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
10 * Portions of this software were developed at the Centre for Advanced Internet
11 * Architectures, Swinburne University of Technology, by Lawrence Stewart,
12 * James Healy and David Hayes, made possible in part by a grant from the Cisco
13 * University Research Program Fund at Community Foundation Silicon Valley.
15 * Portions of this software were developed at the Centre for Advanced
16 * Internet Architectures, Swinburne University of Technology, Melbourne,
17 * Australia by David Hayes under sponsorship from the FreeBSD Foundation.
19 * Redistribution and use in source and binary forms, with or without
20 * modification, are permitted provided that the following conditions
22 * 1. Redistributions of source code must retain the above copyright
23 * notice, this list of conditions and the following disclaimer.
24 * 2. Redistributions in binary form must reproduce the above copyright
25 * notice, this list of conditions and the following disclaimer in the
26 * documentation and/or other materials provided with the distribution.
27 * 4. Neither the name of the University nor the names of its contributors
28 * may be used to endorse or promote products derived from this software
29 * without specific prior written permission.
31 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
32 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
33 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
34 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
35 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
36 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
37 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
38 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
39 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
40 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
43 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95
46 #include <sys/cdefs.h>
47 __FBSDID("$FreeBSD$");
49 #include "opt_ipfw.h" /* for ipfw_fwd */
51 #include "opt_inet6.h"
52 #include "opt_ipsec.h"
53 #include "opt_tcpdebug.h"
55 #include <sys/param.h>
56 #include <sys/kernel.h>
57 #include <sys/hhook.h>
58 #include <sys/malloc.h>
60 #include <sys/proc.h> /* for proc0 declaration */
61 #include <sys/protosw.h>
62 #include <sys/signalvar.h>
63 #include <sys/socket.h>
64 #include <sys/socketvar.h>
65 #include <sys/sysctl.h>
66 #include <sys/syslog.h>
67 #include <sys/systm.h>
69 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
74 #include <net/route.h>
77 #define TCPSTATES /* for logging */
79 #include <netinet/cc.h>
80 #include <netinet/in.h>
81 #include <netinet/in_pcb.h>
82 #include <netinet/in_systm.h>
83 #include <netinet/in_var.h>
84 #include <netinet/ip.h>
85 #include <netinet/ip_icmp.h> /* required for icmp_var.h */
86 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */
87 #include <netinet/ip_var.h>
88 #include <netinet/ip_options.h>
89 #include <netinet/ip6.h>
90 #include <netinet/icmp6.h>
91 #include <netinet6/in6_pcb.h>
92 #include <netinet6/ip6_var.h>
93 #include <netinet6/nd6.h>
94 #include <netinet/tcp_fsm.h>
95 #include <netinet/tcp_seq.h>
96 #include <netinet/tcp_timer.h>
97 #include <netinet/tcp_var.h>
98 #include <netinet6/tcp6_var.h>
99 #include <netinet/tcpip.h>
100 #include <netinet/tcp_syncache.h>
102 #include <netinet/tcp_debug.h>
103 #endif /* TCPDEBUG */
106 #include <netipsec/ipsec.h>
107 #include <netipsec/ipsec6.h>
110 #include <machine/in_cksum.h>
112 #include <security/mac/mac_framework.h>
114 const int tcprexmtthresh = 3;
116 VNET_DEFINE(struct tcpstat, tcpstat);
117 SYSCTL_VNET_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW,
118 &VNET_NAME(tcpstat), tcpstat,
119 "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
121 int tcp_log_in_vain = 0;
122 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
124 "Log all incoming TCP segments to closed ports");
126 VNET_DEFINE(int, blackhole) = 0;
127 #define V_blackhole VNET(blackhole)
128 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW,
129 &VNET_NAME(blackhole), 0,
130 "Do not send RST on segments to closed ports");
132 VNET_DEFINE(int, tcp_delack_enabled) = 1;
133 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW,
134 &VNET_NAME(tcp_delack_enabled), 0,
135 "Delay ACK to try and piggyback it onto a data packet");
137 VNET_DEFINE(int, drop_synfin) = 0;
138 #define V_drop_synfin VNET(drop_synfin)
139 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW,
140 &VNET_NAME(drop_synfin), 0,
141 "Drop TCP packets with SYN+FIN set");
143 VNET_DEFINE(int, tcp_do_rfc3042) = 1;
144 #define V_tcp_do_rfc3042 VNET(tcp_do_rfc3042)
145 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_RW,
146 &VNET_NAME(tcp_do_rfc3042), 0,
147 "Enable RFC 3042 (Limited Transmit)");
149 VNET_DEFINE(int, tcp_do_rfc3390) = 1;
150 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW,
151 &VNET_NAME(tcp_do_rfc3390), 0,
152 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
154 VNET_DEFINE(int, tcp_do_rfc3465) = 1;
155 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_RW,
156 &VNET_NAME(tcp_do_rfc3465), 0,
157 "Enable RFC 3465 (Appropriate Byte Counting)");
159 VNET_DEFINE(int, tcp_abc_l_var) = 2;
160 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_RW,
161 &VNET_NAME(tcp_abc_l_var), 2,
162 "Cap the max cwnd increment during slow-start to this number of segments");
164 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, CTLFLAG_RW, 0, "TCP ECN");
166 VNET_DEFINE(int, tcp_do_ecn) = 0;
167 SYSCTL_VNET_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_RW,
168 &VNET_NAME(tcp_do_ecn), 0,
171 VNET_DEFINE(int, tcp_ecn_maxretries) = 1;
172 SYSCTL_VNET_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_RW,
173 &VNET_NAME(tcp_ecn_maxretries), 0,
174 "Max retries before giving up on ECN");
176 VNET_DEFINE(int, tcp_insecure_rst) = 0;
177 #define V_tcp_insecure_rst VNET(tcp_insecure_rst)
178 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_RW,
179 &VNET_NAME(tcp_insecure_rst), 0,
180 "Follow the old (insecure) criteria for accepting RST packets");
182 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1;
183 #define V_tcp_do_autorcvbuf VNET(tcp_do_autorcvbuf)
184 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_RW,
185 &VNET_NAME(tcp_do_autorcvbuf), 0,
186 "Enable automatic receive buffer sizing");
188 VNET_DEFINE(int, tcp_autorcvbuf_inc) = 16*1024;
189 #define V_tcp_autorcvbuf_inc VNET(tcp_autorcvbuf_inc)
190 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_RW,
191 &VNET_NAME(tcp_autorcvbuf_inc), 0,
192 "Incrementor step size of automatic receive buffer");
194 VNET_DEFINE(int, tcp_autorcvbuf_max) = 256*1024;
195 #define V_tcp_autorcvbuf_max VNET(tcp_autorcvbuf_max)
196 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_RW,
197 &VNET_NAME(tcp_autorcvbuf_max), 0,
198 "Max size of automatic receive buffer");
200 int tcp_read_locking = 1;
201 SYSCTL_INT(_net_inet_tcp, OID_AUTO, read_locking, CTLFLAG_RW,
202 &tcp_read_locking, 0, "Enable read locking strategy");
204 VNET_DEFINE(struct inpcbhead, tcb);
205 #define tcb6 tcb /* for KAME src sync over BSD*'s */
206 VNET_DEFINE(struct inpcbinfo, tcbinfo);
208 static void tcp_dooptions(struct tcpopt *, u_char *, int, int);
209 static void tcp_do_segment(struct mbuf *, struct tcphdr *,
210 struct socket *, struct tcpcb *, int, int, uint8_t,
212 static void tcp_dropwithreset(struct mbuf *, struct tcphdr *,
213 struct tcpcb *, int, int);
214 static void tcp_pulloutofband(struct socket *,
215 struct tcphdr *, struct mbuf *, int);
216 static void tcp_xmit_timer(struct tcpcb *, int);
217 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *);
218 static void inline cc_ack_received(struct tcpcb *tp, struct tcphdr *th,
220 static void inline cc_conn_init(struct tcpcb *tp);
221 static void inline cc_post_recovery(struct tcpcb *tp, struct tcphdr *th);
222 static void inline tcp_fields_to_host(struct tcphdr *);
223 static void inline hhook_run_tcp_est_in(struct tcpcb *tp,
224 struct tcphdr *th, struct tcpopt *to);
226 static void inline tcp_fields_to_net(struct tcphdr *);
227 static int inline tcp_signature_verify_input(struct mbuf *, int, int,
228 int, struct tcpopt *, struct tcphdr *, u_int);
232 * Kernel module interface for updating tcpstat. The argument is an index
233 * into tcpstat treated as an array of u_long. While this encodes the
234 * general layout of tcpstat into the caller, it doesn't encode its location,
235 * so that future changes to add, for example, per-CPU stats support won't
236 * cause binary compatibility problems for kernel modules.
239 kmod_tcpstat_inc(int statnum)
242 (*((u_long *)&V_tcpstat + statnum))++;
246 * Wrapper for the TCP established input helper hook.
249 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to)
251 struct tcp_hhook_data hhook_data;
253 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) {
258 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data,
264 * CC wrapper hook functions
267 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t type)
269 INP_WLOCK_ASSERT(tp->t_inpcb);
271 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th);
272 if (tp->snd_cwnd <= tp->snd_wnd)
273 tp->ccv->flags |= CCF_CWND_LIMITED;
275 tp->ccv->flags &= ~CCF_CWND_LIMITED;
277 if (type == CC_ACK) {
278 if (tp->snd_cwnd > tp->snd_ssthresh) {
279 tp->t_bytes_acked += min(tp->ccv->bytes_this_ack,
280 V_tcp_abc_l_var * tp->t_maxseg);
281 if (tp->t_bytes_acked >= tp->snd_cwnd) {
282 tp->t_bytes_acked -= tp->snd_cwnd;
283 tp->ccv->flags |= CCF_ABC_SENTAWND;
286 tp->ccv->flags &= ~CCF_ABC_SENTAWND;
287 tp->t_bytes_acked = 0;
291 if (CC_ALGO(tp)->ack_received != NULL) {
292 /* XXXLAS: Find a way to live without this */
293 tp->ccv->curack = th->th_ack;
294 CC_ALGO(tp)->ack_received(tp->ccv, type);
299 cc_conn_init(struct tcpcb *tp)
301 struct hc_metrics_lite metrics;
302 struct inpcb *inp = tp->t_inpcb;
305 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
308 INP_WLOCK_ASSERT(tp->t_inpcb);
310 tcp_hc_get(&inp->inp_inc, &metrics);
312 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) {
314 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
315 TCPSTAT_INC(tcps_usedrtt);
316 if (metrics.rmx_rttvar) {
317 tp->t_rttvar = metrics.rmx_rttvar;
318 TCPSTAT_INC(tcps_usedrttvar);
320 /* default variation is +- 1 rtt */
322 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
324 TCPT_RANGESET(tp->t_rxtcur,
325 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
326 tp->t_rttmin, TCPTV_REXMTMAX);
328 if (metrics.rmx_ssthresh) {
330 * There's some sort of gateway or interface
331 * buffer limit on the path. Use this to set
332 * the slow start threshhold, but set the
333 * threshold to no less than 2*mss.
335 tp->snd_ssthresh = max(2 * tp->t_maxseg, metrics.rmx_ssthresh);
336 TCPSTAT_INC(tcps_usedssthresh);
340 * Set the slow-start flight size depending on whether this
341 * is a local network or not.
343 * Extend this so we cache the cwnd too and retrieve it here.
344 * Make cwnd even bigger than RFC3390 suggests but only if we
345 * have previous experience with the remote host. Be careful
346 * not make cwnd bigger than remote receive window or our own
347 * send socket buffer. Maybe put some additional upper bound
348 * on the retrieved cwnd. Should do incremental updates to
349 * hostcache when cwnd collapses so next connection doesn't
350 * overloads the path again.
352 * XXXAO: Initializing the CWND from the hostcache is broken
353 * and in its current form not RFC conformant. It is disabled
354 * until fixed or removed entirely.
356 * RFC3390 says only do this if SYN or SYN/ACK didn't got lost.
357 * We currently check only in syncache_socket for that.
