th_ack;
446 CC_ALGO(tp)->post_recovery(tp->ccv);
448 /* XXXLAS: EXIT_RECOVERY ? */
449 tp->t_bytes_acked = 0;
454 tcp_signature_verify_input(struct mbuf *m, int off0, int tlen, int optlen,
455 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag)
459 tcp_fields_to_net(th);
460 ret = tcp_signature_verify(m, off0, tlen, optlen, to, th, tcpbflag);
461 tcp_fields_to_host(th);
466 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
468 #define ND6_HINT(tp) \
470 if ((tp) && (tp)->t_inpcb && \
471 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \
472 nd6_nud_hint(NULL, NULL, 0); \
479 * Indicate whether this ack should be delayed. We can delay the ack if
480 * - there is no delayed ack timer in progress and
481 * - our last ack wasn't a 0-sized window. We never want to delay
482 * the ack that opens up a 0-sized window and
483 * - delayed acks are enabled or
484 * - this is a half-synchronized T/TCP connection.
485 * - the segment size is not larger than the MSS and LRO wasn't used
488 #define DELAY_ACK(tp, tlen) \
489 ((!tcp_timer_active(tp, TT_DELACK) && \
490 (tp->t_flags & TF_RXWIN0SENT) == 0) && \
491 (tlen <= tp->t_maxopd) && \
492 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
495 * TCP input handling is split into multiple parts:
496 * tcp6_input is a thin wrapper around tcp_input for the extended
497 * ip6_protox[] call format in ip6_input
498 * tcp_input handles primary segment validation, inpcb lookup and
499 * SYN processing on listen sockets
500 * tcp_do_segment processes the ACK and text of the segment for
501 * establishing, established and closing connections
505 tcp6_input(struct mbuf **mp, int *offp, int proto)
507 struct mbuf *m = *mp;
508 struct in6_ifaddr *ia6;
510 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
513 * draft-itojun-ipv6-tcp-to-anycast
514 * better place to put this in?
516 ia6 = ip6_getdstifaddr(m);
517 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
520 ifa_free(&ia6->ia_ifa);
521 ip6 = mtod(m, struct ip6_hdr *);
522 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
523 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
527 ifa_free(&ia6->ia_ifa);
535 tcp_input(struct mbuf *m, int off0)
537 struct tcphdr *th = NULL;
538 struct ip *ip = NULL;
539 struct inpcb *inp = NULL;
540 struct tcpcb *tp = NULL;
541 struct socket *so = NULL;
550 int rstreason = 0; /* For badport_bandlim accounting purposes */
552 uint8_t sig_checked = 0;
555 struct m_tag *fwd_tag = NULL;
557 struct ip6_hdr *ip6 = NULL;
560 const void *ip6 = NULL;
562 struct tcpopt to; /* options in this segment */
563 char *s = NULL; /* address and port logging */
565 #define TI_UNLOCKED 1
570 * The size of tcp_saveipgen must be the size of the max ip header,
573 u_char tcp_saveipgen[IP6_HDR_LEN];
574 struct tcphdr tcp_savetcp;
579 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
583 TCPSTAT_INC(tcps_rcvtotal);
587 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */
589 if (m->m_len < (sizeof(*ip6) + sizeof(*th))) {
590 m = m_pullup(m, sizeof(*ip6) + sizeof(*th));
592 TCPSTAT_INC(tcps_rcvshort);
597 ip6 = mtod(m, struct ip6_hdr *);
598 th = (struct tcphdr *)((caddr_t)ip6 + off0);
599 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
600 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) {
601 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
602 th->th_sum = m->m_pkthdr.csum_data;
604 th->th_sum = in6_cksum_pseudo(ip6, tlen,
605 IPPROTO_TCP, m->m_pkthdr.csum_data);
606 th->th_sum ^= 0xffff;
608 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen);
610 TCPSTAT_INC(tcps_rcvbadsum);
615 * Be proactive about unspecified IPv6 address in source.
616 * As we use all-zero to indicate unbounded/unconnected pcb,
617 * unspecified IPv6 address can be used to confuse us.
619 * Note that packets with unspecified IPv6 destination is
620 * already dropped in ip6_input.
622 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
628 #if defined(INET) && defined(INET6)
634 * Get IP and TCP header together in first mbuf.
635 * Note: IP leaves IP header in first mbuf.
637 if (off0 > sizeof (struct ip)) {
639 off0 = sizeof(struct ip);
641 if (m->m_len < sizeof (struct tcpiphdr)) {
642 if ((m = m_pullup(m, sizeof (struct tcpiphdr)))
644 TCPSTAT_INC(tcps_rcvshort);
648 ip = mtod(m, struct ip *);
649 th = (struct tcphdr *)((caddr_t)ip + off0);
650 tlen = ntohs(ip->ip_len) - off0;
652 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
653 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
654 th->th_sum = m->m_pkthdr.csum_data;
656 th->th_sum = in_pseudo(ip->ip_src.s_addr,
658 htonl(m->m_pkthdr.csum_data + tlen +
660 th->th_sum ^= 0xffff;
662 struct ipovly *ipov = (struct ipovly *)ip;
665 * Checksum extended TCP header and data.
668 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
669 ipov->ih_len = htons(tlen);
670 th->th_sum = in_cksum(m, len);
671 /* Reset length for SDT probes. */
672 ip->ip_len = htons(tlen + off0);
676 TCPSTAT_INC(tcps_rcvbadsum);
679 /* Re-initialization for later version check */
680 ip->ip_v = IPVERSION;
686 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
688 #if defined(INET) && defined(INET6)
696 * Check that TCP offset makes sense,
697 * pull out TCP options and adjust length. XXX
699 off = th->th_off << 2;
700 if (off < sizeof (struct tcphdr) || off > tlen) {
701 TCPSTAT_INC(tcps_rcvbadoff);
704 tlen -= off; /* tlen is used instead of ti->ti_len */
705 if (off > sizeof (struct tcphdr)) {
708 IP6_EXTHDR_CHECK(m, off0, off, );
709 ip6 = mtod(m, struct ip6_hdr *);
710 th = (struct tcphdr *)((caddr_t)ip6 + off0);
713 #if defined(INET) && defined(INET6)
718 if (m->m_len < sizeof(struct ip) + off) {
719 if ((m = m_pullup(m, sizeof (struct ip) + off))
721 TCPSTAT_INC(tcps_rcvshort);
724 ip = mtod(m, struct ip *);
725 th = (struct tcphdr *)((caddr_t)ip + off0);
729 optlen = off - sizeof (struct tcphdr);
730 optp = (u_char *)(th + 1);
732 thflags = th->th_flags;
735 * Convert TCP protocol specific fields to host format.
737 tcp_fields_to_host(th);
740 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options.
742 drop_hdrlen = off0 + off;
745 * Locate pcb for segment; if we're likely to add or remove a
746 * connection then first acquire pcbinfo lock. There are two cases
747 * where we might discover later we need a write lock despite the
748 * flags: ACKs moving a connection out of the syncache, and ACKs for
749 * a connection in TIMEWAIT.
751 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0) {
752 INP_INFO_WLOCK(&V_tcbinfo);
753 ti_locked = TI_WLOCKED;
755 ti_locked = TI_UNLOCKED;
758 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
762 (isipv6 && (m->m_flags & M_IP6_NEXTHOP))
764 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP))
767 #if defined(INET) && !defined(INET6)
768 (m->m_flags & M_IP_NEXTHOP)
771 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
775 if (ti_locked == TI_WLOCKED) {
776 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
778 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
782 if (isipv6 && fwd_tag != NULL) {
783 struct sockaddr_in6 *next_hop6;
785 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1);
787 * Transparently forwarded. Pretend to be the destination.
788 * Already got one like this?
790 inp = in6_pcblookup_mbuf(&V_tcbinfo,
791 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport,
792 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m);
795 * It's new. Try to find the ambushing socket.
796 * Because we've rewritten the destination address,
797 * any hardware-generated hash is ignored.
799 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src,
800 th->th_sport, &next_hop6->sin6_addr,
801 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) :
802 th->th_dport, INPLOOKUP_WILDCARD |
803 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
806 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src,
807 th->th_sport, &ip6->ip6_dst, th->th_dport,
808 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
809 m->m_pkthdr.rcvif, m);
812 #if defined(INET6) && defined(INET)
816 if (fwd_tag != NULL) {
817 struct sockaddr_in *next_hop;
819 next_hop = (struct sockaddr_in *)(fwd_tag+1);
821 * Transparently forwarded. Pretend to be the destination.
822 * already got one like this?
824 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport,
825 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB,
826 m->m_pkthdr.rcvif, m);
829 * It's new. Try to find the ambushing socket.
830 * Because we've rewritten the destination address,
831 * any hardware-generated hash is ignored.
833 inp = in_pcblookup(&V_tcbinfo, ip->ip_src,
834 th->th_sport, next_hop->sin_addr,
835 next_hop->sin_port ? ntohs(next_hop->sin_port) :
836 th->th_dport, INPLOOKUP_WILDCARD |
837 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
840 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src,
841 th->th_sport, ip->ip_dst, th->th_dport,
842 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
843 m->m_pkthdr.rcvif, m);
847 * If the INPCB does not exist then all data in the incoming
848 * segment is discarded and an appropriate RST is sent back.
849 * XXX MRT Send RST using which routing table?
853 * Log communication attempts to ports that are not
856 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) ||
857 tcp_log_in_vain == 2) {
858 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6)))
859 log(LOG_INFO, "%s; %s: Connection attempt "
860 "to closed port\n", s, __func__);
863 * When blackholing do not respond with a RST but
864 * completely ignore the segment and drop it.
866 if ((V_blackhole == 1 && (thflags & TH_SYN)) ||
870 rstreason = BANDLIM_RST_CLOSEDPORT;
873 INP_WLOCK_ASSERT(inp);
874 if ((inp->inp_flowtype == M_HASHTYPE_NONE) &&
875 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) &&
876 ((inp->inp_socket == NULL) ||
877 (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0)) {
878 inp->inp_flowid = m->m_pkthdr.flowid;
879 inp->inp_flowtype = M_HASHTYPE_GET(m);
883 if (isipv6 && ipsec6_in_reject(m, inp)) {
884 IPSEC6STAT_INC(ips_in_polvio);
888 if (ipsec4_in_reject(m, inp) != 0) {
889 IPSECSTAT_INC(ips_in_polvio);
895 * Check the minimum TTL for socket.
897 if (inp->inp_ip_minttl != 0) {
899 if (isipv6 && inp->inp_ip_minttl > ip6->ip6_hlim)
903 if (inp->inp_ip_minttl > ip->ip_ttl)
908 * A previous connection in TIMEWAIT state is supposed to catch stray
909 * or duplicate segments arriving late. If this segment was a
910 * legitimate new connection attempt, the old INPCB gets removed and
911 * we can try again to find a listening socket.
913 * At this point, due to earlier optimism, we may hold only an inpcb
914 * lock, and not the inpcbinfo write lock. If so, we need to try to
915 * acquire it, or if that fails, acquire a reference on the inpcb,
916 * drop all locks, acquire a global write lock, and then re-acquire
917 * the inpcb lock. We may at that point discover that another thread
918 * has tried to free the inpcb, in which case we need to loop back
919 * and try to find a new inpcb to deliver to.
921 * XXXRW: It may be time to rethink timewait locking.
