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 int tcp_do_rfc3390 = 1;
171 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW,
173 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
175 static int tcp_do_eifel_detect = 1;
176 SYSCTL_INT(_net_inet_tcp, OID_AUTO, eifel, CTLFLAG_RW,
177 &tcp_do_eifel_detect, 0, "Eifel detection algorithm (RFC 3522)");
179 static int tcp_do_abc = 1;
180 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc, CTLFLAG_RW,
182 "TCP Appropriate Byte Counting (RFC 3465)");
185 * Define as tunable for easy testing with SACK on and off.
186 * Warning: do not change setting in the middle of an existing active TCP flow,
187 * else strange things might happen to that flow.
190 SYSCTL_INT(_net_inet_tcp, OID_AUTO, sack, CTLFLAG_RW,
191 &tcp_do_sack, 0, "Enable SACK Algorithms");
193 int tcp_do_smartsack = 1;
194 SYSCTL_INT(_net_inet_tcp, OID_AUTO, smartsack, CTLFLAG_RW,
195 &tcp_do_smartsack, 0, "Enable Smart SACK Algorithms");
197 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0,
198 "TCP Segment Reassembly Queue");
200 int tcp_reass_maxseg = 0;
201 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RD,
202 &tcp_reass_maxseg, 0,
203 "Global maximum number of TCP Segments in Reassembly Queue");
205 int tcp_reass_qsize = 0;
206 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD,
208 "Global number of TCP Segments currently in Reassembly Queue");
210 static int tcp_reass_overflows = 0;
211 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD,
212 &tcp_reass_overflows, 0,
213 "Global number of TCP Segment Reassembly Queue Overflows");
215 int tcp_do_autorcvbuf = 1;
216 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_RW,
217 &tcp_do_autorcvbuf, 0, "Enable automatic receive buffer sizing");
219 int tcp_autorcvbuf_inc = 16*1024;
220 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_RW,
221 &tcp_autorcvbuf_inc, 0,
222 "Incrementor step size of automatic receive buffer");
224 int tcp_autorcvbuf_max = 2*1024*1024;
225 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_RW,
226 &tcp_autorcvbuf_max, 0, "Max size of automatic receive buffer");
228 int tcp_sosnd_agglim = 2;
229 SYSCTL_INT(_net_inet_tcp, OID_AUTO, sosnd_agglim, CTLFLAG_RW,
230 &tcp_sosnd_agglim, 0, "TCP sosend mbuf aggregation limit");
232 int tcp_sosnd_async = 1;
233 SYSCTL_INT(_net_inet_tcp, OID_AUTO, sosnd_async, CTLFLAG_RW,
234 &tcp_sosnd_async, 0, "TCP asynchronized pru_send");
236 static void tcp_dooptions(struct tcpopt *, u_char *, int, boolean_t);
237 static void tcp_pulloutofband(struct socket *,
238 struct tcphdr *, struct mbuf *, int);
239 static int tcp_reass(struct tcpcb *, struct tcphdr *, int *,
241 static void tcp_xmit_timer(struct tcpcb *, int);
242 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *, int);
243 static void tcp_sack_rexmt(struct tcpcb *, struct tcphdr *);
244 static int tcp_rmx_msl(const struct tcpcb *);
245 static void tcp_established(struct tcpcb *);
247 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
249 #define ND6_HINT(tp) \
251 if ((tp) && (tp)->t_inpcb && \
252 ((tp)->t_inpcb->inp_vflag & INP_IPV6) && \
253 (tp)->t_inpcb->in6p_route.ro_rt) \
254 nd6_nud_hint((tp)->t_inpcb->in6p_route.ro_rt, NULL, 0); \
261 * Indicate whether this ack should be delayed. We can delay the ack if
262 * - delayed acks are enabled and
263 * - there is no delayed ack timer in progress and
264 * - our last ack wasn't a 0-sized window. We never want to delay
265 * the ack that opens up a 0-sized window.
267 #define DELAY_ACK(tp) \
268 (tcp_delack_enabled && !tcp_callout_pending(tp, tp->tt_delack) && \
269 !(tp->t_flags & TF_RXWIN0SENT))
271 #define acceptable_window_update(tp, th, tiwin) \
272 (SEQ_LT(tp->snd_wl1, th->th_seq) || \
273 (tp->snd_wl1 == th->th_seq && \
274 (SEQ_LT(tp->snd_wl2, th->th_ack) || \
275 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))
278 tcp_reass(struct tcpcb *tp, struct tcphdr *th, int *tlenp, struct mbuf *m)
281 struct tseg_qent *p = NULL;
282 struct tseg_qent *te;
283 struct socket *so = tp->t_inpcb->inp_socket;
287 * Call with th == NULL after become established to
288 * force pre-ESTABLISHED data up to user socket.
294 * Limit the number of segments in the reassembly queue to prevent
295 * holding on to too many segments (and thus running out of mbufs).
296 * Make sure to let the missing segment through which caused this
297 * queue. Always keep one global queue entry spare to be able to
298 * process the missing segment.
300 if (th->th_seq != tp->rcv_nxt &&
301 tcp_reass_qsize + 1 >= tcp_reass_maxseg) {
302 tcp_reass_overflows++;
303 tcpstat.tcps_rcvmemdrop++;
305 /* no SACK block to report */
306 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
310 /* Allocate a new queue entry. */
311 te = kmalloc(sizeof(struct tseg_qent), M_TSEGQ, M_INTWAIT | M_NULLOK);
313 tcpstat.tcps_rcvmemdrop++;
315 /* no SACK block to report */
316 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
319 atomic_add_int(&tcp_reass_qsize, 1);
322 * Find a segment which begins after this one does.
324 LIST_FOREACH(q, &tp->t_segq, tqe_q) {
325 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
331 * If there is a preceding segment, it may provide some of
332 * our data already. If so, drop the data from the incoming
333 * segment. If it provides all of our data, drop us.
338 /* conversion to int (in i) handles seq wraparound */
339 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
340 if (i > 0) { /* overlaps preceding segment */
341 tp->t_flags |= (TF_DUPSEG | TF_ENCLOSESEG);
342 /* enclosing block starts w/ preceding segment */
343 tp->encloseblk.rblk_start = p->tqe_th->th_seq;
345 /* preceding encloses incoming segment */
346 tp->encloseblk.rblk_end = TCP_SACK_BLKEND(
347 p->tqe_th->th_seq + p->tqe_len,
348 p->tqe_th->th_flags);
349 tcpstat.tcps_rcvduppack++;
350 tcpstat.tcps_rcvdupbyte += *tlenp;
353 atomic_add_int(&tcp_reass_qsize, -1);
355 * Try to present any queued data
356 * at the left window edge to the user.
357 * This is needed after the 3-WHS
360 goto present; /* ??? */
365 /* incoming segment end is enclosing block end */
366 tp->encloseblk.rblk_end = TCP_SACK_BLKEND(
367 th->th_seq + *tlenp, th->th_flags);
368 /* trim end of reported D-SACK block */
369 tp->reportblk.rblk_end = th->th_seq;
372 tcpstat.tcps_rcvoopack++;
373 tcpstat.tcps_rcvoobyte += *tlenp;
376 * While we overlap succeeding segments trim them or,
377 * if they are completely covered, dequeue them.
380 tcp_seq_diff_t i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
381 tcp_seq qend = q->tqe_th->th_seq + q->tqe_len;
382 tcp_seq qend_sack = TCP_SACK_BLKEND(qend, q->tqe_th->th_flags);
383 struct tseg_qent *nq;
387 if (!(tp->t_flags & TF_DUPSEG)) { /* first time through */
388 tp->t_flags |= (TF_DUPSEG | TF_ENCLOSESEG);
389 tp->encloseblk = tp->reportblk;
390 /* report trailing duplicate D-SACK segment */
391 tp->reportblk.rblk_start = q->tqe_th->th_seq;
393 if ((tp->t_flags & TF_ENCLOSESEG) &&
394 SEQ_GT(qend_sack, tp->encloseblk.rblk_end)) {
395 /* extend enclosing block if one exists */
396 tp->encloseblk.rblk_end = qend_sack;
398 if (i < q->tqe_len) {
399 q->tqe_th->th_seq += i;
405 nq = LIST_NEXT(q, tqe_q);
406 LIST_REMOVE(q, tqe_q);
409 atomic_add_int(&tcp_reass_qsize, -1);
413 /* Insert the new segment queue entry into place. */
416 te->tqe_len = *tlenp;
418 /* check if can coalesce with following segment */
419 if (q != NULL && (th->th_seq + *tlenp == q->tqe_th->th_seq)) {
420 tcp_seq tend = te->tqe_th->th_seq + te->tqe_len;
421 tcp_seq tend_sack = TCP_SACK_BLKEND(tend, te->tqe_th->th_flags);
423 te->tqe_len += q->tqe_len;
424 if (q->tqe_th->th_flags & TH_FIN)
425 te->tqe_th->th_flags |= TH_FIN;
426 m_cat(te->tqe_m, q->tqe_m);
427 tp->encloseblk.rblk_end = tend_sack;
429 * When not reporting a duplicate segment, use
430 * the larger enclosing block as the SACK block.
432 if (!(tp->t_flags & TF_DUPSEG))
433 tp->reportblk.rblk_end = tend_sack;
434 LIST_REMOVE(q, tqe_q);
436 atomic_add_int(&tcp_reass_qsize, -1);
440 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
442 /* check if can coalesce with preceding segment */
443 if (p->tqe_th->th_seq + p->tqe_len == th->th_seq) {
444 p->tqe_len += te->tqe_len;
445 m_cat(p->tqe_m, te->tqe_m);
446 tp->encloseblk.rblk_start = p->tqe_th->th_seq;
448 * When not reporting a duplicate segment, use
449 * the larger enclosing block as the SACK block.
451 if (!(tp->t_flags & TF_DUPSEG))
452 tp->reportblk.rblk_start = p->tqe_th->th_seq;
454 atomic_add_int(&tcp_reass_qsize, -1);
456 LIST_INSERT_AFTER(p, te, tqe_q);
462 * Present data to user, advancing rcv_nxt through
463 * completed sequence space.
465 if (!TCPS_HAVEESTABLISHED(tp->t_state))
467 q = LIST_FIRST(&tp->t_segq);
468 if (q == NULL || q->tqe_th->th_seq != tp->rcv_nxt)
470 tp->rcv_nxt += q->tqe_len;
471 if (!(tp->t_flags & TF_DUPSEG)) {
472 /* no SACK block to report since ACK advanced */
473 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
475 /* no enclosing block to report since ACK advanced */
476 tp->t_flags &= ~TF_ENCLOSESEG;
477 flags = q->tqe_th->th_flags & TH_FIN;
478 LIST_REMOVE(q, tqe_q);
479 KASSERT(LIST_EMPTY(&tp->t_segq) ||
480 LIST_FIRST(&tp->t_segq)->tqe_th->th_seq != tp->rcv_nxt,
481 ("segment not coalesced"));
482 if (so->so_state & SS_CANTRCVMORE) {
485 lwkt_gettoken(&so->so_rcv.ssb_token);
486 ssb_appendstream(&so->so_rcv, q->tqe_m);
487 lwkt_reltoken(&so->so_rcv.ssb_token);
490 atomic_add_int(&tcp_reass_qsize, -1);
497 * TCP input routine, follows pages 65-76 of the
498 * protocol specification dated September, 1981 very closely.
502 tcp6_input(struct mbuf **mp, int *offp, int proto)
504 struct mbuf *m = *mp;
505 struct in6_ifaddr *ia6;
507 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
510 * draft-itojun-ipv6-tcp-to-anycast
511 * better place to put this in?