359 /* #define TCP_METRICS_CWND */
360 #ifdef TCP_METRICS_CWND
361 if (metrics.rmx_cwnd)
362 tp->snd_cwnd = max(tp->t_maxseg, min(metrics.rmx_cwnd / 2,
363 min(tp->snd_wnd, so->so_snd.sb_hiwat)));
366 if (V_tcp_do_rfc3390)
367 tp->snd_cwnd = min(4 * tp->t_maxseg,
368 max(2 * tp->t_maxseg, 4380));
370 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) ||
371 (!isipv6 && in_localaddr(inp->inp_faddr)))
373 else if (in_localaddr(inp->inp_faddr))
375 tp->snd_cwnd = tp->t_maxseg * V_ss_fltsz_local;
377 tp->snd_cwnd = tp->t_maxseg * V_ss_fltsz;
379 if (CC_ALGO(tp)->conn_init != NULL)
380 CC_ALGO(tp)->conn_init(tp->ccv);
384 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type)
386 INP_WLOCK_ASSERT(tp->t_inpcb);
390 if (!IN_FASTRECOVERY(tp->t_flags)) {
391 tp->snd_recover = tp->snd_max;
392 if (tp->t_flags & TF_ECN_PERMIT)
393 tp->t_flags |= TF_ECN_SND_CWR;
397 if (!IN_CONGRECOVERY(tp->t_flags)) {
398 TCPSTAT_INC(tcps_ecn_rcwnd);
399 tp->snd_recover = tp->snd_max;
400 if (tp->t_flags & TF_ECN_PERMIT)
401 tp->t_flags |= TF_ECN_SND_CWR;
406 tp->t_bytes_acked = 0;
407 EXIT_RECOVERY(tp->t_flags);
408 tp->snd_ssthresh = max(2, min(tp->snd_wnd, tp->snd_cwnd) / 2 /
409 tp->t_maxseg) * tp->t_maxseg;
410 tp->snd_cwnd = tp->t_maxseg;
413 TCPSTAT_INC(tcps_sndrexmitbad);
414 /* RTO was unnecessary, so reset everything. */
415 tp->snd_cwnd = tp->snd_cwnd_prev;
416 tp->snd_ssthresh = tp->snd_ssthresh_prev;
417 tp->snd_recover = tp->snd_recover_prev;
418 if (tp->t_flags & TF_WASFRECOVERY)
419 ENTER_FASTRECOVERY(tp->t_flags);
420 if (tp->t_flags & TF_WASCRECOVERY)
421 ENTER_CONGRECOVERY(tp->t_flags);
422 tp->snd_nxt = tp->snd_max;
423 tp->t_flags &= ~TF_PREVVALID;
428 if (CC_ALGO(tp)->cong_signal != NULL) {
430 tp->ccv->curack = th->th_ack;
431 CC_ALGO(tp)->cong_signal(tp->ccv, type);
436 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th)
438 INP_WLOCK_ASSERT(tp->t_inpcb);
440 /* XXXLAS: KASSERT that we're in recovery? */
442 if (CC_ALGO(tp)->post_recovery != NULL) {
443 tp->ccv->curack = th->th_ack;
444 CC_ALGO(tp)->post_recovery(tp->ccv);
446 /* XXXLAS: EXIT_RECOVERY ? */
447 tp->t_bytes_acked = 0;
451 tcp_fields_to_host(struct tcphdr *th)
454 th->th_seq = ntohl(th->th_seq);
455 th->th_ack = ntohl(th->th_ack);
456 th->th_win = ntohs(th->th_win);
457 th->th_urp = ntohs(th->th_urp);
462 tcp_fields_to_net(struct tcphdr *th)
465 th->th_seq = htonl(th->th_seq);
466 th->th_ack = htonl(th->th_ack);
467 th->th_win = htons(th->th_win);
468 th->th_urp = htons(th->th_urp);
472 tcp_signature_verify_input(struct mbuf *m, int off0, int tlen, int optlen,
473 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag)
477 tcp_fields_to_net(th);
478 ret = tcp_signature_verify(m, off0, tlen, optlen, to, th, tcpbflag);
479 tcp_fields_to_host(th);
484 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
486 #define ND6_HINT(tp) \
488 if ((tp) && (tp)->t_inpcb && \
489 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \
490 nd6_nud_hint(NULL, NULL, 0); \
497 * Indicate whether this ack should be delayed. We can delay the ack if
498 * - there is no delayed ack timer in progress and
499 * - our last ack wasn't a 0-sized window. We never want to delay
500 * the ack that opens up a 0-sized window and
501 * - delayed acks are enabled or
502 * - this is a half-synchronized T/TCP connection.
504 #define DELAY_ACK(tp) \
505 ((!tcp_timer_active(tp, TT_DELACK) && \
506 (tp->t_flags & TF_RXWIN0SENT) == 0) && \
507 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
510 * TCP input handling is split into multiple parts:
511 * tcp6_input is a thin wrapper around tcp_input for the extended
512 * ip6_protox[] call format in ip6_input
513 * tcp_input handles primary segment validation, inpcb lookup and
514 * SYN processing on listen sockets
515 * tcp_do_segment processes the ACK and text of the segment for
516 * establishing, established and closing connections
520 tcp6_input(struct mbuf **mp, int *offp, int proto)
522 struct mbuf *m = *mp;
523 struct in6_ifaddr *ia6;
525 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
528 * draft-itojun-ipv6-tcp-to-anycast
529 * better place to put this in?
531 ia6 = ip6_getdstifaddr(m);
532 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
535 ifa_free(&ia6->ia_ifa);
536 ip6 = mtod(m, struct ip6_hdr *);
537 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
538 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
542 ifa_free(&ia6->ia_ifa);
550 tcp_input(struct mbuf *m, int off0)
553 struct ip *ip = NULL;
555 struct inpcb *inp = NULL;
556 struct tcpcb *tp = NULL;
557 struct socket *so = NULL;
563 int rstreason = 0; /* For badport_bandlim accounting purposes */
566 uint8_t sig_checked = 0;
568 #ifdef IPFIREWALL_FORWARD
569 struct m_tag *fwd_tag;
572 struct ip6_hdr *ip6 = NULL;
575 const void *ip6 = NULL;
576 const int isipv6 = 0;
578 struct tcpopt to; /* options in this segment */
579 char *s = NULL; /* address and port logging */
581 #define TI_UNLOCKED 1
587 * The size of tcp_saveipgen must be the size of the max ip header,
590 u_char tcp_saveipgen[IP6_HDR_LEN];
591 struct tcphdr tcp_savetcp;
596 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
600 TCPSTAT_INC(tcps_rcvtotal);
604 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */
605 ip6 = mtod(m, struct ip6_hdr *);
606 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
607 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) {
608 TCPSTAT_INC(tcps_rcvbadsum);
611 th = (struct tcphdr *)((caddr_t)ip6 + off0);
614 * Be proactive about unspecified IPv6 address in source.
615 * As we use all-zero to indicate unbounded/unconnected pcb,
616 * unspecified IPv6 address can be used to confuse us.
618 * Note that packets with unspecified IPv6 destination is
619 * already dropped in ip6_input.
621 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
626 th = NULL; /* XXX: Avoid compiler warning. */
630 * Get IP and TCP header together in first mbuf.
631 * Note: IP leaves IP header in first mbuf.
633 if (off0 > sizeof (struct ip)) {
634 ip_stripoptions(m, (struct mbuf *)0);
635 off0 = sizeof(struct ip);
637 if (m->m_len < sizeof (struct tcpiphdr)) {
638 if ((m = m_pullup(m, sizeof (struct tcpiphdr)))
640 TCPSTAT_INC(tcps_rcvshort);
644 ip = mtod(m, struct ip *);
645 ipov = (struct ipovly *)ip;
646 th = (struct tcphdr *)((caddr_t)ip + off0);
649 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
650 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
651 th->th_sum = m->m_pkthdr.csum_data;
653 th->th_sum = in_pseudo(ip->ip_src.s_addr,
655 htonl(m->m_pkthdr.csum_data +
658 th->th_sum ^= 0xffff;
660 ipov->ih_len = (u_short)tlen;
661 ipov->ih_len = htons(ipov->ih_len);
665 * Checksum extended TCP header and data.
667 len = sizeof (struct ip) + tlen;
668 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
669 ipov->ih_len = (u_short)tlen;
670 ipov->ih_len = htons(ipov->ih_len);
671 th->th_sum = in_cksum(m, len);
674 TCPSTAT_INC(tcps_rcvbadsum);
677 /* Re-initialization for later version check */
678 ip->ip_v = IPVERSION;
683 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
689 * Check that TCP offset makes sense,
690 * pull out TCP options and adjust length. XXX
692 off = th->th_off << 2;
693 if (off < sizeof (struct tcphdr) || off > tlen) {
694 TCPSTAT_INC(tcps_rcvbadoff);
697 tlen -= off; /* tlen is used instead of ti->ti_len */
698 if (off > sizeof (struct tcphdr)) {
701 IP6_EXTHDR_CHECK(m, off0, off, );
702 ip6 = mtod(m, struct ip6_hdr *);
703 th = (struct tcphdr *)((caddr_t)ip6 + off0);
706 if (m->m_len < sizeof(struct ip) + off) {
707 if ((m = m_pullup(m, sizeof (struct ip) + off))
709 TCPSTAT_INC(tcps_rcvshort);
712 ip = mtod(m, struct ip *);
713 ipov = (struct ipovly *)ip;
714 th = (struct tcphdr *)((caddr_t)ip + off0);
717 optlen = off - sizeof (struct tcphdr);
718 optp = (u_char *)(th + 1);
720 thflags = th->th_flags;
723 * Convert TCP protocol specific fields to host format.
725 tcp_fields_to_host(th);
728 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options.
730 drop_hdrlen = off0 + off;
733 * Locate pcb for segment, which requires a lock on tcbinfo.
734 * Optimisticaly acquire a global read lock rather than a write lock
735 * unless header flags necessarily imply a state change. There are
736 * two cases where we might discover later we need a write lock
737 * despite the flags: ACKs moving a connection out of the syncache,
738 * and ACKs for a connection in TIMEWAIT.
740 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 ||
741 tcp_read_locking == 0) {
742 INP_INFO_WLOCK(&V_tcbinfo);
743 ti_locked = TI_WLOCKED;
745 INP_INFO_RLOCK(&V_tcbinfo);
746 ti_locked = TI_RLOCKED;
751 if (ti_locked == TI_RLOCKED)
752 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
753 else if (ti_locked == TI_WLOCKED)
754 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
756 panic("%s: findpcb ti_locked %d\n", __func__, ti_locked);
759 #ifdef IPFIREWALL_FORWARD
761 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
763 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
765 if (fwd_tag != NULL && isipv6 == 0) { /* IPv6 support is not yet */
766 struct sockaddr_in *next_hop;
768 next_hop = (struct sockaddr_in *)(fwd_tag+1);
770 * Transparently forwarded. Pretend to be the destination.
771 * already got one like this?
773 inp = in_pcblookup_hash(&V_tcbinfo,
774 ip->ip_src, th->th_sport,
775 ip->ip_dst, th->th_dport,
776 0, m->m_pkthdr.rcvif);
778 /* It's new. Try to find the ambushing socket. */
779 inp = in_pcblookup_hash(&V_tcbinfo,
780 ip->ip_src, th->th_sport,
783 ntohs(next_hop->sin_port) :
788 /* Remove the tag from the packet. We don't need it anymore. */
789 m_tag_delete(m, fwd_tag);
791 #endif /* IPFIREWALL_FORWARD */
795 inp = in6_pcblookup_hash(&V_tcbinfo,
796 &ip6->ip6_src, th->th_sport,
797 &ip6->ip6_dst, th->th_dport,
802 inp = in_pcblookup_hash(&V_tcbinfo,
803 ip->ip_src, th->th_sport,
804 ip->ip_dst, th->th_dport,
810 * If the INPCB does not exist then all data in the incoming
811 * segment is discarded and an appropriate RST is sent back.
812 * XXX MRT Send RST using which routing table?
816 * Log communication attempts to ports that are not
819 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) ||
820 tcp_log_in_vain == 2) {
821 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6)))
822 log(LOG_INFO, "%s; %s: Connection attempt "
823 "to closed port\n", s, __func__);
826 * When blackholing do not respond with a RST but
827 * completely ignore the segment and drop it.
829 if ((V_blackhole == 1 && (thflags & TH_SYN)) ||
833 rstreason = BANDLIM_RST_CLOSEDPORT;
837 if (!(inp->inp_flags & INP_HW_FLOWID)
838 && (m->m_flags & M_FLOWID)
839 && ((inp->inp_socket == NULL)
840 || !(inp->inp_socket->so_options & SO_ACCEPTCONN))) {
841 inp->inp_flags |= INP_HW_FLOWID;
842 inp->inp_flags &= ~INP_SW_FLOWID;
843 inp->inp_flowid = m->m_pkthdr.flowid;
847 if (isipv6 && ipsec6_in_reject(m, inp)) {
848 V_ipsec6stat.in_polvio++;
852 if (ipsec4_in_reject(m, inp) != 0) {
853 V_ipsec4stat.in_polvio++;
859 * Check the minimum TTL for socket.
861 if (inp->inp_ip_minttl != 0) {
863 if (isipv6 && inp->inp_ip_minttl > ip6->ip6_hlim)
867 if (inp->inp_ip_minttl > ip->ip_ttl)
872 * A previous connection in TIMEWAIT state is supposed to catch stray
873 * or duplicate segments arriving late. If this segment was a
874 * legitimate new connection attempt the old INPCB gets removed and
875 * we can try again to find a listening socket.
877 * At this point, due to earlier optimism, we may hold a read lock on
878 * the inpcbinfo, rather than a write lock. If so, we need to
879 * upgrade, or if that fails, acquire a reference on the inpcb, drop
880 * all locks, acquire a global write lock, and then re-acquire the
881 * inpcb lock. We may at that point discover that another thread has
882 * tried to free the inpcb, in which case we need to loop back and
883 * try to find a new inpcb to deliver to.
886 if (inp->inp_flags & INP_TIMEWAIT) {
887 KASSERT(ti_locked == TI_RLOCKED || ti_locked == TI_WLOCKED,
888 ("%s: INP_TIMEWAIT ti_locked %d", __func__, ti_locked));
890 if (ti_locked == TI_RLOCKED) {
891 if (INP_INFO_TRY_UPGRADE(&V_tcbinfo) == 0) {
894 INP_INFO_RUNLOCK(&V_tcbinfo);
895 INP_INFO_WLOCK(&V_tcbinfo);
896 ti_locked = TI_WLOCKED;
898 if (in_pcbrele(inp)) {
903 ti_locked = TI_WLOCKED;
905 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
907 if (thflags & TH_SYN)
908 tcp_dooptions(&to, optp, optlen, TO_SYN);
910 * NB: tcp_twcheck unlocks the INP and frees the mbuf.