924 if (inp->inp_flags & INP_TIMEWAIT) {
925 if (ti_locked == TI_UNLOCKED) {
926 if (INP_INFO_TRY_WLOCK(&V_tcbinfo) == 0) {
929 INP_INFO_WLOCK(&V_tcbinfo);
930 ti_locked = TI_WLOCKED;
932 if (in_pcbrele_wlocked(inp)) {
937 ti_locked = TI_WLOCKED;
939 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
941 if (thflags & TH_SYN)
942 tcp_dooptions(&to, optp, optlen, TO_SYN);
944 * NB: tcp_twcheck unlocks the INP and frees the mbuf.
946 if (tcp_twcheck(inp, &to, th, m, tlen))
948 INP_INFO_WUNLOCK(&V_tcbinfo);
952 * The TCPCB may no longer exist if the connection is winding
953 * down or it is in the CLOSED state. Either way we drop the
954 * segment and send an appropriate response.
957 if (tp == NULL || tp->t_state == TCPS_CLOSED) {
958 rstreason = BANDLIM_RST_CLOSEDPORT;
963 if (tp->t_flags & TF_TOE) {
964 tcp_offload_input(tp, m);
965 m = NULL; /* consumed by the TOE driver */
971 * We've identified a valid inpcb, but it could be that we need an
972 * inpcbinfo write lock but don't hold it. In this case, attempt to
973 * acquire using the same strategy as the TIMEWAIT case above. If we
974 * relock, we have to jump back to 'relocked' as the connection might
975 * now be in TIMEWAIT.
978 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0)
979 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
981 if (tp->t_state != TCPS_ESTABLISHED) {
982 if (ti_locked == TI_UNLOCKED) {
983 if (INP_INFO_TRY_WLOCK(&V_tcbinfo) == 0) {
986 INP_INFO_WLOCK(&V_tcbinfo);
987 ti_locked = TI_WLOCKED;
989 if (in_pcbrele_wlocked(inp)) {
995 ti_locked = TI_WLOCKED;
997 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1001 INP_WLOCK_ASSERT(inp);
1002 if (mac_inpcb_check_deliver(inp, m))
1005 so = inp->inp_socket;
1006 KASSERT(so != NULL, ("%s: so == NULL", __func__));
1008 if (so->so_options & SO_DEBUG) {
1009 ostate = tp->t_state;
1012 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
1015 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
1018 #endif /* TCPDEBUG */
1020 * When the socket is accepting connections (the INPCB is in LISTEN
1021 * state) we look into the SYN cache if this is a new connection
1022 * attempt or the completion of a previous one. Because listen
1023 * sockets are never in TCPS_ESTABLISHED, the V_tcbinfo lock will be
1024 * held in this case.
1026 if (so->so_options & SO_ACCEPTCONN) {
1027 struct in_conninfo inc;
1029 KASSERT(tp->t_state == TCPS_LISTEN, ("%s: so accepting but "
1030 "tp not listening", __func__));
1031 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1033 bzero(&inc, sizeof(inc));
1036 inc.inc_flags |= INC_ISIPV6;
1037 inc.inc6_faddr = ip6->ip6_src;
1038 inc.inc6_laddr = ip6->ip6_dst;
1042 inc.inc_faddr = ip->ip_src;
1043 inc.inc_laddr = ip->ip_dst;
1045 inc.inc_fport = th->th_sport;
1046 inc.inc_lport = th->th_dport;
1047 inc.inc_fibnum = so->so_fibnum;
1050 * Check for an existing connection attempt in syncache if
1051 * the flag is only ACK. A successful lookup creates a new
1052 * socket appended to the listen queue in SYN_RECEIVED state.
1054 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
1056 * Parse the TCP options here because
1057 * syncookies need access to the reflected
1060 tcp_dooptions(&to, optp, optlen, 0);
1062 * NB: syncache_expand() doesn't unlock
1063 * inp and tcpinfo locks.
1065 if (!syncache_expand(&inc, &to, th, &so, m)) {
1067 * No syncache entry or ACK was not
1068 * for our SYN/ACK. Send a RST.
1069 * NB: syncache did its own logging
1070 * of the failure cause.
1072 rstreason = BANDLIM_RST_OPENPORT;
1077 * We completed the 3-way handshake
1078 * but could not allocate a socket
1079 * either due to memory shortage,
1080 * listen queue length limits or
1081 * global socket limits. Send RST
1082 * or wait and have the remote end
1083 * retransmit the ACK for another
1086 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1087 log(LOG_DEBUG, "%s; %s: Listen socket: "
1088 "Socket allocation failed due to "
1089 "limits or memory shortage, %s\n",
1091 V_tcp_sc_rst_sock_fail ?
1092 "sending RST" : "try again");
1093 if (V_tcp_sc_rst_sock_fail) {
1094 rstreason = BANDLIM_UNLIMITED;
1100 * Socket is created in state SYN_RECEIVED.
1101 * Unlock the listen socket, lock the newly
1102 * created socket and update the tp variable.
1104 INP_WUNLOCK(inp); /* listen socket */
1105 inp = sotoinpcb(so);
1106 INP_WLOCK(inp); /* new connection */
1107 tp = intotcpcb(inp);
1108 KASSERT(tp->t_state == TCPS_SYN_RECEIVED,
1109 ("%s: ", __func__));
1110 #ifdef TCP_SIGNATURE
1111 if (sig_checked == 0) {
1112 tcp_dooptions(&to, optp, optlen,
1113 (thflags & TH_SYN) ? TO_SYN : 0);
1114 if (!tcp_signature_verify_input(m, off0, tlen,
1115 optlen, &to, th, tp->t_flags)) {
1118 * In SYN_SENT state if it receives an
1119 * RST, it is allowed for further
1122 if ((thflags & TH_RST) == 0 ||
1123 (tp->t_state == TCPS_SYN_SENT) == 0)
1131 * Process the segment and the data it
1132 * contains. tcp_do_segment() consumes
1133 * the mbuf chain and unlocks the inpcb.
1135 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen,
1137 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1141 * Segment flag validation for new connection attempts:
1143 * Our (SYN|ACK) response was rejected.
1144 * Check with syncache and remove entry to prevent
1147 * NB: syncache_chkrst does its own logging of failure
1150 if (thflags & TH_RST) {
1151 syncache_chkrst(&inc, th);
1155 * We can't do anything without SYN.
1157 if ((thflags & TH_SYN) == 0) {
1158 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1159 log(LOG_DEBUG, "%s; %s: Listen socket: "
1160 "SYN is missing, segment ignored\n",
1162 TCPSTAT_INC(tcps_badsyn);
1166 * (SYN|ACK) is bogus on a listen socket.
1168 if (thflags & TH_ACK) {
1169 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1170 log(LOG_DEBUG, "%s; %s: Listen socket: "
1171 "SYN|ACK invalid, segment rejected\n",
1173 syncache_badack(&inc); /* XXX: Not needed! */
1174 TCPSTAT_INC(tcps_badsyn);
1175 rstreason = BANDLIM_RST_OPENPORT;
1179 * If the drop_synfin option is enabled, drop all
1180 * segments with both the SYN and FIN bits set.
1181 * This prevents e.g. nmap from identifying the
1183 * XXX: Poor reasoning. nmap has other methods
1184 * and is constantly refining its stack detection
1186 * XXX: This is a violation of the TCP specification
1187 * and was used by RFC1644.
1189 if ((thflags & TH_FIN) && V_drop_synfin) {
1190 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1191 log(LOG_DEBUG, "%s; %s: Listen socket: "
1192 "SYN|FIN segment ignored (based on "
1193 "sysctl setting)\n", s, __func__);
1194 TCPSTAT_INC(tcps_badsyn);
1198 * Segment's flags are (SYN) or (SYN|FIN).
1200 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored
1201 * as they do not affect the state of the TCP FSM.
1202 * The data pointed to by TH_URG and th_urp is ignored.
1204 KASSERT((thflags & (TH_RST|TH_ACK)) == 0,
1205 ("%s: Listen socket: TH_RST or TH_ACK set", __func__));
1206 KASSERT(thflags & (TH_SYN),
1207 ("%s: Listen socket: TH_SYN not set", __func__));
1210 * If deprecated address is forbidden,
1211 * we do not accept SYN to deprecated interface
1212 * address to prevent any new inbound connection from
1213 * getting established.
1214 * When we do not accept SYN, we send a TCP RST,
1215 * with deprecated source address (instead of dropping
1216 * it). We compromise it as it is much better for peer
1217 * to send a RST, and RST will be the final packet
1220 * If we do not forbid deprecated addresses, we accept
1221 * the SYN packet. RFC2462 does not suggest dropping
1223 * If we decipher RFC2462 5.5.4, it says like this:
1224 * 1. use of deprecated addr with existing
1225 * communication is okay - "SHOULD continue to be
1227 * 2. use of it with new communication:
1228 * (2a) "SHOULD NOT be used if alternate address
1229 * with sufficient scope is available"
1230 * (2b) nothing mentioned otherwise.
1231 * Here we fall into (2b) case as we have no choice in
1232 * our source address selection - we must obey the peer.
1234 * The wording in RFC2462 is confusing, and there are
1235 * multiple description text for deprecated address
1236 * handling - worse, they are not exactly the same.
1237 * I believe 5.5.4 is the best one, so we follow 5.5.4.
1239 if (isipv6 && !V_ip6_use_deprecated) {
1240 struct in6_ifaddr *ia6;
1242 ia6 = ip6_getdstifaddr(m);
1244 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1245 ifa_free(&ia6->ia_ifa);
1246 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1247 log(LOG_DEBUG, "%s; %s: Listen socket: "
1248 "Connection attempt to deprecated "
1249 "IPv6 address rejected\n",
1251 rstreason = BANDLIM_RST_OPENPORT;
1255 ifa_free(&ia6->ia_ifa);
1259 * Basic sanity checks on incoming SYN requests:
1260 * Don't respond if the destination is a link layer
1261 * broadcast according to RFC1122 4.2.3.10, p. 104.
1262 * If it is from this socket it must be forged.
1263 * Don't respond if the source or destination is a
1264 * global or subnet broad- or multicast address.
1265 * Note that it is quite possible to receive unicast
1266 * link-layer packets with a broadcast IP address. Use
1267 * in_broadcast() to find them.
1269 if (m->m_flags & (M_BCAST|M_MCAST)) {
1270 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1271 log(LOG_DEBUG, "%s; %s: Listen socket: "
1272 "Connection attempt from broad- or multicast "
1273 "link layer address ignored\n", s, __func__);
1278 if (th->th_dport == th->th_sport &&
1279 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) {
1280 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1281 log(LOG_DEBUG, "%s; %s: Listen socket: "
1282 "Connection attempt to/from self "
1283 "ignored\n", s, __func__);
1286 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1287 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) {
1288 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1289 log(LOG_DEBUG, "%s; %s: Listen socket: "
1290 "Connection attempt from/to multicast "
1291 "address ignored\n", s, __func__);
1296 #if defined(INET) && defined(INET6)
1301 if (th->th_dport == th->th_sport &&
1302 ip->ip_dst.s_addr == ip->ip_src.s_addr) {
1303 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1304 log(LOG_DEBUG, "%s; %s: Listen socket: "
1305 "Connection attempt from/to self "
1306 "ignored\n", s, __func__);
1309 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1310 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1311 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1312 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
1313 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1314 log(LOG_DEBUG, "%s; %s: Listen socket: "
1315 "Connection attempt from/to broad- "
1316 "or multicast address ignored\n",
1323 * SYN appears to be valid. Create compressed TCP state
1327 if (so->so_options & SO_DEBUG)
1328 tcp_trace(TA_INPUT, ostate, tp,
1329 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1331 tcp_dooptions(&to, optp, optlen, TO_SYN);
1332 syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL);
1334 * Entry added to syncache and mbuf consumed.