513 ia6 = ip6_getdstifaddr(m);
514 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
517 ip6 = mtod(m, struct ip6_hdr *);
518 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
519 offsetof(struct ip6_hdr, ip6_dst));
520 return (IPPROTO_DONE);
523 tcp_input(mp, offp, proto);
524 return (IPPROTO_DONE);
529 tcp_input(struct mbuf **mp, int *offp, int proto)
533 struct ip *ip = NULL;
535 struct inpcb *inp = NULL;
541 struct tcpcb *tp = NULL;
543 struct socket *so = NULL;
545 boolean_t ourfinisacked, needoutput = FALSE;
548 struct tcpopt to; /* options in this segment */
549 struct sockaddr_in *next_hop = NULL;
550 int rstreason; /* For badport_bandlim accounting purposes */
552 struct ip6_hdr *ip6 = NULL;
557 const boolean_t isipv6 = FALSE;
567 tcpstat.tcps_rcvtotal++;
569 if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
572 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
573 KKASSERT(mtag != NULL);
574 next_hop = m_tag_data(mtag);
578 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? TRUE : FALSE;
582 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */
583 ip6 = mtod(m, struct ip6_hdr *);
584 tlen = (sizeof *ip6) + ntohs(ip6->ip6_plen) - off0;
585 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) {
586 tcpstat.tcps_rcvbadsum++;
589 th = (struct tcphdr *)((caddr_t)ip6 + off0);
592 * Be proactive about unspecified IPv6 address in source.
593 * As we use all-zero to indicate unbounded/unconnected pcb,
594 * unspecified IPv6 address can be used to confuse us.
596 * Note that packets with unspecified IPv6 destination is
597 * already dropped in ip6_input.
599 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
605 * Get IP and TCP header together in first mbuf.
606 * Note: IP leaves IP header in first mbuf.
608 if (off0 > sizeof(struct ip)) {
610 off0 = sizeof(struct ip);
612 /* already checked and pulled up in ip_demux() */
613 KASSERT(m->m_len >= sizeof(struct tcpiphdr),
614 ("TCP header not in one mbuf: m->m_len %d", m->m_len));
615 ip = mtod(m, struct ip *);
616 ipov = (struct ipovly *)ip;
617 th = (struct tcphdr *)((caddr_t)ip + off0);
620 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
621 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
622 th->th_sum = m->m_pkthdr.csum_data;
624 th->th_sum = in_pseudo(ip->ip_src.s_addr,
626 htonl(m->m_pkthdr.csum_data +
629 th->th_sum ^= 0xffff;
632 * Checksum extended TCP header and data.
634 len = sizeof(struct ip) + tlen;
635 bzero(ipov->ih_x1, sizeof ipov->ih_x1);
636 ipov->ih_len = (u_short)tlen;
637 ipov->ih_len = htons(ipov->ih_len);
638 th->th_sum = in_cksum(m, len);
641 tcpstat.tcps_rcvbadsum++;
645 /* Re-initialization for later version check */
646 ip->ip_v = IPVERSION;
651 * Check that TCP offset makes sense,
652 * pull out TCP options and adjust length. XXX
654 off = th->th_off << 2;
655 /* already checked and pulled up in ip_demux() */
656 KASSERT(off >= sizeof(struct tcphdr) && off <= tlen,
657 ("bad TCP data offset %d (tlen %d)", off, tlen));
658 tlen -= off; /* tlen is used instead of ti->ti_len */
659 if (off > sizeof(struct tcphdr)) {
661 IP6_EXTHDR_CHECK(m, off0, off, IPPROTO_DONE);
662 ip6 = mtod(m, struct ip6_hdr *);
663 th = (struct tcphdr *)((caddr_t)ip6 + off0);
665 /* already pulled up in ip_demux() */
666 KASSERT(m->m_len >= sizeof(struct ip) + off,
667 ("TCP header and options not in one mbuf: "
668 "m_len %d, off %d", m->m_len, off));
670 optlen = off - sizeof(struct tcphdr);
671 optp = (u_char *)(th + 1);
673 thflags = th->th_flags;
675 #ifdef TCP_DROP_SYNFIN
677 * If the drop_synfin option is enabled, drop all packets with
678 * both the SYN and FIN bits set. This prevents e.g. nmap from
679 * identifying the TCP/IP stack.
681 * This is a violation of the TCP specification.
683 if (drop_synfin && (thflags & (TH_SYN | TH_FIN)) == (TH_SYN | TH_FIN))
688 * Convert TCP protocol specific fields to host format.
690 th->th_seq = ntohl(th->th_seq);
691 th->th_ack = ntohl(th->th_ack);
692 th->th_win = ntohs(th->th_win);
693 th->th_urp = ntohs(th->th_urp);
696 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options,
697 * until after ip6_savecontrol() is called and before other functions
698 * which don't want those proto headers.
699 * Because ip6_savecontrol() is going to parse the mbuf to
700 * search for data to be passed up to user-land, it wants mbuf
701 * parameters to be unchanged.
702 * XXX: the call of ip6_savecontrol() has been obsoleted based on
703 * latest version of the advanced API (20020110).
705 drop_hdrlen = off0 + off;
708 * Locate pcb for segment.
711 /* IPFIREWALL_FORWARD section */
712 if (next_hop != NULL && !isipv6) { /* IPv6 support is not there yet */
714 * Transparently forwarded. Pretend to be the destination.
715 * already got one like this?
717 cpu = mycpu->gd_cpuid;
718 inp = in_pcblookup_hash(&tcbinfo[cpu],
719 ip->ip_src, th->th_sport,
720 ip->ip_dst, th->th_dport,
721 0, m->m_pkthdr.rcvif);
724 * It's new. Try to find the ambushing socket.
728 * The rest of the ipfw code stores the port in
730 * (The IP address is still in network order.)
732 in_port_t dport = next_hop->sin_port ?
733 htons(next_hop->sin_port) :
736 cpu = tcp_addrcpu(ip->ip_src.s_addr, th->th_sport,
737 next_hop->sin_addr.s_addr, dport);
738 inp = in_pcblookup_hash(&tcbinfo[cpu],
739 ip->ip_src, th->th_sport,
740 next_hop->sin_addr, dport,
741 1, m->m_pkthdr.rcvif);
745 inp = in6_pcblookup_hash(&tcbinfo[0],
746 &ip6->ip6_src, th->th_sport,
747 &ip6->ip6_dst, th->th_dport,
748 1, m->m_pkthdr.rcvif);
750 cpu = mycpu->gd_cpuid;
751 inp = in_pcblookup_hash(&tcbinfo[cpu],
752 ip->ip_src, th->th_sport,
753 ip->ip_dst, th->th_dport,
754 1, m->m_pkthdr.rcvif);
759 * If the state is CLOSED (i.e., TCB does not exist) then
760 * all data in the incoming segment is discarded.
761 * If the TCB exists but is in CLOSED state, it is embryonic,
762 * but should either do a listen or a connect soon.
767 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2];
769 char dbuf[sizeof "aaa.bbb.ccc.ddd"];
770 char sbuf[sizeof "aaa.bbb.ccc.ddd"];
774 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst));
777 strcat(sbuf, ip6_sprintf(&ip6->ip6_src));
780 strcpy(dbuf, inet_ntoa(ip->ip_dst));
781 strcpy(sbuf, inet_ntoa(ip->ip_src));
783 switch (log_in_vain) {
785 if (!(thflags & TH_SYN))
789 "Connection attempt to TCP %s:%d "
790 "from %s:%d flags:0x%02x\n",
791 dbuf, ntohs(th->th_dport), sbuf,
792 ntohs(th->th_sport), thflags);
801 if (thflags & TH_SYN)
810 rstreason = BANDLIM_RST_CLOSEDPORT;
816 if (ipsec6_in_reject_so(m, inp->inp_socket)) {
817 ipsec6stat.in_polvio++;
821 if (ipsec4_in_reject_so(m, inp->inp_socket)) {
822 ipsecstat.in_polvio++;
829 if (ipsec6_in_reject(m, inp))
832 if (ipsec4_in_reject(m, inp))
836 /* Check the minimum TTL for socket. */
838 if ((isipv6 ? ip6->ip6_hlim : ip->ip_ttl) < inp->inp_ip_minttl)
844 rstreason = BANDLIM_RST_CLOSEDPORT;
847 if (tp->t_state <= TCPS_CLOSED)
850 /* Unscale the window into a 32-bit value. */
851 if (!(thflags & TH_SYN))
852 tiwin = th->th_win << tp->snd_scale;
856 so = inp->inp_socket;
859 if (so->so_options & SO_DEBUG) {
860 ostate = tp->t_state;
862 bcopy(ip6, tcp_saveipgen, sizeof(*ip6));
864 bcopy(ip, tcp_saveipgen, sizeof(*ip));
869 bzero(&to, sizeof to);
871 if (so->so_options & SO_ACCEPTCONN) {
872 struct in_conninfo inc;
875 inc.inc_isipv6 = (isipv6 == TRUE);
878 inc.inc6_faddr = ip6->ip6_src;
879 inc.inc6_laddr = ip6->ip6_dst;
880 inc.inc6_route.ro_rt = NULL; /* XXX */
882 inc.inc_faddr = ip->ip_src;
883 inc.inc_laddr = ip->ip_dst;
884 inc.inc_route.ro_rt = NULL; /* XXX */
886 inc.inc_fport = th->th_sport;
887 inc.inc_lport = th->th_dport;
890 * If the state is LISTEN then ignore segment if it contains
891 * a RST. If the segment contains an ACK then it is bad and
892 * send a RST. If it does not contain a SYN then it is not
893 * interesting; drop it.
895 * If the state is SYN_RECEIVED (syncache) and seg contains
896 * an ACK, but not for our SYN/ACK, send a RST. If the seg
897 * contains a RST, check the sequence number to see if it
898 * is a valid reset segment.
900 if ((thflags & (TH_RST | TH_ACK | TH_SYN)) != TH_SYN) {
901 if ((thflags & (TH_RST | TH_ACK | TH_SYN)) == TH_ACK) {
902 if (!syncache_expand(&inc, th, &so, m)) {
904 * No syncache entry, or ACK was not
905 * for our SYN/ACK. Send a RST.
907 tcpstat.tcps_badsyn++;
908 rstreason = BANDLIM_RST_OPENPORT;
913 * Could not complete 3-way handshake,
914 * connection is being closed down, and
915 * syncache will free mbuf.
918 return(IPPROTO_DONE);
921 * We must be in the correct protocol thread
922 * for this connection.
924 KKASSERT(so->so_port == &curthread->td_msgport);
927 * Socket is created in state SYN_RECEIVED.
928 * Continue processing segment.
933 * This is what would have happened in
934 * tcp_output() when the SYN,ACK was sent.
936 tp->snd_up = tp->snd_una;
937 tp->snd_max = tp->snd_nxt = tp->iss + 1;
938 tp->last_ack_sent = tp->rcv_nxt;
940 * XXX possible bug - it doesn't appear that tp->snd_wnd is unscaled
941 * until the _second_ ACK is received:
942 * rcv SYN (set wscale opts) --> send SYN/ACK, set snd_wnd = window.
943 * rcv ACK, calculate tiwin --> process SYN_RECEIVED, determine wscale,
944 * move to ESTAB, set snd_wnd to tiwin.
946 tp->snd_wnd = tiwin; /* unscaled */
949 if (thflags & TH_RST) {
950 syncache_chkrst(&inc, th);
953 if (thflags & TH_ACK) {
954 syncache_badack(&inc);
955 tcpstat.tcps_badsyn++;
956 rstreason = BANDLIM_RST_OPENPORT;
963 * Segment's flags are (SYN) or (SYN | FIN).
967 * If deprecated address is forbidden,
968 * we do not accept SYN to deprecated interface
969 * address to prevent any new inbound connection from
970 * getting established.