912 if (tcp_twcheck(inp, &to, th, m, tlen))
914 INP_INFO_WUNLOCK(&V_tcbinfo);
918 * The TCPCB may no longer exist if the connection is winding
919 * down or it is in the CLOSED state. Either way we drop the
920 * segment and send an appropriate response.
923 if (tp == NULL || tp->t_state == TCPS_CLOSED) {
924 rstreason = BANDLIM_RST_CLOSEDPORT;
929 * We've identified a valid inpcb, but it could be that we need an
930 * inpcbinfo write lock and have only a read lock. In this case,
931 * attempt to upgrade/relock using the same strategy as the TIMEWAIT
932 * case above. If we relock, we have to jump back to 'relocked' as
933 * the connection might now be in TIMEWAIT.
935 if (tp->t_state != TCPS_ESTABLISHED ||
936 (thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 ||
937 tcp_read_locking == 0) {
938 KASSERT(ti_locked == TI_RLOCKED || ti_locked == TI_WLOCKED,
939 ("%s: upgrade check ti_locked %d", __func__, ti_locked));
941 if (ti_locked == TI_RLOCKED) {
942 if (INP_INFO_TRY_UPGRADE(&V_tcbinfo) == 0) {
945 INP_INFO_RUNLOCK(&V_tcbinfo);
946 INP_INFO_WLOCK(&V_tcbinfo);
947 ti_locked = TI_WLOCKED;
949 if (in_pcbrele(inp)) {
955 ti_locked = TI_WLOCKED;
957 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
961 INP_WLOCK_ASSERT(inp);
962 if (mac_inpcb_check_deliver(inp, m))
965 so = inp->inp_socket;
966 KASSERT(so != NULL, ("%s: so == NULL", __func__));
968 if (so->so_options & SO_DEBUG) {
969 ostate = tp->t_state;
972 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
975 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
980 * When the socket is accepting connections (the INPCB is in LISTEN
981 * state) we look into the SYN cache if this is a new connection
982 * attempt or the completion of a previous one.
984 if (so->so_options & SO_ACCEPTCONN) {
985 struct in_conninfo inc;
987 KASSERT(tp->t_state == TCPS_LISTEN, ("%s: so accepting but "
988 "tp not listening", __func__));
990 bzero(&inc, sizeof(inc));
993 inc.inc_flags |= INC_ISIPV6;
994 inc.inc6_faddr = ip6->ip6_src;
995 inc.inc6_laddr = ip6->ip6_dst;
999 inc.inc_faddr = ip->ip_src;
1000 inc.inc_laddr = ip->ip_dst;
1002 inc.inc_fport = th->th_sport;
1003 inc.inc_lport = th->th_dport;
1004 inc.inc_fibnum = so->so_fibnum;
1007 * Check for an existing connection attempt in syncache if
1008 * the flag is only ACK. A successful lookup creates a new
1009 * socket appended to the listen queue in SYN_RECEIVED state.
1011 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
1013 * Parse the TCP options here because
1014 * syncookies need access to the reflected
1017 tcp_dooptions(&to, optp, optlen, 0);
1019 * NB: syncache_expand() doesn't unlock
1020 * inp and tcpinfo locks.
1022 if (!syncache_expand(&inc, &to, th, &so, m)) {
1024 * No syncache entry or ACK was not
1025 * for our SYN/ACK. Send a RST.
1026 * NB: syncache did its own logging
1027 * of the failure cause.
1029 rstreason = BANDLIM_RST_OPENPORT;
1034 * We completed the 3-way handshake
1035 * but could not allocate a socket
1036 * either due to memory shortage,
1037 * listen queue length limits or
1038 * global socket limits. Send RST
1039 * or wait and have the remote end
1040 * retransmit the ACK for another
1043 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1044 log(LOG_DEBUG, "%s; %s: Listen socket: "
1045 "Socket allocation failed due to "
1046 "limits or memory shortage, %s\n",
1048 V_tcp_sc_rst_sock_fail ?
1049 "sending RST" : "try again");
1050 if (V_tcp_sc_rst_sock_fail) {
1051 rstreason = BANDLIM_UNLIMITED;
1057 * Socket is created in state SYN_RECEIVED.
1058 * Unlock the listen socket, lock the newly
1059 * created socket and update the tp variable.
1061 INP_WUNLOCK(inp); /* listen socket */
1062 inp = sotoinpcb(so);
1063 INP_WLOCK(inp); /* new connection */
1064 tp = intotcpcb(inp);
1065 KASSERT(tp->t_state == TCPS_SYN_RECEIVED,
1066 ("%s: ", __func__));
1067 #ifdef TCP_SIGNATURE
1068 if (sig_checked == 0) {
1069 tcp_dooptions(&to, optp, optlen,
1070 (thflags & TH_SYN) ? TO_SYN : 0);
1071 if (!tcp_signature_verify_input(m, off0, tlen,
1072 optlen, &to, th, tp->t_flags)) {
1075 * In SYN_SENT state if it receives an
1076 * RST, it is allowed for further
1079 if ((thflags & TH_RST) == 0 ||
1080 (tp->t_state == TCPS_SYN_SENT) == 0)
1088 * Process the segment and the data it
1089 * contains. tcp_do_segment() consumes
1090 * the mbuf chain and unlocks the inpcb.
1092 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen,
1094 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1098 * Segment flag validation for new connection attempts:
1100 * Our (SYN|ACK) response was rejected.
1101 * Check with syncache and remove entry to prevent
1104 * NB: syncache_chkrst does its own logging of failure
1107 if (thflags & TH_RST) {
1108 syncache_chkrst(&inc, th);
1112 * We can't do anything without SYN.
1114 if ((thflags & TH_SYN) == 0) {
1115 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1116 log(LOG_DEBUG, "%s; %s: Listen socket: "
1117 "SYN is missing, segment ignored\n",
1119 TCPSTAT_INC(tcps_badsyn);
1123 * (SYN|ACK) is bogus on a listen socket.
1125 if (thflags & TH_ACK) {
1126 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1127 log(LOG_DEBUG, "%s; %s: Listen socket: "
1128 "SYN|ACK invalid, segment rejected\n",
1130 syncache_badack(&inc); /* XXX: Not needed! */
1131 TCPSTAT_INC(tcps_badsyn);
1132 rstreason = BANDLIM_RST_OPENPORT;
1136 * If the drop_synfin option is enabled, drop all
1137 * segments with both the SYN and FIN bits set.
1138 * This prevents e.g. nmap from identifying the
1140 * XXX: Poor reasoning. nmap has other methods
1141 * and is constantly refining its stack detection
1143 * XXX: This is a violation of the TCP specification
1144 * and was used by RFC1644.
1146 if ((thflags & TH_FIN) && V_drop_synfin) {
1147 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1148 log(LOG_DEBUG, "%s; %s: Listen socket: "
1149 "SYN|FIN segment ignored (based on "
1150 "sysctl setting)\n", s, __func__);
1151 TCPSTAT_INC(tcps_badsyn);
1155 * Segment's flags are (SYN) or (SYN|FIN).
1157 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored
1158 * as they do not affect the state of the TCP FSM.
1159 * The data pointed to by TH_URG and th_urp is ignored.
1161 KASSERT((thflags & (TH_RST|TH_ACK)) == 0,
1162 ("%s: Listen socket: TH_RST or TH_ACK set", __func__));
1163 KASSERT(thflags & (TH_SYN),
1164 ("%s: Listen socket: TH_SYN not set", __func__));
1167 * If deprecated address is forbidden,
1168 * we do not accept SYN to deprecated interface
1169 * address to prevent any new inbound connection from
1170 * getting established.
1171 * When we do not accept SYN, we send a TCP RST,
1172 * with deprecated source address (instead of dropping
1173 * it). We compromise it as it is much better for peer
1174 * to send a RST, and RST will be the final packet
1177 * If we do not forbid deprecated addresses, we accept
1178 * the SYN packet. RFC2462 does not suggest dropping
1180 * If we decipher RFC2462 5.5.4, it says like this:
1181 * 1. use of deprecated addr with existing
1182 * communication is okay - "SHOULD continue to be
1184 * 2. use of it with new communication:
1185 * (2a) "SHOULD NOT be used if alternate address
1186 * with sufficient scope is available"
1187 * (2b) nothing mentioned otherwise.
1188 * Here we fall into (2b) case as we have no choice in
1189 * our source address selection - we must obey the peer.
1191 * The wording in RFC2462 is confusing, and there are
1192 * multiple description text for deprecated address
1193 * handling - worse, they are not exactly the same.
1194 * I believe 5.5.4 is the best one, so we follow 5.5.4.
1196 if (isipv6 && !V_ip6_use_deprecated) {
1197 struct in6_ifaddr *ia6;
1199 ia6 = ip6_getdstifaddr(m);
1201 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1202 ifa_free(&ia6->ia_ifa);
1203 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1204 log(LOG_DEBUG, "%s; %s: Listen socket: "
1205 "Connection attempt to deprecated "
1206 "IPv6 address rejected\n",
1208 rstreason = BANDLIM_RST_OPENPORT;
1212 ifa_free(&ia6->ia_ifa);
1216 * Basic sanity checks on incoming SYN requests:
1217 * Don't respond if the destination is a link layer
1218 * broadcast according to RFC1122 4.2.3.10, p. 104.
1219 * If it is from this socket it must be forged.
1220 * Don't respond if the source or destination is a
1221 * global or subnet broad- or multicast address.
1222 * Note that it is quite possible to receive unicast
1223 * link-layer packets with a broadcast IP address. Use
1224 * in_broadcast() to find them.
1226 if (m->m_flags & (M_BCAST|M_MCAST)) {
1227 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1228 log(LOG_DEBUG, "%s; %s: Listen socket: "
1229 "Connection attempt from broad- or multicast "
1230 "link layer address ignored\n", s, __func__);
1235 if (th->th_dport == th->th_sport &&
1236 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) {
1237 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1238 log(LOG_DEBUG, "%s; %s: Listen socket: "
1239 "Connection attempt to/from self "
1240 "ignored\n", s, __func__);
1243 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1244 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) {
1245 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1246 log(LOG_DEBUG, "%s; %s: Listen socket: "
1247 "Connection attempt from/to multicast "
1248 "address ignored\n", s, __func__);
1253 if (th->th_dport == th->th_sport &&
1254 ip->ip_dst.s_addr == ip->ip_src.s_addr) {
1255 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1256 log(LOG_DEBUG, "%s; %s: Listen socket: "
1257 "Connection attempt from/to self "
1258 "ignored\n", s, __func__);
1261 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1262 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1263 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1264 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
1265 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1266 log(LOG_DEBUG, "%s; %s: Listen socket: "
1267 "Connection attempt from/to broad- "
1268 "or multicast address ignored\n",
1274 * SYN appears to be valid. Create compressed TCP state
1278 if (so->so_options & SO_DEBUG)
1279 tcp_trace(TA_INPUT, ostate, tp,
1280 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1282 tcp_dooptions(&to, optp, optlen, TO_SYN);
1283 syncache_add(&inc, &to, th, inp, &so, m);
1285 * Entry added to syncache and mbuf consumed.
1286 * Everything already unlocked by syncache_add().
1288 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1292 #ifdef TCP_SIGNATURE
1293 if (sig_checked == 0) {
1294 tcp_dooptions(&to, optp, optlen,
1295 (thflags & TH_SYN) ? TO_SYN : 0);
1296 if (!tcp_signature_verify_input(m, off0, tlen, optlen, &to,
1300 * In SYN_SENT state if it receives an RST, it is
1301 * allowed for further processing.
1303 if ((thflags & TH_RST) == 0 ||
1304 (tp->t_state == TCPS_SYN_SENT) == 0)
1312 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later
1313 * state. tcp_do_segment() always consumes the mbuf chain, unlocks
1314 * the inpcb, and unlocks pcbinfo.
1316 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos, ti_locked);
1317 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1321 if (ti_locked == TI_RLOCKED)
1322 INP_INFO_RUNLOCK(&V_tcbinfo);
1323 else if (ti_locked == TI_WLOCKED)
1324 INP_INFO_WUNLOCK(&V_tcbinfo);
1326 panic("%s: dropwithreset ti_locked %d", __func__, ti_locked);
1327 ti_locked = TI_UNLOCKED;
1330 tcp_dropwithreset(m, th, tp, tlen, rstreason);
1333 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
1334 m = NULL; /* mbuf chain got consumed. */
1338 if (ti_locked == TI_RLOCKED)
1339 INP_INFO_RUNLOCK(&V_tcbinfo);
1340 else if (ti_locked == TI_WLOCKED)
1341 INP_INFO_WUNLOCK(&V_tcbinfo);
1343 panic("%s: dropunlock ti_locked %d", __func__, ti_locked);
1344 ti_locked = TI_UNLOCKED;
1350 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1358 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so,
1359 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos,
1362 int thflags, acked, ourfinisacked, needoutput = 0;
1363 int rstreason, todrop, win;
1369 * The size of tcp_saveipgen must be the size of the max ip header,
1372 u_char tcp_saveipgen[IP6_HDR_LEN];
1373 struct tcphdr tcp_savetcp;
1376 thflags = th->th_flags;
1377 tp->sackhint.last_sack_ack = 0;
1380 * If this is either a state-changing packet or current state isn't
1381 * established, we require a write lock on tcbinfo. Otherwise, we
1382 * allow either a read lock or a write lock, as we may have acquired
1383 * a write lock due to a race.