1335 * Everything already unlocked by syncache_add().
1337 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1339 } else if (tp->t_state == TCPS_LISTEN) {
1341 * When a listen socket is torn down the SO_ACCEPTCONN
1342 * flag is removed first while connections are drained
1343 * from the accept queue in a unlock/lock cycle of the
1344 * ACCEPT_LOCK, opening a race condition allowing a SYN
1345 * attempt go through unhandled.
1350 #ifdef TCP_SIGNATURE
1351 if (sig_checked == 0) {
1352 tcp_dooptions(&to, optp, optlen,
1353 (thflags & TH_SYN) ? TO_SYN : 0);
1354 if (!tcp_signature_verify_input(m, off0, tlen, optlen, &to,
1358 * In SYN_SENT state if it receives an RST, it is
1359 * allowed for further processing.
1361 if ((thflags & TH_RST) == 0 ||
1362 (tp->t_state == TCPS_SYN_SENT) == 0)
1369 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th);
1372 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later
1373 * state. tcp_do_segment() always consumes the mbuf chain, unlocks
1374 * the inpcb, and unlocks pcbinfo.
1376 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos, ti_locked);
1377 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1381 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th);
1383 if (ti_locked == TI_WLOCKED) {
1384 INP_INFO_WUNLOCK(&V_tcbinfo);
1385 ti_locked = TI_UNLOCKED;
1389 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset "
1390 "ti_locked: %d", __func__, ti_locked));
1391 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1396 tcp_dropwithreset(m, th, tp, tlen, rstreason);
1399 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
1400 m = NULL; /* mbuf chain got consumed. */
1405 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th);
1407 if (ti_locked == TI_WLOCKED) {
1408 INP_INFO_WUNLOCK(&V_tcbinfo);
1409 ti_locked = TI_UNLOCKED;
1413 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock "
1414 "ti_locked: %d", __func__, ti_locked));
1415 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1423 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1431 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so,
1432 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos,
1435 int thflags, acked, ourfinisacked, needoutput = 0;
1436 int rstreason, todrop, win;
1439 struct in_conninfo *inc;
1445 * The size of tcp_saveipgen must be the size of the max ip header,
1448 u_char tcp_saveipgen[IP6_HDR_LEN];
1449 struct tcphdr tcp_savetcp;
1452 thflags = th->th_flags;
1453 inc = &tp->t_inpcb->inp_inc;
1454 tp->sackhint.last_sack_ack = 0;
1457 * If this is either a state-changing packet or current state isn't
1458 * established, we require a write lock on tcbinfo. Otherwise, we
1459 * allow the tcbinfo to be in either alocked or unlocked, as the
1460 * caller may have unnecessarily acquired a write lock due to a race.
1462 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 ||
1463 tp->t_state != TCPS_ESTABLISHED) {
1464 KASSERT(ti_locked == TI_WLOCKED, ("%s ti_locked %d for "
1465 "SYN/FIN/RST/!EST", __func__, ti_locked));
1466 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1469 if (ti_locked == TI_WLOCKED)
1470 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1472 KASSERT(ti_locked == TI_UNLOCKED, ("%s: EST "
1473 "ti_locked: %d", __func__, ti_locked));
1474 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1478 INP_WLOCK_ASSERT(tp->t_inpcb);
1479 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN",
1481 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT",
1485 * Segment received on connection.
1486 * Reset idle time and keep-alive timer.
1487 * XXX: This should be done after segment
1488 * validation to ignore broken/spoofed segs.
1490 tp->t_rcvtime = ticks;
1491 if (TCPS_HAVEESTABLISHED(tp->t_state))
1492 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
1495 * Unscale the window into a 32-bit value.
1496 * For the SYN_SENT state the scale is zero.
1498 tiwin = th->th_win << tp->snd_scale;
1501 * TCP ECN processing.
1503 if (tp->t_flags & TF_ECN_PERMIT) {
1504 if (thflags & TH_CWR)
1505 tp->t_flags &= ~TF_ECN_SND_ECE;
1506 switch (iptos & IPTOS_ECN_MASK) {
1508 tp->t_flags |= TF_ECN_SND_ECE;
1509 TCPSTAT_INC(tcps_ecn_ce);
1511 case IPTOS_ECN_ECT0:
1512 TCPSTAT_INC(tcps_ecn_ect0);
1514 case IPTOS_ECN_ECT1:
1515 TCPSTAT_INC(tcps_ecn_ect1);
1518 /* Congestion experienced. */
1519 if (thflags & TH_ECE) {
1520 cc_cong_signal(tp, th, CC_ECN);
1525 * Parse options on any incoming segment.
1527 tcp_dooptions(&to, (u_char *)(th + 1),
1528 (th->th_off << 2) - sizeof(struct tcphdr),
1529 (thflags & TH_SYN) ? TO_SYN : 0);
1532 * If echoed timestamp is later than the current time,
1533 * fall back to non RFC1323 RTT calculation. Normalize
1534 * timestamp if syncookies were used when this connection
1537 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1538 to.to_tsecr -= tp->ts_offset;
1539 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks()))
1543 * If timestamps were negotiated during SYN/ACK they should
1544 * appear on every segment during this session and vice versa.
1546 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) {
1547 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1548 log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1549 "no action\n", s, __func__);
1553 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) {
1554 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1555 log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
1556 "no action\n", s, __func__);
1562 * Process options only when we get SYN/ACK back. The SYN case
1563 * for incoming connections is handled in tcp_syncache.
1564 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
1565 * or <SYN,ACK>) segment itself is never scaled.
1566 * XXX this is traditional behavior, may need to be cleaned up.
1568 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1569 if ((to.to_flags & TOF_SCALE) &&
1570 (tp->t_flags & TF_REQ_SCALE)) {
1571 tp->t_flags |= TF_RCVD_SCALE;
1572 tp->snd_scale = to.to_wscale;
1575 * Initial send window. It will be updated with
1576 * the next incoming segment to the scaled value.
1578 tp->snd_wnd = th->th_win;
1579 if (to.to_flags & TOF_TS) {
1580 tp->t_flags |= TF_RCVD_TSTMP;
1581 tp->ts_recent = to.to_tsval;
1582 tp->ts_recent_age = tcp_ts_getticks();
1584 if (to.to_flags & TOF_MSS)
1585 tcp_mss(tp, to.to_mss);
1586 if ((tp->t_flags & TF_SACK_PERMIT) &&
1587 (to.to_flags & TOF_SACKPERM) == 0)
1588 tp->t_flags &= ~TF_SACK_PERMIT;
1592 * Header prediction: check for the two common cases
1593 * of a uni-directional data xfer. If the packet has
1594 * no control flags, is in-sequence, the window didn't
1595 * change and we're not retransmitting, it's a
1596 * candidate. If the length is zero and the ack moved
1597 * forward, we're the sender side of the xfer. Just
1598 * free the data acked & wake any higher level process
1599 * that was blocked waiting for space. If the length
1600 * is non-zero and the ack didn't move, we're the
1601 * receiver side. If we're getting packets in-order
1602 * (the reassembly queue is empty), add the data to
1603 * the socket buffer and note that we need a delayed ack.
1604 * Make sure that the hidden state-flags are also off.
1605 * Since we check for TCPS_ESTABLISHED first, it can only
1608 if (tp->t_state == TCPS_ESTABLISHED &&
1609 th->th_seq == tp->rcv_nxt &&
1610 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1611 tp->snd_nxt == tp->snd_max &&
1612 tiwin && tiwin == tp->snd_wnd &&
1613 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1614 LIST_EMPTY(&tp->t_segq) &&
1615 ((to.to_flags & TOF_TS) == 0 ||
1616 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) {
1619 * If last ACK falls within this segment's sequence numbers,
1620 * record the timestamp.
1621 * NOTE that the test is modified according to the latest
1622 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1624 if ((to.to_flags & TOF_TS) != 0 &&
1625 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1626 tp->ts_recent_age = tcp_ts_getticks();
1627 tp->ts_recent = to.to_tsval;
1631 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1632 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1633 !IN_RECOVERY(tp->t_flags) &&
1634 (to.to_flags & TOF_SACK) == 0 &&
1635 TAILQ_EMPTY(&tp->snd_holes)) {
1637 * This is a pure ack for outstanding data.
1639 if (ti_locked == TI_WLOCKED)
1640 INP_INFO_WUNLOCK(&V_tcbinfo);
1641 ti_locked = TI_UNLOCKED;
1643 TCPSTAT_INC(tcps_predack);
1646 * "bad retransmit" recovery.
1648 if (tp->t_rxtshift == 1 &&
1649 tp->t_flags & TF_PREVVALID &&
1650 (int)(ticks - tp->t_badrxtwin) < 0) {
1651 cc_cong_signal(tp, th, CC_RTO_ERR);
1655 * Recalculate the transmit timer / rtt.
1657 * Some boxes send broken timestamp replies
1658 * during the SYN+ACK phase, ignore
1659 * timestamps of 0 or we could calculate a
1660 * huge RTT and blow up the retransmit timer.
1662 if ((to.to_flags & TOF_TS) != 0 &&
1666 t = tcp_ts_getticks() - to.to_tsecr;
1667 if (!tp->t_rttlow || tp->t_rttlow > t)
1670 TCP_TS_TO_TICKS(t) + 1);
1671 } else if (tp->t_rtttime &&
1672 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1673 if (!tp->t_rttlow ||
1674 tp->t_rttlow > ticks - tp->t_rtttime)
1675 tp->t_rttlow = ticks - tp->t_rtttime;
1677 ticks - tp->t_rtttime);
1679 acked = BYTES_THIS_ACK(tp, th);
1681 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
1682 hhook_run_tcp_est_in(tp, th, &to);
1684 TCPSTAT_INC(tcps_rcvackpack);
1685 TCPSTAT_ADD(tcps_rcvackbyte, acked);
1686 sbdrop(&so->so_snd, acked);
1687 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1688 SEQ_LEQ(th->th_ack, tp->snd_recover))
1689 tp->snd_recover = th->th_ack - 1;
1692 * Let the congestion control algorithm update
1693 * congestion control related information. This
1694 * typically means increasing the congestion
1697 cc_ack_received(tp, th, CC_ACK);
1699 tp->snd_una = th->th_ack;
1701 * Pull snd_wl2 up to prevent seq wrap relative
1704 tp->snd_wl2 = th->th_ack;
1707 ND6_HINT(tp); /* Some progress has been made. */
1710 * If all outstanding data are acked, stop
1711 * retransmit timer, otherwise restart timer
1712 * using current (possibly backed-off) value.
1713 * If process is waiting for space,
1714 * wakeup/selwakeup/signal. If data
1715 * are ready to send, let tcp_output
1716 * decide between more output or persist.
