2 * Copyright (c) 2002, 2003, 2004 Jeffrey M. Hsu. All rights reserved.
3 * Copyright (c) 2002, 2003, 2004 The DragonFly Project. All rights reserved.
5 * This code is derived from software contributed to The DragonFly Project
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. Neither the name of The DragonFly Project nor the names of its
17 * contributors may be used to endorse or promote products derived
18 * from this software without specific, prior written permission.
20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
23 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
24 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
25 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
26 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
27 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
28 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
29 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
30 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
36 * The Regents of the University of California. All rights reserved.
38 * Redistribution and use in source and binary forms, with or without
39 * modification, are permitted provided that the following conditions
41 * 1. Redistributions of source code must retain the above copyright
42 * notice, this list of conditions and the following disclaimer.
43 * 2. Redistributions in binary form must reproduce the above copyright
44 * notice, this list of conditions and the following disclaimer in the
45 * documentation and/or other materials provided with the distribution.
46 * 3. All advertising materials mentioning features or use of this software
47 * must display the following acknowledgement:
48 * This product includes software developed by the University of
49 * California, Berkeley and its contributors.
50 * 4. Neither the name of the University nor the names of its contributors
51 * may be used to endorse or promote products derived from this software
52 * without specific prior written permission.
54 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
55 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
56 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
57 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
58 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
59 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
60 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
61 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
62 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
63 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
66 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95
67 * $FreeBSD: src/sys/netinet/tcp_input.c,v 1.107.2.38 2003/05/21 04:46:41 cjc Exp $
71 #include "opt_inet6.h"
72 #include "opt_ipsec.h"
73 #include "opt_tcpdebug.h"
74 #include "opt_tcp_input.h"
76 #include <sys/param.h>
77 #include <sys/systm.h>
78 #include <sys/kernel.h>
79 #include <sys/sysctl.h>
80 #include <sys/malloc.h>
82 #include <sys/proc.h> /* for proc0 declaration */
83 #include <sys/protosw.h>
84 #include <sys/socket.h>
85 #include <sys/socketvar.h>
86 #include <sys/syslog.h>
87 #include <sys/in_cksum.h>
89 #include <sys/socketvar2.h>
91 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
92 #include <machine/stdarg.h>
95 #include <net/route.h>
97 #include <netinet/in.h>
98 #include <netinet/in_systm.h>
99 #include <netinet/ip.h>
100 #include <netinet/ip_icmp.h> /* for ICMP_BANDLIM */
101 #include <netinet/in_var.h>
102 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */
103 #include <netinet/in_pcb.h>
104 #include <netinet/ip_var.h>
105 #include <netinet/ip6.h>
106 #include <netinet/icmp6.h>
107 #include <netinet6/nd6.h>
108 #include <netinet6/ip6_var.h>
109 #include <netinet6/in6_pcb.h>
110 #include <netinet/tcp.h>
111 #include <netinet/tcp_fsm.h>
112 #include <netinet/tcp_seq.h>
113 #include <netinet/tcp_timer.h>
114 #include <netinet/tcp_timer2.h>
115 #include <netinet/tcp_var.h>
116 #include <netinet6/tcp6_var.h>
117 #include <netinet/tcpip.h>
120 #include <netinet/tcp_debug.h>
122 u_char tcp_saveipgen[40]; /* the size must be of max ip header, now IPv6 */
123 struct tcphdr tcp_savetcp;
127 #include <netproto/ipsec/ipsec.h>
128 #include <netproto/ipsec/ipsec6.h>
132 #include <netinet6/ipsec.h>
133 #include <netinet6/ipsec6.h>
134 #include <netproto/key/key.h>
137 MALLOC_DEFINE(M_TSEGQ, "tseg_qent", "TCP segment queue entry");
139 static int log_in_vain = 0;
140 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
141 &log_in_vain, 0, "Log all incoming TCP connections");
143 static int blackhole = 0;
144 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW,
145 &blackhole, 0, "Do not send RST when dropping refused connections");
147 int tcp_delack_enabled = 1;
148 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW,
149 &tcp_delack_enabled, 0,
150 "Delay ACK to try and piggyback it onto a data packet");
152 #ifdef TCP_DROP_SYNFIN
153 static int drop_synfin = 0;
154 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW,
155 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set");
158 static int tcp_do_limitedtransmit = 1;
159 SYSCTL_INT(_net_inet_tcp, OID_AUTO, limitedtransmit, CTLFLAG_RW,
160 &tcp_do_limitedtransmit, 0, "Enable RFC 3042 (Limited Transmit)");
162 static int tcp_do_early_retransmit = 1;
163 SYSCTL_INT(_net_inet_tcp, OID_AUTO, earlyretransmit, CTLFLAG_RW,
164 &tcp_do_early_retransmit, 0, "Early retransmit");
166 int tcp_aggregate_acks = 1;
167 SYSCTL_INT(_net_inet_tcp, OID_AUTO, aggregate_acks, CTLFLAG_RW,
168 &tcp_aggregate_acks, 0, "Aggregate built-up acks into one ack");
170 static int tcp_do_eifel_detect = 1;
171 SYSCTL_INT(_net_inet_tcp, OID_AUTO, eifel, CTLFLAG_RW,
172 &tcp_do_eifel_detect, 0, "Eifel detection algorithm (RFC 3522)");
174 static int tcp_do_abc = 1;
175 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc, CTLFLAG_RW,
177 "TCP Appropriate Byte Counting (RFC 3465)");
180 * Define as tunable for easy testing with SACK on and off.
181 * Warning: do not change setting in the middle of an existing active TCP flow,
182 * else strange things might happen to that flow.
185 SYSCTL_INT(_net_inet_tcp, OID_AUTO, sack, CTLFLAG_RW,
186 &tcp_do_sack, 0, "Enable SACK Algorithms");
188 int tcp_do_smartsack = 1;
189 SYSCTL_INT(_net_inet_tcp, OID_AUTO, smartsack, CTLFLAG_RW,
190 &tcp_do_smartsack, 0, "Enable Smart SACK Algorithms");
192 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0,
193 "TCP Segment Reassembly Queue");
195 int tcp_reass_maxseg = 0;
196 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RD,
197 &tcp_reass_maxseg, 0,
198 "Global maximum number of TCP Segments in Reassembly Queue");
200 int tcp_reass_qsize = 0;
201 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD,
203 "Global number of TCP Segments currently in Reassembly Queue");
205 static int tcp_reass_overflows = 0;
206 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD,
207 &tcp_reass_overflows, 0,
208 "Global number of TCP Segment Reassembly Queue Overflows");
210 int tcp_do_autorcvbuf = 1;
211 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_RW,
212 &tcp_do_autorcvbuf, 0, "Enable automatic receive buffer sizing");
214 int tcp_autorcvbuf_inc = 16*1024;
215 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_RW,
216 &tcp_autorcvbuf_inc, 0,
217 "Incrementor step size of automatic receive buffer");
219 int tcp_autorcvbuf_max = 2*1024*1024;
220 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_RW,
221 &tcp_autorcvbuf_max, 0, "Max size of automatic receive buffer");
223 int tcp_sosnd_agglim = 2;
224 SYSCTL_INT(_net_inet_tcp, OID_AUTO, sosnd_agglim, CTLFLAG_RW,
225 &tcp_sosnd_agglim, 0, "TCP sosend mbuf aggregation limit");
227 int tcp_sosnd_async = 1;
228 SYSCTL_INT(_net_inet_tcp, OID_AUTO, sosnd_async, CTLFLAG_RW,
229 &tcp_sosnd_async, 0, "TCP asynchronized pru_send");
231 static void tcp_dooptions(struct tcpopt *, u_char *, int, boolean_t);
232 static void tcp_pulloutofband(struct socket *,
233 struct tcphdr *, struct mbuf *, int);
234 static int tcp_reass(struct tcpcb *, struct tcphdr *, int *,
236 static void tcp_xmit_timer(struct tcpcb *, int);
237 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *, int);
238 static void tcp_sack_rexmt(struct tcpcb *, struct tcphdr *);
239 static int tcp_rmx_msl(const struct tcpcb *);
240 static void tcp_established(struct tcpcb *);
242 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
244 #define ND6_HINT(tp) \
246 if ((tp) && (tp)->t_inpcb && \
247 ((tp)->t_inpcb->inp_vflag & INP_IPV6) && \
248 (tp)->t_inpcb->in6p_route.ro_rt) \
249 nd6_nud_hint((tp)->t_inpcb->in6p_route.ro_rt, NULL, 0); \
256 * Indicate whether this ack should be delayed. We can delay the ack if
257 * - delayed acks are enabled and
258 * - there is no delayed ack timer in progress and
259 * - our last ack wasn't a 0-sized window. We never want to delay
260 * the ack that opens up a 0-sized window.
262 #define DELAY_ACK(tp) \
263 (tcp_delack_enabled && !tcp_callout_pending(tp, tp->tt_delack) && \
264 !(tp->t_flags & TF_RXWIN0SENT))
266 #define acceptable_window_update(tp, th, tiwin) \
267 (SEQ_LT(tp->snd_wl1, th->th_seq) || \
268 (tp->snd_wl1 == th->th_seq && \
269 (SEQ_LT(tp->snd_wl2, th->th_ack) || \
270 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))
273 tcp_reass(struct tcpcb *tp, struct tcphdr *th, int *tlenp, struct mbuf *m)
276 struct tseg_qent *p = NULL;
277 struct tseg_qent *te;
278 struct socket *so = tp->t_inpcb->inp_socket;
282 * Call with th == NULL after become established to
283 * force pre-ESTABLISHED data up to user socket.
289 * Limit the number of segments in the reassembly queue to prevent
290 * holding on to too many segments (and thus running out of mbufs).
291 * Make sure to let the missing segment through which caused this
292 * queue. Always keep one global queue entry spare to be able to
293 * process the missing segment.
295 if (th->th_seq != tp->rcv_nxt &&
296 tcp_reass_qsize + 1 >= tcp_reass_maxseg) {
297 tcp_reass_overflows++;
298 tcpstat.tcps_rcvmemdrop++;
300 /* no SACK block to report */
301 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
305 /* Allocate a new queue entry. */
306 te = kmalloc(sizeof(struct tseg_qent), M_TSEGQ, M_INTWAIT | M_NULLOK);
308 tcpstat.tcps_rcvmemdrop++;
310 /* no SACK block to report */
311 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
314 atomic_add_int(&tcp_reass_qsize, 1);
317 * Find a segment which begins after this one does.
319 LIST_FOREACH(q, &tp->t_segq, tqe_q) {
320 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
326 * If there is a preceding segment, it may provide some of
327 * our data already. If so, drop the data from the incoming
328 * segment. If it provides all of our data, drop us.
333 /* conversion to int (in i) handles seq wraparound */
334 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
335 if (i > 0) { /* overlaps preceding segment */
336 tp->t_flags |= (TF_DUPSEG | TF_ENCLOSESEG);
337 /* enclosing block starts w/ preceding segment */
338 tp->encloseblk.rblk_start = p->tqe_th->th_seq;
340 /* preceding encloses incoming segment */
341 tp->encloseblk.rblk_end = TCP_SACK_BLKEND(
342 p->tqe_th->th_seq + p->tqe_len,
343 p->tqe_th->th_flags);
344 tcpstat.tcps_rcvduppack++;
345 tcpstat.tcps_rcvdupbyte += *tlenp;
348 atomic_add_int(&tcp_reass_qsize, -1);
350 * Try to present any queued data
351 * at the left window edge to the user.
352 * This is needed after the 3-WHS
355 goto present; /* ??? */
360 /* incoming segment end is enclosing block end */
361 tp->encloseblk.rblk_end = TCP_SACK_BLKEND(
362 th->th_seq + *tlenp, th->th_flags);
363 /* trim end of reported D-SACK block */
364 tp->reportblk.rblk_end = th->th_seq;
367 tcpstat.tcps_rcvoopack++;
368 tcpstat.tcps_rcvoobyte += *tlenp;
371 * While we overlap succeeding segments trim them or,
372 * if they are completely covered, dequeue them.