971 * When we do not accept SYN, we send a TCP RST,
972 * with deprecated source address (instead of dropping
973 * it). We compromise it as it is much better for peer
974 * to send a RST, and RST will be the final packet
977 * If we do not forbid deprecated addresses, we accept
978 * the SYN packet. RFC2462 does not suggest dropping
980 * If we decipher RFC2462 5.5.4, it says like this:
981 * 1. use of deprecated addr with existing
982 * communication is okay - "SHOULD continue to be
984 * 2. use of it with new communication:
985 * (2a) "SHOULD NOT be used if alternate address
986 * with sufficient scope is available"
987 * (2b) nothing mentioned otherwise.
988 * Here we fall into (2b) case as we have no choice in
989 * our source address selection - we must obey the peer.
991 * The wording in RFC2462 is confusing, and there are
992 * multiple description text for deprecated address
993 * handling - worse, they are not exactly the same.
994 * I believe 5.5.4 is the best one, so we follow 5.5.4.
996 if (isipv6 && !ip6_use_deprecated) {
997 struct in6_ifaddr *ia6;
999 if ((ia6 = ip6_getdstifaddr(m)) &&
1000 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1002 rstreason = BANDLIM_RST_OPENPORT;
1008 * If it is from this socket, drop it, it must be forged.
1009 * Don't bother responding if the destination was a broadcast.
1011 if (th->th_dport == th->th_sport) {
1013 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
1017 if (ip->ip_dst.s_addr == ip->ip_src.s_addr)
1022 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
1024 * Note that it is quite possible to receive unicast
1025 * link-layer packets with a broadcast IP address. Use
1026 * in_broadcast() to find them.
1028 if (m->m_flags & (M_BCAST | M_MCAST))
1031 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1032 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
1035 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1036 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1037 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1038 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
1042 * SYN appears to be valid; create compressed TCP state
1043 * for syncache, or perform t/tcp connection.
1045 if (so->so_qlen <= so->so_qlimit) {
1046 tcp_dooptions(&to, optp, optlen, TRUE);
1047 if (!syncache_add(&inc, &to, th, so, m))
1051 * Entry added to syncache, mbuf used to
1052 * send SYN,ACK packet.
1054 return(IPPROTO_DONE);
1061 * Should not happen - syncache should pick up these connections.
1063 * Once we are past handling listen sockets we must be in the
1064 * correct protocol processing thread.
1066 KASSERT(tp->t_state != TCPS_LISTEN, ("tcp_input: TCPS_LISTEN state"));
1067 KKASSERT(so->so_port == &curthread->td_msgport);
1070 * This is the second part of the MSS DoS prevention code (after
1071 * minmss on the sending side) and it deals with too many too small
1072 * tcp packets in a too short timeframe (1 second).
1074 * XXX Removed. This code was crap. It does not scale to network
1075 * speed, and default values break NFS. Gone.
1080 * Segment received on connection.
1082 * Reset idle time and keep-alive timer. Don't waste time if less
1083 * then a second has elapsed.
1085 if ((int)(ticks - tp->t_rcvtime) > hz)
1086 tcp_timer_keep_activity(tp, thflags);
1090 * XXX this is tradtitional behavior, may need to be cleaned up.
1092 tcp_dooptions(&to, optp, optlen, (thflags & TH_SYN) != 0);
1093 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1094 if (to.to_flags & TOF_SCALE) {
1095 tp->t_flags |= TF_RCVD_SCALE;
1096 tp->requested_s_scale = to.to_requested_s_scale;
1098 if (to.to_flags & TOF_TS) {
1099 tp->t_flags |= TF_RCVD_TSTMP;
1100 tp->ts_recent = to.to_tsval;
1101 tp->ts_recent_age = ticks;
1103 if (!(to.to_flags & TOF_MSS))
1105 tcp_mss(tp, to.to_mss);
1107 * Only set the TF_SACK_PERMITTED per-connection flag
1108 * if we got a SACK_PERMITTED option from the other side
1109 * and the global tcp_do_sack variable is true.
1111 if (tcp_do_sack && (to.to_flags & TOF_SACK_PERMITTED))
1112 tp->t_flags |= TF_SACK_PERMITTED;
1116 * Header prediction: check for the two common cases
1117 * of a uni-directional data xfer. If the packet has
1118 * no control flags, is in-sequence, the window didn't
1119 * change and we're not retransmitting, it's a
1120 * candidate. If the length is zero and the ack moved
1121 * forward, we're the sender side of the xfer. Just
1122 * free the data acked & wake any higher level process
1123 * that was blocked waiting for space. If the length
1124 * is non-zero and the ack didn't move, we're the
1125 * receiver side. If we're getting packets in-order
1126 * (the reassembly queue is empty), add the data to
1127 * the socket buffer and note that we need a delayed ack.
1128 * Make sure that the hidden state-flags are also off.
1129 * Since we check for TCPS_ESTABLISHED above, it can only
1132 if (tp->t_state == TCPS_ESTABLISHED &&
1133 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1134 !(tp->t_flags & (TF_NEEDSYN | TF_NEEDFIN)) &&
1135 (!(to.to_flags & TOF_TS) ||
1136 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) &&
1137 th->th_seq == tp->rcv_nxt &&
1138 tp->snd_nxt == tp->snd_max) {
1141 * If last ACK falls within this segment's sequence numbers,
1142 * record the timestamp.
1143 * NOTE that the test is modified according to the latest
1144 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1146 if ((to.to_flags & TOF_TS) &&
1147 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1148 tp->ts_recent_age = ticks;
1149 tp->ts_recent = to.to_tsval;
1153 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1154 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1155 tp->snd_cwnd >= tp->snd_wnd &&
1156 !IN_FASTRECOVERY(tp)) {
1158 * This is a pure ack for outstanding data.
1160 ++tcpstat.tcps_predack;
1162 * "bad retransmit" recovery
1164 * If Eifel detection applies, then
1165 * it is deterministic, so use it
1166 * unconditionally over the old heuristic.
1167 * Otherwise, fall back to the old heuristic.
1169 if (tcp_do_eifel_detect &&
1170 (to.to_flags & TOF_TS) && to.to_tsecr &&
1171 (tp->t_flags & TF_FIRSTACCACK)) {
1172 /* Eifel detection applicable. */
1173 if (to.to_tsecr < tp->t_rexmtTS) {
1174 tcp_revert_congestion_state(tp);
1175 ++tcpstat.tcps_eifeldetected;
1176 if (tp->t_rxtshift != 1 ||
1177 ticks >= tp->t_badrxtwin)
1178 ++tcpstat.tcps_rttcantdetect;
1180 } else if (tp->t_rxtshift == 1 &&
1181 ticks < tp->t_badrxtwin) {
1182 tcp_revert_congestion_state(tp);
1183 ++tcpstat.tcps_rttdetected;
1185 tp->t_flags &= ~(TF_FIRSTACCACK |
1186 TF_FASTREXMT | TF_EARLYREXMT);
1188 * Recalculate the retransmit timer / rtt.
1190 * Some machines (certain windows boxes)
1191 * send broken timestamp replies during the
1192 * SYN+ACK phase, ignore timestamps of 0.
1194 if ((to.to_flags & TOF_TS) && to.to_tsecr) {
1196 ticks - to.to_tsecr + 1);
1197 } else if (tp->t_rtttime &&
1198 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1200 ticks - tp->t_rtttime);
1202 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1203 acked = th->th_ack - tp->snd_una;
1204 tcpstat.tcps_rcvackpack++;
1205 tcpstat.tcps_rcvackbyte += acked;
1206 sbdrop(&so->so_snd.sb, acked);
1207 tp->snd_recover = th->th_ack - 1;
1208 tp->snd_una = th->th_ack;
1211 * Update window information.
1213 if (tiwin != tp->snd_wnd &&
1214 acceptable_window_update(tp, th, tiwin)) {
1215 /* keep track of pure window updates */
1216 if (tp->snd_wl2 == th->th_ack &&
1217 tiwin > tp->snd_wnd)
1218 tcpstat.tcps_rcvwinupd++;
1219 tp->snd_wnd = tiwin;
1220 tp->snd_wl1 = th->th_seq;
1221 tp->snd_wl2 = th->th_ack;
1222 if (tp->snd_wnd > tp->max_sndwnd)
1223 tp->max_sndwnd = tp->snd_wnd;
1226 ND6_HINT(tp); /* some progress has been done */
1228 * If all outstanding data are acked, stop
1229 * retransmit timer, otherwise restart timer
1230 * using current (possibly backed-off) value.
1231 * If process is waiting for space,
1232 * wakeup/selwakeup/signal. If data
1233 * are ready to send, let tcp_output
1234 * decide between more output or persist.
1236 if (tp->snd_una == tp->snd_max) {
1237 tcp_callout_stop(tp, tp->tt_rexmt);
1238 } else if (!tcp_callout_active(tp,
1240 tcp_callout_reset(tp, tp->tt_rexmt,
1241 tp->t_rxtcur, tcp_timer_rexmt);
1244 if (so->so_snd.ssb_cc > 0)
1246 return(IPPROTO_DONE);
1248 } else if (tiwin == tp->snd_wnd &&
1249 th->th_ack == tp->snd_una &&
1250 LIST_EMPTY(&tp->t_segq) &&
1251 tlen <= ssb_space(&so->so_rcv)) {
1252 u_long newsize = 0; /* automatic sockbuf scaling */
1254 * This is a pure, in-sequence data packet
1255 * with nothing on the reassembly queue and
1256 * we have enough buffer space to take it.
1258 ++tcpstat.tcps_preddat;
1259 tp->rcv_nxt += tlen;
1260 tcpstat.tcps_rcvpack++;
1261 tcpstat.tcps_rcvbyte += tlen;
1262 ND6_HINT(tp); /* some progress has been done */
1264 * Automatic sizing of receive socket buffer. Often the send
1265 * buffer size is not optimally adjusted to the actual network
1266 * conditions at hand (delay bandwidth product). Setting the
1267 * buffer size too small limits throughput on links with high
1268 * bandwidth and high delay (eg. trans-continental/oceanic links).
1270 * On the receive side the socket buffer memory is only rarely
1271 * used to any significant extent. This allows us to be much
1272 * more aggressive in scaling the receive socket buffer. For
1273 * the case that the buffer space is actually used to a large
1274 * extent and we run out of kernel memory we can simply drop
1275 * the new segments; TCP on the sender will just retransmit it
1276 * later. Setting the buffer size too big may only consume too
1277 * much kernel memory if the application doesn't read() from
1278 * the socket or packet loss or reordering makes use of the
1281 * The criteria to step up the receive buffer one notch are:
1282 * 1. the number of bytes received during the time it takes
1283 * one timestamp to be reflected back to us (the RTT);
1284 * 2. received bytes per RTT is within seven eighth of the
1285 * current socket buffer size;
1286 * 3. receive buffer size has not hit maximal automatic size;
1288 * This algorithm does one step per RTT at most and only if
1289 * we receive a bulk stream w/o packet losses or reorderings.
1290 * Shrinking the buffer during idle times is not necessary as
1291 * it doesn't consume any memory when idle.
1293 * TODO: Only step up if the application is actually serving
1294 * the buffer to better manage the socket buffer resources.
1296 if (tcp_do_autorcvbuf &&
1298 (so->so_rcv.ssb_flags & SSB_AUTOSIZE)) {
1299 if (to.to_tsecr > tp->rfbuf_ts &&
1300 to.to_tsecr - tp->rfbuf_ts < hz) {
1302 (so->so_rcv.ssb_hiwat / 8 * 7) &&
1303 so->so_rcv.ssb_hiwat <
1304 tcp_autorcvbuf_max) {
1306 ulmin(so->so_rcv.ssb_hiwat +
1308 tcp_autorcvbuf_max);
1310 /* Start over with next RTT. */
1314 tp->rfbuf_cnt += tlen; /* add up */
1317 * Add data to socket buffer.