1719 if (so->so_options & SO_DEBUG)
1720 tcp_trace(TA_INPUT, ostate, tp,
1721 (void *)tcp_saveipgen,
1724 if (tp->snd_una == tp->snd_max)
1725 tcp_timer_activate(tp, TT_REXMT, 0);
1726 else if (!tcp_timer_active(tp, TT_PERSIST))
1727 tcp_timer_activate(tp, TT_REXMT,
1730 if (so->so_snd.sb_cc)
1731 (void) tcp_output(tp);
1734 } else if (th->th_ack == tp->snd_una &&
1735 tlen <= sbspace(&so->so_rcv)) {
1736 int newsize = 0; /* automatic sockbuf scaling */
1739 * This is a pure, in-sequence data packet with
1740 * nothing on the reassembly queue and we have enough
1741 * buffer space to take it.
1743 if (ti_locked == TI_WLOCKED)
1744 INP_INFO_WUNLOCK(&V_tcbinfo);
1745 ti_locked = TI_UNLOCKED;
1747 /* Clean receiver SACK report if present */
1748 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks)
1749 tcp_clean_sackreport(tp);
1750 TCPSTAT_INC(tcps_preddat);
1751 tp->rcv_nxt += tlen;
1753 * Pull snd_wl1 up to prevent seq wrap relative to
1756 tp->snd_wl1 = th->th_seq;
1758 * Pull rcv_up up to prevent seq wrap relative to
1761 tp->rcv_up = tp->rcv_nxt;
1762 TCPSTAT_INC(tcps_rcvpack);
1763 TCPSTAT_ADD(tcps_rcvbyte, tlen);
1764 ND6_HINT(tp); /* Some progress has been made */
1766 if (so->so_options & SO_DEBUG)
1767 tcp_trace(TA_INPUT, ostate, tp,
1768 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1771 * Automatic sizing of receive socket buffer. Often the send
1772 * buffer size is not optimally adjusted to the actual network
1773 * conditions at hand (delay bandwidth product). Setting the
1774 * buffer size too small limits throughput on links with high
1775 * bandwidth and high delay (eg. trans-continental/oceanic links).
1777 * On the receive side the socket buffer memory is only rarely
1778 * used to any significant extent. This allows us to be much
1779 * more aggressive in scaling the receive socket buffer. For
1780 * the case that the buffer space is actually used to a large
1781 * extent and we run out of kernel memory we can simply drop
1782 * the new segments; TCP on the sender will just retransmit it
1783 * later. Setting the buffer size too big may only consume too
1784 * much kernel memory if the application doesn't read() from
1785 * the socket or packet loss or reordering makes use of the
1788 * The criteria to step up the receive buffer one notch are:
1789 * 1. the number of bytes received during the time it takes
1790 * one timestamp to be reflected back to us (the RTT);
1791 * 2. received bytes per RTT is within seven eighth of the
1792 * current socket buffer size;
1793 * 3. receive buffer size has not hit maximal automatic size;
1795 * This algorithm does one step per RTT at most and only if
1796 * we receive a bulk stream w/o packet losses or reorderings.
1797 * Shrinking the buffer during idle times is not necessary as
1798 * it doesn't consume any memory when idle.
1800 * TODO: Only step up if the application is actually serving
1801 * the buffer to better manage the socket buffer resources.
1803 if (V_tcp_do_autorcvbuf &&
1805 (so->so_rcv.sb_flags & SB_AUTOSIZE)) {
1806 if (TSTMP_GT(to.to_tsecr, tp->rfbuf_ts) &&
1807 to.to_tsecr - tp->rfbuf_ts < hz) {
1809 (so->so_rcv.sb_hiwat / 8 * 7) &&
1810 so->so_rcv.sb_hiwat <
1811 V_tcp_autorcvbuf_max) {
1813 min(so->so_rcv.sb_hiwat +
1814 V_tcp_autorcvbuf_inc,
1815 V_tcp_autorcvbuf_max);
1817 /* Start over with next RTT. */
1821 tp->rfbuf_cnt += tlen; /* add up */
1824 /* Add data to socket buffer. */
1825 SOCKBUF_LOCK(&so->so_rcv);
1826 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1830 * Set new socket buffer size.
1831 * Give up when limit is reached.
1834 if (!sbreserve_locked(&so->so_rcv,
1836 so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1837 m_adj(m, drop_hdrlen); /* delayed header drop */
1838 sbappendstream_locked(&so->so_rcv, m);
1840 /* NB: sorwakeup_locked() does an implicit unlock. */
1841 sorwakeup_locked(so);
1842 if (DELAY_ACK(tp, tlen)) {
1843 tp->t_flags |= TF_DELACK;
1845 tp->t_flags |= TF_ACKNOW;
1853 * Calculate amount of space in receive window,
1854 * and then do TCP input processing.
1855 * Receive window is amount of space in rcv queue,
1856 * but not less than advertised window.
1858 win = sbspace(&so->so_rcv);
1861 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1863 /* Reset receive buffer auto scaling when not in bulk receive mode. */
1867 switch (tp->t_state) {
1870 * If the state is SYN_RECEIVED:
1871 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1873 case TCPS_SYN_RECEIVED:
1874 if ((thflags & TH_ACK) &&
1875 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1876 SEQ_GT(th->th_ack, tp->snd_max))) {
1877 rstreason = BANDLIM_RST_OPENPORT;
1883 * If the state is SYN_SENT:
1884 * if seg contains an ACK, but not for our SYN, drop the input.
1885 * if seg contains a RST, then drop the connection.
1886 * if seg does not contain SYN, then drop it.
1887 * Otherwise this is an acceptable SYN segment
1888 * initialize tp->rcv_nxt and tp->irs
1889 * if seg contains ack then advance tp->snd_una
1890 * if seg contains an ECE and ECN support is enabled, the stream
1892 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1893 * arrange for segment to be acked (eventually)
1894 * continue processing rest of data/controls, beginning with URG
1897 if ((thflags & TH_ACK) &&
1898 (SEQ_LEQ(th->th_ack, tp->iss) ||
1899 SEQ_GT(th->th_ack, tp->snd_max))) {
1900 rstreason = BANDLIM_UNLIMITED;
1903 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) {
1904 TCP_PROBE5(connect__refused, NULL, tp,
1905 mtod(m, const char *), tp, th);
1906 tp = tcp_drop(tp, ECONNREFUSED);
1908 if (thflags & TH_RST)
1910 if (!(thflags & TH_SYN))
1913 tp->irs = th->th_seq;
1915 if (thflags & TH_ACK) {
1916 TCPSTAT_INC(tcps_connects);
1919 mac_socketpeer_set_from_mbuf(m, so);
1921 /* Do window scaling on this connection? */
1922 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1923 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1924 tp->rcv_scale = tp->request_r_scale;
1926 tp->rcv_adv += imin(tp->rcv_wnd,
1927 TCP_MAXWIN << tp->rcv_scale);
1928 tp->snd_una++; /* SYN is acked */
1930 * If there's data, delay ACK; if there's also a FIN
1931 * ACKNOW will be turned on later.
1933 if (DELAY_ACK(tp, tlen) && tlen != 0)
1934 tcp_timer_activate(tp, TT_DELACK,
1937 tp->t_flags |= TF_ACKNOW;
1939 if ((thflags & TH_ECE) && V_tcp_do_ecn) {
1940 tp->t_flags |= TF_ECN_PERMIT;
1941 TCPSTAT_INC(tcps_ecn_shs);
1945 * Received <SYN,ACK> in SYN_SENT[*] state.
1947 * SYN_SENT --> ESTABLISHED
1948 * SYN_SENT* --> FIN_WAIT_1
1950 tp->t_starttime = ticks;
1951 if (tp->t_flags & TF_NEEDFIN) {
1952 tcp_state_change(tp, TCPS_FIN_WAIT_1);
1953 tp->t_flags &= ~TF_NEEDFIN;
1956 tcp_state_change(tp, TCPS_ESTABLISHED);
1957 TCP_PROBE5(connect__established, NULL, tp,
1958 mtod(m, const char *), tp, th);
1960 tcp_timer_activate(tp, TT_KEEP,
1965 * Received initial SYN in SYN-SENT[*] state =>
1966 * simultaneous open. If segment contains CC option
1967 * and there is a cached CC, apply TAO test.
1968 * If it succeeds, connection is * half-synchronized.
1969 * Otherwise, do 3-way handshake:
1970 * SYN-SENT -> SYN-RECEIVED
1971 * SYN-SENT* -> SYN-RECEIVED*
1972 * If there was no CC option, clear cached CC value.
1974 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
1975 tcp_timer_activate(tp, TT_REXMT, 0);
1976 tcp_state_change(tp, TCPS_SYN_RECEIVED);
1979 KASSERT(ti_locked == TI_WLOCKED, ("%s: trimthenstep6: "
1980 "ti_locked %d", __func__, ti_locked));
1981 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1982 INP_WLOCK_ASSERT(tp->t_inpcb);
1985 * Advance th->th_seq to correspond to first data byte.
1986 * If data, trim to stay within window,
1987 * dropping FIN if necessary.
1990 if (tlen > tp->rcv_wnd) {
1991 todrop = tlen - tp->rcv_wnd;
1995 TCPSTAT_INC(tcps_rcvpackafterwin);
1996 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
1998 tp->snd_wl1 = th->th_seq - 1;
1999 tp->rcv_up = th->th_seq;
2001 * Client side of transaction: already sent SYN and data.
2002 * If the remote host used T/TCP to validate the SYN,
2003 * our data will be ACK'd; if so, enter normal data segment
2004 * processing in the middle of step 5, ack processing.
2005 * Otherwise, goto step 6.
2007 if (thflags & TH_ACK)
2013 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
2014 * do normal processing.
2016 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later.
2020 break; /* continue normal processing */
2024 * States other than LISTEN or SYN_SENT.
2025 * First check the RST flag and sequence number since reset segments
2026 * are exempt from the timestamp and connection count tests. This
2027 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
2028 * below which allowed reset segments in half the sequence space
2029 * to fall though and be processed (which gives forged reset
2030 * segments with a random sequence number a 50 percent chance of
2031 * killing a connection).
2032 * Then check timestamp, if present.
2033 * Then check the connection count, if present.
2034 * Then check that at least some bytes of segment are within
2035 * receive window. If segment begins before rcv_nxt,
2036 * drop leading data (and SYN); if nothing left, just ack.
2039 * If the RST bit is set, check the sequence number to see
2040 * if this is a valid reset segment.
2042 * In all states except SYN-SENT, all reset (RST) segments
2043 * are validated by checking their SEQ-fields. A reset is
2044 * valid if its sequence number is in the window.
2045 * Note: this does not take into account delayed ACKs, so
2046 * we should test against last_ack_sent instead of rcv_nxt.
2047 * The sequence number in the reset segment is normally an
2048 * echo of our outgoing acknowlegement numbers, but some hosts
2049 * send a reset with the sequence number at the rightmost edge
2050 * of our receive window, and we have to handle this case.
2051 * Note 2: Paul Watson's paper "Slipping in the Window" has shown
2052 * that brute force RST attacks are possible. To combat this,
2053 * we use a much stricter check while in the ESTABLISHED state,
2054 * only accepting RSTs where the sequence number is equal to
2055 * last_ack_sent. In all other states (the states in which a
2056 * RST is more likely), the more permissive check is used.