375 tcp_seq_diff_t i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
376 tcp_seq qend = q->tqe_th->th_seq + q->tqe_len;
377 tcp_seq qend_sack = TCP_SACK_BLKEND(qend, q->tqe_th->th_flags);
378 struct tseg_qent *nq;
382 if (!(tp->t_flags & TF_DUPSEG)) { /* first time through */
383 tp->t_flags |= (TF_DUPSEG | TF_ENCLOSESEG);
384 tp->encloseblk = tp->reportblk;
385 /* report trailing duplicate D-SACK segment */
386 tp->reportblk.rblk_start = q->tqe_th->th_seq;
388 if ((tp->t_flags & TF_ENCLOSESEG) &&
389 SEQ_GT(qend_sack, tp->encloseblk.rblk_end)) {
390 /* extend enclosing block if one exists */
391 tp->encloseblk.rblk_end = qend_sack;
393 if (i < q->tqe_len) {
394 q->tqe_th->th_seq += i;
400 nq = LIST_NEXT(q, tqe_q);
401 LIST_REMOVE(q, tqe_q);
404 atomic_add_int(&tcp_reass_qsize, -1);
408 /* Insert the new segment queue entry into place. */
411 te->tqe_len = *tlenp;
413 /* check if can coalesce with following segment */
414 if (q != NULL && (th->th_seq + *tlenp == q->tqe_th->th_seq)) {
415 tcp_seq tend = te->tqe_th->th_seq + te->tqe_len;
416 tcp_seq tend_sack = TCP_SACK_BLKEND(tend, te->tqe_th->th_flags);
418 te->tqe_len += q->tqe_len;
419 if (q->tqe_th->th_flags & TH_FIN)
420 te->tqe_th->th_flags |= TH_FIN;
421 m_cat(te->tqe_m, q->tqe_m);
422 tp->encloseblk.rblk_end = tend_sack;
424 * When not reporting a duplicate segment, use
425 * the larger enclosing block as the SACK block.
427 if (!(tp->t_flags & TF_DUPSEG))
428 tp->reportblk.rblk_end = tend_sack;
429 LIST_REMOVE(q, tqe_q);
431 atomic_add_int(&tcp_reass_qsize, -1);
435 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
437 /* check if can coalesce with preceding segment */
438 if (p->tqe_th->th_seq + p->tqe_len == th->th_seq) {
439 p->tqe_len += te->tqe_len;
440 m_cat(p->tqe_m, te->tqe_m);
441 tp->encloseblk.rblk_start = p->tqe_th->th_seq;
443 * When not reporting a duplicate segment, use
444 * the larger enclosing block as the SACK block.
446 if (!(tp->t_flags & TF_DUPSEG))
447 tp->reportblk.rblk_start = p->tqe_th->th_seq;
449 atomic_add_int(&tcp_reass_qsize, -1);
451 LIST_INSERT_AFTER(p, te, tqe_q);
457 * Present data to user, advancing rcv_nxt through
458 * completed sequence space.
460 if (!TCPS_HAVEESTABLISHED(tp->t_state))
462 q = LIST_FIRST(&tp->t_segq);
463 if (q == NULL || q->tqe_th->th_seq != tp->rcv_nxt)
465 tp->rcv_nxt += q->tqe_len;
466 if (!(tp->t_flags & TF_DUPSEG)) {
467 /* no SACK block to report since ACK advanced */
468 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
470 /* no enclosing block to report since ACK advanced */
471 tp->t_flags &= ~TF_ENCLOSESEG;
472 flags = q->tqe_th->th_flags & TH_FIN;
473 LIST_REMOVE(q, tqe_q);
474 KASSERT(LIST_EMPTY(&tp->t_segq) ||
475 LIST_FIRST(&tp->t_segq)->tqe_th->th_seq != tp->rcv_nxt,
476 ("segment not coalesced"));
477 if (so->so_state & SS_CANTRCVMORE) {
480 lwkt_gettoken(&so->so_rcv.ssb_token);
481 ssb_appendstream(&so->so_rcv, q->tqe_m);
482 lwkt_reltoken(&so->so_rcv.ssb_token);
485 atomic_add_int(&tcp_reass_qsize, -1);
492 * TCP input routine, follows pages 65-76 of the
493 * protocol specification dated September, 1981 very closely.
497 tcp6_input(struct mbuf **mp, int *offp, int proto)
499 struct mbuf *m = *mp;
500 struct in6_ifaddr *ia6;
502 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
505 * draft-itojun-ipv6-tcp-to-anycast
506 * better place to put this in?
508 ia6 = ip6_getdstifaddr(m);
509 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
512 ip6 = mtod(m, struct ip6_hdr *);
513 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
514 offsetof(struct ip6_hdr, ip6_dst));
515 return (IPPROTO_DONE);
518 tcp_input(mp, offp, proto);
519 return (IPPROTO_DONE);
524 tcp_input(struct mbuf **mp, int *offp, int proto)
528 struct ip *ip = NULL;
530 struct inpcb *inp = NULL;
536 struct tcpcb *tp = NULL;
538 struct socket *so = NULL;
540 boolean_t ourfinisacked, needoutput = FALSE;
543 struct tcpopt to; /* options in this segment */
544 struct sockaddr_in *next_hop = NULL;
545 int rstreason; /* For badport_bandlim accounting purposes */
547 struct ip6_hdr *ip6 = NULL;
552 const boolean_t isipv6 = FALSE;
562 tcpstat.tcps_rcvtotal++;
564 if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
567 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
568 KKASSERT(mtag != NULL);
569 next_hop = m_tag_data(mtag);
573 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? TRUE : FALSE;
577 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */
578 ip6 = mtod(m, struct ip6_hdr *);
579 tlen = (sizeof *ip6) + ntohs(ip6->ip6_plen) - off0;
580 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) {
581 tcpstat.tcps_rcvbadsum++;
584 th = (struct tcphdr *)((caddr_t)ip6 + off0);
587 * Be proactive about unspecified IPv6 address in source.
588 * As we use all-zero to indicate unbounded/unconnected pcb,
589 * unspecified IPv6 address can be used to confuse us.
591 * Note that packets with unspecified IPv6 destination is
592 * already dropped in ip6_input.
594 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
600 * Get IP and TCP header together in first mbuf.
601 * Note: IP leaves IP header in first mbuf.
603 if (off0 > sizeof(struct ip)) {
605 off0 = sizeof(struct ip);
607 /* already checked and pulled up in ip_demux() */
608 KASSERT(m->m_len >= sizeof(struct tcpiphdr),
609 ("TCP header not in one mbuf: m->m_len %d", m->m_len));
610 ip = mtod(m, struct ip *);
611 ipov = (struct ipovly *)ip;
612 th = (struct tcphdr *)((caddr_t)ip + off0);
615 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
616 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
617 th->th_sum = m->m_pkthdr.csum_data;
619 th->th_sum = in_pseudo(ip->ip_src.s_addr,
621 htonl(m->m_pkthdr.csum_data +
624 th->th_sum ^= 0xffff;
627 * Checksum extended TCP header and data.
629 len = sizeof(struct ip) + tlen;
630 bzero(ipov->ih_x1, sizeof ipov->ih_x1);
631 ipov->ih_len = (u_short)tlen;
632 ipov->ih_len = htons(ipov->ih_len);
633 th->th_sum = in_cksum(m, len);
636 tcpstat.tcps_rcvbadsum++;
640 /* Re-initialization for later version check */
641 ip->ip_v = IPVERSION;
646 * Check that TCP offset makes sense,
647 * pull out TCP options and adjust length. XXX
649 off = th->th_off << 2;
650 /* already checked and pulled up in ip_demux() */
651 KASSERT(off >= sizeof(struct tcphdr) && off <= tlen,
652 ("bad TCP data offset %d (tlen %d)", off, tlen));
653 tlen -= off; /* tlen is used instead of ti->ti_len */
654 if (off > sizeof(struct tcphdr)) {
656 IP6_EXTHDR_CHECK(m, off0, off, IPPROTO_DONE);
657 ip6 = mtod(m, struct ip6_hdr *);
658 th = (struct tcphdr *)((caddr_t)ip6 + off0);
660 /* already pulled up in ip_demux() */
661 KASSERT(m->m_len >= sizeof(struct ip) + off,
662 ("TCP header and options not in one mbuf: "
663 "m_len %d, off %d", m->m_len, off));
665 optlen = off - sizeof(struct tcphdr);
666 optp = (u_char *)(th + 1);
668 thflags = th->th_flags;
670 #ifdef TCP_DROP_SYNFIN
672 * If the drop_synfin option is enabled, drop all packets with
673 * both the SYN and FIN bits set. This prevents e.g. nmap from
674 * identifying the TCP/IP stack.
676 * This is a violation of the TCP specification.
678 if (drop_synfin && (thflags & (TH_SYN | TH_FIN)) == (TH_SYN | TH_FIN))
683 * Convert TCP protocol specific fields to host format.
685 th->th_seq = ntohl(th->th_seq);
686 th->th_ack = ntohl(th->th_ack);
687 th->th_win = ntohs(th->th_win);
688 th->th_urp = ntohs(th->th_urp);
691 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options,
692 * until after ip6_savecontrol() is called and before other functions
693 * which don't want those proto headers.
694 * Because ip6_savecontrol() is going to parse the mbuf to
695 * search for data to be passed up to user-land, it wants mbuf
696 * parameters to be unchanged.
697 * XXX: the call of ip6_savecontrol() has been obsoleted based on
698 * latest version of the advanced API (20020110).
700 drop_hdrlen = off0 + off;
703 * Locate pcb for segment.
706 /* IPFIREWALL_FORWARD section */
707 if (next_hop != NULL && !isipv6) { /* IPv6 support is not there yet */
709 * Transparently forwarded. Pretend to be the destination.
710 * already got one like this?
712 cpu = mycpu->gd_cpuid;
713 inp = in_pcblookup_hash(&tcbinfo[cpu],
714 ip->ip_src, th->th_sport,
715 ip->ip_dst, th->th_dport,
716 0, m->m_pkthdr.rcvif);
719 * It's new. Try to find the ambushing socket.
723 * The rest of the ipfw code stores the port in
725 * (The IP address is still in network order.)
727 in_port_t dport = next_hop->sin_port ?
728 htons(next_hop->sin_port) :
731 cpu = tcp_addrcpu(ip->ip_src.s_addr, th->th_sport,
732 next_hop->sin_addr.s_addr, dport);
733 inp = in_pcblookup_hash(&tcbinfo[cpu],
734 ip->ip_src, th->th_sport,
735 next_hop->sin_addr, dport,
736 1, m->m_pkthdr.rcvif);
740 inp = in6_pcblookup_hash(&tcbinfo[0],
741 &ip6->ip6_src, th->th_sport,
742 &ip6->ip6_dst, th->th_dport,
743 1, m->m_pkthdr.rcvif);
745 cpu = mycpu->gd_cpuid;
746 inp = in_pcblookup_hash(&tcbinfo[cpu],
747 ip->ip_src, th->th_sport,
748 ip->ip_dst, th->th_dport,
749 1, m->m_pkthdr.rcvif);
754 * If the state is CLOSED (i.e., TCB does not exist) then
755 * all data in the incoming segment is discarded.
756 * If the TCB exists but is in CLOSED state, it is embryonic,
757 * but should either do a listen or a connect soon.