1385 * Require a global write lock for SYN/FIN/RST segments or
1386 * non-established connections; otherwise accept either a read or
1387 * write lock, as we may have conservatively acquired a write lock in
1388 * certain cases in tcp_input() (is this still true?). Currently we
1389 * will never enter with no lock, so we try to drop it quickly in the
1390 * common pure ack/pure data cases.
1392 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 ||
1393 tp->t_state != TCPS_ESTABLISHED) {
1394 KASSERT(ti_locked == TI_WLOCKED, ("%s ti_locked %d for "
1395 "SYN/FIN/RST/!EST", __func__, ti_locked));
1396 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1399 if (ti_locked == TI_RLOCKED)
1400 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1401 else if (ti_locked == TI_WLOCKED)
1402 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1404 panic("%s: ti_locked %d for EST", __func__,
1408 INP_WLOCK_ASSERT(tp->t_inpcb);
1409 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN",
1411 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT",
1415 * Segment received on connection.
1416 * Reset idle time and keep-alive timer.
1417 * XXX: This should be done after segment
1418 * validation to ignore broken/spoofed segs.
1420 tp->t_rcvtime = ticks;
1421 if (TCPS_HAVEESTABLISHED(tp->t_state))
1422 tcp_timer_activate(tp, TT_KEEP, tcp_keepidle);
1425 * Unscale the window into a 32-bit value.
1426 * For the SYN_SENT state the scale is zero.
1428 tiwin = th->th_win << tp->snd_scale;
1431 * TCP ECN processing.
1433 if (tp->t_flags & TF_ECN_PERMIT) {
1434 if (thflags & TH_CWR)
1435 tp->t_flags &= ~TF_ECN_SND_ECE;
1436 switch (iptos & IPTOS_ECN_MASK) {
1438 tp->t_flags |= TF_ECN_SND_ECE;
1439 TCPSTAT_INC(tcps_ecn_ce);
1441 case IPTOS_ECN_ECT0:
1442 TCPSTAT_INC(tcps_ecn_ect0);
1444 case IPTOS_ECN_ECT1:
1445 TCPSTAT_INC(tcps_ecn_ect1);
1448 /* Congestion experienced. */
1449 if (thflags & TH_ECE) {
1450 cc_cong_signal(tp, th, CC_ECN);
1455 * Parse options on any incoming segment.
1457 tcp_dooptions(&to, (u_char *)(th + 1),
1458 (th->th_off << 2) - sizeof(struct tcphdr),
1459 (thflags & TH_SYN) ? TO_SYN : 0);
1462 * If echoed timestamp is later than the current time,
1463 * fall back to non RFC1323 RTT calculation. Normalize
1464 * timestamp if syncookies were used when this connection
1467 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1468 to.to_tsecr -= tp->ts_offset;
1469 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks()))
1474 * Process options only when we get SYN/ACK back. The SYN case
1475 * for incoming connections is handled in tcp_syncache.
1476 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
1477 * or <SYN,ACK>) segment itself is never scaled.
1478 * XXX this is traditional behavior, may need to be cleaned up.
1480 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1481 if ((to.to_flags & TOF_SCALE) &&
1482 (tp->t_flags & TF_REQ_SCALE)) {
1483 tp->t_flags |= TF_RCVD_SCALE;
1484 tp->snd_scale = to.to_wscale;
1487 * Initial send window. It will be updated with
1488 * the next incoming segment to the scaled value.
1490 tp->snd_wnd = th->th_win;
1491 if (to.to_flags & TOF_TS) {
1492 tp->t_flags |= TF_RCVD_TSTMP;
1493 tp->ts_recent = to.to_tsval;
1494 tp->ts_recent_age = tcp_ts_getticks();
1496 if (to.to_flags & TOF_MSS)
1497 tcp_mss(tp, to.to_mss);
1498 if ((tp->t_flags & TF_SACK_PERMIT) &&
1499 (to.to_flags & TOF_SACKPERM) == 0)
1500 tp->t_flags &= ~TF_SACK_PERMIT;
1504 * Header prediction: check for the two common cases
1505 * of a uni-directional data xfer. If the packet has
1506 * no control flags, is in-sequence, the window didn't
1507 * change and we're not retransmitting, it's a
1508 * candidate. If the length is zero and the ack moved
1509 * forward, we're the sender side of the xfer. Just
1510 * free the data acked & wake any higher level process
1511 * that was blocked waiting for space. If the length
1512 * is non-zero and the ack didn't move, we're the
1513 * receiver side. If we're getting packets in-order
1514 * (the reassembly queue is empty), add the data to
1515 * the socket buffer and note that we need a delayed ack.
1516 * Make sure that the hidden state-flags are also off.
1517 * Since we check for TCPS_ESTABLISHED first, it can only
1520 if (tp->t_state == TCPS_ESTABLISHED &&
1521 th->th_seq == tp->rcv_nxt &&
1522 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1523 tp->snd_nxt == tp->snd_max &&
1524 tiwin && tiwin == tp->snd_wnd &&
1525 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1526 LIST_EMPTY(&tp->t_segq) &&
1527 ((to.to_flags & TOF_TS) == 0 ||
1528 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) {
1531 * If last ACK falls within this segment's sequence numbers,
1532 * record the timestamp.
1533 * NOTE that the test is modified according to the latest
1534 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1536 if ((to.to_flags & TOF_TS) != 0 &&
1537 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1538 tp->ts_recent_age = tcp_ts_getticks();
1539 tp->ts_recent = to.to_tsval;
1543 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1544 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1545 !IN_RECOVERY(tp->t_flags) &&
1546 (to.to_flags & TOF_SACK) == 0 &&
1547 TAILQ_EMPTY(&tp->snd_holes)) {
1549 * This is a pure ack for outstanding data.
1551 if (ti_locked == TI_RLOCKED)
1552 INP_INFO_RUNLOCK(&V_tcbinfo);
1553 else if (ti_locked == TI_WLOCKED)
1554 INP_INFO_WUNLOCK(&V_tcbinfo);
1556 panic("%s: ti_locked %d on pure ACK",
1557 __func__, ti_locked);
1558 ti_locked = TI_UNLOCKED;
1560 TCPSTAT_INC(tcps_predack);
1563 * "bad retransmit" recovery.
1565 if (tp->t_rxtshift == 1 &&
1566 tp->t_flags & TF_PREVVALID &&
1567 (int)(ticks - tp->t_badrxtwin) < 0) {
1568 cc_cong_signal(tp, th, CC_RTO_ERR);
1572 * Recalculate the transmit timer / rtt.
1574 * Some boxes send broken timestamp replies
1575 * during the SYN+ACK phase, ignore
1576 * timestamps of 0 or we could calculate a
1577 * huge RTT and blow up the retransmit timer.
1579 if ((to.to_flags & TOF_TS) != 0 &&
1583 t = tcp_ts_getticks() - to.to_tsecr;
1584 if (!tp->t_rttlow || tp->t_rttlow > t)
1587 TCP_TS_TO_TICKS(t) + 1);
1588 } else if (tp->t_rtttime &&
1589 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1590 if (!tp->t_rttlow ||
1591 tp->t_rttlow > ticks - tp->t_rtttime)
1592 tp->t_rttlow = ticks - tp->t_rtttime;
1594 ticks - tp->t_rtttime);
1596 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1597 acked = BYTES_THIS_ACK(tp, th);
1599 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
1600 hhook_run_tcp_est_in(tp, th, &to);
1602 TCPSTAT_INC(tcps_rcvackpack);
1603 TCPSTAT_ADD(tcps_rcvackbyte, acked);
1604 sbdrop(&so->so_snd, acked);
1605 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1606 SEQ_LEQ(th->th_ack, tp->snd_recover))
1607 tp->snd_recover = th->th_ack - 1;
1610 * Let the congestion control algorithm update
1611 * congestion control related information. This
1612 * typically means increasing the congestion
1615 cc_ack_received(tp, th, CC_ACK);
1617 tp->snd_una = th->th_ack;
1619 * Pull snd_wl2 up to prevent seq wrap relative
1622 tp->snd_wl2 = th->th_ack;
1625 ND6_HINT(tp); /* Some progress has been made. */
1628 * If all outstanding data are acked, stop
1629 * retransmit timer, otherwise restart timer
1630 * using current (possibly backed-off) value.
1631 * If process is waiting for space,
1632 * wakeup/selwakeup/signal. If data
1633 * are ready to send, let tcp_output
1634 * decide between more output or persist.
1637 if (so->so_options & SO_DEBUG)
1638 tcp_trace(TA_INPUT, ostate, tp,
1639 (void *)tcp_saveipgen,
1642 if (tp->snd_una == tp->snd_max)
1643 tcp_timer_activate(tp, TT_REXMT, 0);
1644 else if (!tcp_timer_active(tp, TT_PERSIST))
1645 tcp_timer_activate(tp, TT_REXMT,
1648 if (so->so_snd.sb_cc)
1649 (void) tcp_output(tp);
1652 } else if (th->th_ack == tp->snd_una &&
1653 tlen <= sbspace(&so->so_rcv)) {
1654 int newsize = 0; /* automatic sockbuf scaling */
1657 * This is a pure, in-sequence data packet with
1658 * nothing on the reassembly queue and we have enough
1659 * buffer space to take it.
1661 if (ti_locked == TI_RLOCKED)
1662 INP_INFO_RUNLOCK(&V_tcbinfo);
1663 else if (ti_locked == TI_WLOCKED)
1664 INP_INFO_WUNLOCK(&V_tcbinfo);
1666 panic("%s: ti_locked %d on pure data "
1667 "segment", __func__, ti_locked);
1668 ti_locked = TI_UNLOCKED;
1670 /* Clean receiver SACK report if present */
1671 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks)
1672 tcp_clean_sackreport(tp);
1673 TCPSTAT_INC(tcps_preddat);
1674 tp->rcv_nxt += tlen;
1676 * Pull snd_wl1 up to prevent seq wrap relative to
1679 tp->snd_wl1 = th->th_seq;
1681 * Pull rcv_up up to prevent seq wrap relative to
1684 tp->rcv_up = tp->rcv_nxt;
1685 TCPSTAT_INC(tcps_rcvpack);
1686 TCPSTAT_ADD(tcps_rcvbyte, tlen);
1687 ND6_HINT(tp); /* Some progress has been made */
1689 if (so->so_options & SO_DEBUG)
1690 tcp_trace(TA_INPUT, ostate, tp,
1691 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1694 * Automatic sizing of receive socket buffer. Often the send
1695 * buffer size is not optimally adjusted to the actual network
1696 * conditions at hand (delay bandwidth product). Setting the
1697 * buffer size too small limits throughput on links with high
1698 * bandwidth and high delay (eg. trans-continental/oceanic links).
1700 * On the receive side the socket buffer memory is only rarely
1701 * used to any significant extent. This allows us to be much
1702 * more aggressive in scaling the receive socket buffer. For
1703 * the case that the buffer space is actually used to a large
1704 * extent and we run out of kernel memory we can simply drop
1705 * the new segments; TCP on the sender will just retransmit it
1706 * later. Setting the buffer size too big may only consume too
1707 * much kernel memory if the application doesn't read() from
1708 * the socket or packet loss or reordering makes use of the
1711 * The criteria to step up the receive buffer one notch are:
1712 * 1. the number of bytes received during the time it takes
1713 * one timestamp to be reflected back to us (the RTT);
1714 * 2. received bytes per RTT is within seven eighth of the
1715 * current socket buffer size;
1716 * 3. receive buffer size has not hit maximal automatic size;
1718 * This algorithm does one step per RTT at most and only if
1719 * we receive a bulk stream w/o packet losses or reorderings.
1720 * Shrinking the buffer during idle times is not necessary as
1721 * it doesn't consume any memory when idle.
1723 * TODO: Only step up if the application is actually serving
1724 * the buffer to better manage the socket buffer resources.
1726 if (V_tcp_do_autorcvbuf &&
1728 (so->so_rcv.sb_flags & SB_AUTOSIZE)) {
1729 if (TSTMP_GT(to.to_tsecr, tp->rfbuf_ts) &&
1730 to.to_tsecr - tp->rfbuf_ts < hz) {
1732 (so->so_rcv.sb_hiwat / 8 * 7) &&
1733 so->so_rcv.sb_hiwat <
1734 V_tcp_autorcvbuf_max) {
1736 min(so->so_rcv.sb_hiwat +
1737 V_tcp_autorcvbuf_inc,
1738 V_tcp_autorcvbuf_max);
1740 /* Start over with next RTT. */
1744 tp->rfbuf_cnt += tlen; /* add up */
1747 /* Add data to socket buffer. */
1748 SOCKBUF_LOCK(&so->so_rcv);
1749 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1753 * Set new socket buffer size.
1754 * Give up when limit is reached.