1319 if (so->so_state & SS_CANTRCVMORE) {
1323 * Set new socket buffer size, give up when
1326 * Adjusting the size can mess up ACK
1327 * sequencing when pure window updates are
1328 * being avoided (which is the default),
1331 lwkt_gettoken(&so->so_rcv.ssb_token);
1333 tp->t_flags |= TF_RXRESIZED;
1334 if (!ssb_reserve(&so->so_rcv, newsize,
1336 atomic_clear_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE);
1339 (TCP_MAXWIN << tp->rcv_scale)) {
1340 atomic_clear_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE);
1343 m_adj(m, drop_hdrlen); /* delayed header drop */
1344 ssb_appendstream(&so->so_rcv, m);
1345 lwkt_reltoken(&so->so_rcv.ssb_token);
1349 * This code is responsible for most of the ACKs
1350 * the TCP stack sends back after receiving a data
1351 * packet. Note that the DELAY_ACK check fails if
1352 * the delack timer is already running, which results
1353 * in an ack being sent every other packet (which is
1356 * We then further aggregate acks by not actually
1357 * sending one until the protocol thread has completed
1358 * processing the current backlog of packets. This
1359 * does not delay the ack any further, but allows us
1360 * to take advantage of the packet aggregation that
1361 * high speed NICs do (usually blocks of 8-10 packets)
1362 * to send a single ack rather then four or five acks,
1363 * greatly reducing the ack rate, the return channel
1364 * bandwidth, and the protocol overhead on both ends.
1366 * Since this also has the effect of slowing down
1367 * the exponential slow-start ramp-up, systems with
1368 * very large bandwidth-delay products might want
1369 * to turn the feature off.
1371 if (DELAY_ACK(tp)) {
1372 tcp_callout_reset(tp, tp->tt_delack,
1373 tcp_delacktime, tcp_timer_delack);
1374 } else if (tcp_aggregate_acks) {
1375 tp->t_flags |= TF_ACKNOW;
1376 if (!(tp->t_flags & TF_ONOUTPUTQ)) {
1377 tp->t_flags |= TF_ONOUTPUTQ;
1378 tp->tt_cpu = mycpu->gd_cpuid;
1380 &tcpcbackq[tp->tt_cpu],
1384 tp->t_flags |= TF_ACKNOW;
1387 return(IPPROTO_DONE);
1392 * Calculate amount of space in receive window,
1393 * and then do TCP input processing.
1394 * Receive window is amount of space in rcv queue,
1395 * but not less than advertised window.
1397 recvwin = ssb_space(&so->so_rcv);
1400 tp->rcv_wnd = imax(recvwin, (int)(tp->rcv_adv - tp->rcv_nxt));
1402 /* Reset receive buffer auto scaling when not in bulk receive mode. */
1406 switch (tp->t_state) {
1408 * If the state is SYN_RECEIVED:
1409 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1411 case TCPS_SYN_RECEIVED:
1412 if ((thflags & TH_ACK) &&
1413 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1414 SEQ_GT(th->th_ack, tp->snd_max))) {
1415 rstreason = BANDLIM_RST_OPENPORT;
1421 * If the state is SYN_SENT:
1422 * if seg contains an ACK, but not for our SYN, drop the input.
1423 * if seg contains a RST, then drop the connection.
1424 * if seg does not contain SYN, then drop it.
1425 * Otherwise this is an acceptable SYN segment
1426 * initialize tp->rcv_nxt and tp->irs
1427 * if seg contains ack then advance tp->snd_una
1428 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1429 * arrange for segment to be acked (eventually)
1430 * continue processing rest of data/controls, beginning with URG
1433 if ((thflags & TH_ACK) &&
1434 (SEQ_LEQ(th->th_ack, tp->iss) ||
1435 SEQ_GT(th->th_ack, tp->snd_max))) {
1436 rstreason = BANDLIM_UNLIMITED;
1439 if (thflags & TH_RST) {
1440 if (thflags & TH_ACK)
1441 tp = tcp_drop(tp, ECONNREFUSED);
1444 if (!(thflags & TH_SYN))
1446 tp->snd_wnd = th->th_win; /* initial send window */
1448 tp->irs = th->th_seq;
1450 if (thflags & TH_ACK) {
1451 /* Our SYN was acked. */
1452 tcpstat.tcps_connects++;
1454 /* Do window scaling on this connection? */
1455 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
1456 (TF_RCVD_SCALE | TF_REQ_SCALE)) {
1457 tp->snd_scale = tp->requested_s_scale;
1458 tp->rcv_scale = tp->request_r_scale;
1460 tp->rcv_adv += tp->rcv_wnd;
1461 tp->snd_una++; /* SYN is acked */
1462 tcp_callout_stop(tp, tp->tt_rexmt);
1464 * If there's data, delay ACK; if there's also a FIN
1465 * ACKNOW will be turned on later.
1467 if (DELAY_ACK(tp) && tlen != 0) {
1468 tcp_callout_reset(tp, tp->tt_delack,
1469 tcp_delacktime, tcp_timer_delack);
1471 tp->t_flags |= TF_ACKNOW;
1474 * Received <SYN,ACK> in SYN_SENT[*] state.
1476 * SYN_SENT --> ESTABLISHED
1477 * SYN_SENT* --> FIN_WAIT_1
1479 tp->t_starttime = ticks;
1480 if (tp->t_flags & TF_NEEDFIN) {
1481 tp->t_state = TCPS_FIN_WAIT_1;
1482 tp->t_flags &= ~TF_NEEDFIN;
1485 tcp_established(tp);
1489 * Received initial SYN in SYN-SENT[*] state =>
1490 * simultaneous open.
1491 * Do 3-way handshake:
1492 * SYN-SENT -> SYN-RECEIVED
1493 * SYN-SENT* -> SYN-RECEIVED*
1495 tp->t_flags |= TF_ACKNOW;
1496 tcp_callout_stop(tp, tp->tt_rexmt);
1497 tp->t_state = TCPS_SYN_RECEIVED;
1501 * Advance th->th_seq to correspond to first data byte.
1502 * If data, trim to stay within window,
1503 * dropping FIN if necessary.
1506 if (tlen > tp->rcv_wnd) {
1507 todrop = tlen - tp->rcv_wnd;
1511 tcpstat.tcps_rcvpackafterwin++;
1512 tcpstat.tcps_rcvbyteafterwin += todrop;
1514 tp->snd_wl1 = th->th_seq - 1;
1515 tp->rcv_up = th->th_seq;
1517 * Client side of transaction: already sent SYN and data.
1518 * If the remote host used T/TCP to validate the SYN,
1519 * our data will be ACK'd; if so, enter normal data segment
1520 * processing in the middle of step 5, ack processing.
1521 * Otherwise, goto step 6.
1523 if (thflags & TH_ACK)
1529 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
1530 * do normal processing (we no longer bother with T/TCP).
1534 case TCPS_TIME_WAIT:
1535 break; /* continue normal processing */
1539 * States other than LISTEN or SYN_SENT.
1540 * First check the RST flag and sequence number since reset segments
1541 * are exempt from the timestamp and connection count tests. This
1542 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
1543 * below which allowed reset segments in half the sequence space
1544 * to fall though and be processed (which gives forged reset
1545 * segments with a random sequence number a 50 percent chance of
1546 * killing a connection).
1547 * Then check timestamp, if present.
1548 * Then check the connection count, if present.
1549 * Then check that at least some bytes of segment are within
1550 * receive window. If segment begins before rcv_nxt,
1551 * drop leading data (and SYN); if nothing left, just ack.
1554 * If the RST bit is set, check the sequence number to see
1555 * if this is a valid reset segment.
1557 * In all states except SYN-SENT, all reset (RST) segments
1558 * are validated by checking their SEQ-fields. A reset is
1559 * valid if its sequence number is in the window.
1560 * Note: this does not take into account delayed ACKs, so
1561 * we should test against last_ack_sent instead of rcv_nxt.
1562 * The sequence number in the reset segment is normally an
1563 * echo of our outgoing acknowledgement numbers, but some hosts
1564 * send a reset with the sequence number at the rightmost edge
1565 * of our receive window, and we have to handle this case.
1566 * If we have multiple segments in flight, the intial reset
1567 * segment sequence numbers will be to the left of last_ack_sent,
1568 * but they will eventually catch up.
1569 * In any case, it never made sense to trim reset segments to
1570 * fit the receive window since RFC 1122 says:
1571 * 4.2.2.12 RST Segment: RFC-793 Section 3.4
1573 * A TCP SHOULD allow a received RST segment to include data.
1576 * It has been suggested that a RST segment could contain
1577 * ASCII text that encoded and explained the cause of the
1578 * RST. No standard has yet been established for such
1581 * If the reset segment passes the sequence number test examine
1583 * SYN_RECEIVED STATE:
1584 * If passive open, return to LISTEN state.
1585 * If active open, inform user that connection was refused.
1586 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
1587 * Inform user that connection was reset, and close tcb.
1588 * CLOSING, LAST_ACK STATES:
1591 * Drop the segment - see Stevens, vol. 2, p. 964 and
1594 if (thflags & TH_RST) {
1595 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
1596 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
1597 switch (tp->t_state) {
1599 case TCPS_SYN_RECEIVED:
1600 so->so_error = ECONNREFUSED;
1603 case TCPS_ESTABLISHED:
1604 case TCPS_FIN_WAIT_1:
1605 case TCPS_FIN_WAIT_2:
1606 case TCPS_CLOSE_WAIT:
1607 so->so_error = ECONNRESET;
1609 tp->t_state = TCPS_CLOSED;
1610 tcpstat.tcps_drops++;
1619 case TCPS_TIME_WAIT:
1627 * RFC 1323 PAWS: If we have a timestamp reply on this segment
1628 * and it's less than ts_recent, drop it.
1630 if ((to.to_flags & TOF_TS) && tp->ts_recent != 0 &&
1631 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
1633 /* Check to see if ts_recent is over 24 days old. */
1634 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1636 * Invalidate ts_recent. If this segment updates
1637 * ts_recent, the age will be reset later and ts_recent
1638 * will get a valid value. If it does not, setting
1639 * ts_recent to zero will at least satisfy the
1640 * requirement that zero be placed in the timestamp
1641 * echo reply when ts_recent isn't valid. The
1642 * age isn't reset until we get a valid ts_recent
1643 * because we don't want out-of-order segments to be
1644 * dropped when ts_recent is old.
1648 tcpstat.tcps_rcvduppack++;
1649 tcpstat.tcps_rcvdupbyte += tlen;
1650 tcpstat.tcps_pawsdrop++;
1658 * In the SYN-RECEIVED state, validate that the packet belongs to
1659 * this connection before trimming the data to fit the receive
1660 * window. Check the sequence number versus IRS since we know
1661 * the sequence numbers haven't wrapped. This is a partial fix
1662 * for the "LAND" DoS attack.
1664 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
1665 rstreason = BANDLIM_RST_OPENPORT;
1669 todrop = tp->rcv_nxt - th->th_seq;
1671 if (TCP_DO_SACK(tp)) {
1672 /* Report duplicate segment at head of packet. */
1673 tp->reportblk.rblk_start = th->th_seq;
1674 tp->reportblk.rblk_end = TCP_SACK_BLKEND(
1675 th->th_seq + tlen, thflags);
1676 if (SEQ_GT(tp->reportblk.rblk_end, tp->rcv_nxt))
1677 tp->reportblk.rblk_end = tp->rcv_nxt;
1678 tp->t_flags |= (TF_DUPSEG | TF_SACKLEFT | TF_ACKNOW);
1680 if (thflags & TH_SYN) {
1690 * Following if statement from Stevens, vol. 2, p. 960.