762 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2];
764 char dbuf[sizeof "aaa.bbb.ccc.ddd"];
765 char sbuf[sizeof "aaa.bbb.ccc.ddd"];
769 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst));
772 strcat(sbuf, ip6_sprintf(&ip6->ip6_src));
775 strcpy(dbuf, inet_ntoa(ip->ip_dst));
776 strcpy(sbuf, inet_ntoa(ip->ip_src));
778 switch (log_in_vain) {
780 if (!(thflags & TH_SYN))
784 "Connection attempt to TCP %s:%d "
785 "from %s:%d flags:0x%02x\n",
786 dbuf, ntohs(th->th_dport), sbuf,
787 ntohs(th->th_sport), thflags);
796 if (thflags & TH_SYN)
805 rstreason = BANDLIM_RST_CLOSEDPORT;
811 if (ipsec6_in_reject_so(m, inp->inp_socket)) {
812 ipsec6stat.in_polvio++;
816 if (ipsec4_in_reject_so(m, inp->inp_socket)) {
817 ipsecstat.in_polvio++;
824 if (ipsec6_in_reject(m, inp))
827 if (ipsec4_in_reject(m, inp))
831 /* Check the minimum TTL for socket. */
833 if ((isipv6 ? ip6->ip6_hlim : ip->ip_ttl) < inp->inp_ip_minttl)
839 rstreason = BANDLIM_RST_CLOSEDPORT;
842 if (tp->t_state <= TCPS_CLOSED)
845 /* Unscale the window into a 32-bit value. */
846 if (!(thflags & TH_SYN))
847 tiwin = th->th_win << tp->snd_scale;
851 so = inp->inp_socket;
854 if (so->so_options & SO_DEBUG) {
855 ostate = tp->t_state;
857 bcopy(ip6, tcp_saveipgen, sizeof(*ip6));
859 bcopy(ip, tcp_saveipgen, sizeof(*ip));
864 bzero(&to, sizeof to);
866 if (so->so_options & SO_ACCEPTCONN) {
867 struct in_conninfo inc;
870 inc.inc_isipv6 = (isipv6 == TRUE);
873 inc.inc6_faddr = ip6->ip6_src;
874 inc.inc6_laddr = ip6->ip6_dst;
875 inc.inc6_route.ro_rt = NULL; /* XXX */
877 inc.inc_faddr = ip->ip_src;
878 inc.inc_laddr = ip->ip_dst;
879 inc.inc_route.ro_rt = NULL; /* XXX */
881 inc.inc_fport = th->th_sport;
882 inc.inc_lport = th->th_dport;
885 * If the state is LISTEN then ignore segment if it contains
886 * a RST. If the segment contains an ACK then it is bad and
887 * send a RST. If it does not contain a SYN then it is not
888 * interesting; drop it.
890 * If the state is SYN_RECEIVED (syncache) and seg contains
891 * an ACK, but not for our SYN/ACK, send a RST. If the seg
892 * contains a RST, check the sequence number to see if it
893 * is a valid reset segment.
895 if ((thflags & (TH_RST | TH_ACK | TH_SYN)) != TH_SYN) {
896 if ((thflags & (TH_RST | TH_ACK | TH_SYN)) == TH_ACK) {
897 if (!syncache_expand(&inc, th, &so, m)) {
899 * No syncache entry, or ACK was not
900 * for our SYN/ACK. Send a RST.
902 tcpstat.tcps_badsyn++;
903 rstreason = BANDLIM_RST_OPENPORT;
908 * Could not complete 3-way handshake,
909 * connection is being closed down, and
910 * syncache will free mbuf.
913 return(IPPROTO_DONE);
916 * We must be in the correct protocol thread
917 * for this connection.
919 KKASSERT(so->so_port == &curthread->td_msgport);
922 * Socket is created in state SYN_RECEIVED.
923 * Continue processing segment.
928 * This is what would have happened in
929 * tcp_output() when the SYN,ACK was sent.
931 tp->snd_up = tp->snd_una;
932 tp->snd_max = tp->snd_nxt = tp->iss + 1;
933 tp->last_ack_sent = tp->rcv_nxt;
935 * XXX possible bug - it doesn't appear that tp->snd_wnd is unscaled
936 * until the _second_ ACK is received:
937 * rcv SYN (set wscale opts) --> send SYN/ACK, set snd_wnd = window.
938 * rcv ACK, calculate tiwin --> process SYN_RECEIVED, determine wscale,
939 * move to ESTAB, set snd_wnd to tiwin.
941 tp->snd_wnd = tiwin; /* unscaled */
944 if (thflags & TH_RST) {
945 syncache_chkrst(&inc, th);
948 if (thflags & TH_ACK) {
949 syncache_badack(&inc);
950 tcpstat.tcps_badsyn++;
951 rstreason = BANDLIM_RST_OPENPORT;
958 * Segment's flags are (SYN) or (SYN | FIN).
962 * If deprecated address is forbidden,
963 * we do not accept SYN to deprecated interface
964 * address to prevent any new inbound connection from
965 * getting established.
966 * When we do not accept SYN, we send a TCP RST,
967 * with deprecated source address (instead of dropping
968 * it). We compromise it as it is much better for peer
969 * to send a RST, and RST will be the final packet
972 * If we do not forbid deprecated addresses, we accept
973 * the SYN packet. RFC2462 does not suggest dropping
975 * If we decipher RFC2462 5.5.4, it says like this:
976 * 1. use of deprecated addr with existing
977 * communication is okay - "SHOULD continue to be
979 * 2. use of it with new communication:
980 * (2a) "SHOULD NOT be used if alternate address
981 * with sufficient scope is available"
982 * (2b) nothing mentioned otherwise.
983 * Here we fall into (2b) case as we have no choice in
984 * our source address selection - we must obey the peer.
986 * The wording in RFC2462 is confusing, and there are
987 * multiple description text for deprecated address
988 * handling - worse, they are not exactly the same.
989 * I believe 5.5.4 is the best one, so we follow 5.5.4.
991 if (isipv6 && !ip6_use_deprecated) {
992 struct in6_ifaddr *ia6;
994 if ((ia6 = ip6_getdstifaddr(m)) &&
995 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
997 rstreason = BANDLIM_RST_OPENPORT;
1003 * If it is from this socket, drop it, it must be forged.
1004 * Don't bother responding if the destination was a broadcast.
1006 if (th->th_dport == th->th_sport) {
1008 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
1012 if (ip->ip_dst.s_addr == ip->ip_src.s_addr)
1017 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
1019 * Note that it is quite possible to receive unicast
1020 * link-layer packets with a broadcast IP address. Use
1021 * in_broadcast() to find them.
1023 if (m->m_flags & (M_BCAST | M_MCAST))
1026 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1027 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
1030 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1031 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1032 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1033 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
1037 * SYN appears to be valid; create compressed TCP state
1038 * for syncache, or perform t/tcp connection.
1040 if (so->so_qlen <= so->so_qlimit) {
1041 tcp_dooptions(&to, optp, optlen, TRUE);
1042 if (!syncache_add(&inc, &to, th, so, m))
1046 * Entry added to syncache, mbuf used to
1047 * send SYN,ACK packet.
1049 return(IPPROTO_DONE);
1056 * Should not happen - syncache should pick up these connections.
1058 * Once we are past handling listen sockets we must be in the
1059 * correct protocol processing thread.
1061 KASSERT(tp->t_state != TCPS_LISTEN, ("tcp_input: TCPS_LISTEN state"));
1062 KKASSERT(so->so_port == &curthread->td_msgport);
1065 * This is the second part of the MSS DoS prevention code (after
1066 * minmss on the sending side) and it deals with too many too small
1067 * tcp packets in a too short timeframe (1 second).
1069 * XXX Removed. This code was crap. It does not scale to network
1070 * speed, and default values break NFS. Gone.
1075 * Segment received on connection.
1077 * Reset idle time and keep-alive timer. Don't waste time if less
1078 * then a second has elapsed.
1080 if ((int)(ticks - tp->t_rcvtime) > hz)
1081 tcp_timer_keep_activity(tp, thflags);
1085 * XXX this is tradtitional behavior, may need to be cleaned up.
1087 tcp_dooptions(&to, optp, optlen, (thflags & TH_SYN) != 0);
1088 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1089 if (to.to_flags & TOF_SCALE) {
1090 tp->t_flags |= TF_RCVD_SCALE;
1091 tp->requested_s_scale = to.to_requested_s_scale;
1093 if (to.to_flags & TOF_TS) {
1094 tp->t_flags |= TF_RCVD_TSTMP;
1095 tp->ts_recent = to.to_tsval;
1096 tp->ts_recent_age = ticks;
1098 if (!(to.to_flags & TOF_MSS))
1100 tcp_mss(tp, to.to_mss);
1102 * Only set the TF_SACK_PERMITTED per-connection flag
1103 * if we got a SACK_PERMITTED option from the other side
1104 * and the global tcp_do_sack variable is true.
1106 if (tcp_do_sack && (to.to_flags & TOF_SACK_PERMITTED))
1107 tp->t_flags |= TF_SACK_PERMITTED;
1111 * Header prediction: check for the two common cases
1112 * of a uni-directional data xfer. If the packet has
1113 * no control flags, is in-sequence, the window didn't
1114 * change and we're not retransmitting, it's a
1115 * candidate. If the length is zero and the ack moved
1116 * forward, we're the sender side of the xfer. Just
1117 * free the data acked & wake any higher level process
1118 * that was blocked waiting for space. If the length
1119 * is non-zero and the ack didn't move, we're the
1120 * receiver side. If we're getting packets in-order
1121 * (the reassembly queue is empty), add the data to
1122 * the socket buffer and note that we need a delayed ack.
1123 * Make sure that the hidden state-flags are also off.
1124 * Since we check for TCPS_ESTABLISHED above, it can only
1127 if (tp->t_state == TCPS_ESTABLISHED &&
1128 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1129 !(tp->t_flags & (TF_NEEDSYN | TF_NEEDFIN)) &&
1130 (!(to.to_flags & TOF_TS) ||
1131 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) &&
1132 th->th_seq == tp->rcv_nxt &&
1133 tp->snd_nxt == tp->snd_max) {
1136 * If last ACK falls within this segment's sequence numbers,
1137 * record the timestamp.
1138 * NOTE that the test is modified according to the latest
1139 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1141 if ((to.to_flags & TOF_TS) &&
1142 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1143 tp->ts_recent_age = ticks;
1144 tp->ts_recent = to.to_tsval;
1148 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1149 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1150 tp->snd_cwnd >= tp->snd_wnd &&
1151 !IN_FASTRECOVERY(tp)) {
1153 * This is a pure ack for outstanding data.
1155 ++tcpstat.tcps_predack;
1157 * "bad retransmit" recovery
1159 * If Eifel detection applies, then
1160 * it is deterministic, so use it
1161 * unconditionally over the old heuristic.
1162 * Otherwise, fall back to the old heuristic.
1164 if (tcp_do_eifel_detect &&
1165 (to.to_flags & TOF_TS) && to.to_tsecr &&
1166 (tp->t_flags & TF_FIRSTACCACK)) {
1167 /* Eifel detection applicable. */
1168 if (to.to_tsecr < tp->t_rexmtTS) {
1169 tcp_revert_congestion_state(tp);
1170 ++tcpstat.tcps_eifeldetected;
1171 if (tp->t_rxtshift != 1 ||
1172 ticks >= tp->t_badrxtwin)
1173 ++tcpstat.tcps_rttcantdetect;
1175 } else if (tp->t_rxtshift == 1 &&
1176 ticks < tp->t_badrxtwin) {
1177 tcp_revert_congestion_state(tp);
1178 ++tcpstat.tcps_rttdetected;
1180 tp->t_flags &= ~(TF_FIRSTACCACK |
1181 TF_FASTREXMT | TF_EARLYREXMT);
1183 * Recalculate the retransmit timer / rtt.
1185 * Some machines (certain windows boxes)
1186 * send broken timestamp replies during the
1187 * SYN+ACK phase, ignore timestamps of 0.
1189 if ((to.to_flags & TOF_TS) && to.to_tsecr) {
1191 ticks - to.to_tsecr + 1);
1192 } else if (tp->t_rtttime &&
1193 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1195 ticks - tp->t_rtttime);
1197 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1198 acked = th->th_ack - tp->snd_una;
1199 tcpstat.tcps_rcvackpack++;
1200 tcpstat.tcps_rcvackbyte += acked;
1201 sbdrop(&so->so_snd.sb, acked);
1202 tp->snd_recover = th->th_ack - 1;
1203 tp->snd_una = th->th_ack;
1206 * Update window information.
1208 if (tiwin != tp->snd_wnd &&
1209 acceptable_window_update(tp, th, tiwin)) {
1210 /* keep track of pure window updates */
1211 if (tp->snd_wl2 == th->th_ack &&
1212 tiwin > tp->snd_wnd)
1213 tcpstat.tcps_rcvwinupd++;
1214 tp->snd_wnd = tiwin;
1215 tp->snd_wl1 = th->th_seq;
1216 tp->snd_wl2 = th->th_ack;
1217 if (tp->snd_wnd > tp->max_sndwnd)
1218 tp->max_sndwnd = tp->snd_wnd;
1221 ND6_HINT(tp); /* some progress has been done */
1223 * If all outstanding data are acked, stop
1224 * retransmit timer, otherwise restart timer
1225 * using current (possibly backed-off) value.
1226 * If process is waiting for space,
1227 * wakeup/selwakeup/signal. If data
1228 * are ready to send, let tcp_output
1229 * decide between more output or persist.