1757 if (!sbreserve_locked(&so->so_rcv,
1759 so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1760 m_adj(m, drop_hdrlen); /* delayed header drop */
1761 sbappendstream_locked(&so->so_rcv, m);
1763 /* NB: sorwakeup_locked() does an implicit unlock. */
1764 sorwakeup_locked(so);
1765 if (DELAY_ACK(tp)) {
1766 tp->t_flags |= TF_DELACK;
1768 tp->t_flags |= TF_ACKNOW;
1776 * Calculate amount of space in receive window,
1777 * and then do TCP input processing.
1778 * Receive window is amount of space in rcv queue,
1779 * but not less than advertised window.
1781 win = sbspace(&so->so_rcv);
1784 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1786 /* Reset receive buffer auto scaling when not in bulk receive mode. */
1790 switch (tp->t_state) {
1793 * If the state is SYN_RECEIVED:
1794 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1796 case TCPS_SYN_RECEIVED:
1797 if ((thflags & TH_ACK) &&
1798 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1799 SEQ_GT(th->th_ack, tp->snd_max))) {
1800 rstreason = BANDLIM_RST_OPENPORT;
1806 * If the state is SYN_SENT:
1807 * if seg contains an ACK, but not for our SYN, drop the input.
1808 * if seg contains a RST, then drop the connection.
1809 * if seg does not contain SYN, then drop it.
1810 * Otherwise this is an acceptable SYN segment
1811 * initialize tp->rcv_nxt and tp->irs
1812 * if seg contains ack then advance tp->snd_una
1813 * if seg contains an ECE and ECN support is enabled, the stream
1815 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1816 * arrange for segment to be acked (eventually)
1817 * continue processing rest of data/controls, beginning with URG
1820 if ((thflags & TH_ACK) &&
1821 (SEQ_LEQ(th->th_ack, tp->iss) ||
1822 SEQ_GT(th->th_ack, tp->snd_max))) {
1823 rstreason = BANDLIM_UNLIMITED;
1826 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST))
1827 tp = tcp_drop(tp, ECONNREFUSED);
1828 if (thflags & TH_RST)
1830 if (!(thflags & TH_SYN))
1833 tp->irs = th->th_seq;
1835 if (thflags & TH_ACK) {
1836 TCPSTAT_INC(tcps_connects);
1839 mac_socketpeer_set_from_mbuf(m, so);
1841 /* Do window scaling on this connection? */
1842 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1843 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1844 tp->rcv_scale = tp->request_r_scale;
1846 tp->rcv_adv += imin(tp->rcv_wnd,
1847 TCP_MAXWIN << tp->rcv_scale);
1848 tp->snd_una++; /* SYN is acked */
1850 * If there's data, delay ACK; if there's also a FIN
1851 * ACKNOW will be turned on later.
1853 if (DELAY_ACK(tp) && tlen != 0)
1854 tcp_timer_activate(tp, TT_DELACK,
1857 tp->t_flags |= TF_ACKNOW;
1859 if ((thflags & TH_ECE) && V_tcp_do_ecn) {
1860 tp->t_flags |= TF_ECN_PERMIT;
1861 TCPSTAT_INC(tcps_ecn_shs);
1865 * Received <SYN,ACK> in SYN_SENT[*] state.
1867 * SYN_SENT --> ESTABLISHED
1868 * SYN_SENT* --> FIN_WAIT_1
1870 tp->t_starttime = ticks;
1871 if (tp->t_flags & TF_NEEDFIN) {
1872 tp->t_state = TCPS_FIN_WAIT_1;
1873 tp->t_flags &= ~TF_NEEDFIN;
1876 tp->t_state = TCPS_ESTABLISHED;
1878 tcp_timer_activate(tp, TT_KEEP, tcp_keepidle);
1882 * Received initial SYN in SYN-SENT[*] state =>
1883 * simultaneous open. If segment contains CC option
1884 * and there is a cached CC, apply TAO test.
1885 * If it succeeds, connection is * half-synchronized.
1886 * Otherwise, do 3-way handshake:
1887 * SYN-SENT -> SYN-RECEIVED
1888 * SYN-SENT* -> SYN-RECEIVED*
1889 * If there was no CC option, clear cached CC value.
1891 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
1892 tcp_timer_activate(tp, TT_REXMT, 0);
1893 tp->t_state = TCPS_SYN_RECEIVED;
1896 KASSERT(ti_locked == TI_WLOCKED, ("%s: trimthenstep6: "
1897 "ti_locked %d", __func__, ti_locked));
1898 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1899 INP_WLOCK_ASSERT(tp->t_inpcb);
1902 * Advance th->th_seq to correspond to first data byte.
1903 * If data, trim to stay within window,
1904 * dropping FIN if necessary.
1907 if (tlen > tp->rcv_wnd) {
1908 todrop = tlen - tp->rcv_wnd;
1912 TCPSTAT_INC(tcps_rcvpackafterwin);
1913 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
1915 tp->snd_wl1 = th->th_seq - 1;
1916 tp->rcv_up = th->th_seq;
1918 * Client side of transaction: already sent SYN and data.
1919 * If the remote host used T/TCP to validate the SYN,
1920 * our data will be ACK'd; if so, enter normal data segment
1921 * processing in the middle of step 5, ack processing.
1922 * Otherwise, goto step 6.
1924 if (thflags & TH_ACK)
1930 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
1931 * do normal processing.
1933 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later.
1937 break; /* continue normal processing */
1941 * States other than LISTEN or SYN_SENT.
1942 * First check the RST flag and sequence number since reset segments
1943 * are exempt from the timestamp and connection count tests. This
1944 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
1945 * below which allowed reset segments in half the sequence space
1946 * to fall though and be processed (which gives forged reset
1947 * segments with a random sequence number a 50 percent chance of
1948 * killing a connection).
1949 * Then check timestamp, if present.
1950 * Then check the connection count, if present.
1951 * Then check that at least some bytes of segment are within
1952 * receive window. If segment begins before rcv_nxt,
1953 * drop leading data (and SYN); if nothing left, just ack.
1956 * If the RST bit is set, check the sequence number to see
1957 * if this is a valid reset segment.
1959 * In all states except SYN-SENT, all reset (RST) segments
1960 * are validated by checking their SEQ-fields. A reset is
1961 * valid if its sequence number is in the window.
1962 * Note: this does not take into account delayed ACKs, so
1963 * we should test against last_ack_sent instead of rcv_nxt.
1964 * The sequence number in the reset segment is normally an
1965 * echo of our outgoing acknowlegement numbers, but some hosts
1966 * send a reset with the sequence number at the rightmost edge
1967 * of our receive window, and we have to handle this case.
1968 * Note 2: Paul Watson's paper "Slipping in the Window" has shown
1969 * that brute force RST attacks are possible. To combat this,
1970 * we use a much stricter check while in the ESTABLISHED state,
1971 * only accepting RSTs where the sequence number is equal to
1972 * last_ack_sent. In all other states (the states in which a
1973 * RST is more likely), the more permissive check is used.
1974 * If we have multiple segments in flight, the initial reset
1975 * segment sequence numbers will be to the left of last_ack_sent,
1976 * but they will eventually catch up.
1977 * In any case, it never made sense to trim reset segments to
1978 * fit the receive window since RFC 1122 says:
1979 * 4.2.2.12 RST Segment: RFC-793 Section 3.4
1981 * A TCP SHOULD allow a received RST segment to include data.
1984 * It has been suggested that a RST segment could contain
1985 * ASCII text that encoded and explained the cause of the
1986 * RST. No standard has yet been established for such
1989 * If the reset segment passes the sequence number test examine
1991 * SYN_RECEIVED STATE:
1992 * If passive open, return to LISTEN state.
1993 * If active open, inform user that connection was refused.
1994 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
1995 * Inform user that connection was reset, and close tcb.
1996 * CLOSING, LAST_ACK STATES:
1999 * Drop the segment - see Stevens, vol. 2, p. 964 and
2002 if (thflags & TH_RST) {
2003 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) &&
2004 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
2005 switch (tp->t_state) {
2007 case TCPS_SYN_RECEIVED:
2008 so->so_error = ECONNREFUSED;
2011 case TCPS_ESTABLISHED:
2012 if (V_tcp_insecure_rst == 0 &&
2013 !(SEQ_GEQ(th->th_seq, tp->rcv_nxt - 1) &&
2014 SEQ_LEQ(th->th_seq, tp->rcv_nxt + 1)) &&
2015 !(SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) &&
2016 SEQ_LEQ(th->th_seq, tp->last_ack_sent + 1))) {
2017 TCPSTAT_INC(tcps_badrst);
2021 case TCPS_FIN_WAIT_1:
2022 case TCPS_FIN_WAIT_2:
2023 case TCPS_CLOSE_WAIT:
2024 so->so_error = ECONNRESET;
2026 KASSERT(ti_locked == TI_WLOCKED,
2027 ("tcp_do_segment: TH_RST 1 ti_locked %d",
2029 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2031 tp->t_state = TCPS_CLOSED;
2032 TCPSTAT_INC(tcps_drops);
2038 KASSERT(ti_locked == TI_WLOCKED,
2039 ("tcp_do_segment: TH_RST 2 ti_locked %d",
2041 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2051 * RFC 1323 PAWS: If we have a timestamp reply on this segment
2052 * and it's less than ts_recent, drop it.
2054 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
2055 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
2057 /* Check to see if ts_recent is over 24 days old. */
2058 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) {
2060 * Invalidate ts_recent. If this segment updates
2061 * ts_recent, the age will be reset later and ts_recent
2062 * will get a valid value. If it does not, setting
2063 * ts_recent to zero will at least satisfy the
2064 * requirement that zero be placed in the timestamp
2065 * echo reply when ts_recent isn't valid. The
2066 * age isn't reset until we get a valid ts_recent
2067 * because we don't want out-of-order segments to be
2068 * dropped when ts_recent is old.
2072 TCPSTAT_INC(tcps_rcvduppack);
2073 TCPSTAT_ADD(tcps_rcvdupbyte, tlen);
2074 TCPSTAT_INC(tcps_pawsdrop);
2082 * In the SYN-RECEIVED state, validate that the packet belongs to
2083 * this connection before trimming the data to fit the receive
2084 * window. Check the sequence number versus IRS since we know
2085 * the sequence numbers haven't wrapped. This is a partial fix
2086 * for the "LAND" DoS attack.
2088 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
2089 rstreason = BANDLIM_RST_OPENPORT;
2093 todrop = tp->rcv_nxt - th->th_seq;
2095 if (thflags & TH_SYN) {
2105 * Following if statement from Stevens, vol. 2, p. 960.
2108 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
2110 * Any valid FIN must be to the left of the window.
2111 * At this point the FIN must be a duplicate or out
2112 * of sequence; drop it.
2117 * Send an ACK to resynchronize and drop any data.
2118 * But keep on processing for RST or ACK.
2120 tp->t_flags |= TF_ACKNOW;
2122 TCPSTAT_INC(tcps_rcvduppack);
2123 TCPSTAT_ADD(tcps_rcvdupbyte, todrop);
2125 TCPSTAT_INC(tcps_rcvpartduppack);
2126 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop);
2128 drop_hdrlen += todrop; /* drop from the top afterwards */
2129 th->th_seq += todrop;
2131 if (th->th_urp > todrop)
2132 th->th_urp -= todrop;
2140 * If new data are received on a connection after the
2141 * user processes are gone, then RST the other end.
2143 if ((so->so_state & SS_NOFDREF) &&
2144 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
2147 KASSERT(ti_locked == TI_WLOCKED, ("%s: SS_NOFDEREF && "
2148 "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked));
2149 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2151 if ((s = tcp_log_addrs(&tp->t_inpcb->inp_inc, th, NULL, NULL))) {
2152 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data after socket "
2153 "was closed, sending RST and removing tcpcb\n",
2154 s, __func__, tcpstates[tp->t_state], tlen);
2158 TCPSTAT_INC(tcps_rcvafterclose);
2159 rstreason = BANDLIM_UNLIMITED;
2164 * If segment ends after window, drop trailing data
2165 * (and PUSH and FIN); if nothing left, just ACK.
2167 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
2169 TCPSTAT_INC(tcps_rcvpackafterwin);
2170 if (todrop >= tlen) {
2171 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen);
2173 * If window is closed can only take segments at
2174 * window edge, and have to drop data and PUSH from
2175 * incoming segments. Continue processing, but
2176 * remember to ack. Otherwise, drop segment
2179 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2180 tp->t_flags |= TF_ACKNOW;
2181 TCPSTAT_INC(tcps_rcvwinprobe);
2185 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2188 thflags &= ~(TH_PUSH|TH_FIN);
2192 * If last ACK falls within this segment's sequence numbers,
2193 * record its timestamp.
2195 * 1) That the test incorporates suggestions from the latest
2196 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
2197 * 2) That updating only on newer timestamps interferes with
2198 * our earlier PAWS tests, so this check should be solely
2199 * predicated on the sequence space of this segment.
2200 * 3) That we modify the segment boundary check to be
2201 * Last.ACK.Sent <= SEG.SEQ + SEG.Len
2202 * instead of RFC1323's
2203 * Last.ACK.Sent < SEG.SEQ + SEG.Len,
2204 * This modified check allows us to overcome RFC1323's
2205 * limitations as described in Stevens TCP/IP Illustrated
2206 * Vol. 2 p.869. In such cases, we can still calculate the
2207 * RTT correctly when RCV.NXT == Last.ACK.Sent.