1692 if (todrop > tlen ||
1693 (todrop == tlen && !(thflags & TH_FIN))) {
1695 * Any valid FIN must be to the left of the window.
1696 * At this point the FIN must be a duplicate or out
1697 * of sequence; drop it.
1702 * Send an ACK to resynchronize and drop any data.
1703 * But keep on processing for RST or ACK.
1705 tp->t_flags |= TF_ACKNOW;
1707 tcpstat.tcps_rcvduppack++;
1708 tcpstat.tcps_rcvdupbyte += todrop;
1710 tcpstat.tcps_rcvpartduppack++;
1711 tcpstat.tcps_rcvpartdupbyte += todrop;
1713 drop_hdrlen += todrop; /* drop from the top afterwards */
1714 th->th_seq += todrop;
1716 if (th->th_urp > todrop)
1717 th->th_urp -= todrop;
1725 * If new data are received on a connection after the
1726 * user processes are gone, then RST the other end.
1728 if ((so->so_state & SS_NOFDREF) &&
1729 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1731 tcpstat.tcps_rcvafterclose++;
1732 rstreason = BANDLIM_UNLIMITED;
1737 * If segment ends after window, drop trailing data
1738 * (and PUSH and FIN); if nothing left, just ACK.
1740 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
1742 tcpstat.tcps_rcvpackafterwin++;
1743 if (todrop >= tlen) {
1744 tcpstat.tcps_rcvbyteafterwin += tlen;
1746 * If a new connection request is received
1747 * while in TIME_WAIT, drop the old connection
1748 * and start over if the sequence numbers
1749 * are above the previous ones.
1751 if (thflags & TH_SYN &&
1752 tp->t_state == TCPS_TIME_WAIT &&
1753 SEQ_GT(th->th_seq, tp->rcv_nxt)) {
1758 * If window is closed can only take segments at
1759 * window edge, and have to drop data and PUSH from
1760 * incoming segments. Continue processing, but
1761 * remember to ack. Otherwise, drop segment
1764 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1765 tp->t_flags |= TF_ACKNOW;
1766 tcpstat.tcps_rcvwinprobe++;
1770 tcpstat.tcps_rcvbyteafterwin += todrop;
1773 thflags &= ~(TH_PUSH | TH_FIN);
1777 * If last ACK falls within this segment's sequence numbers,
1778 * record its timestamp.
1780 * 1) That the test incorporates suggestions from the latest
1781 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1782 * 2) That updating only on newer timestamps interferes with
1783 * our earlier PAWS tests, so this check should be solely
1784 * predicated on the sequence space of this segment.
1785 * 3) That we modify the segment boundary check to be
1786 * Last.ACK.Sent <= SEG.SEQ + SEG.LEN
1787 * instead of RFC1323's
1788 * Last.ACK.Sent < SEG.SEQ + SEG.LEN,
1789 * This modified check allows us to overcome RFC1323's
1790 * limitations as described in Stevens TCP/IP Illustrated
1791 * Vol. 2 p.869. In such cases, we can still calculate the
1792 * RTT correctly when RCV.NXT == Last.ACK.Sent.
1794 if ((to.to_flags & TOF_TS) && SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
1795 SEQ_LEQ(tp->last_ack_sent, (th->th_seq + tlen
1796 + ((thflags & TH_SYN) != 0)
1797 + ((thflags & TH_FIN) != 0)))) {
1798 tp->ts_recent_age = ticks;
1799 tp->ts_recent = to.to_tsval;
1803 * If a SYN is in the window, then this is an
1804 * error and we send an RST and drop the connection.
1806 if (thflags & TH_SYN) {
1807 tp = tcp_drop(tp, ECONNRESET);
1808 rstreason = BANDLIM_UNLIMITED;
1813 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
1814 * flag is on (half-synchronized state), then queue data for
1815 * later processing; else drop segment and return.
1817 if (!(thflags & TH_ACK)) {
1818 if (tp->t_state == TCPS_SYN_RECEIVED ||
1819 (tp->t_flags & TF_NEEDSYN))
1828 switch (tp->t_state) {
1830 * In SYN_RECEIVED state, the ACK acknowledges our SYN, so enter
1831 * ESTABLISHED state and continue processing.
1832 * The ACK was checked above.
1834 case TCPS_SYN_RECEIVED:
1836 tcpstat.tcps_connects++;
1838 /* Do window scaling? */
1839 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
1840 (TF_RCVD_SCALE | TF_REQ_SCALE)) {
1841 tp->snd_scale = tp->requested_s_scale;
1842 tp->rcv_scale = tp->request_r_scale;
1846 * SYN-RECEIVED -> ESTABLISHED
1847 * SYN-RECEIVED* -> FIN-WAIT-1
1849 tp->t_starttime = ticks;
1850 if (tp->t_flags & TF_NEEDFIN) {
1851 tp->t_state = TCPS_FIN_WAIT_1;
1852 tp->t_flags &= ~TF_NEEDFIN;
1854 tcp_established(tp);
1857 * If segment contains data or ACK, will call tcp_reass()
1858 * later; if not, do so now to pass queued data to user.
1860 if (tlen == 0 && !(thflags & TH_FIN))
1861 tcp_reass(tp, NULL, NULL, NULL);
1865 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1866 * ACKs. If the ack is in the range
1867 * tp->snd_una < th->th_ack <= tp->snd_max
1868 * then advance tp->snd_una to th->th_ack and drop
1869 * data from the retransmission queue. If this ACK reflects
1870 * more up to date window information we update our window information.
1872 case TCPS_ESTABLISHED:
1873 case TCPS_FIN_WAIT_1:
1874 case TCPS_FIN_WAIT_2:
1875 case TCPS_CLOSE_WAIT:
1878 case TCPS_TIME_WAIT:
1880 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
1881 if (TCP_DO_SACK(tp))
1882 tcp_sack_update_scoreboard(tp, &to);
1883 if (tlen != 0 || tiwin != tp->snd_wnd) {
1887 tcpstat.tcps_rcvdupack++;
1888 if (!tcp_callout_active(tp, tp->tt_rexmt) ||
1889 th->th_ack != tp->snd_una) {
1894 * We have outstanding data (other than
1895 * a window probe), this is a completely
1896 * duplicate ack (ie, window info didn't
1897 * change), the ack is the biggest we've
1898 * seen and we've seen exactly our rexmt
1899 * threshhold of them, so assume a packet
1900 * has been dropped and retransmit it.
1901 * Kludge snd_nxt & the congestion
1902 * window so we send only this one
1905 if (IN_FASTRECOVERY(tp)) {
1906 if (TCP_DO_SACK(tp)) {
1907 /* No artifical cwnd inflation. */
1908 tcp_sack_rexmt(tp, th);
1911 * Dup acks mean that packets
1912 * have left the network
1913 * (they're now cached at the
1914 * receiver) so bump cwnd by
1915 * the amount in the receiver
1916 * to keep a constant cwnd
1917 * packets in the network.
1919 tp->snd_cwnd += tp->t_maxseg;
1922 } else if (SEQ_LT(th->th_ack, tp->snd_recover)) {
1925 } else if (++tp->t_dupacks == tcprexmtthresh) {
1926 tcp_seq old_snd_nxt;
1930 if (tcp_do_eifel_detect &&
1931 (tp->t_flags & TF_RCVD_TSTMP)) {
1932 tcp_save_congestion_state(tp);
1933 tp->t_flags |= TF_FASTREXMT;
1936 * We know we're losing at the current
1937 * window size, so do congestion avoidance:
1938 * set ssthresh to half the current window
1939 * and pull our congestion window back to the
1942 win = min(tp->snd_wnd, tp->snd_cwnd) / 2 /
1946 tp->snd_ssthresh = win * tp->t_maxseg;
1947 ENTER_FASTRECOVERY(tp);
1948 tp->snd_recover = tp->snd_max;
1949 tcp_callout_stop(tp, tp->tt_rexmt);
1951 old_snd_nxt = tp->snd_nxt;
1952 tp->snd_nxt = th->th_ack;
1953 tp->snd_cwnd = tp->t_maxseg;
1955 ++tcpstat.tcps_sndfastrexmit;
1956 tp->snd_cwnd = tp->snd_ssthresh;
1957 tp->rexmt_high = tp->snd_nxt;
1958 if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
1959 tp->snd_nxt = old_snd_nxt;
1960 KASSERT(tp->snd_limited <= 2,
1961 ("tp->snd_limited too big"));
1962 if (TCP_DO_SACK(tp))
1963 tcp_sack_rexmt(tp, th);
1965 tp->snd_cwnd += tp->t_maxseg *
1966 (tp->t_dupacks - tp->snd_limited);
1967 } else if (tcp_do_limitedtransmit) {
1968 u_long oldcwnd = tp->snd_cwnd;
1969 tcp_seq oldsndmax = tp->snd_max;
1970 tcp_seq oldsndnxt = tp->snd_nxt;
1971 /* outstanding data */
1972 uint32_t ownd = tp->snd_max - tp->snd_una;
1975 #define iceildiv(n, d) (((n)+(d)-1) / (d))
1977 KASSERT(tp->t_dupacks == 1 ||
1979 ("dupacks not 1 or 2"));
1980 if (tp->t_dupacks == 1)
1981 tp->snd_limited = 0;
1982 tp->snd_nxt = tp->snd_max;
1983 tp->snd_cwnd = ownd +
1984 (tp->t_dupacks - tp->snd_limited) *
1989 * Other acks may have been processed,
1990 * snd_nxt cannot be reset to a value less
1993 if (SEQ_LT(oldsndnxt, oldsndmax)) {
1994 if (SEQ_GT(oldsndnxt, tp->snd_una))
1995 tp->snd_nxt = oldsndnxt;
1997 tp->snd_nxt = tp->snd_una;
1999 tp->snd_cwnd = oldcwnd;
2000 sent = tp->snd_max - oldsndmax;
2001 if (sent > tp->t_maxseg) {
2002 KASSERT((tp->t_dupacks == 2 &&
2003 tp->snd_limited == 0) ||
2004 (sent == tp->t_maxseg + 1 &&
2005 tp->t_flags & TF_SENTFIN),
2007 KASSERT(sent <= tp->t_maxseg * 2,
2008 ("sent too many segments"));
2009 tp->snd_limited = 2;
2010 tcpstat.tcps_sndlimited += 2;
2011 } else if (sent > 0) {
2013 ++tcpstat.tcps_sndlimited;
2014 } else if (tcp_do_early_retransmit &&
2015 (tcp_do_eifel_detect &&
2016 (tp->t_flags & TF_RCVD_TSTMP)) &&
2017 ownd < 4 * tp->t_maxseg &&
2018 tp->t_dupacks + 1 >=
2019 iceildiv(ownd, tp->t_maxseg) &&
2020 (!TCP_DO_SACK(tp) ||
2021 ownd <= tp->t_maxseg ||
2022 tcp_sack_has_sacked(&tp->scb,
2023 ownd - tp->t_maxseg))) {
2024 ++tcpstat.tcps_sndearlyrexmit;
2025 tp->t_flags |= TF_EARLYREXMT;
2026 goto fastretransmit;
2032 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una"));
2034 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2036 * Detected optimistic ACK attack.
2037 * Force slow-start to de-synchronize attack.
2039 tp->snd_cwnd = tp->t_maxseg;
2042 tcpstat.tcps_rcvacktoomuch++;
2046 * If we reach this point, ACK is not a duplicate,
2047 * i.e., it ACKs something we sent.
2049 if (tp->t_flags & TF_NEEDSYN) {
2051 * T/TCP: Connection was half-synchronized, and our
2052 * SYN has been ACK'd (so connection is now fully
2053 * synchronized). Go to non-starred state,
2054 * increment snd_una for ACK of SYN, and check if
2055 * we can do window scaling.