1231 if (tp->snd_una == tp->snd_max) {
1232 tcp_callout_stop(tp, tp->tt_rexmt);
1233 } else if (!tcp_callout_active(tp,
1235 tcp_callout_reset(tp, tp->tt_rexmt,
1236 tp->t_rxtcur, tcp_timer_rexmt);
1239 if (so->so_snd.ssb_cc > 0)
1241 return(IPPROTO_DONE);
1243 } else if (tiwin == tp->snd_wnd &&
1244 th->th_ack == tp->snd_una &&
1245 LIST_EMPTY(&tp->t_segq) &&
1246 tlen <= ssb_space(&so->so_rcv)) {
1247 u_long newsize = 0; /* automatic sockbuf scaling */
1249 * This is a pure, in-sequence data packet
1250 * with nothing on the reassembly queue and
1251 * we have enough buffer space to take it.
1253 ++tcpstat.tcps_preddat;
1254 tp->rcv_nxt += tlen;
1255 tcpstat.tcps_rcvpack++;
1256 tcpstat.tcps_rcvbyte += tlen;
1257 ND6_HINT(tp); /* some progress has been done */
1259 * Automatic sizing of receive socket buffer. Often the send
1260 * buffer size is not optimally adjusted to the actual network
1261 * conditions at hand (delay bandwidth product). Setting the
1262 * buffer size too small limits throughput on links with high
1263 * bandwidth and high delay (eg. trans-continental/oceanic links).
1265 * On the receive side the socket buffer memory is only rarely
1266 * used to any significant extent. This allows us to be much
1267 * more aggressive in scaling the receive socket buffer. For
1268 * the case that the buffer space is actually used to a large
1269 * extent and we run out of kernel memory we can simply drop
1270 * the new segments; TCP on the sender will just retransmit it
1271 * later. Setting the buffer size too big may only consume too
1272 * much kernel memory if the application doesn't read() from
1273 * the socket or packet loss or reordering makes use of the
1276 * The criteria to step up the receive buffer one notch are:
1277 * 1. the number of bytes received during the time it takes
1278 * one timestamp to be reflected back to us (the RTT);
1279 * 2. received bytes per RTT is within seven eighth of the
1280 * current socket buffer size;
1281 * 3. receive buffer size has not hit maximal automatic size;
1283 * This algorithm does one step per RTT at most and only if
1284 * we receive a bulk stream w/o packet losses or reorderings.
1285 * Shrinking the buffer during idle times is not necessary as
1286 * it doesn't consume any memory when idle.
1288 * TODO: Only step up if the application is actually serving
1289 * the buffer to better manage the socket buffer resources.
1291 if (tcp_do_autorcvbuf &&
1293 (so->so_rcv.ssb_flags & SSB_AUTOSIZE)) {
1294 if (to.to_tsecr > tp->rfbuf_ts &&
1295 to.to_tsecr - tp->rfbuf_ts < hz) {
1297 (so->so_rcv.ssb_hiwat / 8 * 7) &&
1298 so->so_rcv.ssb_hiwat <
1299 tcp_autorcvbuf_max) {
1301 ulmin(so->so_rcv.ssb_hiwat +
1303 tcp_autorcvbuf_max);
1305 /* Start over with next RTT. */
1309 tp->rfbuf_cnt += tlen; /* add up */
1312 * Add data to socket buffer.
1314 if (so->so_state & SS_CANTRCVMORE) {
1318 * Set new socket buffer size, give up when
1321 * Adjusting the size can mess up ACK
1322 * sequencing when pure window updates are
1323 * being avoided (which is the default),
1326 lwkt_gettoken(&so->so_rcv.ssb_token);
1328 tp->t_flags |= TF_RXRESIZED;
1329 if (!ssb_reserve(&so->so_rcv, newsize,
1331 atomic_clear_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE);
1334 (TCP_MAXWIN << tp->rcv_scale)) {
1335 atomic_clear_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE);
1338 m_adj(m, drop_hdrlen); /* delayed header drop */
1339 ssb_appendstream(&so->so_rcv, m);
1340 lwkt_reltoken(&so->so_rcv.ssb_token);
1344 * This code is responsible for most of the ACKs
1345 * the TCP stack sends back after receiving a data
1346 * packet. Note that the DELAY_ACK check fails if
1347 * the delack timer is already running, which results
1348 * in an ack being sent every other packet (which is
1351 * We then further aggregate acks by not actually
1352 * sending one until the protocol thread has completed
1353 * processing the current backlog of packets. This
1354 * does not delay the ack any further, but allows us
1355 * to take advantage of the packet aggregation that
1356 * high speed NICs do (usually blocks of 8-10 packets)
1357 * to send a single ack rather then four or five acks,
1358 * greatly reducing the ack rate, the return channel
1359 * bandwidth, and the protocol overhead on both ends.
1361 * Since this also has the effect of slowing down
1362 * the exponential slow-start ramp-up, systems with
1363 * very large bandwidth-delay products might want
1364 * to turn the feature off.
1366 if (DELAY_ACK(tp)) {
1367 tcp_callout_reset(tp, tp->tt_delack,
1368 tcp_delacktime, tcp_timer_delack);
1369 } else if (tcp_aggregate_acks) {
1370 tp->t_flags |= TF_ACKNOW;
1371 if (!(tp->t_flags & TF_ONOUTPUTQ)) {
1372 tp->t_flags |= TF_ONOUTPUTQ;
1373 tp->tt_cpu = mycpu->gd_cpuid;
1375 &tcpcbackq[tp->tt_cpu],
1379 tp->t_flags |= TF_ACKNOW;
1382 return(IPPROTO_DONE);
1387 * Calculate amount of space in receive window,
1388 * and then do TCP input processing.
1389 * Receive window is amount of space in rcv queue,
1390 * but not less than advertised window.
1392 recvwin = ssb_space(&so->so_rcv);
1395 tp->rcv_wnd = imax(recvwin, (int)(tp->rcv_adv - tp->rcv_nxt));
1397 /* Reset receive buffer auto scaling when not in bulk receive mode. */
1401 switch (tp->t_state) {
1403 * If the state is SYN_RECEIVED:
1404 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1406 case TCPS_SYN_RECEIVED:
1407 if ((thflags & TH_ACK) &&
1408 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1409 SEQ_GT(th->th_ack, tp->snd_max))) {
1410 rstreason = BANDLIM_RST_OPENPORT;
1416 * If the state is SYN_SENT:
1417 * if seg contains an ACK, but not for our SYN, drop the input.
1418 * if seg contains a RST, then drop the connection.
1419 * if seg does not contain SYN, then drop it.
1420 * Otherwise this is an acceptable SYN segment
1421 * initialize tp->rcv_nxt and tp->irs
1422 * if seg contains ack then advance tp->snd_una
1423 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1424 * arrange for segment to be acked (eventually)
1425 * continue processing rest of data/controls, beginning with URG
1428 if ((thflags & TH_ACK) &&
1429 (SEQ_LEQ(th->th_ack, tp->iss) ||
1430 SEQ_GT(th->th_ack, tp->snd_max))) {
1431 rstreason = BANDLIM_UNLIMITED;
1434 if (thflags & TH_RST) {
1435 if (thflags & TH_ACK)
1436 tp = tcp_drop(tp, ECONNREFUSED);
1439 if (!(thflags & TH_SYN))
1441 tp->snd_wnd = th->th_win; /* initial send window */
1443 tp->irs = th->th_seq;
1445 if (thflags & TH_ACK) {
1446 /* Our SYN was acked. */
1447 tcpstat.tcps_connects++;
1449 /* Do window scaling on this connection? */
1450 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
1451 (TF_RCVD_SCALE | TF_REQ_SCALE)) {
1452 tp->snd_scale = tp->requested_s_scale;
1453 tp->rcv_scale = tp->request_r_scale;
1455 tp->rcv_adv += tp->rcv_wnd;
1456 tp->snd_una++; /* SYN is acked */
1457 tcp_callout_stop(tp, tp->tt_rexmt);
1459 * If there's data, delay ACK; if there's also a FIN
1460 * ACKNOW will be turned on later.
1462 if (DELAY_ACK(tp) && tlen != 0) {
1463 tcp_callout_reset(tp, tp->tt_delack,
1464 tcp_delacktime, tcp_timer_delack);
1466 tp->t_flags |= TF_ACKNOW;
1469 * Received <SYN,ACK> in SYN_SENT[*] state.
1471 * SYN_SENT --> ESTABLISHED
1472 * SYN_SENT* --> FIN_WAIT_1
1474 tp->t_starttime = ticks;
1475 if (tp->t_flags & TF_NEEDFIN) {
1476 tp->t_state = TCPS_FIN_WAIT_1;
1477 tp->t_flags &= ~TF_NEEDFIN;
1480 tcp_established(tp);
1484 * Received initial SYN in SYN-SENT[*] state =>
1485 * simultaneous open.
1486 * Do 3-way handshake:
1487 * SYN-SENT -> SYN-RECEIVED
1488 * SYN-SENT* -> SYN-RECEIVED*
1490 tp->t_flags |= TF_ACKNOW;
1491 tcp_callout_stop(tp, tp->tt_rexmt);
1492 tp->t_state = TCPS_SYN_RECEIVED;
1496 * Advance th->th_seq to correspond to first data byte.
1497 * If data, trim to stay within window,
1498 * dropping FIN if necessary.
1501 if (tlen > tp->rcv_wnd) {
1502 todrop = tlen - tp->rcv_wnd;
1506 tcpstat.tcps_rcvpackafterwin++;
1507 tcpstat.tcps_rcvbyteafterwin += todrop;
1509 tp->snd_wl1 = th->th_seq - 1;
1510 tp->rcv_up = th->th_seq;
1512 * Client side of transaction: already sent SYN and data.
1513 * If the remote host used T/TCP to validate the SYN,
1514 * our data will be ACK'd; if so, enter normal data segment
1515 * processing in the middle of step 5, ack processing.
1516 * Otherwise, goto step 6.
1518 if (thflags & TH_ACK)
1524 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
1525 * do normal processing (we no longer bother with T/TCP).
1529 case TCPS_TIME_WAIT:
1530 break; /* continue normal processing */
1534 * States other than LISTEN or SYN_SENT.
1535 * First check the RST flag and sequence number since reset segments
1536 * are exempt from the timestamp and connection count tests. This
1537 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
1538 * below which allowed reset segments in half the sequence space
1539 * to fall though and be processed (which gives forged reset
1540 * segments with a random sequence number a 50 percent chance of
1541 * killing a connection).
1542 * Then check timestamp, if present.
1543 * Then check the connection count, if present.
1544 * Then check that at least some bytes of segment are within
1545 * receive window. If segment begins before rcv_nxt,
1546 * drop leading data (and SYN); if nothing left, just ack.
1549 * If the RST bit is set, check the sequence number to see
1550 * if this is a valid reset segment.
1552 * In all states except SYN-SENT, all reset (RST) segments
1553 * are validated by checking their SEQ-fields. A reset is
1554 * valid if its sequence number is in the window.
1555 * Note: this does not take into account delayed ACKs, so
1556 * we should test against last_ack_sent instead of rcv_nxt.
1557 * The sequence number in the reset segment is normally an
1558 * echo of our outgoing acknowledgement numbers, but some hosts
1559 * send a reset with the sequence number at the rightmost edge
1560 * of our receive window, and we have to handle this case.
1561 * If we have multiple segments in flight, the intial reset
1562 * segment sequence numbers will be to the left of last_ack_sent,
1563 * but they will eventually catch up.
1564 * In any case, it never made sense to trim reset segments to
1565 * fit the receive window since RFC 1122 says:
1566 * 4.2.2.12 RST Segment: RFC-793 Section 3.4
1568 * A TCP SHOULD allow a received RST segment to include data.
1571 * It has been suggested that a RST segment could contain
1572 * ASCII text that encoded and explained the cause of the
1573 * RST. No standard has yet been established for such
1576 * If the reset segment passes the sequence number test examine
1578 * SYN_RECEIVED STATE:
1579 * If passive open, return to LISTEN state.
1580 * If active open, inform user that connection was refused.
1581 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
1582 * Inform user that connection was reset, and close tcb.