2209 if ((to.to_flags & TOF_TS) != 0 &&
2210 SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2211 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
2212 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
2213 tp->ts_recent_age = tcp_ts_getticks();
2214 tp->ts_recent = to.to_tsval;
2218 * If a SYN is in the window, then this is an
2219 * error and we send an RST and drop the connection.
2221 if (thflags & TH_SYN) {
2222 KASSERT(ti_locked == TI_WLOCKED,
2223 ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked));
2224 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2226 tp = tcp_drop(tp, ECONNRESET);
2227 rstreason = BANDLIM_UNLIMITED;
2232 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
2233 * flag is on (half-synchronized state), then queue data for
2234 * later processing; else drop segment and return.
2236 if ((thflags & TH_ACK) == 0) {
2237 if (tp->t_state == TCPS_SYN_RECEIVED ||
2238 (tp->t_flags & TF_NEEDSYN))
2240 else if (tp->t_flags & TF_ACKNOW)
2249 switch (tp->t_state) {
2252 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
2253 * ESTABLISHED state and continue processing.
2254 * The ACK was checked above.
2256 case TCPS_SYN_RECEIVED:
2258 TCPSTAT_INC(tcps_connects);
2260 /* Do window scaling? */
2261 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2262 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2263 tp->rcv_scale = tp->request_r_scale;
2264 tp->snd_wnd = tiwin;
2268 * SYN-RECEIVED -> ESTABLISHED
2269 * SYN-RECEIVED* -> FIN-WAIT-1
2271 tp->t_starttime = ticks;
2272 if (tp->t_flags & TF_NEEDFIN) {
2273 tp->t_state = TCPS_FIN_WAIT_1;
2274 tp->t_flags &= ~TF_NEEDFIN;
2276 tp->t_state = TCPS_ESTABLISHED;
2278 tcp_timer_activate(tp, TT_KEEP, tcp_keepidle);
2281 * If segment contains data or ACK, will call tcp_reass()
2282 * later; if not, do so now to pass queued data to user.
2284 if (tlen == 0 && (thflags & TH_FIN) == 0)
2285 (void) tcp_reass(tp, (struct tcphdr *)0, 0,
2287 tp->snd_wl1 = th->th_seq - 1;
2291 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2292 * ACKs. If the ack is in the range
2293 * tp->snd_una < th->th_ack <= tp->snd_max
2294 * then advance tp->snd_una to th->th_ack and drop
2295 * data from the retransmission queue. If this ACK reflects
2296 * more up to date window information we update our window information.
2298 case TCPS_ESTABLISHED:
2299 case TCPS_FIN_WAIT_1:
2300 case TCPS_FIN_WAIT_2:
2301 case TCPS_CLOSE_WAIT:
2304 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2305 TCPSTAT_INC(tcps_rcvacktoomuch);
2308 if ((tp->t_flags & TF_SACK_PERMIT) &&
2309 ((to.to_flags & TOF_SACK) ||
2310 !TAILQ_EMPTY(&tp->snd_holes)))
2311 tcp_sack_doack(tp, &to, th->th_ack);
2313 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
2314 hhook_run_tcp_est_in(tp, th, &to);
2316 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2317 if (tlen == 0 && tiwin == tp->snd_wnd) {
2318 TCPSTAT_INC(tcps_rcvdupack);
2320 * If we have outstanding data (other than
2321 * a window probe), this is a completely
2322 * duplicate ack (ie, window info didn't
2323 * change), the ack is the biggest we've
2324 * seen and we've seen exactly our rexmt
2325 * threshhold of them, assume a packet
2326 * has been dropped and retransmit it.
2327 * Kludge snd_nxt & the congestion
2328 * window so we send only this one
2331 * We know we're losing at the current
2332 * window size so do congestion avoidance
2333 * (set ssthresh to half the current window
2334 * and pull our congestion window back to
2335 * the new ssthresh).
2337 * Dup acks mean that packets have left the
2338 * network (they're now cached at the receiver)
2339 * so bump cwnd by the amount in the receiver
2340 * to keep a constant cwnd packets in the
2343 * When using TCP ECN, notify the peer that
2344 * we reduced the cwnd.
2346 if (!tcp_timer_active(tp, TT_REXMT) ||
2347 th->th_ack != tp->snd_una)
2349 else if (++tp->t_dupacks > tcprexmtthresh ||
2350 IN_FASTRECOVERY(tp->t_flags)) {
2351 cc_ack_received(tp, th, CC_DUPACK);
2352 if ((tp->t_flags & TF_SACK_PERMIT) &&
2353 IN_FASTRECOVERY(tp->t_flags)) {
2357 * Compute the amount of data in flight first.
2358 * We can inject new data into the pipe iff
2359 * we have less than 1/2 the original window's
2360 * worth of data in flight.
2362 awnd = (tp->snd_nxt - tp->snd_fack) +
2363 tp->sackhint.sack_bytes_rexmit;
2364 if (awnd < tp->snd_ssthresh) {
2365 tp->snd_cwnd += tp->t_maxseg;
2366 if (tp->snd_cwnd > tp->snd_ssthresh)
2367 tp->snd_cwnd = tp->snd_ssthresh;
2370 tp->snd_cwnd += tp->t_maxseg;
2371 if ((thflags & TH_FIN) &&
2372 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2374 * If its a fin we need to process
2375 * it to avoid a race where both
2376 * sides enter FIN-WAIT and send FIN|ACK
2381 (void) tcp_output(tp);
2383 } else if (tp->t_dupacks == tcprexmtthresh) {
2384 tcp_seq onxt = tp->snd_nxt;
2387 * If we're doing sack, check to
2388 * see if we're already in sack
2389 * recovery. If we're not doing sack,
2390 * check to see if we're in newreno
2393 if (tp->t_flags & TF_SACK_PERMIT) {
2394 if (IN_FASTRECOVERY(tp->t_flags)) {
2399 if (SEQ_LEQ(th->th_ack,
2405 /* Congestion signal before ack. */
2406 cc_cong_signal(tp, th, CC_NDUPACK);
2407 cc_ack_received(tp, th, CC_DUPACK);
2408 tcp_timer_activate(tp, TT_REXMT, 0);
2410 if (tp->t_flags & TF_SACK_PERMIT) {
2412 tcps_sack_recovery_episode);
2413 tp->sack_newdata = tp->snd_nxt;
2414 tp->snd_cwnd = tp->t_maxseg;
2415 (void) tcp_output(tp);
2418 tp->snd_nxt = th->th_ack;
2419 tp->snd_cwnd = tp->t_maxseg;
2420 if ((thflags & TH_FIN) &&
2421 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2423 * If its a fin we need to process
2424 * it to avoid a race where both
2425 * sides enter FIN-WAIT and send FIN|ACK
2430 (void) tcp_output(tp);
2431 KASSERT(tp->snd_limited <= 2,
2432 ("%s: tp->snd_limited too big",
2434 tp->snd_cwnd = tp->snd_ssthresh +
2436 (tp->t_dupacks - tp->snd_limited);
2437 if (SEQ_GT(onxt, tp->snd_nxt))
2440 } else if (V_tcp_do_rfc3042) {
2441 cc_ack_received(tp, th, CC_DUPACK);
2442 u_long oldcwnd = tp->snd_cwnd;
2443 tcp_seq oldsndmax = tp->snd_max;
2446 KASSERT(tp->t_dupacks == 1 ||
2448 ("%s: dupacks not 1 or 2",
2450 if (tp->t_dupacks == 1)
2451 tp->snd_limited = 0;
2453 (tp->snd_nxt - tp->snd_una) +
2454 (tp->t_dupacks - tp->snd_limited) *
2456 if ((thflags & TH_FIN) &&
2457 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2459 * If its a fin we need to process
2460 * it to avoid a race where both
2461 * sides enter FIN-WAIT and send FIN|ACK
2466 (void) tcp_output(tp);
2467 sent = tp->snd_max - oldsndmax;
2468 if (sent > tp->t_maxseg) {
2469 KASSERT((tp->t_dupacks == 2 &&
2470 tp->snd_limited == 0) ||
2471 (sent == tp->t_maxseg + 1 &&
2472 tp->t_flags & TF_SENTFIN),
2473 ("%s: sent too much",
2475 tp->snd_limited = 2;
2476 } else if (sent > 0)
2478 tp->snd_cwnd = oldcwnd;
2486 KASSERT(SEQ_GT(th->th_ack, tp->snd_una),
2487 ("%s: th_ack <= snd_una", __func__));
2490 * If the congestion window was inflated to account
2491 * for the other side's cached packets, retract it.
2493 if (IN_FASTRECOVERY(tp->t_flags)) {
2494 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2495 if (tp->t_flags & TF_SACK_PERMIT)
2496 tcp_sack_partialack(tp, th);
2498 tcp_newreno_partial_ack(tp, th);
2500 cc_post_recovery(tp, th);
2504 * If we reach this point, ACK is not a duplicate,
2505 * i.e., it ACKs something we sent.
2507 if (tp->t_flags & TF_NEEDSYN) {
2509 * T/TCP: Connection was half-synchronized, and our
2510 * SYN has been ACK'd (so connection is now fully
2511 * synchronized). Go to non-starred state,
2512 * increment snd_una for ACK of SYN, and check if
2513 * we can do window scaling.
2515 tp->t_flags &= ~TF_NEEDSYN;
2517 /* Do window scaling? */
2518 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2519 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2520 tp->rcv_scale = tp->request_r_scale;
2521 /* Send window already scaled. */
2526 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
2527 KASSERT(ti_locked == TI_RLOCKED || ti_locked == TI_WLOCKED,
2528 ("tcp_input: process_ACK ti_locked %d", ti_locked));
2529 INP_WLOCK_ASSERT(tp->t_inpcb);
2531 acked = BYTES_THIS_ACK(tp, th);
2532 TCPSTAT_INC(tcps_rcvackpack);
2533 TCPSTAT_ADD(tcps_rcvackbyte, acked);
2536 * If we just performed our first retransmit, and the ACK
2537 * arrives within our recovery window, then it was a mistake
2538 * to do the retransmit in the first place. Recover our
2539 * original cwnd and ssthresh, and proceed to transmit where
2542 if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID &&
2543 (int)(ticks - tp->t_badrxtwin) < 0)
2544 cc_cong_signal(tp, th, CC_RTO_ERR);
2547 * If we have a timestamp reply, update smoothed
2548 * round trip time. If no timestamp is present but
2549 * transmit timer is running and timed sequence
2550 * number was acked, update smoothed round trip time.
2551 * Since we now have an rtt measurement, cancel the
2552 * timer backoff (cf., Phil Karn's retransmit alg.).
2553 * Recompute the initial retransmit timer.
2555 * Some boxes send broken timestamp replies
2556 * during the SYN+ACK phase, ignore
2557 * timestamps of 0 or we could calculate a
2558 * huge RTT and blow up the retransmit timer.
2560 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) {
2563 t = tcp_ts_getticks() - to.to_tsecr;
2564 if (!tp->t_rttlow || tp->t_rttlow > t)
2566 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1);
2567 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2568 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2569 tp->t_rttlow = ticks - tp->t_rtttime;
2570 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2572 tcp_xmit_bandwidth_limit(tp, th->th_ack);
2575 * If all outstanding data is acked, stop retransmit
2576 * timer and remember to restart (more output or persist).
2577 * If there is more data to be acked, restart retransmit
2578 * timer, using current (possibly backed-off) value.
2580 if (th->th_ack == tp->snd_max) {
2581 tcp_timer_activate(tp, TT_REXMT, 0);
2583 } else if (!tcp_timer_active(tp, TT_PERSIST))
2584 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
2587 * If no data (only SYN) was ACK'd,
2588 * skip rest of ACK processing.
2594 * Let the congestion control algorithm update congestion
2595 * control related information. This typically means increasing
2596 * the congestion window.
2598 cc_ack_received(tp, th, CC_ACK);
2600 SOCKBUF_LOCK(&so->so_snd);
2601 if (acked > so->so_snd.sb_cc) {
2602 tp->snd_wnd -= so->so_snd.sb_cc;
2603 sbdrop_locked(&so->so_snd, (int)so->so_snd.sb_cc);
2606 sbdrop_locked(&so->so_snd, acked);
2607 tp->snd_wnd -= acked;
2610 /* NB: sowwakeup_locked() does an implicit unlock. */
2611 sowwakeup_locked(so);
2612 /* Detect una wraparound. */
2613 if (!IN_RECOVERY(tp->t_flags) &&
2614 SEQ_GT(tp->snd_una, tp->snd_recover) &&
2615 SEQ_LEQ(th->th_ack, tp->snd_recover))
2616 tp->snd_recover = th->th_ack - 1;
2617 /* XXXLAS: Can this be moved up into cc_post_recovery? */
2618 if (IN_RECOVERY(tp->t_flags) &&
2619 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2620 EXIT_RECOVERY(tp->t_flags);
2622 tp->snd_una = th->th_ack;
2623 if (tp->t_flags & TF_SACK_PERMIT) {
2624 if (SEQ_GT(tp->snd_una, tp->snd_recover))
2625 tp->snd_recover = tp->snd_una;
2627 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2628 tp->snd_nxt = tp->snd_una;
2630 switch (tp->t_state) {
2633 * In FIN_WAIT_1 STATE in addition to the processing
2634 * for the ESTABLISHED state if our FIN is now acknowledged
2635 * then enter FIN_WAIT_2.