2057 * If we have multiple segments in flight, the initial reset
2058 * segment sequence numbers will be to the left of last_ack_sent,
2059 * but they will eventually catch up.
2060 * In any case, it never made sense to trim reset segments to
2061 * fit the receive window since RFC 1122 says:
2062 * 4.2.2.12 RST Segment: RFC-793 Section 3.4
2064 * A TCP SHOULD allow a received RST segment to include data.
2067 * It has been suggested that a RST segment could contain
2068 * ASCII text that encoded and explained the cause of the
2069 * RST. No standard has yet been established for such
2072 * If the reset segment passes the sequence number test examine
2074 * SYN_RECEIVED STATE:
2075 * If passive open, return to LISTEN state.
2076 * If active open, inform user that connection was refused.
2077 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
2078 * Inform user that connection was reset, and close tcb.
2079 * CLOSING, LAST_ACK STATES:
2082 * Drop the segment - see Stevens, vol. 2, p. 964 and
2085 if (thflags & TH_RST) {
2086 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) &&
2087 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
2088 switch (tp->t_state) {
2090 case TCPS_SYN_RECEIVED:
2091 so->so_error = ECONNREFUSED;
2094 case TCPS_ESTABLISHED:
2095 if (V_tcp_insecure_rst == 0 &&
2096 !(SEQ_GEQ(th->th_seq, tp->rcv_nxt - 1) &&
2097 SEQ_LEQ(th->th_seq, tp->rcv_nxt + 1)) &&
2098 !(SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) &&
2099 SEQ_LEQ(th->th_seq, tp->last_ack_sent + 1))) {
2100 TCPSTAT_INC(tcps_badrst);
2104 case TCPS_FIN_WAIT_1:
2105 case TCPS_FIN_WAIT_2:
2106 case TCPS_CLOSE_WAIT:
2107 so->so_error = ECONNRESET;
2109 KASSERT(ti_locked == TI_WLOCKED,
2110 ("tcp_do_segment: TH_RST 1 ti_locked %d",
2112 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2114 tcp_state_change(tp, TCPS_CLOSED);
2115 TCPSTAT_INC(tcps_drops);
2121 KASSERT(ti_locked == TI_WLOCKED,
2122 ("tcp_do_segment: TH_RST 2 ti_locked %d",
2124 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2134 * RFC 1323 PAWS: If we have a timestamp reply on this segment
2135 * and it's less than ts_recent, drop it.
2137 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
2138 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
2140 /* Check to see if ts_recent is over 24 days old. */
2141 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) {
2143 * Invalidate ts_recent. If this segment updates
2144 * ts_recent, the age will be reset later and ts_recent
2145 * will get a valid value. If it does not, setting
2146 * ts_recent to zero will at least satisfy the
2147 * requirement that zero be placed in the timestamp
2148 * echo reply when ts_recent isn't valid. The
2149 * age isn't reset until we get a valid ts_recent
2150 * because we don't want out-of-order segments to be
2151 * dropped when ts_recent is old.
2155 TCPSTAT_INC(tcps_rcvduppack);
2156 TCPSTAT_ADD(tcps_rcvdupbyte, tlen);
2157 TCPSTAT_INC(tcps_pawsdrop);
2165 * In the SYN-RECEIVED state, validate that the packet belongs to
2166 * this connection before trimming the data to fit the receive
2167 * window. Check the sequence number versus IRS since we know
2168 * the sequence numbers haven't wrapped. This is a partial fix
2169 * for the "LAND" DoS attack.
2171 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
2172 rstreason = BANDLIM_RST_OPENPORT;
2176 todrop = tp->rcv_nxt - th->th_seq;
2178 if (thflags & TH_SYN) {
2188 * Following if statement from Stevens, vol. 2, p. 960.
2191 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
2193 * Any valid FIN must be to the left of the window.
2194 * At this point the FIN must be a duplicate or out
2195 * of sequence; drop it.
2200 * Send an ACK to resynchronize and drop any data.
2201 * But keep on processing for RST or ACK.
2203 tp->t_flags |= TF_ACKNOW;
2205 TCPSTAT_INC(tcps_rcvduppack);
2206 TCPSTAT_ADD(tcps_rcvdupbyte, todrop);
2208 TCPSTAT_INC(tcps_rcvpartduppack);
2209 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop);
2211 drop_hdrlen += todrop; /* drop from the top afterwards */
2212 th->th_seq += todrop;
2214 if (th->th_urp > todrop)
2215 th->th_urp -= todrop;
2223 * If new data are received on a connection after the
2224 * user processes are gone, then RST the other end.
2226 if ((so->so_state & SS_NOFDREF) &&
2227 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
2228 KASSERT(ti_locked == TI_WLOCKED, ("%s: SS_NOFDEREF && "
2229 "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked));
2230 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2232 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
2233 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data "
2234 "after socket was closed, "
2235 "sending RST and removing tcpcb\n",
2236 s, __func__, tcpstates[tp->t_state], tlen);
2240 TCPSTAT_INC(tcps_rcvafterclose);
2241 rstreason = BANDLIM_UNLIMITED;
2246 * If segment ends after window, drop trailing data
2247 * (and PUSH and FIN); if nothing left, just ACK.
2249 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
2251 TCPSTAT_INC(tcps_rcvpackafterwin);
2252 if (todrop >= tlen) {
2253 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen);
2255 * If window is closed can only take segments at
2256 * window edge, and have to drop data and PUSH from
2257 * incoming segments. Continue processing, but
2258 * remember to ack. Otherwise, drop segment
2261 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2262 tp->t_flags |= TF_ACKNOW;
2263 TCPSTAT_INC(tcps_rcvwinprobe);
2267 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2270 thflags &= ~(TH_PUSH|TH_FIN);
2274 * If last ACK falls within this segment's sequence numbers,
2275 * record its timestamp.
2277 * 1) That the test incorporates suggestions from the latest
2278 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
2279 * 2) That updating only on newer timestamps interferes with
2280 * our earlier PAWS tests, so this check should be solely
2281 * predicated on the sequence space of this segment.
2282 * 3) That we modify the segment boundary check to be
2283 * Last.ACK.Sent <= SEG.SEQ + SEG.Len
2284 * instead of RFC1323's
2285 * Last.ACK.Sent < SEG.SEQ + SEG.Len,
2286 * This modified check allows us to overcome RFC1323's
2287 * limitations as described in Stevens TCP/IP Illustrated
2288 * Vol. 2 p.869. In such cases, we can still calculate the
2289 * RTT correctly when RCV.NXT == Last.ACK.Sent.
2291 if ((to.to_flags & TOF_TS) != 0 &&
2292 SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2293 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
2294 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
2295 tp->ts_recent_age = tcp_ts_getticks();
2296 tp->ts_recent = to.to_tsval;
2300 * If a SYN is in the window, then this is an
2301 * error and we send an RST and drop the connection.
2303 if (thflags & TH_SYN) {
2304 KASSERT(ti_locked == TI_WLOCKED,
2305 ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked));
2306 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2308 tp = tcp_drop(tp, ECONNRESET);
2309 rstreason = BANDLIM_UNLIMITED;
2314 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
2315 * flag is on (half-synchronized state), then queue data for
2316 * later processing; else drop segment and return.
2318 if ((thflags & TH_ACK) == 0) {
2319 if (tp->t_state == TCPS_SYN_RECEIVED ||
2320 (tp->t_flags & TF_NEEDSYN))
2322 else if (tp->t_flags & TF_ACKNOW)
2331 switch (tp->t_state) {
2334 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
2335 * ESTABLISHED state and continue processing.
2336 * The ACK was checked above.
2338 case TCPS_SYN_RECEIVED:
2340 TCPSTAT_INC(tcps_connects);
2342 /* Do window scaling? */
2343 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2344 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2345 tp->rcv_scale = tp->request_r_scale;
2346 tp->snd_wnd = tiwin;
2350 * SYN-RECEIVED -> ESTABLISHED
2351 * SYN-RECEIVED* -> FIN-WAIT-1
2353 tp->t_starttime = ticks;
2354 if (tp->t_flags & TF_NEEDFIN) {
2355 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2356 tp->t_flags &= ~TF_NEEDFIN;
2358 tcp_state_change(tp, TCPS_ESTABLISHED);
2359 TCP_PROBE5(accept__established, NULL, tp,
2360 mtod(m, const char *), tp, th);
2362 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
2365 * If segment contains data or ACK, will call tcp_reass()
2366 * later; if not, do so now to pass queued data to user.
2368 if (tlen == 0 && (thflags & TH_FIN) == 0)
2369 (void) tcp_reass(tp, (struct tcphdr *)0, 0,
2371 tp->snd_wl1 = th->th_seq - 1;
2375 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2376 * ACKs. If the ack is in the range
2377 * tp->snd_una < th->th_ack <= tp->snd_max
2378 * then advance tp->snd_una to th->th_ack and drop
2379 * data from the retransmission queue. If this ACK reflects
2380 * more up to date window information we update our window information.
2382 case TCPS_ESTABLISHED:
2383 case TCPS_FIN_WAIT_1:
2384 case TCPS_FIN_WAIT_2:
2385 case TCPS_CLOSE_WAIT:
2388 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2389 TCPSTAT_INC(tcps_rcvacktoomuch);
2392 if ((tp->t_flags & TF_SACK_PERMIT) &&
2393 ((to.to_flags & TOF_SACK) ||
2394 !TAILQ_EMPTY(&tp->snd_holes)))
2395 tcp_sack_doack(tp, &to, th->th_ack);
2397 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
2398 hhook_run_tcp_est_in(tp, th, &to);
2400 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2401 if (tlen == 0 && tiwin == tp->snd_wnd) {
2402 TCPSTAT_INC(tcps_rcvdupack);
2404 * If we have outstanding data (other than
2405 * a window probe), this is a completely
2406 * duplicate ack (ie, window info didn't
2407 * change), the ack is the biggest we've
2408 * seen and we've seen exactly our rexmt
2409 * threshhold of them, assume a packet
2410 * has been dropped and retransmit it.
2411 * Kludge snd_nxt & the congestion
2412 * window so we send only this one
2415 * We know we're losing at the current
2416 * window size so do congestion avoidance
2417 * (set ssthresh to half the current window
2418 * and pull our congestion window back to
2419 * the new ssthresh).
2421 * Dup acks mean that packets have left the
2422 * network (they're now cached at the receiver)
2423 * so bump cwnd by the amount in the receiver
2424 * to keep a constant cwnd packets in the
2427 * When using TCP ECN, notify the peer that
2428 * we reduced the cwnd.
2430 if (!tcp_timer_active(tp, TT_REXMT) ||
2431 th->th_ack != tp->snd_una)
2433 else if (++tp->t_dupacks > tcprexmtthresh ||
2434 IN_FASTRECOVERY(tp->t_flags)) {
2435 cc_ack_received(tp, th, CC_DUPACK);
2436 if ((tp->t_flags & TF_SACK_PERMIT) &&
2437 IN_FASTRECOVERY(tp->t_flags)) {
2441 * Compute the amount of data in flight first.
2442 * We can inject new data into the pipe iff
2443 * we have less than 1/2 the original window's
2444 * worth of data in flight.