1583 * CLOSING, LAST_ACK STATES:
1586 * Drop the segment - see Stevens, vol. 2, p. 964 and
1589 if (thflags & TH_RST) {
1590 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
1591 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
1592 switch (tp->t_state) {
1594 case TCPS_SYN_RECEIVED:
1595 so->so_error = ECONNREFUSED;
1598 case TCPS_ESTABLISHED:
1599 case TCPS_FIN_WAIT_1:
1600 case TCPS_FIN_WAIT_2:
1601 case TCPS_CLOSE_WAIT:
1602 so->so_error = ECONNRESET;
1604 tp->t_state = TCPS_CLOSED;
1605 tcpstat.tcps_drops++;
1614 case TCPS_TIME_WAIT:
1622 * RFC 1323 PAWS: If we have a timestamp reply on this segment
1623 * and it's less than ts_recent, drop it.
1625 if ((to.to_flags & TOF_TS) && tp->ts_recent != 0 &&
1626 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
1628 /* Check to see if ts_recent is over 24 days old. */
1629 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1631 * Invalidate ts_recent. If this segment updates
1632 * ts_recent, the age will be reset later and ts_recent
1633 * will get a valid value. If it does not, setting
1634 * ts_recent to zero will at least satisfy the
1635 * requirement that zero be placed in the timestamp
1636 * echo reply when ts_recent isn't valid. The
1637 * age isn't reset until we get a valid ts_recent
1638 * because we don't want out-of-order segments to be
1639 * dropped when ts_recent is old.
1643 tcpstat.tcps_rcvduppack++;
1644 tcpstat.tcps_rcvdupbyte += tlen;
1645 tcpstat.tcps_pawsdrop++;
1653 * In the SYN-RECEIVED state, validate that the packet belongs to
1654 * this connection before trimming the data to fit the receive
1655 * window. Check the sequence number versus IRS since we know
1656 * the sequence numbers haven't wrapped. This is a partial fix
1657 * for the "LAND" DoS attack.
1659 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
1660 rstreason = BANDLIM_RST_OPENPORT;
1664 todrop = tp->rcv_nxt - th->th_seq;
1666 if (TCP_DO_SACK(tp)) {
1667 /* Report duplicate segment at head of packet. */
1668 tp->reportblk.rblk_start = th->th_seq;
1669 tp->reportblk.rblk_end = TCP_SACK_BLKEND(
1670 th->th_seq + tlen, thflags);
1671 if (SEQ_GT(tp->reportblk.rblk_end, tp->rcv_nxt))
1672 tp->reportblk.rblk_end = tp->rcv_nxt;
1673 tp->t_flags |= (TF_DUPSEG | TF_SACKLEFT | TF_ACKNOW);
1675 if (thflags & TH_SYN) {
1685 * Following if statement from Stevens, vol. 2, p. 960.
1687 if (todrop > tlen ||
1688 (todrop == tlen && !(thflags & TH_FIN))) {
1690 * Any valid FIN must be to the left of the window.
1691 * At this point the FIN must be a duplicate or out
1692 * of sequence; drop it.
1697 * Send an ACK to resynchronize and drop any data.
1698 * But keep on processing for RST or ACK.
1700 tp->t_flags |= TF_ACKNOW;
1702 tcpstat.tcps_rcvduppack++;
1703 tcpstat.tcps_rcvdupbyte += todrop;
1705 tcpstat.tcps_rcvpartduppack++;
1706 tcpstat.tcps_rcvpartdupbyte += todrop;
1708 drop_hdrlen += todrop; /* drop from the top afterwards */
1709 th->th_seq += todrop;
1711 if (th->th_urp > todrop)
1712 th->th_urp -= todrop;
1720 * If new data are received on a connection after the
1721 * user processes are gone, then RST the other end.
1723 if ((so->so_state & SS_NOFDREF) &&
1724 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1726 tcpstat.tcps_rcvafterclose++;
1727 rstreason = BANDLIM_UNLIMITED;
1732 * If segment ends after window, drop trailing data
1733 * (and PUSH and FIN); if nothing left, just ACK.
1735 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
1737 tcpstat.tcps_rcvpackafterwin++;
1738 if (todrop >= tlen) {
1739 tcpstat.tcps_rcvbyteafterwin += tlen;
1741 * If a new connection request is received
1742 * while in TIME_WAIT, drop the old connection
1743 * and start over if the sequence numbers
1744 * are above the previous ones.
1746 if (thflags & TH_SYN &&
1747 tp->t_state == TCPS_TIME_WAIT &&
1748 SEQ_GT(th->th_seq, tp->rcv_nxt)) {
1753 * If window is closed can only take segments at
1754 * window edge, and have to drop data and PUSH from
1755 * incoming segments. Continue processing, but
1756 * remember to ack. Otherwise, drop segment
1759 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1760 tp->t_flags |= TF_ACKNOW;
1761 tcpstat.tcps_rcvwinprobe++;
1765 tcpstat.tcps_rcvbyteafterwin += todrop;
1768 thflags &= ~(TH_PUSH | TH_FIN);
1772 * If last ACK falls within this segment's sequence numbers,
1773 * record its timestamp.
1775 * 1) That the test incorporates suggestions from the latest
1776 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1777 * 2) That updating only on newer timestamps interferes with
1778 * our earlier PAWS tests, so this check should be solely
1779 * predicated on the sequence space of this segment.
1780 * 3) That we modify the segment boundary check to be
1781 * Last.ACK.Sent <= SEG.SEQ + SEG.LEN
1782 * instead of RFC1323's
1783 * Last.ACK.Sent < SEG.SEQ + SEG.LEN,
1784 * This modified check allows us to overcome RFC1323's
1785 * limitations as described in Stevens TCP/IP Illustrated
1786 * Vol. 2 p.869. In such cases, we can still calculate the
1787 * RTT correctly when RCV.NXT == Last.ACK.Sent.
1789 if ((to.to_flags & TOF_TS) && SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
1790 SEQ_LEQ(tp->last_ack_sent, (th->th_seq + tlen
1791 + ((thflags & TH_SYN) != 0)
1792 + ((thflags & TH_FIN) != 0)))) {
1793 tp->ts_recent_age = ticks;
1794 tp->ts_recent = to.to_tsval;
1798 * If a SYN is in the window, then this is an
1799 * error and we send an RST and drop the connection.
1801 if (thflags & TH_SYN) {
1802 tp = tcp_drop(tp, ECONNRESET);
1803 rstreason = BANDLIM_UNLIMITED;
1808 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
1809 * flag is on (half-synchronized state), then queue data for
1810 * later processing; else drop segment and return.
1812 if (!(thflags & TH_ACK)) {
1813 if (tp->t_state == TCPS_SYN_RECEIVED ||
1814 (tp->t_flags & TF_NEEDSYN))
1823 switch (tp->t_state) {
1825 * In SYN_RECEIVED state, the ACK acknowledges our SYN, so enter
1826 * ESTABLISHED state and continue processing.
1827 * The ACK was checked above.
1829 case TCPS_SYN_RECEIVED:
1831 tcpstat.tcps_connects++;
1833 /* Do window scaling? */
1834 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
1835 (TF_RCVD_SCALE | TF_REQ_SCALE)) {
1836 tp->snd_scale = tp->requested_s_scale;
1837 tp->rcv_scale = tp->request_r_scale;
1841 * SYN-RECEIVED -> ESTABLISHED
1842 * SYN-RECEIVED* -> FIN-WAIT-1
1844 tp->t_starttime = ticks;
1845 if (tp->t_flags & TF_NEEDFIN) {
1846 tp->t_state = TCPS_FIN_WAIT_1;
1847 tp->t_flags &= ~TF_NEEDFIN;
1849 tcp_established(tp);
1852 * If segment contains data or ACK, will call tcp_reass()
1853 * later; if not, do so now to pass queued data to user.
1855 if (tlen == 0 && !(thflags & TH_FIN))
1856 tcp_reass(tp, NULL, NULL, NULL);
1860 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1861 * ACKs. If the ack is in the range
1862 * tp->snd_una < th->th_ack <= tp->snd_max
1863 * then advance tp->snd_una to th->th_ack and drop
1864 * data from the retransmission queue. If this ACK reflects
1865 * more up to date window information we update our window information.
1867 case TCPS_ESTABLISHED:
1868 case TCPS_FIN_WAIT_1:
1869 case TCPS_FIN_WAIT_2:
1870 case TCPS_CLOSE_WAIT:
1873 case TCPS_TIME_WAIT:
1875 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
1876 if (TCP_DO_SACK(tp))
1877 tcp_sack_update_scoreboard(tp, &to);
1878 if (tlen != 0 || tiwin != tp->snd_wnd) {
1882 tcpstat.tcps_rcvdupack++;
1883 if (!tcp_callout_active(tp, tp->tt_rexmt) ||
1884 th->th_ack != tp->snd_una) {
1889 * We have outstanding data (other than
1890 * a window probe), this is a completely
1891 * duplicate ack (ie, window info didn't
1892 * change), the ack is the biggest we've
1893 * seen and we've seen exactly our rexmt
1894 * threshhold of them, so assume a packet
1895 * has been dropped and retransmit it.
1896 * Kludge snd_nxt & the congestion
1897 * window so we send only this one
1900 if (IN_FASTRECOVERY(tp)) {
1901 if (TCP_DO_SACK(tp)) {
1902 /* No artifical cwnd inflation. */
1903 tcp_sack_rexmt(tp, th);
1906 * Dup acks mean that packets
1907 * have left the network
1908 * (they're now cached at the
1909 * receiver) so bump cwnd by
1910 * the amount in the receiver
1911 * to keep a constant cwnd
1912 * packets in the network.
1914 tp->snd_cwnd += tp->t_maxseg;
1917 } else if (SEQ_LT(th->th_ack, tp->snd_recover)) {
1920 } else if (++tp->t_dupacks == tcprexmtthresh) {
1921 tcp_seq old_snd_nxt;
1925 if (tcp_do_eifel_detect &&
1926 (tp->t_flags & TF_RCVD_TSTMP)) {
1927 tcp_save_congestion_state(tp);
1928 tp->t_flags |= TF_FASTREXMT;
1931 * We know we're losing at the current
1932 * window size, so do congestion avoidance:
1933 * set ssthresh to half the current window
1934 * and pull our congestion window back to the
1937 win = min(tp->snd_wnd, tp->snd_cwnd) / 2 /
1941 tp->snd_ssthresh = win * tp->t_maxseg;
1942 ENTER_FASTRECOVERY(tp);
1943 tp->snd_recover = tp->snd_max;
1944 tcp_callout_stop(tp, tp->tt_rexmt);
1946 old_snd_nxt = tp->snd_nxt;
1947 tp->snd_nxt = th->th_ack;
1948 tp->snd_cwnd = tp->t_maxseg;
1950 ++tcpstat.tcps_sndfastrexmit;
1951 tp->snd_cwnd = tp->snd_ssthresh;
1952 tp->rexmt_high = tp->snd_nxt;
1953 if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
1954 tp->snd_nxt = old_snd_nxt;
1955 KASSERT(tp->snd_limited <= 2,
1956 ("tp->snd_limited too big"));
1957 if (TCP_DO_SACK(tp))
1958 tcp_sack_rexmt(tp, th);
1960 tp->snd_cwnd += tp->t_maxseg *
1961 (tp->t_dupacks - tp->snd_limited);
1962 } else if (tcp_do_limitedtransmit) {
1963 u_long oldcwnd = tp->snd_cwnd;
1964 tcp_seq oldsndmax = tp->snd_max;
1965 tcp_seq oldsndnxt = tp->snd_nxt;
1966 /* outstanding data */
1967 uint32_t ownd = tp->snd_max - tp->snd_una;
1970 #define iceildiv(n, d) (((n)+(d)-1) / (d))
1972 KASSERT(tp->t_dupacks == 1 ||
1974 ("dupacks not 1 or 2"));
1975 if (tp->t_dupacks == 1)
1976 tp->snd_limited = 0;
1977 tp->snd_nxt = tp->snd_max;
1978 tp->snd_cwnd = ownd +
1979 (tp->t_dupacks - tp->snd_limited) *
1983 if (SEQ_LT(oldsndnxt, oldsndmax)) {
1984 KASSERT(SEQ_GEQ(oldsndnxt, tp->snd_una),
1985 ("snd_una moved in other threads"));
1986 tp->snd_nxt = oldsndnxt;
1988 tp->snd_cwnd = oldcwnd;
1989 sent = tp->snd_max - oldsndmax;
1990 if (sent > tp->t_maxseg) {
1991 KASSERT((tp->t_dupacks == 2 &&
1992 tp->snd_limited == 0) ||
1993 (sent == tp->t_maxseg + 1 &&
1994 tp->t_flags & TF_SENTFIN),
1996 KASSERT(sent <= tp->t_maxseg * 2,
1997 ("sent too many segments"));
1998 tp->snd_limited = 2;
1999 tcpstat.tcps_sndlimited += 2;
2000 } else if (sent > 0) {
2002 ++tcpstat.tcps_sndlimited;
2003 } else if (tcp_do_early_retransmit &&
2004 (tcp_do_eifel_detect &&
2005 (tp->t_flags & TF_RCVD_TSTMP)) &&
2006 ownd < 4 * tp->t_maxseg &&
2007 tp->t_dupacks + 1 >=
2008 iceildiv(ownd, tp->t_maxseg) &&
2009 (!TCP_DO_SACK(tp) ||
2010 ownd <= tp->t_maxseg ||
2011 tcp_sack_has_sacked(&tp->scb,
2012 ownd - tp->t_maxseg))) {
2013 ++tcpstat.tcps_sndearlyrexmit;
2014 tp->t_flags |= TF_EARLYREXMT;
2015 goto fastretransmit;
2021 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una"));
2023 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2025 * Detected optimistic ACK attack.