2637 case TCPS_FIN_WAIT_1:
2638 if (ourfinisacked) {
2640 * If we can't receive any more
2641 * data, then closing user can proceed.
2642 * Starting the timer is contrary to the
2643 * specification, but if we don't get a FIN
2644 * we'll hang forever.
2647 * we should release the tp also, and use a
2650 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2653 soisdisconnected(so);
2654 timeout = (tcp_fast_finwait2_recycle) ?
2655 tcp_finwait2_timeout : tcp_maxidle;
2656 tcp_timer_activate(tp, TT_2MSL, timeout);
2658 tp->t_state = TCPS_FIN_WAIT_2;
2663 * In CLOSING STATE in addition to the processing for
2664 * the ESTABLISHED state if the ACK acknowledges our FIN
2665 * then enter the TIME-WAIT state, otherwise ignore
2669 if (ourfinisacked) {
2670 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2672 INP_INFO_WUNLOCK(&V_tcbinfo);
2679 * In LAST_ACK, we may still be waiting for data to drain
2680 * and/or to be acked, as well as for the ack of our FIN.
2681 * If our FIN is now acknowledged, delete the TCB,
2682 * enter the closed state and return.
2685 if (ourfinisacked) {
2686 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2695 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
2696 KASSERT(ti_locked == TI_RLOCKED || ti_locked == TI_WLOCKED,
2697 ("tcp_do_segment: step6 ti_locked %d", ti_locked));
2698 INP_WLOCK_ASSERT(tp->t_inpcb);
2701 * Update window information.
2702 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2704 if ((thflags & TH_ACK) &&
2705 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2706 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2707 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2708 /* keep track of pure window updates */
2710 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2711 TCPSTAT_INC(tcps_rcvwinupd);
2712 tp->snd_wnd = tiwin;
2713 tp->snd_wl1 = th->th_seq;
2714 tp->snd_wl2 = th->th_ack;
2715 if (tp->snd_wnd > tp->max_sndwnd)
2716 tp->max_sndwnd = tp->snd_wnd;
2721 * Process segments with URG.
2723 if ((thflags & TH_URG) && th->th_urp &&
2724 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2726 * This is a kludge, but if we receive and accept
2727 * random urgent pointers, we'll crash in
2728 * soreceive. It's hard to imagine someone
2729 * actually wanting to send this much urgent data.
2731 SOCKBUF_LOCK(&so->so_rcv);
2732 if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2733 th->th_urp = 0; /* XXX */
2734 thflags &= ~TH_URG; /* XXX */
2735 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */
2736 goto dodata; /* XXX */
2739 * If this segment advances the known urgent pointer,
2740 * then mark the data stream. This should not happen
2741 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2742 * a FIN has been received from the remote side.
2743 * In these states we ignore the URG.
2745 * According to RFC961 (Assigned Protocols),
2746 * the urgent pointer points to the last octet
2747 * of urgent data. We continue, however,
2748 * to consider it to indicate the first octet
2749 * of data past the urgent section as the original
2750 * spec states (in one of two places).
2752 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2753 tp->rcv_up = th->th_seq + th->th_urp;
2754 so->so_oobmark = so->so_rcv.sb_cc +
2755 (tp->rcv_up - tp->rcv_nxt) - 1;
2756 if (so->so_oobmark == 0)
2757 so->so_rcv.sb_state |= SBS_RCVATMARK;
2759 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2761 SOCKBUF_UNLOCK(&so->so_rcv);
2763 * Remove out of band data so doesn't get presented to user.
2764 * This can happen independent of advancing the URG pointer,
2765 * but if two URG's are pending at once, some out-of-band
2766 * data may creep in... ick.
2768 if (th->th_urp <= (u_long)tlen &&
2769 !(so->so_options & SO_OOBINLINE)) {
2770 /* hdr drop is delayed */
2771 tcp_pulloutofband(so, th, m, drop_hdrlen);
2775 * If no out of band data is expected,
2776 * pull receive urgent pointer along
2777 * with the receive window.
2779 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2780 tp->rcv_up = tp->rcv_nxt;
2783 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
2784 KASSERT(ti_locked == TI_RLOCKED || ti_locked == TI_WLOCKED,
2785 ("tcp_do_segment: dodata ti_locked %d", ti_locked));
2786 INP_WLOCK_ASSERT(tp->t_inpcb);
2789 * Process the segment text, merging it into the TCP sequencing queue,
2790 * and arranging for acknowledgment of receipt if necessary.
2791 * This process logically involves adjusting tp->rcv_wnd as data
2792 * is presented to the user (this happens in tcp_usrreq.c,
2793 * case PRU_RCVD). If a FIN has already been received on this
2794 * connection then we just ignore the text.
2796 if ((tlen || (thflags & TH_FIN)) &&
2797 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2798 tcp_seq save_start = th->th_seq;
2799 m_adj(m, drop_hdrlen); /* delayed header drop */
2801 * Insert segment which includes th into TCP reassembly queue
2802 * with control block tp. Set thflags to whether reassembly now
2803 * includes a segment with FIN. This handles the common case
2804 * inline (segment is the next to be received on an established
2805 * connection, and the queue is empty), avoiding linkage into
2806 * and removal from the queue and repetition of various
2808 * Set DELACK for segments received in order, but ack
2809 * immediately when segments are out of order (so
2810 * fast retransmit can work).
2812 if (th->th_seq == tp->rcv_nxt &&
2813 LIST_EMPTY(&tp->t_segq) &&
2814 TCPS_HAVEESTABLISHED(tp->t_state)) {
2816 tp->t_flags |= TF_DELACK;
2818 tp->t_flags |= TF_ACKNOW;
2819 tp->rcv_nxt += tlen;
2820 thflags = th->th_flags & TH_FIN;
2821 TCPSTAT_INC(tcps_rcvpack);
2822 TCPSTAT_ADD(tcps_rcvbyte, tlen);
2824 SOCKBUF_LOCK(&so->so_rcv);
2825 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
2828 sbappendstream_locked(&so->so_rcv, m);
2829 /* NB: sorwakeup_locked() does an implicit unlock. */
2830 sorwakeup_locked(so);
2833 * XXX: Due to the header drop above "th" is
2834 * theoretically invalid by now. Fortunately
2835 * m_adj() doesn't actually frees any mbufs
2836 * when trimming from the head.
2838 thflags = tcp_reass(tp, th, &tlen, m);
2839 tp->t_flags |= TF_ACKNOW;
2841 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT))
2842 tcp_update_sack_list(tp, save_start, save_start + tlen);
2845 * Note the amount of data that peer has sent into
2846 * our window, in order to estimate the sender's
2850 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
2851 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2853 len = so->so_rcv.sb_hiwat;
2861 * If FIN is received ACK the FIN and let the user know
2862 * that the connection is closing.
2864 if (thflags & TH_FIN) {
2865 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2868 * If connection is half-synchronized
2869 * (ie NEEDSYN flag on) then delay ACK,
2870 * so it may be piggybacked when SYN is sent.
2871 * Otherwise, since we received a FIN then no
2872 * more input can be expected, send ACK now.
2874 if (tp->t_flags & TF_NEEDSYN)
2875 tp->t_flags |= TF_DELACK;
2877 tp->t_flags |= TF_ACKNOW;
2880 switch (tp->t_state) {
2883 * In SYN_RECEIVED and ESTABLISHED STATES
2884 * enter the CLOSE_WAIT state.
2886 case TCPS_SYN_RECEIVED:
2887 tp->t_starttime = ticks;
2889 case TCPS_ESTABLISHED:
2890 tp->t_state = TCPS_CLOSE_WAIT;
2894 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2895 * enter the CLOSING state.
2897 case TCPS_FIN_WAIT_1:
2898 tp->t_state = TCPS_CLOSING;
2902 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2903 * starting the time-wait timer, turning off the other
2906 case TCPS_FIN_WAIT_2:
2907 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2908 KASSERT(ti_locked == TI_WLOCKED, ("%s: dodata "
2909 "TCP_FIN_WAIT_2 ti_locked: %d", __func__,
2913 INP_INFO_WUNLOCK(&V_tcbinfo);
2917 if (ti_locked == TI_RLOCKED)
2918 INP_INFO_RUNLOCK(&V_tcbinfo);
2919 else if (ti_locked == TI_WLOCKED)
2920 INP_INFO_WUNLOCK(&V_tcbinfo);
2922 panic("%s: dodata epilogue ti_locked %d", __func__,
2924 ti_locked = TI_UNLOCKED;
2927 if (so->so_options & SO_DEBUG)
2928 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
2933 * Return any desired output.
2935 if (needoutput || (tp->t_flags & TF_ACKNOW))
2936 (void) tcp_output(tp);
2939 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d",
2940 __func__, ti_locked));
2941 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
2942 INP_WLOCK_ASSERT(tp->t_inpcb);
2944 if (tp->t_flags & TF_DELACK) {
2945 tp->t_flags &= ~TF_DELACK;
2946 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
2948 INP_WUNLOCK(tp->t_inpcb);
2952 KASSERT(ti_locked == TI_RLOCKED || ti_locked == TI_WLOCKED,
2953 ("tcp_do_segment: dropafterack ti_locked %d", ti_locked));
2956 * Generate an ACK dropping incoming segment if it occupies
2957 * sequence space, where the ACK reflects our state.
2959 * We can now skip the test for the RST flag since all
2960 * paths to this code happen after packets containing
2961 * RST have been dropped.
2963 * In the SYN-RECEIVED state, don't send an ACK unless the
2964 * segment we received passes the SYN-RECEIVED ACK test.
2965 * If it fails send a RST. This breaks the loop in the
2966 * "LAND" DoS attack, and also prevents an ACK storm
2967 * between two listening ports that have been sent forged
2968 * SYN segments, each with the source address of the other.
2970 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
2971 (SEQ_GT(tp->snd_una, th->th_ack) ||
2972 SEQ_GT(th->th_ack, tp->snd_max)) ) {
2973 rstreason = BANDLIM_RST_OPENPORT;
2977 if (so->so_options & SO_DEBUG)
2978 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2981 if (ti_locked == TI_RLOCKED)
2982 INP_INFO_RUNLOCK(&V_tcbinfo);
2983 else if (ti_locked == TI_WLOCKED)
2984 INP_INFO_WUNLOCK(&V_tcbinfo);
2986 panic("%s: dropafterack epilogue ti_locked %d", __func__,
2988 ti_locked = TI_UNLOCKED;
2990 tp->t_flags |= TF_ACKNOW;
2991 (void) tcp_output(tp);
2992 INP_WUNLOCK(tp->t_inpcb);
2997 if (ti_locked == TI_RLOCKED)
2998 INP_INFO_RUNLOCK(&V_tcbinfo);
2999 else if (ti_locked == TI_WLOCKED)
3000 INP_INFO_WUNLOCK(&V_tcbinfo);
3002 panic("%s: dropwithreset ti_locked %d", __func__, ti_locked);
3003 ti_locked = TI_UNLOCKED;
3006 tcp_dropwithreset(m, th, tp, tlen, rstreason);
3007 INP_WUNLOCK(tp->t_inpcb);
3009 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
3013 if (ti_locked == TI_RLOCKED)
3014 INP_INFO_RUNLOCK(&V_tcbinfo);
3015 else if (ti_locked == TI_WLOCKED)
3016 INP_INFO_WUNLOCK(&V_tcbinfo);
3019 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
3021 ti_locked = TI_UNLOCKED;
3024 * Drop space held by incoming segment and return.
3027 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
3028 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3032 INP_WUNLOCK(tp->t_inpcb);
3037 * Issue RST and make ACK acceptable to originator of segment.
3038 * The mbuf must still include the original packet header.
3042 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
3043 int tlen, int rstreason)
3047 struct ip6_hdr *ip6;
3051 INP_WLOCK_ASSERT(tp->t_inpcb);
3054 /* Don't bother if destination was broadcast/multicast. */
3055 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
3058 if (mtod(m, struct ip *)->ip_v == 6) {
3059 ip6 = mtod(m, struct ip6_hdr *);
3060 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3061 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3063 /* IPv6 anycast check is done at tcp6_input() */
3067 ip = mtod(m, struct ip *);
3068 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3069 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3070 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3071 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3075 /* Perform bandwidth limiting. */
3076 if (badport_bandlim(rstreason) < 0)
3079 /* tcp_respond consumes the mbuf chain. */
3080 if (th->th_flags & TH_ACK) {
3081 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0,
3082 th->th_ack, TH_RST);
3084 if (th->th_flags & TH_SYN)
3086 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
3087 (tcp_seq)0, TH_RST|TH_ACK);
3095 * Parse TCP options and place in tcpopt.