2446 awnd = (tp->snd_nxt - tp->snd_fack) +
2447 tp->sackhint.sack_bytes_rexmit;
2448 if (awnd < tp->snd_ssthresh) {
2449 tp->snd_cwnd += tp->t_maxseg;
2450 if (tp->snd_cwnd > tp->snd_ssthresh)
2451 tp->snd_cwnd = tp->snd_ssthresh;
2454 tp->snd_cwnd += tp->t_maxseg;
2455 if ((thflags & TH_FIN) &&
2456 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2458 * If its a fin we need to process
2459 * it to avoid a race where both
2460 * sides enter FIN-WAIT and send FIN|ACK
2465 (void) tcp_output(tp);
2467 } else if (tp->t_dupacks == tcprexmtthresh) {
2468 tcp_seq onxt = tp->snd_nxt;
2471 * If we're doing sack, check to
2472 * see if we're already in sack
2473 * recovery. If we're not doing sack,
2474 * check to see if we're in newreno
2477 if (tp->t_flags & TF_SACK_PERMIT) {
2478 if (IN_FASTRECOVERY(tp->t_flags)) {
2483 if (SEQ_LEQ(th->th_ack,
2489 /* Congestion signal before ack. */
2490 cc_cong_signal(tp, th, CC_NDUPACK);
2491 cc_ack_received(tp, th, CC_DUPACK);
2492 tcp_timer_activate(tp, TT_REXMT, 0);
2494 if (tp->t_flags & TF_SACK_PERMIT) {
2496 tcps_sack_recovery_episode);
2497 tp->sack_newdata = tp->snd_nxt;
2498 tp->snd_cwnd = tp->t_maxseg;
2499 (void) tcp_output(tp);
2502 tp->snd_nxt = th->th_ack;
2503 tp->snd_cwnd = tp->t_maxseg;
2504 if ((thflags & TH_FIN) &&
2505 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2507 * If its a fin we need to process
2508 * it to avoid a race where both
2509 * sides enter FIN-WAIT and send FIN|ACK
2514 (void) tcp_output(tp);
2515 KASSERT(tp->snd_limited <= 2,
2516 ("%s: tp->snd_limited too big",
2518 tp->snd_cwnd = tp->snd_ssthresh +
2520 (tp->t_dupacks - tp->snd_limited);
2521 if (SEQ_GT(onxt, tp->snd_nxt))
2524 } else if (V_tcp_do_rfc3042) {
2525 cc_ack_received(tp, th, CC_DUPACK);
2526 u_long oldcwnd = tp->snd_cwnd;
2527 tcp_seq oldsndmax = tp->snd_max;
2531 KASSERT(tp->t_dupacks == 1 ||
2533 ("%s: dupacks not 1 or 2",
2535 if (tp->t_dupacks == 1)
2536 tp->snd_limited = 0;
2538 (tp->snd_nxt - tp->snd_una) +
2539 (tp->t_dupacks - tp->snd_limited) *
2541 if ((thflags & TH_FIN) &&
2542 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2544 * If its a fin we need to process
2545 * it to avoid a race where both
2546 * sides enter FIN-WAIT and send FIN|ACK
2552 * Only call tcp_output when there
2553 * is new data available to be sent.
2554 * Otherwise we would send pure ACKs.
2556 SOCKBUF_LOCK(&so->so_snd);
2557 avail = so->so_snd.sb_cc -
2558 (tp->snd_nxt - tp->snd_una);
2559 SOCKBUF_UNLOCK(&so->so_snd);
2561 (void) tcp_output(tp);
2562 sent = tp->snd_max - oldsndmax;
2563 if (sent > tp->t_maxseg) {
2564 KASSERT((tp->t_dupacks == 2 &&
2565 tp->snd_limited == 0) ||
2566 (sent == tp->t_maxseg + 1 &&
2567 tp->t_flags & TF_SENTFIN),
2568 ("%s: sent too much",
2570 tp->snd_limited = 2;
2571 } else if (sent > 0)
2573 tp->snd_cwnd = oldcwnd;
2581 KASSERT(SEQ_GT(th->th_ack, tp->snd_una),
2582 ("%s: th_ack <= snd_una", __func__));
2585 * If the congestion window was inflated to account
2586 * for the other side's cached packets, retract it.
2588 if (IN_FASTRECOVERY(tp->t_flags)) {
2589 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2590 if (tp->t_flags & TF_SACK_PERMIT)
2591 tcp_sack_partialack(tp, th);
2593 tcp_newreno_partial_ack(tp, th);
2595 cc_post_recovery(tp, th);
2599 * If we reach this point, ACK is not a duplicate,
2600 * i.e., it ACKs something we sent.
2602 if (tp->t_flags & TF_NEEDSYN) {
2604 * T/TCP: Connection was half-synchronized, and our
2605 * SYN has been ACK'd (so connection is now fully
2606 * synchronized). Go to non-starred state,
2607 * increment snd_una for ACK of SYN, and check if
2608 * we can do window scaling.
2610 tp->t_flags &= ~TF_NEEDSYN;
2612 /* Do window scaling? */
2613 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2614 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2615 tp->rcv_scale = tp->request_r_scale;
2616 /* Send window already scaled. */
2621 INP_WLOCK_ASSERT(tp->t_inpcb);
2623 acked = BYTES_THIS_ACK(tp, th);
2624 TCPSTAT_INC(tcps_rcvackpack);
2625 TCPSTAT_ADD(tcps_rcvackbyte, acked);
2628 * If we just performed our first retransmit, and the ACK
2629 * arrives within our recovery window, then it was a mistake
2630 * to do the retransmit in the first place. Recover our
2631 * original cwnd and ssthresh, and proceed to transmit where
2634 if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID &&
2635 (int)(ticks - tp->t_badrxtwin) < 0)
2636 cc_cong_signal(tp, th, CC_RTO_ERR);
2639 * If we have a timestamp reply, update smoothed
2640 * round trip time. If no timestamp is present but
2641 * transmit timer is running and timed sequence
2642 * number was acked, update smoothed round trip time.
2643 * Since we now have an rtt measurement, cancel the
2644 * timer backoff (cf., Phil Karn's retransmit alg.).
2645 * Recompute the initial retransmit timer.
2647 * Some boxes send broken timestamp replies
2648 * during the SYN+ACK phase, ignore
2649 * timestamps of 0 or we could calculate a
2650 * huge RTT and blow up the retransmit timer.
2652 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) {
2655 t = tcp_ts_getticks() - to.to_tsecr;
2656 if (!tp->t_rttlow || tp->t_rttlow > t)
2658 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1);
2659 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2660 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2661 tp->t_rttlow = ticks - tp->t_rtttime;
2662 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2666 * If all outstanding data is acked, stop retransmit
2667 * timer and remember to restart (more output or persist).
2668 * If there is more data to be acked, restart retransmit
2669 * timer, using current (possibly backed-off) value.
2671 if (th->th_ack == tp->snd_max) {
2672 tcp_timer_activate(tp, TT_REXMT, 0);
2674 } else if (!tcp_timer_active(tp, TT_PERSIST))
2675 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
2678 * If no data (only SYN) was ACK'd,
2679 * skip rest of ACK processing.
2685 * Let the congestion control algorithm update congestion
2686 * control related information. This typically means increasing
2687 * the congestion window.
2689 cc_ack_received(tp, th, CC_ACK);
2691 SOCKBUF_LOCK(&so->so_snd);
2692 if (acked > so->so_snd.sb_cc) {
2693 tp->snd_wnd -= so->so_snd.sb_cc;
2694 mfree = sbcut_locked(&so->so_snd,
2695 (int)so->so_snd.sb_cc);
2698 mfree = sbcut_locked(&so->so_snd, acked);
2699 tp->snd_wnd -= acked;
2702 /* NB: sowwakeup_locked() does an implicit unlock. */
2703 sowwakeup_locked(so);
2705 /* Detect una wraparound. */
2706 if (!IN_RECOVERY(tp->t_flags) &&
2707 SEQ_GT(tp->snd_una, tp->snd_recover) &&
2708 SEQ_LEQ(th->th_ack, tp->snd_recover))
2709 tp->snd_recover = th->th_ack - 1;
2710 /* XXXLAS: Can this be moved up into cc_post_recovery? */
2711 if (IN_RECOVERY(tp->t_flags) &&
2712 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2713 EXIT_RECOVERY(tp->t_flags);
2715 tp->snd_una = th->th_ack;
2716 if (tp->t_flags & TF_SACK_PERMIT) {
2717 if (SEQ_GT(tp->snd_una, tp->snd_recover))
2718 tp->snd_recover = tp->snd_una;
2720 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2721 tp->snd_nxt = tp->snd_una;
2723 switch (tp->t_state) {
2726 * In FIN_WAIT_1 STATE in addition to the processing
2727 * for the ESTABLISHED state if our FIN is now acknowledged
2728 * then enter FIN_WAIT_2.
2730 case TCPS_FIN_WAIT_1:
2731 if (ourfinisacked) {
2733 * If we can't receive any more
2734 * data, then closing user can proceed.
2735 * Starting the timer is contrary to the
2736 * specification, but if we don't get a FIN
2737 * we'll hang forever.
2740 * we should release the tp also, and use a
2743 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2744 soisdisconnected(so);
2745 tcp_timer_activate(tp, TT_2MSL,
2746 (tcp_fast_finwait2_recycle ?
2747 tcp_finwait2_timeout :
2750 tcp_state_change(tp, TCPS_FIN_WAIT_2);
2755 * In CLOSING STATE in addition to the processing for
2756 * the ESTABLISHED state if the ACK acknowledges our FIN
2757 * then enter the TIME-WAIT state, otherwise ignore
2761 if (ourfinisacked) {
2762 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2764 INP_INFO_WUNLOCK(&V_tcbinfo);
2771 * In LAST_ACK, we may still be waiting for data to drain
2772 * and/or to be acked, as well as for the ack of our FIN.
2773 * If our FIN is now acknowledged, delete the TCB,
2774 * enter the closed state and return.
2777 if (ourfinisacked) {
2778 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2787 INP_WLOCK_ASSERT(tp->t_inpcb);
2790 * Update window information.
2791 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2793 if ((thflags & TH_ACK) &&
2794 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2795 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2796 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2797 /* keep track of pure window updates */
2799 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2800 TCPSTAT_INC(tcps_rcvwinupd);
2801 tp->snd_wnd = tiwin;
2802 tp->snd_wl1 = th->th_seq;
2803 tp->snd_wl2 = th->th_ack;
2804 if (tp->snd_wnd > tp->max_sndwnd)
2805 tp->max_sndwnd = tp->snd_wnd;
2810 * Process segments with URG.
2812 if ((thflags & TH_URG) && th->th_urp &&
2813 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2815 * This is a kludge, but if we receive and accept
2816 * random urgent pointers, we'll crash in
2817 * soreceive. It's hard to imagine someone
2818 * actually wanting to send this much urgent data.
2820 SOCKBUF_LOCK(&so->so_rcv);
2821 if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2822 th->th_urp = 0; /* XXX */
2823 thflags &= ~TH_URG; /* XXX */
2824 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */
2825 goto dodata; /* XXX */
2828 * If this segment advances the known urgent pointer,
2829 * then mark the data stream. This should not happen
2830 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2831 * a FIN has been received from the remote side.
2832 * In these states we ignore the URG.
2834 * According to RFC961 (Assigned Protocols),
2835 * the urgent pointer points to the last octet
2836 * of urgent data. We continue, however,
2837 * to consider it to indicate the first octet
2838 * of data past the urgent section as the original
2839 * spec states (in one of two places).