2026 * Force slow-start to de-synchronize attack.
2028 tp->snd_cwnd = tp->t_maxseg;
2031 tcpstat.tcps_rcvacktoomuch++;
2035 * If we reach this point, ACK is not a duplicate,
2036 * i.e., it ACKs something we sent.
2038 if (tp->t_flags & TF_NEEDSYN) {
2040 * T/TCP: Connection was half-synchronized, and our
2041 * SYN has been ACK'd (so connection is now fully
2042 * synchronized). Go to non-starred state,
2043 * increment snd_una for ACK of SYN, and check if
2044 * we can do window scaling.
2046 tp->t_flags &= ~TF_NEEDSYN;
2048 /* Do window scaling? */
2049 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
2050 (TF_RCVD_SCALE | TF_REQ_SCALE)) {
2051 tp->snd_scale = tp->requested_s_scale;
2052 tp->rcv_scale = tp->request_r_scale;
2057 acked = th->th_ack - tp->snd_una;
2058 tcpstat.tcps_rcvackpack++;
2059 tcpstat.tcps_rcvackbyte += acked;
2061 if (tcp_do_eifel_detect && acked > 0 &&
2062 (to.to_flags & TOF_TS) && (to.to_tsecr != 0) &&
2063 (tp->t_flags & TF_FIRSTACCACK)) {
2064 /* Eifel detection applicable. */
2065 if (to.to_tsecr < tp->t_rexmtTS) {
2066 ++tcpstat.tcps_eifeldetected;
2067 tcp_revert_congestion_state(tp);
2068 if (tp->t_rxtshift != 1 ||
2069 ticks >= tp->t_badrxtwin)
2070 ++tcpstat.tcps_rttcantdetect;
2072 } else if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) {
2074 * If we just performed our first retransmit,
2075 * and the ACK arrives within our recovery window,
2076 * then it was a mistake to do the retransmit
2077 * in the first place. Recover our original cwnd
2078 * and ssthresh, and proceed to transmit where we
2081 tcp_revert_congestion_state(tp);
2082 ++tcpstat.tcps_rttdetected;
2086 * If we have a timestamp reply, update smoothed
2087 * round trip time. If no timestamp is present but
2088 * transmit timer is running and timed sequence
2089 * number was acked, update smoothed round trip time.
2090 * Since we now have an rtt measurement, cancel the
2091 * timer backoff (cf., Phil Karn's retransmit alg.).
2092 * Recompute the initial retransmit timer.
2094 * Some machines (certain windows boxes) send broken
2095 * timestamp replies during the SYN+ACK phase, ignore
2098 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0))
2099 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1);
2100 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq))
2101 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2102 tcp_xmit_bandwidth_limit(tp, th->th_ack);
2105 * If no data (only SYN) was ACK'd,
2106 * skip rest of ACK processing.
2111 /* Stop looking for an acceptable ACK since one was received. */
2112 tp->t_flags &= ~(TF_FIRSTACCACK | TF_FASTREXMT | TF_EARLYREXMT);
2114 if (acked > so->so_snd.ssb_cc) {
2115 tp->snd_wnd -= so->so_snd.ssb_cc;
2116 sbdrop(&so->so_snd.sb, (int)so->so_snd.ssb_cc);
2117 ourfinisacked = TRUE;
2119 sbdrop(&so->so_snd.sb, acked);
2120 tp->snd_wnd -= acked;
2121 ourfinisacked = FALSE;
2126 * Update window information.
2128 if (acceptable_window_update(tp, th, tiwin)) {
2129 /* keep track of pure window updates */
2130 if (tlen == 0 && tp->snd_wl2 == th->th_ack &&
2131 tiwin > tp->snd_wnd)
2132 tcpstat.tcps_rcvwinupd++;
2133 tp->snd_wnd = tiwin;
2134 tp->snd_wl1 = th->th_seq;
2135 tp->snd_wl2 = th->th_ack;
2136 if (tp->snd_wnd > tp->max_sndwnd)
2137 tp->max_sndwnd = tp->snd_wnd;
2141 tp->snd_una = th->th_ack;
2142 if (TCP_DO_SACK(tp))
2143 tcp_sack_update_scoreboard(tp, &to);
2144 if (IN_FASTRECOVERY(tp)) {
2145 if (SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2146 EXIT_FASTRECOVERY(tp);
2149 * If the congestion window was inflated
2150 * to account for the other side's
2151 * cached packets, retract it.
2153 if (!TCP_DO_SACK(tp))
2154 tp->snd_cwnd = tp->snd_ssthresh;
2157 * Window inflation should have left us
2158 * with approximately snd_ssthresh outstanding
2159 * data. But, in case we would be inclined
2160 * to send a burst, better do it using
2163 if (SEQ_GT(th->th_ack + tp->snd_cwnd,
2164 tp->snd_max + 2 * tp->t_maxseg))
2166 (tp->snd_max - tp->snd_una) +
2171 if (TCP_DO_SACK(tp)) {
2172 tp->snd_max_rexmt = tp->snd_max;
2173 tcp_sack_rexmt(tp, th);
2175 tcp_newreno_partial_ack(tp, th, acked);
2181 * Open the congestion window. When in slow-start,
2182 * open exponentially: maxseg per packet. Otherwise,
2183 * open linearly: maxseg per window.
2185 if (tp->snd_cwnd <= tp->snd_ssthresh) {
2187 (SEQ_LT(tp->snd_nxt, tp->snd_max) ?
2188 tp->t_maxseg : 2 * tp->t_maxseg);
2191 tp->snd_cwnd += tcp_do_abc ?
2192 min(acked, abc_sslimit) : tp->t_maxseg;
2194 /* linear increase */
2195 tp->snd_wacked += tcp_do_abc ? acked :
2197 if (tp->snd_wacked >= tp->snd_cwnd) {
2198 tp->snd_wacked -= tp->snd_cwnd;
2199 tp->snd_cwnd += tp->t_maxseg;
2202 tp->snd_cwnd = min(tp->snd_cwnd,
2203 TCP_MAXWIN << tp->snd_scale);
2204 tp->snd_recover = th->th_ack - 1;
2206 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2207 tp->snd_nxt = tp->snd_una;
2210 * If all outstanding data is acked, stop retransmit
2211 * timer and remember to restart (more output or persist).
2212 * If there is more data to be acked, restart retransmit
2213 * timer, using current (possibly backed-off) value.
2215 if (th->th_ack == tp->snd_max) {
2216 tcp_callout_stop(tp, tp->tt_rexmt);
2218 } else if (!tcp_callout_active(tp, tp->tt_persist)) {
2219 tcp_callout_reset(tp, tp->tt_rexmt, tp->t_rxtcur,
2223 switch (tp->t_state) {
2225 * In FIN_WAIT_1 STATE in addition to the processing
2226 * for the ESTABLISHED state if our FIN is now acknowledged
2227 * then enter FIN_WAIT_2.
2229 case TCPS_FIN_WAIT_1:
2230 if (ourfinisacked) {
2232 * If we can't receive any more
2233 * data, then closing user can proceed.
2234 * Starting the timer is contrary to the
2235 * specification, but if we don't get a FIN
2236 * we'll hang forever.
2238 if (so->so_state & SS_CANTRCVMORE) {
2239 soisdisconnected(so);
2240 tcp_callout_reset(tp, tp->tt_2msl,
2241 tp->t_maxidle, tcp_timer_2msl);
2243 tp->t_state = TCPS_FIN_WAIT_2;
2248 * In CLOSING STATE in addition to the processing for
2249 * the ESTABLISHED state if the ACK acknowledges our FIN
2250 * then enter the TIME-WAIT state, otherwise ignore
2254 if (ourfinisacked) {
2255 tp->t_state = TCPS_TIME_WAIT;
2256 tcp_canceltimers(tp);
2257 tcp_callout_reset(tp, tp->tt_2msl,
2258 2 * tcp_rmx_msl(tp),
2260 soisdisconnected(so);
2265 * In LAST_ACK, we may still be waiting for data to drain
2266 * and/or to be acked, as well as for the ack of our FIN.
2267 * If our FIN is now acknowledged, delete the TCB,
2268 * enter the closed state and return.
2271 if (ourfinisacked) {
2278 * In TIME_WAIT state the only thing that should arrive
2279 * is a retransmission of the remote FIN. Acknowledge
2280 * it and restart the finack timer.
2282 case TCPS_TIME_WAIT:
2283 tcp_callout_reset(tp, tp->tt_2msl, 2 * tcp_rmx_msl(tp),
2291 * Update window information.
2292 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2294 if ((thflags & TH_ACK) &&
2295 acceptable_window_update(tp, th, tiwin)) {
2296 /* keep track of pure window updates */
2297 if (tlen == 0 && tp->snd_wl2 == th->th_ack &&
2298 tiwin > tp->snd_wnd)
2299 tcpstat.tcps_rcvwinupd++;
2300 tp->snd_wnd = tiwin;
2301 tp->snd_wl1 = th->th_seq;
2302 tp->snd_wl2 = th->th_ack;
2303 if (tp->snd_wnd > tp->max_sndwnd)
2304 tp->max_sndwnd = tp->snd_wnd;
2309 * Process segments with URG.
2311 if ((thflags & TH_URG) && th->th_urp &&
2312 !TCPS_HAVERCVDFIN(tp->t_state)) {
2314 * This is a kludge, but if we receive and accept
2315 * random urgent pointers, we'll crash in
2316 * soreceive. It's hard to imagine someone
2317 * actually wanting to send this much urgent data.
2319 if (th->th_urp + so->so_rcv.ssb_cc > sb_max) {
2320 th->th_urp = 0; /* XXX */
2321 thflags &= ~TH_URG; /* XXX */
2322 goto dodata; /* XXX */
2325 * If this segment advances the known urgent pointer,
2326 * then mark the data stream. This should not happen
2327 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2328 * a FIN has been received from the remote side.
2329 * In these states we ignore the URG.
2331 * According to RFC961 (Assigned Protocols),
2332 * the urgent pointer points to the last octet
2333 * of urgent data. We continue, however,
2334 * to consider it to indicate the first octet
2335 * of data past the urgent section as the original
2336 * spec states (in one of two places).
2338 if (SEQ_GT(th->th_seq + th->th_urp, tp->rcv_up)) {
2339 tp->rcv_up = th->th_seq + th->th_urp;
2340 so->so_oobmark = so->so_rcv.ssb_cc +
2341 (tp->rcv_up - tp->rcv_nxt) - 1;
2342 if (so->so_oobmark == 0)
2343 sosetstate(so, SS_RCVATMARK);
2345 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2348 * Remove out of band data so doesn't get presented to user.
2349 * This can happen independent of advancing the URG pointer,
2350 * but if two URG's are pending at once, some out-of-band
2351 * data may creep in... ick.
2353 if (th->th_urp <= (u_long)tlen &&
2354 !(so->so_options & SO_OOBINLINE)) {
2355 /* hdr drop is delayed */
2356 tcp_pulloutofband(so, th, m, drop_hdrlen);
2360 * If no out of band data is expected,
2361 * pull receive urgent pointer along
2362 * with the receive window.