2057 tp->t_flags &= ~TF_NEEDSYN;
2059 /* Do window scaling? */
2060 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
2061 (TF_RCVD_SCALE | TF_REQ_SCALE)) {
2062 tp->snd_scale = tp->requested_s_scale;
2063 tp->rcv_scale = tp->request_r_scale;
2068 acked = th->th_ack - tp->snd_una;
2069 tcpstat.tcps_rcvackpack++;
2070 tcpstat.tcps_rcvackbyte += acked;
2072 if (tcp_do_eifel_detect && acked > 0 &&
2073 (to.to_flags & TOF_TS) && (to.to_tsecr != 0) &&
2074 (tp->t_flags & TF_FIRSTACCACK)) {
2075 /* Eifel detection applicable. */
2076 if (to.to_tsecr < tp->t_rexmtTS) {
2077 ++tcpstat.tcps_eifeldetected;
2078 tcp_revert_congestion_state(tp);
2079 if (tp->t_rxtshift != 1 ||
2080 ticks >= tp->t_badrxtwin)
2081 ++tcpstat.tcps_rttcantdetect;
2083 } else if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) {
2085 * If we just performed our first retransmit,
2086 * and the ACK arrives within our recovery window,
2087 * then it was a mistake to do the retransmit
2088 * in the first place. Recover our original cwnd
2089 * and ssthresh, and proceed to transmit where we
2092 tcp_revert_congestion_state(tp);
2093 ++tcpstat.tcps_rttdetected;
2097 * If we have a timestamp reply, update smoothed
2098 * round trip time. If no timestamp is present but
2099 * transmit timer is running and timed sequence
2100 * number was acked, update smoothed round trip time.
2101 * Since we now have an rtt measurement, cancel the
2102 * timer backoff (cf., Phil Karn's retransmit alg.).
2103 * Recompute the initial retransmit timer.
2105 * Some machines (certain windows boxes) send broken
2106 * timestamp replies during the SYN+ACK phase, ignore
2109 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0))
2110 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1);
2111 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq))
2112 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2113 tcp_xmit_bandwidth_limit(tp, th->th_ack);
2116 * If no data (only SYN) was ACK'd,
2117 * skip rest of ACK processing.
2122 /* Stop looking for an acceptable ACK since one was received. */
2123 tp->t_flags &= ~(TF_FIRSTACCACK | TF_FASTREXMT | TF_EARLYREXMT);
2125 if (acked > so->so_snd.ssb_cc) {
2126 tp->snd_wnd -= so->so_snd.ssb_cc;
2127 sbdrop(&so->so_snd.sb, (int)so->so_snd.ssb_cc);
2128 ourfinisacked = TRUE;
2130 sbdrop(&so->so_snd.sb, acked);
2131 tp->snd_wnd -= acked;
2132 ourfinisacked = FALSE;
2137 * Update window information.
2139 if (acceptable_window_update(tp, th, tiwin)) {
2140 /* keep track of pure window updates */
2141 if (tlen == 0 && tp->snd_wl2 == th->th_ack &&
2142 tiwin > tp->snd_wnd)
2143 tcpstat.tcps_rcvwinupd++;
2144 tp->snd_wnd = tiwin;
2145 tp->snd_wl1 = th->th_seq;
2146 tp->snd_wl2 = th->th_ack;
2147 if (tp->snd_wnd > tp->max_sndwnd)
2148 tp->max_sndwnd = tp->snd_wnd;
2152 tp->snd_una = th->th_ack;
2153 if (TCP_DO_SACK(tp))
2154 tcp_sack_update_scoreboard(tp, &to);
2155 if (IN_FASTRECOVERY(tp)) {
2156 if (SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2157 EXIT_FASTRECOVERY(tp);
2160 * If the congestion window was inflated
2161 * to account for the other side's
2162 * cached packets, retract it.
2164 if (!TCP_DO_SACK(tp))
2165 tp->snd_cwnd = tp->snd_ssthresh;
2168 * Window inflation should have left us
2169 * with approximately snd_ssthresh outstanding
2170 * data. But, in case we would be inclined
2171 * to send a burst, better do it using
2174 if (SEQ_GT(th->th_ack + tp->snd_cwnd,
2175 tp->snd_max + 2 * tp->t_maxseg))
2177 (tp->snd_max - tp->snd_una) +
2182 if (TCP_DO_SACK(tp)) {
2183 tp->snd_max_rexmt = tp->snd_max;
2184 tcp_sack_rexmt(tp, th);
2186 tcp_newreno_partial_ack(tp, th, acked);
2192 * Open the congestion window. When in slow-start,
2193 * open exponentially: maxseg per packet. Otherwise,
2194 * open linearly: maxseg per window.
2196 if (tp->snd_cwnd <= tp->snd_ssthresh) {
2198 (SEQ_LT(tp->snd_nxt, tp->snd_max) ?
2199 tp->t_maxseg : 2 * tp->t_maxseg);
2202 tp->snd_cwnd += tcp_do_abc ?
2203 min(acked, abc_sslimit) : tp->t_maxseg;
2205 /* linear increase */
2206 tp->snd_wacked += tcp_do_abc ? acked :
2208 if (tp->snd_wacked >= tp->snd_cwnd) {
2209 tp->snd_wacked -= tp->snd_cwnd;
2210 tp->snd_cwnd += tp->t_maxseg;
2213 tp->snd_cwnd = min(tp->snd_cwnd,
2214 TCP_MAXWIN << tp->snd_scale);
2215 tp->snd_recover = th->th_ack - 1;
2217 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2218 tp->snd_nxt = tp->snd_una;
2221 * If all outstanding data is acked, stop retransmit
2222 * timer and remember to restart (more output or persist).
2223 * If there is more data to be acked, restart retransmit
2224 * timer, using current (possibly backed-off) value.
2226 if (th->th_ack == tp->snd_max) {
2227 tcp_callout_stop(tp, tp->tt_rexmt);
2229 } else if (!tcp_callout_active(tp, tp->tt_persist)) {
2230 tcp_callout_reset(tp, tp->tt_rexmt, tp->t_rxtcur,
2234 switch (tp->t_state) {
2236 * In FIN_WAIT_1 STATE in addition to the processing
2237 * for the ESTABLISHED state if our FIN is now acknowledged
2238 * then enter FIN_WAIT_2.
2240 case TCPS_FIN_WAIT_1:
2241 if (ourfinisacked) {
2243 * If we can't receive any more
2244 * data, then closing user can proceed.
2245 * Starting the timer is contrary to the
2246 * specification, but if we don't get a FIN
2247 * we'll hang forever.
2249 if (so->so_state & SS_CANTRCVMORE) {
2250 soisdisconnected(so);
2251 tcp_callout_reset(tp, tp->tt_2msl,
2252 tp->t_maxidle, tcp_timer_2msl);
2254 tp->t_state = TCPS_FIN_WAIT_2;
2259 * In CLOSING STATE in addition to the processing for
2260 * the ESTABLISHED state if the ACK acknowledges our FIN
2261 * then enter the TIME-WAIT state, otherwise ignore
2265 if (ourfinisacked) {
2266 tp->t_state = TCPS_TIME_WAIT;
2267 tcp_canceltimers(tp);
2268 tcp_callout_reset(tp, tp->tt_2msl,
2269 2 * tcp_rmx_msl(tp),
2271 soisdisconnected(so);
2276 * In LAST_ACK, we may still be waiting for data to drain
2277 * and/or to be acked, as well as for the ack of our FIN.
2278 * If our FIN is now acknowledged, delete the TCB,
2279 * enter the closed state and return.
2282 if (ourfinisacked) {
2289 * In TIME_WAIT state the only thing that should arrive
2290 * is a retransmission of the remote FIN. Acknowledge
2291 * it and restart the finack timer.
2293 case TCPS_TIME_WAIT:
2294 tcp_callout_reset(tp, tp->tt_2msl, 2 * tcp_rmx_msl(tp),
2302 * Update window information.
2303 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2305 if ((thflags & TH_ACK) &&
2306 acceptable_window_update(tp, th, tiwin)) {
2307 /* keep track of pure window updates */
2308 if (tlen == 0 && tp->snd_wl2 == th->th_ack &&
2309 tiwin > tp->snd_wnd)
2310 tcpstat.tcps_rcvwinupd++;
2311 tp->snd_wnd = tiwin;
2312 tp->snd_wl1 = th->th_seq;
2313 tp->snd_wl2 = th->th_ack;
2314 if (tp->snd_wnd > tp->max_sndwnd)
2315 tp->max_sndwnd = tp->snd_wnd;
2320 * Process segments with URG.
2322 if ((thflags & TH_URG) && th->th_urp &&
2323 !TCPS_HAVERCVDFIN(tp->t_state)) {
2325 * This is a kludge, but if we receive and accept
2326 * random urgent pointers, we'll crash in
2327 * soreceive. It's hard to imagine someone
2328 * actually wanting to send this much urgent data.
2330 if (th->th_urp + so->so_rcv.ssb_cc > sb_max) {
2331 th->th_urp = 0; /* XXX */
2332 thflags &= ~TH_URG; /* XXX */
2333 goto dodata; /* XXX */
2336 * If this segment advances the known urgent pointer,
2337 * then mark the data stream. This should not happen
2338 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2339 * a FIN has been received from the remote side.
2340 * In these states we ignore the URG.
2342 * According to RFC961 (Assigned Protocols),
2343 * the urgent pointer points to the last octet
2344 * of urgent data. We continue, however,
2345 * to consider it to indicate the first octet
2346 * of data past the urgent section as the original
2347 * spec states (in one of two places).
2349 if (SEQ_GT(th->th_seq + th->th_urp, tp->rcv_up)) {
2350 tp->rcv_up = th->th_seq + th->th_urp;
2351 so->so_oobmark = so->so_rcv.ssb_cc +
2352 (tp->rcv_up - tp->rcv_nxt) - 1;
2353 if (so->so_oobmark == 0)
2354 sosetstate(so, SS_RCVATMARK);
2356 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2359 * Remove out of band data so doesn't get presented to user.
2360 * This can happen independent of advancing the URG pointer,
2361 * but if two URG's are pending at once, some out-of-band
2362 * data may creep in... ick.
2364 if (th->th_urp <= (u_long)tlen &&
2365 !(so->so_options & SO_OOBINLINE)) {
2366 /* hdr drop is delayed */
2367 tcp_pulloutofband(so, th, m, drop_hdrlen);
2371 * If no out of band data is expected,
2372 * pull receive urgent pointer along
2373 * with the receive window.
2375 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2376 tp->rcv_up = tp->rcv_nxt;
2381 * Process the segment text, merging it into the TCP sequencing queue,
2382 * and arranging for acknowledgment of receipt if necessary.
2383 * This process logically involves adjusting tp->rcv_wnd as data
2384 * is presented to the user (this happens in tcp_usrreq.c,
2385 * case PRU_RCVD). If a FIN has already been received on this
2386 * connection then we just ignore the text.
2388 if ((tlen || (thflags & TH_FIN)) && !TCPS_HAVERCVDFIN(tp->t_state)) {
2389 m_adj(m, drop_hdrlen); /* delayed header drop */
2391 * Insert segment which includes th into TCP reassembly queue
2392 * with control block tp. Set thflags to whether reassembly now
2393 * includes a segment with FIN. This handles the common case
2394 * inline (segment is the next to be received on an established
2395 * connection, and the queue is empty), avoiding linkage into
2396 * and removal from the queue and repetition of various
2398 * Set DELACK for segments received in order, but ack
2399 * immediately when segments are out of order (so
2400 * fast retransmit can work).