3098 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags)
3103 for (; cnt > 0; cnt -= optlen, cp += optlen) {
3105 if (opt == TCPOPT_EOL)
3107 if (opt == TCPOPT_NOP)
3113 if (optlen < 2 || optlen > cnt)
3118 if (optlen != TCPOLEN_MAXSEG)
3120 if (!(flags & TO_SYN))
3122 to->to_flags |= TOF_MSS;
3123 bcopy((char *)cp + 2,
3124 (char *)&to->to_mss, sizeof(to->to_mss));
3125 to->to_mss = ntohs(to->to_mss);
3128 if (optlen != TCPOLEN_WINDOW)
3130 if (!(flags & TO_SYN))
3132 to->to_flags |= TOF_SCALE;
3133 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT);
3135 case TCPOPT_TIMESTAMP:
3136 if (optlen != TCPOLEN_TIMESTAMP)
3138 to->to_flags |= TOF_TS;
3139 bcopy((char *)cp + 2,
3140 (char *)&to->to_tsval, sizeof(to->to_tsval));
3141 to->to_tsval = ntohl(to->to_tsval);
3142 bcopy((char *)cp + 6,
3143 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
3144 to->to_tsecr = ntohl(to->to_tsecr);
3146 #ifdef TCP_SIGNATURE
3148 * XXX In order to reply to a host which has set the
3149 * TCP_SIGNATURE option in its initial SYN, we have to
3150 * record the fact that the option was observed here
3151 * for the syncache code to perform the correct response.
3153 case TCPOPT_SIGNATURE:
3154 if (optlen != TCPOLEN_SIGNATURE)
3156 to->to_flags |= TOF_SIGNATURE;
3157 to->to_signature = cp + 2;
3160 case TCPOPT_SACK_PERMITTED:
3161 if (optlen != TCPOLEN_SACK_PERMITTED)
3163 if (!(flags & TO_SYN))
3167 to->to_flags |= TOF_SACKPERM;
3170 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
3174 to->to_flags |= TOF_SACK;
3175 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
3176 to->to_sacks = cp + 2;
3177 TCPSTAT_INC(tcps_sack_rcv_blocks);
3186 * Pull out of band byte out of a segment so
3187 * it doesn't appear in the user's data queue.
3188 * It is still reflected in the segment length for
3189 * sequencing purposes.
3192 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m,
3195 int cnt = off + th->th_urp - 1;
3198 if (m->m_len > cnt) {
3199 char *cp = mtod(m, caddr_t) + cnt;
3200 struct tcpcb *tp = sototcpcb(so);
3202 INP_WLOCK_ASSERT(tp->t_inpcb);
3205 tp->t_oobflags |= TCPOOB_HAVEDATA;
3206 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3208 if (m->m_flags & M_PKTHDR)
3217 panic("tcp_pulloutofband");
3221 * Collect new round-trip time estimate
3222 * and update averages and current timeout.
3225 tcp_xmit_timer(struct tcpcb *tp, int rtt)
3229 INP_WLOCK_ASSERT(tp->t_inpcb);
3231 TCPSTAT_INC(tcps_rttupdated);
3233 if (tp->t_srtt != 0) {
3235 * srtt is stored as fixed point with 5 bits after the
3236 * binary point (i.e., scaled by 8). The following magic
3237 * is equivalent to the smoothing algorithm in rfc793 with
3238 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3239 * point). Adjust rtt to origin 0.
3241 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
3242 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
3244 if ((tp->t_srtt += delta) <= 0)
3248 * We accumulate a smoothed rtt variance (actually, a
3249 * smoothed mean difference), then set the retransmit
3250 * timer to smoothed rtt + 4 times the smoothed variance.
3251 * rttvar is stored as fixed point with 4 bits after the
3252 * binary point (scaled by 16). The following is
3253 * equivalent to rfc793 smoothing with an alpha of .75
3254 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
3255 * rfc793's wired-in beta.
3259 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
3260 if ((tp->t_rttvar += delta) <= 0)
3262 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
3263 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3266 * No rtt measurement yet - use the unsmoothed rtt.
3267 * Set the variance to half the rtt (so our first
3268 * retransmit happens at 3*rtt).
3270 tp->t_srtt = rtt << TCP_RTT_SHIFT;
3271 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
3272 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3278 * the retransmit should happen at rtt + 4 * rttvar.
3279 * Because of the way we do the smoothing, srtt and rttvar
3280 * will each average +1/2 tick of bias. When we compute
3281 * the retransmit timer, we want 1/2 tick of rounding and
3282 * 1 extra tick because of +-1/2 tick uncertainty in the
3283 * firing of the timer. The bias will give us exactly the
3284 * 1.5 tick we need. But, because the bias is
3285 * statistical, we have to test that we don't drop below
3286 * the minimum feasible timer (which is 2 ticks).
3288 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3289 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3292 * We received an ack for a packet that wasn't retransmitted;
3293 * it is probably safe to discard any error indications we've
3294 * received recently. This isn't quite right, but close enough
3295 * for now (a route might have failed after we sent a segment,
3296 * and the return path might not be symmetrical).
3298 tp->t_softerror = 0;
3302 * Determine a reasonable value for maxseg size.
3303 * If the route is known, check route for mtu.
3304 * If none, use an mss that can be handled on the outgoing
3305 * interface without forcing IP to fragment; if bigger than
3306 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
3307 * to utilize large mbufs. If no route is found, route has no mtu,
3308 * or the destination isn't local, use a default, hopefully conservative
3309 * size (usually 512 or the default IP max size, but no more than the mtu
3310 * of the interface), as we can't discover anything about intervening
3311 * gateways or networks. We also initialize the congestion/slow start
3312 * window to be a single segment if the destination isn't local.
3313 * While looking at the routing entry, we also initialize other path-dependent
3314 * parameters from pre-set or cached values in the routing entry.
3316 * Also take into account the space needed for options that we
3317 * send regularly. Make maxseg shorter by that amount to assure
3318 * that we can send maxseg amount of data even when the options
3319 * are present. Store the upper limit of the length of options plus
3322 * NOTE that this routine is only called when we process an incoming
3323 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS
3324 * settings are handled in tcp_mssopt().
3327 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer,
3328 struct hc_metrics_lite *metricptr, int *mtuflags)
3332 struct inpcb *inp = tp->t_inpcb;
3333 struct hc_metrics_lite metrics;
3336 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
3337 size_t min_protoh = isipv6 ?
3338 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
3339 sizeof (struct tcpiphdr);
3341 const size_t min_protoh = sizeof(struct tcpiphdr);
3344 INP_WLOCK_ASSERT(tp->t_inpcb);
3346 if (mtuoffer != -1) {
3347 KASSERT(offer == -1, ("%s: conflict", __func__));
3348 offer = mtuoffer - min_protoh;
3355 maxmtu = tcp_maxmtu6(&inp->inp_inc, mtuflags);
3356 tp->t_maxopd = tp->t_maxseg = V_tcp_v6mssdflt;
3360 maxmtu = tcp_maxmtu(&inp->inp_inc, mtuflags);
3361 tp->t_maxopd = tp->t_maxseg = V_tcp_mssdflt;
3365 * No route to sender, stay with default mss and return.
3369 * In case we return early we need to initialize metrics
3370 * to a defined state as tcp_hc_get() would do for us
3371 * if there was no cache hit.
3373 if (metricptr != NULL)
3374 bzero(metricptr, sizeof(struct hc_metrics_lite));
3378 /* What have we got? */
3382 * Offer == 0 means that there was no MSS on the SYN
3383 * segment, in this case we use tcp_mssdflt as
3384 * already assigned to t_maxopd above.
3386 offer = tp->t_maxopd;
3391 * Offer == -1 means that we didn't receive SYN yet.
3397 * Prevent DoS attack with too small MSS. Round up
3398 * to at least minmss.
3400 offer = max(offer, V_tcp_minmss);
3404 * rmx information is now retrieved from tcp_hostcache.
3406 tcp_hc_get(&inp->inp_inc, &metrics);
3407 if (metricptr != NULL)
3408 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite));
3411 * If there's a discovered mtu int tcp hostcache, use it
3412 * else, use the link mtu.
3414 if (metrics.rmx_mtu)
3415 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
3419 mss = maxmtu - min_protoh;
3420 if (!V_path_mtu_discovery &&
3421 !in6_localaddr(&inp->in6p_faddr))
3422 mss = min(mss, V_tcp_v6mssdflt);
3426 mss = maxmtu - min_protoh;
3427 if (!V_path_mtu_discovery &&
3428 !in_localaddr(inp->inp_faddr))
3429 mss = min(mss, V_tcp_mssdflt);
3432 * XXX - The above conditional (mss = maxmtu - min_protoh)
3433 * probably violates the TCP spec.
3434 * The problem is that, since we don't know the
3435 * other end's MSS, we are supposed to use a conservative
3436 * default. But, if we do that, then MTU discovery will
3437 * never actually take place, because the conservative
3438 * default is much less than the MTUs typically seen
3439 * on the Internet today. For the moment, we'll sweep
3440 * this under the carpet.
3442 * The conservative default might not actually be a problem
3443 * if the only case this occurs is when sending an initial
3444 * SYN with options and data to a host we've never talked
3445 * to before. Then, they will reply with an MSS value which
3446 * will get recorded and the new parameters should get
3447 * recomputed. For Further Study.
3450 mss = min(mss, offer);
3453 * Sanity check: make sure that maxopd will be large
3454 * enough to allow some data on segments even if the
3455 * all the option space is used (40bytes). Otherwise
3456 * funny things may happen in tcp_output.
3461 * maxopd stores the maximum length of data AND options
3462 * in a segment; maxseg is the amount of data in a normal
3463 * segment. We need to store this value (maxopd) apart
3464 * from maxseg, because now every segment carries options
3465 * and thus we normally have somewhat less data in segments.
3470 * origoffer==-1 indicates that no segments were received yet.
3471 * In this case we just guess.
3473 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
3475 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
3476 mss -= TCPOLEN_TSTAMP_APPA;
3478 #if (MCLBYTES & (MCLBYTES - 1)) == 0
3480 mss &= ~(MCLBYTES-1);
3483 mss = mss / MCLBYTES * MCLBYTES;
3489 tcp_mss(struct tcpcb *tp, int offer)
3495 struct hc_metrics_lite metrics;
3498 KASSERT(tp != NULL, ("%s: tp == NULL", __func__));
3500 tcp_mss_update(tp, offer, -1, &metrics, &mtuflags);
3506 * If there's a pipesize, change the socket buffer to that size,
3507 * don't change if sb_hiwat is different than default (then it
3508 * has been changed on purpose with setsockopt).
3509 * Make the socket buffers an integral number of mss units;
3510 * if the mss is larger than the socket buffer, decrease the mss.
3512 so = inp->inp_socket;
3513 SOCKBUF_LOCK(&so->so_snd);
3514 if ((so->so_snd.sb_hiwat == tcp_sendspace) && metrics.rmx_sendpipe)
3515 bufsize = metrics.rmx_sendpipe;
3517 bufsize = so->so_snd.sb_hiwat;
3521 bufsize = roundup(bufsize, mss);
3522 if (bufsize > sb_max)
3524 if (bufsize > so->so_snd.sb_hiwat)
3525 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
3527 SOCKBUF_UNLOCK(&so->so_snd);
3530 SOCKBUF_LOCK(&so->so_rcv);
3531 if ((so->so_rcv.sb_hiwat == tcp_recvspace) && metrics.rmx_recvpipe)
3532 bufsize = metrics.rmx_recvpipe;
3534 bufsize = so->so_rcv.sb_hiwat;
3535 if (bufsize > mss) {
3536 bufsize = roundup(bufsize, mss);
3537 if (bufsize > sb_max)
3539 if (bufsize > so->so_rcv.sb_hiwat)
3540 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
3542 SOCKBUF_UNLOCK(&so->so_rcv);
3544 /* Check the interface for TSO capabilities. */
3545 if (mtuflags & CSUM_TSO)
3546 tp->t_flags |= TF_TSO;
3550 * Determine the MSS option to send on an outgoing SYN.
3553 tcp_mssopt(struct in_conninfo *inc)
3560 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3563 if (inc->inc_flags & INC_ISIPV6) {
3564 mss = V_tcp_v6mssdflt;
3565 maxmtu = tcp_maxmtu6(inc, NULL);
3566 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3567 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3571 mss = V_tcp_mssdflt;
3572 maxmtu = tcp_maxmtu(inc, NULL);
3573 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3574 min_protoh = sizeof(struct tcpiphdr);
3576 if (maxmtu && thcmtu)
3577 mss = min(maxmtu, thcmtu) - min_protoh;
3578 else if (maxmtu || thcmtu)
3579 mss = max(maxmtu, thcmtu) - min_protoh;
3586 * On a partial ack arrives, force the retransmission of the
3587 * next unacknowledged segment. Do not clear tp->t_dupacks.
3588 * By setting snd_nxt to ti_ack, this forces retransmission timer to
3592 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th)
3594 tcp_seq onxt = tp->snd_nxt;
3595 u_long ocwnd = tp->snd_cwnd;
3597 INP_WLOCK_ASSERT(tp->t_inpcb);
3599 tcp_timer_activate(tp, TT_REXMT, 0);
3601 tp->snd_nxt = th->th_ack;
3603 * Set snd_cwnd to one segment beyond acknowledged offset.
3604 * (tp->snd_una has not yet been updated when this function is called.)
3606 tp->snd_cwnd = tp->t_maxseg + BYTES_THIS_ACK(tp, th);
3607 tp->t_flags |= TF_ACKNOW;
3608 (void) tcp_output(tp);
3609 tp->snd_cwnd = ocwnd;
3610 if (SEQ_GT(onxt, tp->snd_nxt))
3613 * Partial window deflation. Relies on fact that tp->snd_una
3616 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th))
3617 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th);
3620 tp->snd_cwnd += tp->t_maxseg;