2364 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2365 tp->rcv_up = tp->rcv_nxt;
2370 * Process the segment text, merging it into the TCP sequencing queue,
2371 * and arranging for acknowledgment of receipt if necessary.
2372 * This process logically involves adjusting tp->rcv_wnd as data
2373 * is presented to the user (this happens in tcp_usrreq.c,
2374 * case PRU_RCVD). If a FIN has already been received on this
2375 * connection then we just ignore the text.
2377 if ((tlen || (thflags & TH_FIN)) && !TCPS_HAVERCVDFIN(tp->t_state)) {
2378 m_adj(m, drop_hdrlen); /* delayed header drop */
2380 * Insert segment which includes th into TCP reassembly queue
2381 * with control block tp. Set thflags to whether reassembly now
2382 * includes a segment with FIN. This handles the common case
2383 * inline (segment is the next to be received on an established
2384 * connection, and the queue is empty), avoiding linkage into
2385 * and removal from the queue and repetition of various
2387 * Set DELACK for segments received in order, but ack
2388 * immediately when segments are out of order (so
2389 * fast retransmit can work).
2391 if (th->th_seq == tp->rcv_nxt &&
2392 LIST_EMPTY(&tp->t_segq) &&
2393 TCPS_HAVEESTABLISHED(tp->t_state)) {
2394 if (DELAY_ACK(tp)) {
2395 tcp_callout_reset(tp, tp->tt_delack,
2396 tcp_delacktime, tcp_timer_delack);
2398 tp->t_flags |= TF_ACKNOW;
2400 tp->rcv_nxt += tlen;
2401 thflags = th->th_flags & TH_FIN;
2402 tcpstat.tcps_rcvpack++;
2403 tcpstat.tcps_rcvbyte += tlen;
2405 if (so->so_state & SS_CANTRCVMORE) {
2408 lwkt_gettoken(&so->so_rcv.ssb_token);
2409 ssb_appendstream(&so->so_rcv, m);
2410 lwkt_reltoken(&so->so_rcv.ssb_token);
2414 if (!(tp->t_flags & TF_DUPSEG)) {
2415 /* Initialize SACK report block. */
2416 tp->reportblk.rblk_start = th->th_seq;
2417 tp->reportblk.rblk_end = TCP_SACK_BLKEND(
2418 th->th_seq + tlen, thflags);
2420 thflags = tcp_reass(tp, th, &tlen, m);
2421 tp->t_flags |= TF_ACKNOW;
2425 * Note the amount of data that peer has sent into
2426 * our window, in order to estimate the sender's
2429 len = so->so_rcv.ssb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2436 * If FIN is received ACK the FIN and let the user know
2437 * that the connection is closing.
2439 if (thflags & TH_FIN) {
2440 if (!TCPS_HAVERCVDFIN(tp->t_state)) {
2443 * If connection is half-synchronized
2444 * (ie NEEDSYN flag on) then delay ACK,
2445 * so it may be piggybacked when SYN is sent.
2446 * Otherwise, since we received a FIN then no
2447 * more input can be expected, send ACK now.
2449 if (DELAY_ACK(tp) && (tp->t_flags & TF_NEEDSYN)) {
2450 tcp_callout_reset(tp, tp->tt_delack,
2451 tcp_delacktime, tcp_timer_delack);
2453 tp->t_flags |= TF_ACKNOW;
2458 switch (tp->t_state) {
2460 * In SYN_RECEIVED and ESTABLISHED STATES
2461 * enter the CLOSE_WAIT state.
2463 case TCPS_SYN_RECEIVED:
2464 tp->t_starttime = ticks;
2466 case TCPS_ESTABLISHED:
2467 tp->t_state = TCPS_CLOSE_WAIT;
2471 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2472 * enter the CLOSING state.
2474 case TCPS_FIN_WAIT_1:
2475 tp->t_state = TCPS_CLOSING;
2479 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2480 * starting the time-wait timer, turning off the other
2483 case TCPS_FIN_WAIT_2:
2484 tp->t_state = TCPS_TIME_WAIT;
2485 tcp_canceltimers(tp);
2486 tcp_callout_reset(tp, tp->tt_2msl, 2 * tcp_rmx_msl(tp),
2488 soisdisconnected(so);
2492 * In TIME_WAIT state restart the 2 MSL time_wait timer.
2494 case TCPS_TIME_WAIT:
2495 tcp_callout_reset(tp, tp->tt_2msl, 2 * tcp_rmx_msl(tp),
2502 if (so->so_options & SO_DEBUG)
2503 tcp_trace(TA_INPUT, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2507 * Return any desired output.
2509 if (needoutput || (tp->t_flags & TF_ACKNOW))
2511 tcp_sack_report_cleanup(tp);
2512 return(IPPROTO_DONE);
2516 * Generate an ACK dropping incoming segment if it occupies
2517 * sequence space, where the ACK reflects our state.
2519 * We can now skip the test for the RST flag since all
2520 * paths to this code happen after packets containing
2521 * RST have been dropped.
2523 * In the SYN-RECEIVED state, don't send an ACK unless the
2524 * segment we received passes the SYN-RECEIVED ACK test.
2525 * If it fails send a RST. This breaks the loop in the
2526 * "LAND" DoS attack, and also prevents an ACK storm
2527 * between two listening ports that have been sent forged
2528 * SYN segments, each with the source address of the other.
2530 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
2531 (SEQ_GT(tp->snd_una, th->th_ack) ||
2532 SEQ_GT(th->th_ack, tp->snd_max)) ) {
2533 rstreason = BANDLIM_RST_OPENPORT;
2537 if (so->so_options & SO_DEBUG)
2538 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2541 tp->t_flags |= TF_ACKNOW;
2543 tcp_sack_report_cleanup(tp);
2544 return(IPPROTO_DONE);
2548 * Generate a RST, dropping incoming segment.
2549 * Make ACK acceptable to originator of segment.
2550 * Don't bother to respond if destination was broadcast/multicast.
2552 if ((thflags & TH_RST) || m->m_flags & (M_BCAST | M_MCAST))
2555 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
2556 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
2559 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
2560 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
2561 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
2562 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2565 /* IPv6 anycast check is done at tcp6_input() */
2568 * Perform bandwidth limiting.
2571 if (badport_bandlim(rstreason) < 0)
2576 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2577 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2579 if (thflags & TH_ACK)
2580 /* mtod() below is safe as long as hdr dropping is delayed */
2581 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack,
2584 if (thflags & TH_SYN)
2586 /* mtod() below is safe as long as hdr dropping is delayed */
2587 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq + tlen,
2588 (tcp_seq)0, TH_RST | TH_ACK);
2591 tcp_sack_report_cleanup(tp);
2592 return(IPPROTO_DONE);
2596 * Drop space held by incoming segment and return.
2599 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2600 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2604 tcp_sack_report_cleanup(tp);
2605 return(IPPROTO_DONE);
2609 * Parse TCP options and place in tcpopt.
2612 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, boolean_t is_syn)
2617 for (; cnt > 0; cnt -= optlen, cp += optlen) {
2619 if (opt == TCPOPT_EOL)
2621 if (opt == TCPOPT_NOP)
2627 if (optlen < 2 || optlen > cnt)
2632 if (optlen != TCPOLEN_MAXSEG)
2636 to->to_flags |= TOF_MSS;
2637 bcopy(cp + 2, &to->to_mss, sizeof to->to_mss);
2638 to->to_mss = ntohs(to->to_mss);
2641 if (optlen != TCPOLEN_WINDOW)
2645 to->to_flags |= TOF_SCALE;
2646 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
2648 case TCPOPT_TIMESTAMP:
2649 if (optlen != TCPOLEN_TIMESTAMP)
2651 to->to_flags |= TOF_TS;
2652 bcopy(cp + 2, &to->to_tsval, sizeof to->to_tsval);
2653 to->to_tsval = ntohl(to->to_tsval);
2654 bcopy(cp + 6, &to->to_tsecr, sizeof to->to_tsecr);
2655 to->to_tsecr = ntohl(to->to_tsecr);
2657 * If echoed timestamp is later than the current time,
2658 * fall back to non RFC1323 RTT calculation.
2660 if (to->to_tsecr != 0 && TSTMP_GT(to->to_tsecr, ticks))
2663 case TCPOPT_SACK_PERMITTED:
2664 if (optlen != TCPOLEN_SACK_PERMITTED)
2668 to->to_flags |= TOF_SACK_PERMITTED;
2671 if ((optlen - 2) & 0x07) /* not multiple of 8 */
2673 to->to_nsackblocks = (optlen - 2) / 8;
2674 to->to_sackblocks = (struct raw_sackblock *) (cp + 2);
2675 to->to_flags |= TOF_SACK;
2676 for (i = 0; i < to->to_nsackblocks; i++) {
2677 struct raw_sackblock *r = &to->to_sackblocks[i];
2679 r->rblk_start = ntohl(r->rblk_start);
2680 r->rblk_end = ntohl(r->rblk_end);
2682 if (SEQ_LEQ(r->rblk_end, r->rblk_start)) {
2684 * Invalid SACK block; discard all
2687 tcpstat.tcps_rcvbadsackopt++;
2688 to->to_nsackblocks = 0;
2689 to->to_sackblocks = NULL;
2690 to->to_flags &= ~TOF_SACK;
2695 #ifdef TCP_SIGNATURE
2697 * XXX In order to reply to a host which has set the
2698 * TCP_SIGNATURE option in its initial SYN, we have to
2699 * record the fact that the option was observed here
2700 * for the syncache code to perform the correct response.
2702 case TCPOPT_SIGNATURE:
2703 if (optlen != TCPOLEN_SIGNATURE)
2705 to->to_flags |= (TOF_SIGNATURE | TOF_SIGLEN);
2707 #endif /* TCP_SIGNATURE */
2715 * Pull out of band byte out of a segment so
2716 * it doesn't appear in the user's data queue.
2717 * It is still reflected in the segment length for
2718 * sequencing purposes.
2719 * "off" is the delayed to be dropped hdrlen.
2722 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, int off)
2724 int cnt = off + th->th_urp - 1;
2727 if (m->m_len > cnt) {
2728 char *cp = mtod(m, caddr_t) + cnt;
2729 struct tcpcb *tp = sototcpcb(so);
2732 tp->t_oobflags |= TCPOOB_HAVEDATA;
2733 bcopy(cp + 1, cp, m->m_len - cnt - 1);
2735 if (m->m_flags & M_PKTHDR)
2744 panic("tcp_pulloutofband");
2748 * Collect new round-trip time estimate
2749 * and update averages and current timeout.
2752 tcp_xmit_timer(struct tcpcb *tp, int rtt)
2756 tcpstat.tcps_rttupdated++;
2758 if (tp->t_srtt != 0) {
2760 * srtt is stored as fixed point with 5 bits after the
2761 * binary point (i.e., scaled by 8). The following magic
2762 * is equivalent to the smoothing algorithm in rfc793 with
2763 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2764 * point). Adjust rtt to origin 0.
2766 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
2767 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
2769 if ((tp->t_srtt += delta) <= 0)
2773 * We accumulate a smoothed rtt variance (actually, a
2774 * smoothed mean difference), then set the retransmit
2775 * timer to smoothed rtt + 4 times the smoothed variance.
2776 * rttvar is stored as fixed point with 4 bits after the
2777 * binary point (scaled by 16). The following is
2778 * equivalent to rfc793 smoothing with an alpha of .75
2779 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
2780 * rfc793's wired-in beta.
2784 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
2785 if ((tp->t_rttvar += delta) <= 0)
2787 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
2788 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2791 * No rtt measurement yet - use the unsmoothed rtt.
2792 * Set the variance to half the rtt (so our first
2793 * retransmit happens at 3*rtt).
2795 tp->t_srtt = rtt << TCP_RTT_SHIFT;
2796 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
2797 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2803 * the retransmit should happen at rtt + 4 * rttvar.
2804 * Because of the way we do the smoothing, srtt and rttvar
2805 * will each average +1/2 tick of bias. When we compute
2806 * the retransmit timer, we want 1/2 tick of rounding and
2807 * 1 extra tick because of +-1/2 tick uncertainty in the
2808 * firing of the timer. The bias will give us exactly the
2809 * 1.5 tick we need. But, because the bias is
2810 * statistical, we have to test that we don't drop below
2811 * the minimum feasible timer (which is 2 ticks).