2402 if (th->th_seq == tp->rcv_nxt &&
2403 LIST_EMPTY(&tp->t_segq) &&
2404 TCPS_HAVEESTABLISHED(tp->t_state)) {
2405 if (DELAY_ACK(tp)) {
2406 tcp_callout_reset(tp, tp->tt_delack,
2407 tcp_delacktime, tcp_timer_delack);
2409 tp->t_flags |= TF_ACKNOW;
2411 tp->rcv_nxt += tlen;
2412 thflags = th->th_flags & TH_FIN;
2413 tcpstat.tcps_rcvpack++;
2414 tcpstat.tcps_rcvbyte += tlen;
2416 if (so->so_state & SS_CANTRCVMORE) {
2419 lwkt_gettoken(&so->so_rcv.ssb_token);
2420 ssb_appendstream(&so->so_rcv, m);
2421 lwkt_reltoken(&so->so_rcv.ssb_token);
2425 if (!(tp->t_flags & TF_DUPSEG)) {
2426 /* Initialize SACK report block. */
2427 tp->reportblk.rblk_start = th->th_seq;
2428 tp->reportblk.rblk_end = TCP_SACK_BLKEND(
2429 th->th_seq + tlen, thflags);
2431 thflags = tcp_reass(tp, th, &tlen, m);
2432 tp->t_flags |= TF_ACKNOW;
2436 * Note the amount of data that peer has sent into
2437 * our window, in order to estimate the sender's
2440 len = so->so_rcv.ssb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2447 * If FIN is received ACK the FIN and let the user know
2448 * that the connection is closing.
2450 if (thflags & TH_FIN) {
2451 if (!TCPS_HAVERCVDFIN(tp->t_state)) {
2454 * If connection is half-synchronized
2455 * (ie NEEDSYN flag on) then delay ACK,
2456 * so it may be piggybacked when SYN is sent.
2457 * Otherwise, since we received a FIN then no
2458 * more input can be expected, send ACK now.
2460 if (DELAY_ACK(tp) && (tp->t_flags & TF_NEEDSYN)) {
2461 tcp_callout_reset(tp, tp->tt_delack,
2462 tcp_delacktime, tcp_timer_delack);
2464 tp->t_flags |= TF_ACKNOW;
2469 switch (tp->t_state) {
2471 * In SYN_RECEIVED and ESTABLISHED STATES
2472 * enter the CLOSE_WAIT state.
2474 case TCPS_SYN_RECEIVED:
2475 tp->t_starttime = ticks;
2477 case TCPS_ESTABLISHED:
2478 tp->t_state = TCPS_CLOSE_WAIT;
2482 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2483 * enter the CLOSING state.
2485 case TCPS_FIN_WAIT_1:
2486 tp->t_state = TCPS_CLOSING;
2490 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2491 * starting the time-wait timer, turning off the other
2494 case TCPS_FIN_WAIT_2:
2495 tp->t_state = TCPS_TIME_WAIT;
2496 tcp_canceltimers(tp);
2497 tcp_callout_reset(tp, tp->tt_2msl, 2 * tcp_rmx_msl(tp),
2499 soisdisconnected(so);
2503 * In TIME_WAIT state restart the 2 MSL time_wait timer.
2505 case TCPS_TIME_WAIT:
2506 tcp_callout_reset(tp, tp->tt_2msl, 2 * tcp_rmx_msl(tp),
2513 if (so->so_options & SO_DEBUG)
2514 tcp_trace(TA_INPUT, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2518 * Return any desired output.
2520 if (needoutput || (tp->t_flags & TF_ACKNOW))
2522 tcp_sack_report_cleanup(tp);
2523 return(IPPROTO_DONE);
2527 * Generate an ACK dropping incoming segment if it occupies
2528 * sequence space, where the ACK reflects our state.
2530 * We can now skip the test for the RST flag since all
2531 * paths to this code happen after packets containing
2532 * RST have been dropped.
2534 * In the SYN-RECEIVED state, don't send an ACK unless the
2535 * segment we received passes the SYN-RECEIVED ACK test.
2536 * If it fails send a RST. This breaks the loop in the
2537 * "LAND" DoS attack, and also prevents an ACK storm
2538 * between two listening ports that have been sent forged
2539 * SYN segments, each with the source address of the other.
2541 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
2542 (SEQ_GT(tp->snd_una, th->th_ack) ||
2543 SEQ_GT(th->th_ack, tp->snd_max)) ) {
2544 rstreason = BANDLIM_RST_OPENPORT;
2548 if (so->so_options & SO_DEBUG)
2549 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2552 tp->t_flags |= TF_ACKNOW;
2554 tcp_sack_report_cleanup(tp);
2555 return(IPPROTO_DONE);
2559 * Generate a RST, dropping incoming segment.
2560 * Make ACK acceptable to originator of segment.
2561 * Don't bother to respond if destination was broadcast/multicast.
2563 if ((thflags & TH_RST) || m->m_flags & (M_BCAST | M_MCAST))
2566 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
2567 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
2570 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
2571 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
2572 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
2573 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2576 /* IPv6 anycast check is done at tcp6_input() */
2579 * Perform bandwidth limiting.
2582 if (badport_bandlim(rstreason) < 0)
2587 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2588 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2590 if (thflags & TH_ACK)
2591 /* mtod() below is safe as long as hdr dropping is delayed */
2592 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack,
2595 if (thflags & TH_SYN)
2597 /* mtod() below is safe as long as hdr dropping is delayed */
2598 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq + tlen,
2599 (tcp_seq)0, TH_RST | TH_ACK);
2602 tcp_sack_report_cleanup(tp);
2603 return(IPPROTO_DONE);
2607 * Drop space held by incoming segment and return.
2610 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2611 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2615 tcp_sack_report_cleanup(tp);
2616 return(IPPROTO_DONE);
2620 * Parse TCP options and place in tcpopt.
2623 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, boolean_t is_syn)
2628 for (; cnt > 0; cnt -= optlen, cp += optlen) {
2630 if (opt == TCPOPT_EOL)
2632 if (opt == TCPOPT_NOP)
2638 if (optlen < 2 || optlen > cnt)
2643 if (optlen != TCPOLEN_MAXSEG)
2647 to->to_flags |= TOF_MSS;
2648 bcopy(cp + 2, &to->to_mss, sizeof to->to_mss);
2649 to->to_mss = ntohs(to->to_mss);
2652 if (optlen != TCPOLEN_WINDOW)
2656 to->to_flags |= TOF_SCALE;
2657 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
2659 case TCPOPT_TIMESTAMP:
2660 if (optlen != TCPOLEN_TIMESTAMP)
2662 to->to_flags |= TOF_TS;
2663 bcopy(cp + 2, &to->to_tsval, sizeof to->to_tsval);
2664 to->to_tsval = ntohl(to->to_tsval);
2665 bcopy(cp + 6, &to->to_tsecr, sizeof to->to_tsecr);
2666 to->to_tsecr = ntohl(to->to_tsecr);
2668 * If echoed timestamp is later than the current time,
2669 * fall back to non RFC1323 RTT calculation.
2671 if (to->to_tsecr != 0 && TSTMP_GT(to->to_tsecr, ticks))
2674 case TCPOPT_SACK_PERMITTED:
2675 if (optlen != TCPOLEN_SACK_PERMITTED)
2679 to->to_flags |= TOF_SACK_PERMITTED;
2682 if ((optlen - 2) & 0x07) /* not multiple of 8 */
2684 to->to_nsackblocks = (optlen - 2) / 8;
2685 to->to_sackblocks = (struct raw_sackblock *) (cp + 2);
2686 to->to_flags |= TOF_SACK;
2687 for (i = 0; i < to->to_nsackblocks; i++) {
2688 struct raw_sackblock *r = &to->to_sackblocks[i];
2690 r->rblk_start = ntohl(r->rblk_start);
2691 r->rblk_end = ntohl(r->rblk_end);
2693 if (SEQ_LEQ(r->rblk_end, r->rblk_start)) {
2695 * Invalid SACK block; discard all
2698 tcpstat.tcps_rcvbadsackopt++;
2699 to->to_nsackblocks = 0;
2700 to->to_sackblocks = NULL;
2701 to->to_flags &= ~TOF_SACK;
2706 #ifdef TCP_SIGNATURE
2708 * XXX In order to reply to a host which has set the
2709 * TCP_SIGNATURE option in its initial SYN, we have to
2710 * record the fact that the option was observed here
2711 * for the syncache code to perform the correct response.
2713 case TCPOPT_SIGNATURE:
2714 if (optlen != TCPOLEN_SIGNATURE)
2716 to->to_flags |= (TOF_SIGNATURE | TOF_SIGLEN);
2718 #endif /* TCP_SIGNATURE */
2726 * Pull out of band byte out of a segment so
2727 * it doesn't appear in the user's data queue.
2728 * It is still reflected in the segment length for
2729 * sequencing purposes.
2730 * "off" is the delayed to be dropped hdrlen.
2733 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, int off)
2735 int cnt = off + th->th_urp - 1;
2738 if (m->m_len > cnt) {
2739 char *cp = mtod(m, caddr_t) + cnt;
2740 struct tcpcb *tp = sototcpcb(so);
2743 tp->t_oobflags |= TCPOOB_HAVEDATA;
2744 bcopy(cp + 1, cp, m->m_len - cnt - 1);
2746 if (m->m_flags & M_PKTHDR)
2755 panic("tcp_pulloutofband");
2759 * Collect new round-trip time estimate
2760 * and update averages and current timeout.
2763 tcp_xmit_timer(struct tcpcb *tp, int rtt)
2767 tcpstat.tcps_rttupdated++;
2769 if (tp->t_srtt != 0) {
2771 * srtt is stored as fixed point with 5 bits after the
2772 * binary point (i.e., scaled by 8). The following magic
2773 * is equivalent to the smoothing algorithm in rfc793 with
2774 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2775 * point). Adjust rtt to origin 0.
2777 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
2778 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
2780 if ((tp->t_srtt += delta) <= 0)
2784 * We accumulate a smoothed rtt variance (actually, a
2785 * smoothed mean difference), then set the retransmit
2786 * timer to smoothed rtt + 4 times the smoothed variance.
2787 * rttvar is stored as fixed point with 4 bits after the
2788 * binary point (scaled by 16). The following is
2789 * equivalent to rfc793 smoothing with an alpha of .75
2790 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
2791 * rfc793's wired-in beta.
2795 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
2796 if ((tp->t_rttvar += delta) <= 0)
2798 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
2799 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2802 * No rtt measurement yet - use the unsmoothed rtt.
2803 * Set the variance to half the rtt (so our first
2804 * retransmit happens at 3*rtt).
2806 tp->t_srtt = rtt << TCP_RTT_SHIFT;
2807 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
2808 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2814 * the retransmit should happen at rtt + 4 * rttvar.
2815 * Because of the way we do the smoothing, srtt and rttvar
2816 * will each average +1/2 tick of bias. When we compute
2817 * the retransmit timer, we want 1/2 tick of rounding and
2818 * 1 extra tick because of +-1/2 tick uncertainty in the
2819 * firing of the timer. The bias will give us exactly the
2820 * 1.5 tick we need. But, because the bias is
2821 * statistical, we have to test that we don't drop below
2822 * the minimum feasible timer (which is 2 ticks).
2824 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
2825 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
2828 * We received an ack for a packet that wasn't retransmitted;
2829 * it is probably safe to discard any error indications we've
2830 * received recently. This isn't quite right, but close enough
2831 * for now (a route might have failed after we sent a segment,
2832 * and the return path might not be symmetrical).
2834 tp->t_softerror = 0;
2838 * Determine a reasonable value for maxseg size.
2839 * If the route is known, check route for mtu.
2840 * If none, use an mss that can be handled on the outgoing
2841 * interface without forcing IP to fragment; if bigger than
2842 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2843 * to utilize large mbufs. If no route is found, route has no mtu,
2844 * or the destination isn't local, use a default, hopefully conservative
2845 * size (usually 512 or the default IP max size, but no more than the mtu
2846 * of the interface), as we can't discover anything about intervening
2847 * gateways or networks. We also initialize the congestion/slow start
2848 * window to be a single segment if the destination isn't local.