2841 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2842 tp->rcv_up = th->th_seq + th->th_urp;
2843 so->so_oobmark = so->so_rcv.sb_cc +
2844 (tp->rcv_up - tp->rcv_nxt) - 1;
2845 if (so->so_oobmark == 0)
2846 so->so_rcv.sb_state |= SBS_RCVATMARK;
2848 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2850 SOCKBUF_UNLOCK(&so->so_rcv);
2852 * Remove out of band data so doesn't get presented to user.
2853 * This can happen independent of advancing the URG pointer,
2854 * but if two URG's are pending at once, some out-of-band
2855 * data may creep in... ick.
2857 if (th->th_urp <= (u_long)tlen &&
2858 !(so->so_options & SO_OOBINLINE)) {
2859 /* hdr drop is delayed */
2860 tcp_pulloutofband(so, th, m, drop_hdrlen);
2864 * If no out of band data is expected,
2865 * pull receive urgent pointer along
2866 * with the receive window.
2868 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2869 tp->rcv_up = tp->rcv_nxt;
2872 INP_WLOCK_ASSERT(tp->t_inpcb);
2875 * Process the segment text, merging it into the TCP sequencing queue,
2876 * and arranging for acknowledgment of receipt if necessary.
2877 * This process logically involves adjusting tp->rcv_wnd as data
2878 * is presented to the user (this happens in tcp_usrreq.c,
2879 * case PRU_RCVD). If a FIN has already been received on this
2880 * connection then we just ignore the text.
2882 if ((tlen || (thflags & TH_FIN)) &&
2883 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2884 tcp_seq save_start = th->th_seq;
2885 m_adj(m, drop_hdrlen); /* delayed header drop */
2887 * Insert segment which includes th into TCP reassembly queue
2888 * with control block tp. Set thflags to whether reassembly now
2889 * includes a segment with FIN. This handles the common case
2890 * inline (segment is the next to be received on an established
2891 * connection, and the queue is empty), avoiding linkage into
2892 * and removal from the queue and repetition of various
2894 * Set DELACK for segments received in order, but ack
2895 * immediately when segments are out of order (so
2896 * fast retransmit can work).
2898 if (th->th_seq == tp->rcv_nxt &&
2899 LIST_EMPTY(&tp->t_segq) &&
2900 TCPS_HAVEESTABLISHED(tp->t_state)) {
2901 if (DELAY_ACK(tp, tlen))
2902 tp->t_flags |= TF_DELACK;
2904 tp->t_flags |= TF_ACKNOW;
2905 tp->rcv_nxt += tlen;
2906 thflags = th->th_flags & TH_FIN;
2907 TCPSTAT_INC(tcps_rcvpack);
2908 TCPSTAT_ADD(tcps_rcvbyte, tlen);
2910 SOCKBUF_LOCK(&so->so_rcv);
2911 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
2914 sbappendstream_locked(&so->so_rcv, m);
2915 /* NB: sorwakeup_locked() does an implicit unlock. */
2916 sorwakeup_locked(so);
2919 * XXX: Due to the header drop above "th" is
2920 * theoretically invalid by now. Fortunately
2921 * m_adj() doesn't actually frees any mbufs
2922 * when trimming from the head.
2924 thflags = tcp_reass(tp, th, &tlen, m);
2925 tp->t_flags |= TF_ACKNOW;
2927 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT))
2928 tcp_update_sack_list(tp, save_start, save_start + tlen);
2931 * Note the amount of data that peer has sent into
2932 * our window, in order to estimate the sender's
2936 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
2937 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2939 len = so->so_rcv.sb_hiwat;
2947 * If FIN is received ACK the FIN and let the user know
2948 * that the connection is closing.
2950 if (thflags & TH_FIN) {
2951 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2954 * If connection is half-synchronized
2955 * (ie NEEDSYN flag on) then delay ACK,
2956 * so it may be piggybacked when SYN is sent.
2957 * Otherwise, since we received a FIN then no
2958 * more input can be expected, send ACK now.
2960 if (tp->t_flags & TF_NEEDSYN)
2961 tp->t_flags |= TF_DELACK;
2963 tp->t_flags |= TF_ACKNOW;
2966 switch (tp->t_state) {
2969 * In SYN_RECEIVED and ESTABLISHED STATES
2970 * enter the CLOSE_WAIT state.
2972 case TCPS_SYN_RECEIVED:
2973 tp->t_starttime = ticks;
2975 case TCPS_ESTABLISHED:
2976 tcp_state_change(tp, TCPS_CLOSE_WAIT);
2980 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2981 * enter the CLOSING state.
2983 case TCPS_FIN_WAIT_1:
2984 tcp_state_change(tp, TCPS_CLOSING);
2988 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2989 * starting the time-wait timer, turning off the other
2992 case TCPS_FIN_WAIT_2:
2993 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2994 KASSERT(ti_locked == TI_WLOCKED, ("%s: dodata "
2995 "TCP_FIN_WAIT_2 ti_locked: %d", __func__,
2999 INP_INFO_WUNLOCK(&V_tcbinfo);
3003 if (ti_locked == TI_WLOCKED)
3004 INP_INFO_WUNLOCK(&V_tcbinfo);
3005 ti_locked = TI_UNLOCKED;
3008 if (so->so_options & SO_DEBUG)
3009 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
3014 * Return any desired output.
3016 if (needoutput || (tp->t_flags & TF_ACKNOW))
3017 (void) tcp_output(tp);
3020 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d",
3021 __func__, ti_locked));
3022 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
3023 INP_WLOCK_ASSERT(tp->t_inpcb);
3025 if (tp->t_flags & TF_DELACK) {
3026 tp->t_flags &= ~TF_DELACK;
3027 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
3029 INP_WUNLOCK(tp->t_inpcb);
3034 * Generate an ACK dropping incoming segment if it occupies
3035 * sequence space, where the ACK reflects our state.
3037 * We can now skip the test for the RST flag since all
3038 * paths to this code happen after packets containing
3039 * RST have been dropped.
3041 * In the SYN-RECEIVED state, don't send an ACK unless the
3042 * segment we received passes the SYN-RECEIVED ACK test.
3043 * If it fails send a RST. This breaks the loop in the
3044 * "LAND" DoS attack, and also prevents an ACK storm
3045 * between two listening ports that have been sent forged
3046 * SYN segments, each with the source address of the other.
3048 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
3049 (SEQ_GT(tp->snd_una, th->th_ack) ||
3050 SEQ_GT(th->th_ack, tp->snd_max)) ) {
3051 rstreason = BANDLIM_RST_OPENPORT;
3055 if (so->so_options & SO_DEBUG)
3056 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3059 if (ti_locked == TI_WLOCKED)
3060 INP_INFO_WUNLOCK(&V_tcbinfo);
3061 ti_locked = TI_UNLOCKED;
3063 tp->t_flags |= TF_ACKNOW;
3064 (void) tcp_output(tp);
3065 INP_WUNLOCK(tp->t_inpcb);
3070 if (ti_locked == TI_WLOCKED)
3071 INP_INFO_WUNLOCK(&V_tcbinfo);
3072 ti_locked = TI_UNLOCKED;
3075 tcp_dropwithreset(m, th, tp, tlen, rstreason);
3076 INP_WUNLOCK(tp->t_inpcb);
3078 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
3082 if (ti_locked == TI_WLOCKED) {
3083 INP_INFO_WUNLOCK(&V_tcbinfo);
3084 ti_locked = TI_UNLOCKED;
3088 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
3092 * Drop space held by incoming segment and return.
3095 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
3096 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3100 INP_WUNLOCK(tp->t_inpcb);
3105 * Issue RST and make ACK acceptable to originator of segment.
3106 * The mbuf must still include the original packet header.
3110 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
3111 int tlen, int rstreason)
3117 struct ip6_hdr *ip6;
3121 INP_WLOCK_ASSERT(tp->t_inpcb);
3124 /* Don't bother if destination was broadcast/multicast. */
3125 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
3128 if (mtod(m, struct ip *)->ip_v == 6) {
3129 ip6 = mtod(m, struct ip6_hdr *);
3130 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3131 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3133 /* IPv6 anycast check is done at tcp6_input() */
3136 #if defined(INET) && defined(INET6)
3141 ip = mtod(m, struct ip *);
3142 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3143 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3144 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3145 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3150 /* Perform bandwidth limiting. */
3151 if (badport_bandlim(rstreason) < 0)
3154 /* tcp_respond consumes the mbuf chain. */
3155 if (th->th_flags & TH_ACK) {
3156 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0,
3157 th->th_ack, TH_RST);
3159 if (th->th_flags & TH_SYN)
3161 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
3162 (tcp_seq)0, TH_RST|TH_ACK);
3170 * Parse TCP options and place in tcpopt.
3173 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags)
3178 for (; cnt > 0; cnt -= optlen, cp += optlen) {
3180 if (opt == TCPOPT_EOL)
3182 if (opt == TCPOPT_NOP)
3188 if (optlen < 2 || optlen > cnt)
3193 if (optlen != TCPOLEN_MAXSEG)
3195 if (!(flags & TO_SYN))
3197 to->to_flags |= TOF_MSS;
3198 bcopy((char *)cp + 2,
3199 (char *)&to->to_mss, sizeof(to->to_mss));
3200 to->to_mss = ntohs(to->to_mss);
3203 if (optlen != TCPOLEN_WINDOW)
3205 if (!(flags & TO_SYN))
3207 to->to_flags |= TOF_SCALE;
3208 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT);
3210 case TCPOPT_TIMESTAMP:
3211 if (optlen != TCPOLEN_TIMESTAMP)
3213 to->to_flags |= TOF_TS;
3214 bcopy((char *)cp + 2,
3215 (char *)&to->to_tsval, sizeof(to->to_tsval));
3216 to->to_tsval = ntohl(to->to_tsval);
3217 bcopy((char *)cp + 6,
3218 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
3219 to->to_tsecr = ntohl(to->to_tsecr);
3221 #ifdef TCP_SIGNATURE
3223 * XXX In order to reply to a host which has set the
3224 * TCP_SIGNATURE option in its initial SYN, we have to
3225 * record the fact that the option was observed here
3226 * for the syncache code to perform the correct response.
3228 case TCPOPT_SIGNATURE:
3229 if (optlen != TCPOLEN_SIGNATURE)
3231 to->to_flags |= TOF_SIGNATURE;
3232 to->to_signature = cp + 2;
3235 case TCPOPT_SACK_PERMITTED:
3236 if (optlen != TCPOLEN_SACK_PERMITTED)
3238 if (!(flags & TO_SYN))
3242 to->to_flags |= TOF_SACKPERM;
3245 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
3249 to->to_flags |= TOF_SACK;
3250 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
3251 to->to_sacks = cp + 2;
3252 TCPSTAT_INC(tcps_sack_rcv_blocks);
3261 * Pull out of band byte out of a segment so
3262 * it doesn't appear in the user's data queue.
3263 * It is still reflected in the segment length for
3264 * sequencing purposes.
3267 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m,
3270 int cnt = off + th->th_urp - 1;
3273 if (m->m_len > cnt) {
3274 char *cp = mtod(m, caddr_t) + cnt;
3275 struct tcpcb *tp = sototcpcb(so);
3277 INP_WLOCK_ASSERT(tp->t_inpcb);
3280 tp->t_oobflags |= TCPOOB_HAVEDATA;
3281 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3283 if (m->m_flags & M_PKTHDR)
3292 panic("tcp_pulloutofband");
3296 * Collect new round-trip time estimate
3297 * and update averages and current timeout.