2813 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
2814 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
2817 * We received an ack for a packet that wasn't retransmitted;
2818 * it is probably safe to discard any error indications we've
2819 * received recently. This isn't quite right, but close enough
2820 * for now (a route might have failed after we sent a segment,
2821 * and the return path might not be symmetrical).
2823 tp->t_softerror = 0;
2827 * Determine a reasonable value for maxseg size.
2828 * If the route is known, check route for mtu.
2829 * If none, use an mss that can be handled on the outgoing
2830 * interface without forcing IP to fragment; if bigger than
2831 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2832 * to utilize large mbufs. If no route is found, route has no mtu,
2833 * or the destination isn't local, use a default, hopefully conservative
2834 * size (usually 512 or the default IP max size, but no more than the mtu
2835 * of the interface), as we can't discover anything about intervening
2836 * gateways or networks. We also initialize the congestion/slow start
2837 * window to be a single segment if the destination isn't local.
2838 * While looking at the routing entry, we also initialize other path-dependent
2839 * parameters from pre-set or cached values in the routing entry.
2841 * Also take into account the space needed for options that we
2842 * send regularly. Make maxseg shorter by that amount to assure
2843 * that we can send maxseg amount of data even when the options
2844 * are present. Store the upper limit of the length of options plus
2847 * NOTE that this routine is only called when we process an incoming
2848 * segment, for outgoing segments only tcp_mssopt is called.
2851 tcp_mss(struct tcpcb *tp, int offer)
2857 struct inpcb *inp = tp->t_inpcb;
2860 boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) ? TRUE : FALSE);
2861 size_t min_protoh = isipv6 ?
2862 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
2863 sizeof(struct tcpiphdr);
2865 const boolean_t isipv6 = FALSE;
2866 const size_t min_protoh = sizeof(struct tcpiphdr);
2870 rt = tcp_rtlookup6(&inp->inp_inc);
2872 rt = tcp_rtlookup(&inp->inp_inc);
2874 tp->t_maxopd = tp->t_maxseg =
2875 (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
2879 so = inp->inp_socket;
2882 * Offer == 0 means that there was no MSS on the SYN segment,
2883 * in this case we use either the interface mtu or tcp_mssdflt.
2885 * An offer which is too large will be cut down later.
2889 if (in6_localaddr(&inp->in6p_faddr)) {
2890 offer = ND_IFINFO(rt->rt_ifp)->linkmtu -
2893 offer = tcp_v6mssdflt;
2896 if (in_localaddr(inp->inp_faddr))
2897 offer = ifp->if_mtu - min_protoh;
2899 offer = tcp_mssdflt;
2904 * Prevent DoS attack with too small MSS. Round up
2905 * to at least minmss.
2907 * Sanity check: make sure that maxopd will be large
2908 * enough to allow some data on segments even is the
2909 * all the option space is used (40bytes). Otherwise
2910 * funny things may happen in tcp_output.
2912 offer = max(offer, tcp_minmss);
2913 offer = max(offer, 64);
2915 rt->rt_rmx.rmx_mssopt = offer;
2918 * While we're here, check if there's an initial rtt
2919 * or rttvar. Convert from the route-table units
2920 * to scaled multiples of the slow timeout timer.
2922 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
2924 * XXX the lock bit for RTT indicates that the value
2925 * is also a minimum value; this is subject to time.
2927 if (rt->rt_rmx.rmx_locks & RTV_RTT)
2928 tp->t_rttmin = rtt / (RTM_RTTUNIT / hz);
2929 tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE));
2930 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
2931 tcpstat.tcps_usedrtt++;
2932 if (rt->rt_rmx.rmx_rttvar) {
2933 tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
2934 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE));
2935 tcpstat.tcps_usedrttvar++;
2937 /* default variation is +- 1 rtt */
2939 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
2941 TCPT_RANGESET(tp->t_rxtcur,
2942 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
2943 tp->t_rttmin, TCPTV_REXMTMAX);
2947 * if there's an mtu associated with the route, use it
2948 * else, use the link mtu. Take the smaller of mss or offer
2951 if (rt->rt_rmx.rmx_mtu) {
2952 mss = rt->rt_rmx.rmx_mtu - min_protoh;
2955 mss = ND_IFINFO(rt->rt_ifp)->linkmtu - min_protoh;
2957 mss = ifp->if_mtu - min_protoh;
2959 mss = min(mss, offer);
2962 * maxopd stores the maximum length of data AND options
2963 * in a segment; maxseg is the amount of data in a normal
2964 * segment. We need to store this value (maxopd) apart
2965 * from maxseg, because now every segment carries options
2966 * and thus we normally have somewhat less data in segments.
2970 if ((tp->t_flags & (TF_REQ_TSTMP | TF_NOOPT)) == TF_REQ_TSTMP &&
2971 ((tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
2972 mss -= TCPOLEN_TSTAMP_APPA;
2974 #if (MCLBYTES & (MCLBYTES - 1)) == 0
2976 mss &= ~(MCLBYTES-1);
2979 mss = mss / MCLBYTES * MCLBYTES;
2982 * If there's a pipesize, change the socket buffer
2983 * to that size. Make the socket buffers an integral
2984 * number of mss units; if the mss is larger than
2985 * the socket buffer, decrease the mss.
2988 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0)
2990 bufsize = so->so_snd.ssb_hiwat;
2994 bufsize = roundup(bufsize, mss);
2995 if (bufsize > sb_max)
2997 if (bufsize > so->so_snd.ssb_hiwat)
2998 ssb_reserve(&so->so_snd, bufsize, so, NULL);
3003 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0)
3005 bufsize = so->so_rcv.ssb_hiwat;
3006 if (bufsize > mss) {
3007 bufsize = roundup(bufsize, mss);
3008 if (bufsize > sb_max)
3010 if (bufsize > so->so_rcv.ssb_hiwat) {
3011 lwkt_gettoken(&so->so_rcv.ssb_token);
3012 ssb_reserve(&so->so_rcv, bufsize, so, NULL);
3013 lwkt_reltoken(&so->so_rcv.ssb_token);
3018 * Set the slow-start flight size
3020 * NOTE: t_maxseg must have been configured!
3022 tp->snd_cwnd = tcp_initial_window(tp);
3024 if (rt->rt_rmx.rmx_ssthresh) {
3026 * There's some sort of gateway or interface
3027 * buffer limit on the path. Use this to set
3028 * the slow start threshhold, but set the
3029 * threshold to no less than 2*mss.
3031 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
3032 tcpstat.tcps_usedssthresh++;
3037 * Determine the MSS option to send on an outgoing SYN.
3040 tcp_mssopt(struct tcpcb *tp)
3045 ((tp->t_inpcb->inp_vflag & INP_IPV6) ? TRUE : FALSE);
3046 int min_protoh = isipv6 ?
3047 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
3048 sizeof(struct tcpiphdr);
3050 const boolean_t isipv6 = FALSE;
3051 const size_t min_protoh = sizeof(struct tcpiphdr);
3055 rt = tcp_rtlookup6(&tp->t_inpcb->inp_inc);
3057 rt = tcp_rtlookup(&tp->t_inpcb->inp_inc);
3059 return (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
3061 return (rt->rt_ifp->if_mtu - min_protoh);
3065 * When a partial ack arrives, force the retransmission of the
3066 * next unacknowledged segment. Do not exit Fast Recovery.
3068 * Implement the Slow-but-Steady variant of NewReno by restarting the
3069 * the retransmission timer. Turn it off here so it can be restarted
3070 * later in tcp_output().
3073 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th, int acked)
3075 tcp_seq old_snd_nxt = tp->snd_nxt;
3076 u_long ocwnd = tp->snd_cwnd;
3078 tcp_callout_stop(tp, tp->tt_rexmt);
3080 tp->snd_nxt = th->th_ack;
3081 /* Set snd_cwnd to one segment beyond acknowledged offset. */
3082 tp->snd_cwnd = tp->t_maxseg;
3083 tp->t_flags |= TF_ACKNOW;
3085 if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
3086 tp->snd_nxt = old_snd_nxt;
3087 /* partial window deflation */
3089 tp->snd_cwnd = ocwnd - acked + tp->t_maxseg;
3091 tp->snd_cwnd = tp->t_maxseg;
3095 * In contrast to the Slow-but-Steady NewReno variant,
3096 * we do not reset the retransmission timer for SACK retransmissions,
3097 * except when retransmitting snd_una.
3100 tcp_sack_rexmt(struct tcpcb *tp, struct tcphdr *th)
3102 uint32_t pipe, seglen;
3105 tcp_seq old_snd_nxt = tp->snd_nxt;
3106 u_long ocwnd = tp->snd_cwnd;
3107 int nseg = 0; /* consecutive new segments */
3108 #define MAXBURST 4 /* limit burst of new packets on partial ack */
3111 pipe = tcp_sack_compute_pipe(tp);
3112 while ((tcp_seq_diff_t)(ocwnd - pipe) >= (tcp_seq_diff_t)tp->t_maxseg &&
3113 (!tcp_do_smartsack || nseg < MAXBURST) &&
3114 tcp_sack_nextseg(tp, &nextrexmt, &seglen, &lostdup)) {
3116 tcp_seq old_snd_max;
3119 if (nextrexmt == tp->snd_max)
3121 tp->snd_nxt = nextrexmt;
3122 tp->snd_cwnd = nextrexmt - tp->snd_una + seglen;
3123 old_snd_max = tp->snd_max;
3124 if (nextrexmt == tp->snd_una)
3125 tcp_callout_stop(tp, tp->tt_rexmt);
3126 error = tcp_output(tp);
3129 sent = tp->snd_nxt - nextrexmt;
3134 tcpstat.tcps_sndsackpack++;
3135 tcpstat.tcps_sndsackbyte += sent;
3136 if (SEQ_LT(nextrexmt, old_snd_max) &&
3137 SEQ_LT(tp->rexmt_high, tp->snd_nxt))
3138 tp->rexmt_high = seq_min(tp->snd_nxt, old_snd_max);
3140 if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
3141 tp->snd_nxt = old_snd_nxt;
3142 tp->snd_cwnd = ocwnd;
3146 * Reset idle time and keep-alive timer, typically called when a valid
3147 * tcp packet is received but may also be called when FASTKEEP is set
3148 * to prevent the previous long-timeout from calculating to a drop.
3150 * Only update t_rcvtime for non-SYN packets.
3152 * Handle the case where one side thinks the connection is established
3153 * but the other side has, say, rebooted without cleaning out the
3154 * connection. The SYNs could be construed as an attack and wind
3155 * up ignored, but in case it isn't an attack we can validate the
3156 * connection by forcing a keepalive.
3159 tcp_timer_keep_activity(struct tcpcb *tp, int thflags)
3161 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
3162 if ((thflags & (TH_SYN | TH_ACK)) == TH_SYN) {
3163 tp->t_flags |= TF_KEEPALIVE;
3164 tcp_callout_reset(tp, tp->tt_keep, hz / 2,
3167 tp->t_rcvtime = ticks;
3168 tp->t_flags &= ~TF_KEEPALIVE;
3169 tcp_callout_reset(tp, tp->tt_keep,
3170 tcp_getkeepidle(tp),
3177 tcp_rmx_msl(const struct tcpcb *tp)
3180 struct inpcb *inp = tp->t_inpcb;
3183 boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) ? TRUE : FALSE);
3185 const boolean_t isipv6 = FALSE;
3189 rt = tcp_rtlookup6(&inp->inp_inc);
3191 rt = tcp_rtlookup(&inp->inp_inc);
3192 if (rt == NULL || rt->rt_rmx.rmx_msl == 0)
3195 msl = (rt->rt_rmx.rmx_msl * hz) / 1000;
3203 tcp_established(struct tcpcb *tp)
3205 tp->t_state = TCPS_ESTABLISHED;
3206 tcp_callout_reset(tp, tp->tt_keep, tcp_getkeepidle(tp), tcp_timer_keep);
3208 if (tp->t_rxtsyn > 0) {
3211 * "If the timer expires awaiting the ACK of a SYN segment
3212 * and the TCP implementation is using an RTO less than 3
3213 * seconds, the RTO MUST be re-initialized to 3 seconds
3214 * when data transmission begins"
3216 if (tp->t_rxtcur < TCPTV_RTOBASE3)
3217 tp->t_rxtcur = TCPTV_RTOBASE3;