2849 * While looking at the routing entry, we also initialize other path-dependent
2850 * parameters from pre-set or cached values in the routing entry.
2852 * Also take into account the space needed for options that we
2853 * send regularly. Make maxseg shorter by that amount to assure
2854 * that we can send maxseg amount of data even when the options
2855 * are present. Store the upper limit of the length of options plus
2858 * NOTE that this routine is only called when we process an incoming
2859 * segment, for outgoing segments only tcp_mssopt is called.
2862 tcp_mss(struct tcpcb *tp, int offer)
2868 struct inpcb *inp = tp->t_inpcb;
2871 boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) ? TRUE : FALSE);
2872 size_t min_protoh = isipv6 ?
2873 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
2874 sizeof(struct tcpiphdr);
2876 const boolean_t isipv6 = FALSE;
2877 const size_t min_protoh = sizeof(struct tcpiphdr);
2881 rt = tcp_rtlookup6(&inp->inp_inc);
2883 rt = tcp_rtlookup(&inp->inp_inc);
2885 tp->t_maxopd = tp->t_maxseg =
2886 (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
2890 so = inp->inp_socket;
2893 * Offer == 0 means that there was no MSS on the SYN segment,
2894 * in this case we use either the interface mtu or tcp_mssdflt.
2896 * An offer which is too large will be cut down later.
2900 if (in6_localaddr(&inp->in6p_faddr)) {
2901 offer = ND_IFINFO(rt->rt_ifp)->linkmtu -
2904 offer = tcp_v6mssdflt;
2907 if (in_localaddr(inp->inp_faddr))
2908 offer = ifp->if_mtu - min_protoh;
2910 offer = tcp_mssdflt;
2915 * Prevent DoS attack with too small MSS. Round up
2916 * to at least minmss.
2918 * Sanity check: make sure that maxopd will be large
2919 * enough to allow some data on segments even is the
2920 * all the option space is used (40bytes). Otherwise
2921 * funny things may happen in tcp_output.
2923 offer = max(offer, tcp_minmss);
2924 offer = max(offer, 64);
2926 rt->rt_rmx.rmx_mssopt = offer;
2929 * While we're here, check if there's an initial rtt
2930 * or rttvar. Convert from the route-table units
2931 * to scaled multiples of the slow timeout timer.
2933 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
2935 * XXX the lock bit for RTT indicates that the value
2936 * is also a minimum value; this is subject to time.
2938 if (rt->rt_rmx.rmx_locks & RTV_RTT)
2939 tp->t_rttmin = rtt / (RTM_RTTUNIT / hz);
2940 tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE));
2941 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
2942 tcpstat.tcps_usedrtt++;
2943 if (rt->rt_rmx.rmx_rttvar) {
2944 tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
2945 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE));
2946 tcpstat.tcps_usedrttvar++;
2948 /* default variation is +- 1 rtt */
2950 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
2952 TCPT_RANGESET(tp->t_rxtcur,
2953 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
2954 tp->t_rttmin, TCPTV_REXMTMAX);
2958 * if there's an mtu associated with the route, use it
2959 * else, use the link mtu. Take the smaller of mss or offer
2962 if (rt->rt_rmx.rmx_mtu) {
2963 mss = rt->rt_rmx.rmx_mtu - min_protoh;
2966 mss = ND_IFINFO(rt->rt_ifp)->linkmtu - min_protoh;
2968 mss = ifp->if_mtu - min_protoh;
2970 mss = min(mss, offer);
2973 * maxopd stores the maximum length of data AND options
2974 * in a segment; maxseg is the amount of data in a normal
2975 * segment. We need to store this value (maxopd) apart
2976 * from maxseg, because now every segment carries options
2977 * and thus we normally have somewhat less data in segments.
2981 if ((tp->t_flags & (TF_REQ_TSTMP | TF_NOOPT)) == TF_REQ_TSTMP &&
2982 ((tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
2983 mss -= TCPOLEN_TSTAMP_APPA;
2985 #if (MCLBYTES & (MCLBYTES - 1)) == 0
2987 mss &= ~(MCLBYTES-1);
2990 mss = mss / MCLBYTES * MCLBYTES;
2993 * If there's a pipesize, change the socket buffer
2994 * to that size. Make the socket buffers an integral
2995 * number of mss units; if the mss is larger than
2996 * the socket buffer, decrease the mss.
2999 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0)
3001 bufsize = so->so_snd.ssb_hiwat;
3005 bufsize = roundup(bufsize, mss);
3006 if (bufsize > sb_max)
3008 if (bufsize > so->so_snd.ssb_hiwat)
3009 ssb_reserve(&so->so_snd, bufsize, so, NULL);
3014 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0)
3016 bufsize = so->so_rcv.ssb_hiwat;
3017 if (bufsize > mss) {
3018 bufsize = roundup(bufsize, mss);
3019 if (bufsize > sb_max)
3021 if (bufsize > so->so_rcv.ssb_hiwat) {
3022 lwkt_gettoken(&so->so_rcv.ssb_token);
3023 ssb_reserve(&so->so_rcv, bufsize, so, NULL);
3024 lwkt_reltoken(&so->so_rcv.ssb_token);
3029 * Set the slow-start flight size depending on whether this
3030 * is a local network or not.
3033 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380));
3037 if (rt->rt_rmx.rmx_ssthresh) {
3039 * There's some sort of gateway or interface
3040 * buffer limit on the path. Use this to set
3041 * the slow start threshhold, but set the
3042 * threshold to no less than 2*mss.
3044 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
3045 tcpstat.tcps_usedssthresh++;
3050 * Determine the MSS option to send on an outgoing SYN.
3053 tcp_mssopt(struct tcpcb *tp)
3058 ((tp->t_inpcb->inp_vflag & INP_IPV6) ? TRUE : FALSE);
3059 int min_protoh = isipv6 ?
3060 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
3061 sizeof(struct tcpiphdr);
3063 const boolean_t isipv6 = FALSE;
3064 const size_t min_protoh = sizeof(struct tcpiphdr);
3068 rt = tcp_rtlookup6(&tp->t_inpcb->inp_inc);
3070 rt = tcp_rtlookup(&tp->t_inpcb->inp_inc);
3072 return (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
3074 return (rt->rt_ifp->if_mtu - min_protoh);
3078 * When a partial ack arrives, force the retransmission of the
3079 * next unacknowledged segment. Do not exit Fast Recovery.
3081 * Implement the Slow-but-Steady variant of NewReno by restarting the
3082 * the retransmission timer. Turn it off here so it can be restarted
3083 * later in tcp_output().
3086 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th, int acked)
3088 tcp_seq old_snd_nxt = tp->snd_nxt;
3089 u_long ocwnd = tp->snd_cwnd;
3091 tcp_callout_stop(tp, tp->tt_rexmt);
3093 tp->snd_nxt = th->th_ack;
3094 /* Set snd_cwnd to one segment beyond acknowledged offset. */
3095 tp->snd_cwnd = tp->t_maxseg;
3096 tp->t_flags |= TF_ACKNOW;
3098 if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
3099 tp->snd_nxt = old_snd_nxt;
3100 /* partial window deflation */
3102 tp->snd_cwnd = ocwnd - acked + tp->t_maxseg;
3104 tp->snd_cwnd = tp->t_maxseg;
3108 * In contrast to the Slow-but-Steady NewReno variant,
3109 * we do not reset the retransmission timer for SACK retransmissions,
3110 * except when retransmitting snd_una.
3113 tcp_sack_rexmt(struct tcpcb *tp, struct tcphdr *th)
3115 uint32_t pipe, seglen;
3118 tcp_seq old_snd_nxt = tp->snd_nxt;
3119 u_long ocwnd = tp->snd_cwnd;
3120 int nseg = 0; /* consecutive new segments */
3121 #define MAXBURST 4 /* limit burst of new packets on partial ack */
3124 pipe = tcp_sack_compute_pipe(tp);
3125 while ((tcp_seq_diff_t)(ocwnd - pipe) >= (tcp_seq_diff_t)tp->t_maxseg &&
3126 (!tcp_do_smartsack || nseg < MAXBURST) &&
3127 tcp_sack_nextseg(tp, &nextrexmt, &seglen, &lostdup)) {
3129 tcp_seq old_snd_max;
3132 if (nextrexmt == tp->snd_max)
3134 tp->snd_nxt = nextrexmt;
3135 tp->snd_cwnd = nextrexmt - tp->snd_una + seglen;
3136 old_snd_max = tp->snd_max;
3137 if (nextrexmt == tp->snd_una)
3138 tcp_callout_stop(tp, tp->tt_rexmt);
3139 error = tcp_output(tp);
3142 sent = tp->snd_nxt - nextrexmt;
3147 tcpstat.tcps_sndsackpack++;
3148 tcpstat.tcps_sndsackbyte += sent;
3149 if (SEQ_LT(nextrexmt, old_snd_max) &&
3150 SEQ_LT(tp->rexmt_high, tp->snd_nxt))
3151 tp->rexmt_high = seq_min(tp->snd_nxt, old_snd_max);
3153 if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
3154 tp->snd_nxt = old_snd_nxt;
3155 tp->snd_cwnd = ocwnd;
3159 * Reset idle time and keep-alive timer, typically called when a valid
3160 * tcp packet is received but may also be called when FASTKEEP is set
3161 * to prevent the previous long-timeout from calculating to a drop.
3163 * Only update t_rcvtime for non-SYN packets.
3165 * Handle the case where one side thinks the connection is established
3166 * but the other side has, say, rebooted without cleaning out the
3167 * connection. The SYNs could be construed as an attack and wind
3168 * up ignored, but in case it isn't an attack we can validate the
3169 * connection by forcing a keepalive.
3172 tcp_timer_keep_activity(struct tcpcb *tp, int thflags)
3174 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
3175 if ((thflags & (TH_SYN | TH_ACK)) == TH_SYN) {
3176 tp->t_flags |= TF_KEEPALIVE;
3177 tcp_callout_reset(tp, tp->tt_keep, hz / 2,
3180 tp->t_rcvtime = ticks;
3181 tp->t_flags &= ~TF_KEEPALIVE;
3182 tcp_callout_reset(tp, tp->tt_keep,
3183 tcp_getkeepidle(tp),
3190 tcp_rmx_msl(const struct tcpcb *tp)
3193 struct inpcb *inp = tp->t_inpcb;
3196 boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) ? TRUE : FALSE);
3198 const boolean_t isipv6 = FALSE;
3202 rt = tcp_rtlookup6(&inp->inp_inc);
3204 rt = tcp_rtlookup(&inp->inp_inc);
3205 if (rt == NULL || rt->rt_rmx.rmx_msl == 0)
3208 msl = (rt->rt_rmx.rmx_msl * hz) / 1000;
3216 tcp_established(struct tcpcb *tp)
3218 tp->t_state = TCPS_ESTABLISHED;
3219 tcp_callout_reset(tp, tp->tt_keep, tcp_getkeepidle(tp), tcp_timer_keep);
3221 if (tp->t_flags & TF_SYN_WASLOST) {
3224 * "If the SYN or SYN/ACK is lost, the initial window used by
3225 * a sender after a correctly transmitted SYN MUST be one
3226 * segment consisting of MSS bytes."
3228 tp->snd_cwnd = tp->t_maxseg;
3232 * "If the timer expires awaiting the ACK of a SYN segment
3233 * and the TCP implementation is using an RTO less than 3
3234 * seconds, the RTO MUST be re-initialized to 3 seconds
3235 * when data transmission begins"
3237 if (tp->t_rxtcur < TCPTV_RTOBASE3)
3238 tp->t_rxtcur = TCPTV_RTOBASE3;