3300 tcp_xmit_timer(struct tcpcb *tp, int rtt)
3304 INP_WLOCK_ASSERT(tp->t_inpcb);
3306 TCPSTAT_INC(tcps_rttupdated);
3308 if (tp->t_srtt != 0) {
3310 * srtt is stored as fixed point with 5 bits after the
3311 * binary point (i.e., scaled by 8). The following magic
3312 * is equivalent to the smoothing algorithm in rfc793 with
3313 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3314 * point). Adjust rtt to origin 0.
3316 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
3317 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
3319 if ((tp->t_srtt += delta) <= 0)
3323 * We accumulate a smoothed rtt variance (actually, a
3324 * smoothed mean difference), then set the retransmit
3325 * timer to smoothed rtt + 4 times the smoothed variance.
3326 * rttvar is stored as fixed point with 4 bits after the
3327 * binary point (scaled by 16). The following is
3328 * equivalent to rfc793 smoothing with an alpha of .75
3329 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
3330 * rfc793's wired-in beta.
3334 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
3335 if ((tp->t_rttvar += delta) <= 0)
3337 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
3338 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3341 * No rtt measurement yet - use the unsmoothed rtt.
3342 * Set the variance to half the rtt (so our first
3343 * retransmit happens at 3*rtt).
3345 tp->t_srtt = rtt << TCP_RTT_SHIFT;
3346 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
3347 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3353 * the retransmit should happen at rtt + 4 * rttvar.
3354 * Because of the way we do the smoothing, srtt and rttvar
3355 * will each average +1/2 tick of bias. When we compute
3356 * the retransmit timer, we want 1/2 tick of rounding and
3357 * 1 extra tick because of +-1/2 tick uncertainty in the
3358 * firing of the timer. The bias will give us exactly the
3359 * 1.5 tick we need. But, because the bias is
3360 * statistical, we have to test that we don't drop below
3361 * the minimum feasible timer (which is 2 ticks).
3363 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3364 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3367 * We received an ack for a packet that wasn't retransmitted;
3368 * it is probably safe to discard any error indications we've
3369 * received recently. This isn't quite right, but close enough
3370 * for now (a route might have failed after we sent a segment,
3371 * and the return path might not be symmetrical).
3373 tp->t_softerror = 0;
3377 * Determine a reasonable value for maxseg size.
3378 * If the route is known, check route for mtu.
3379 * If none, use an mss that can be handled on the outgoing interface
3380 * without forcing IP to fragment. If no route is found, route has no mtu,
3381 * or the destination isn't local, use a default, hopefully conservative
3382 * size (usually 512 or the default IP max size, but no more than the mtu
3383 * of the interface), as we can't discover anything about intervening
3384 * gateways or networks. We also initialize the congestion/slow start
3385 * window to be a single segment if the destination isn't local.
3386 * While looking at the routing entry, we also initialize other path-dependent
3387 * parameters from pre-set or cached values in the routing entry.
3389 * Also take into account the space needed for options that we
3390 * send regularly. Make maxseg shorter by that amount to assure
3391 * that we can send maxseg amount of data even when the options
3392 * are present. Store the upper limit of the length of options plus
3395 * NOTE that this routine is only called when we process an incoming
3396 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS
3397 * settings are handled in tcp_mssopt().
3400 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer,
3401 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap)
3405 struct inpcb *inp = tp->t_inpcb;
3406 struct hc_metrics_lite metrics;
3409 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
3410 size_t min_protoh = isipv6 ?
3411 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
3412 sizeof (struct tcpiphdr);
3414 const size_t min_protoh = sizeof(struct tcpiphdr);
3417 INP_WLOCK_ASSERT(tp->t_inpcb);
3419 if (mtuoffer != -1) {
3420 KASSERT(offer == -1, ("%s: conflict", __func__));
3421 offer = mtuoffer - min_protoh;
3428 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap);
3429 tp->t_maxopd = tp->t_maxseg = V_tcp_v6mssdflt;
3432 #if defined(INET) && defined(INET6)
3437 maxmtu = tcp_maxmtu(&inp->inp_inc, cap);
3438 tp->t_maxopd = tp->t_maxseg = V_tcp_mssdflt;
3443 * No route to sender, stay with default mss and return.
3447 * In case we return early we need to initialize metrics
3448 * to a defined state as tcp_hc_get() would do for us
3449 * if there was no cache hit.
3451 if (metricptr != NULL)
3452 bzero(metricptr, sizeof(struct hc_metrics_lite));
3456 /* What have we got? */
3460 * Offer == 0 means that there was no MSS on the SYN
3461 * segment, in this case we use tcp_mssdflt as
3462 * already assigned to t_maxopd above.
3464 offer = tp->t_maxopd;
3469 * Offer == -1 means that we didn't receive SYN yet.
3475 * Prevent DoS attack with too small MSS. Round up
3476 * to at least minmss.
3478 offer = max(offer, V_tcp_minmss);
3482 * rmx information is now retrieved from tcp_hostcache.
3484 tcp_hc_get(&inp->inp_inc, &metrics);
3485 if (metricptr != NULL)
3486 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite));
3489 * If there's a discovered mtu int tcp hostcache, use it
3490 * else, use the link mtu.
3492 if (metrics.rmx_mtu)
3493 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
3497 mss = maxmtu - min_protoh;
3498 if (!V_path_mtu_discovery &&
3499 !in6_localaddr(&inp->in6p_faddr))
3500 mss = min(mss, V_tcp_v6mssdflt);
3503 #if defined(INET) && defined(INET6)
3508 mss = maxmtu - min_protoh;
3509 if (!V_path_mtu_discovery &&
3510 !in_localaddr(inp->inp_faddr))
3511 mss = min(mss, V_tcp_mssdflt);
3515 * XXX - The above conditional (mss = maxmtu - min_protoh)
3516 * probably violates the TCP spec.
3517 * The problem is that, since we don't know the
3518 * other end's MSS, we are supposed to use a conservative
3519 * default. But, if we do that, then MTU discovery will
3520 * never actually take place, because the conservative
3521 * default is much less than the MTUs typically seen
3522 * on the Internet today. For the moment, we'll sweep
3523 * this under the carpet.
3525 * The conservative default might not actually be a problem
3526 * if the only case this occurs is when sending an initial
3527 * SYN with options and data to a host we've never talked
3528 * to before. Then, they will reply with an MSS value which
3529 * will get recorded and the new parameters should get
3530 * recomputed. For Further Study.
3533 mss = min(mss, offer);
3536 * Sanity check: make sure that maxopd will be large
3537 * enough to allow some data on segments even if the
3538 * all the option space is used (40bytes). Otherwise
3539 * funny things may happen in tcp_output.
3544 * maxopd stores the maximum length of data AND options
3545 * in a segment; maxseg is the amount of data in a normal
3546 * segment. We need to store this value (maxopd) apart
3547 * from maxseg, because now every segment carries options
3548 * and thus we normally have somewhat less data in segments.
3553 * origoffer==-1 indicates that no segments were received yet.
3554 * In this case we just guess.
3556 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
3558 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
3559 mss -= TCPOLEN_TSTAMP_APPA;
3565 tcp_mss(struct tcpcb *tp, int offer)
3571 struct hc_metrics_lite metrics;
3572 struct tcp_ifcap cap;
3574 KASSERT(tp != NULL, ("%s: tp == NULL", __func__));
3576 bzero(&cap, sizeof(cap));
3577 tcp_mss_update(tp, offer, -1, &metrics, &cap);
3583 * If there's a pipesize, change the socket buffer to that size,
3584 * don't change if sb_hiwat is different than default (then it
3585 * has been changed on purpose with setsockopt).
3586 * Make the socket buffers an integral number of mss units;
3587 * if the mss is larger than the socket buffer, decrease the mss.
3589 so = inp->inp_socket;
3590 SOCKBUF_LOCK(&so->so_snd);
3591 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe)
3592 bufsize = metrics.rmx_sendpipe;
3594 bufsize = so->so_snd.sb_hiwat;
3598 bufsize = roundup(bufsize, mss);
3599 if (bufsize > sb_max)
3601 if (bufsize > so->so_snd.sb_hiwat)
3602 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
3604 SOCKBUF_UNLOCK(&so->so_snd);
3607 SOCKBUF_LOCK(&so->so_rcv);
3608 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe)
3609 bufsize = metrics.rmx_recvpipe;
3611 bufsize = so->so_rcv.sb_hiwat;
3612 if (bufsize > mss) {
3613 bufsize = roundup(bufsize, mss);
3614 if (bufsize > sb_max)
3616 if (bufsize > so->so_rcv.sb_hiwat)
3617 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
3619 SOCKBUF_UNLOCK(&so->so_rcv);
3621 /* Check the interface for TSO capabilities. */
3622 if (cap.ifcap & CSUM_TSO) {
3623 tp->t_flags |= TF_TSO;
3624 tp->t_tsomax = cap.tsomax;
3625 tp->t_tsomaxsegcount = cap.tsomaxsegcount;
3626 tp->t_tsomaxsegsize = cap.tsomaxsegsize;
3631 * Determine the MSS option to send on an outgoing SYN.
3634 tcp_mssopt(struct in_conninfo *inc)
3641 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3644 if (inc->inc_flags & INC_ISIPV6) {
3645 mss = V_tcp_v6mssdflt;
3646 maxmtu = tcp_maxmtu6(inc, NULL);
3647 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3650 #if defined(INET) && defined(INET6)
3655 mss = V_tcp_mssdflt;
3656 maxmtu = tcp_maxmtu(inc, NULL);
3657 min_protoh = sizeof(struct tcpiphdr);
3660 #if defined(INET6) || defined(INET)
3661 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3664 if (maxmtu && thcmtu)
3665 mss = min(maxmtu, thcmtu) - min_protoh;
3666 else if (maxmtu || thcmtu)
3667 mss = max(maxmtu, thcmtu) - min_protoh;
3674 * On a partial ack arrives, force the retransmission of the
3675 * next unacknowledged segment. Do not clear tp->t_dupacks.
3676 * By setting snd_nxt to ti_ack, this forces retransmission timer to
3680 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th)
3682 tcp_seq onxt = tp->snd_nxt;
3683 u_long ocwnd = tp->snd_cwnd;
3685 INP_WLOCK_ASSERT(tp->t_inpcb);
3687 tcp_timer_activate(tp, TT_REXMT, 0);
3689 tp->snd_nxt = th->th_ack;
3691 * Set snd_cwnd to one segment beyond acknowledged offset.
3692 * (tp->snd_una has not yet been updated when this function is called.)
3694 tp->snd_cwnd = tp->t_maxseg + BYTES_THIS_ACK(tp, th);
3695 tp->t_flags |= TF_ACKNOW;
3696 (void) tcp_output(tp);
3697 tp->snd_cwnd = ocwnd;
3698 if (SEQ_GT(onxt, tp->snd_nxt))
3701 * Partial window deflation. Relies on fact that tp->snd_una
3704 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th))
3705 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th);
3708 tp->snd_cwnd += tp->t_maxseg;
projects / FreeBSD / stable / 10.git / blob