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 = p->tqe_th->th_seq +
348 tcpstat.tcps_rcvduppack++;
349 tcpstat.tcps_rcvdupbyte += *tlenp;
352 atomic_add_int(&tcp_reass_qsize, -1);
354 * Try to present any queued data
355 * at the left window edge to the user.
356 * This is needed after the 3-WHS
359 goto present; /* ??? */
364 /* incoming segment end is enclosing block end */
365 tp->encloseblk.rblk_end = th->th_seq + *tlenp +
366 ((th->th_flags & TH_FIN) != 0);
367 /* trim end of reported D-SACK block */
368 tp->reportblk.rblk_end = th->th_seq;
371 tcpstat.tcps_rcvoopack++;
372 tcpstat.tcps_rcvoobyte += *tlenp;
375 * While we overlap succeeding segments trim them or,
376 * if they are completely covered, dequeue them.
379 tcp_seq_diff_t i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
380 tcp_seq qend = q->tqe_th->th_seq + q->tqe_len;
381 struct tseg_qent *nq;
385 if (!(tp->t_flags & TF_DUPSEG)) { /* first time through */
386 tp->t_flags |= (TF_DUPSEG | TF_ENCLOSESEG);
387 tp->encloseblk = tp->reportblk;
388 /* report trailing duplicate D-SACK segment */
389 tp->reportblk.rblk_start = q->tqe_th->th_seq;
391 if ((tp->t_flags & TF_ENCLOSESEG) &&
392 SEQ_GT(qend, tp->encloseblk.rblk_end)) {
393 /* extend enclosing block if one exists */
394 tp->encloseblk.rblk_end = qend;
396 if (i < q->tqe_len) {
397 q->tqe_th->th_seq += i;
403 nq = LIST_NEXT(q, tqe_q);
404 LIST_REMOVE(q, tqe_q);
407 atomic_add_int(&tcp_reass_qsize, -1);
411 /* Insert the new segment queue entry into place. */
414 te->tqe_len = *tlenp;
416 /* check if can coalesce with following segment */
417 if (q != NULL && (th->th_seq + *tlenp == q->tqe_th->th_seq)) {
418 tcp_seq tend = te->tqe_th->th_seq + te->tqe_len;
420 te->tqe_len += q->tqe_len;
421 if (q->tqe_th->th_flags & TH_FIN)
422 te->tqe_th->th_flags |= TH_FIN;
423 m_cat(te->tqe_m, q->tqe_m);
424 tp->encloseblk.rblk_end = tend;
426 * When not reporting a duplicate segment, use
427 * the larger enclosing block as the SACK block.
429 if (!(tp->t_flags & TF_DUPSEG))
430 tp->reportblk.rblk_end = tend;
431 LIST_REMOVE(q, tqe_q);
433 atomic_add_int(&tcp_reass_qsize, -1);
437 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
439 /* check if can coalesce with preceding segment */
440 if (p->tqe_th->th_seq + p->tqe_len == th->th_seq) {
441 p->tqe_len += te->tqe_len;
442 m_cat(p->tqe_m, te->tqe_m);
443 tp->encloseblk.rblk_start = p->tqe_th->th_seq;
445 * When not reporting a duplicate segment, use
446 * the larger enclosing block as the SACK block.
448 if (!(tp->t_flags & TF_DUPSEG))
449 tp->reportblk.rblk_start = p->tqe_th->th_seq;
451 atomic_add_int(&tcp_reass_qsize, -1);
453 LIST_INSERT_AFTER(p, te, tqe_q);
459 * Present data to user, advancing rcv_nxt through
460 * completed sequence space.
462 if (!TCPS_HAVEESTABLISHED(tp->t_state))
464 q = LIST_FIRST(&tp->t_segq);
465 if (q == NULL || q->tqe_th->th_seq != tp->rcv_nxt)
467 tp->rcv_nxt += q->tqe_len;
468 if (!(tp->t_flags & TF_DUPSEG)) {
469 /* no SACK block to report since ACK advanced */
470 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
472 /* no enclosing block to report since ACK advanced */
473 tp->t_flags &= ~TF_ENCLOSESEG;
474 flags = q->tqe_th->th_flags & TH_FIN;
475 LIST_REMOVE(q, tqe_q);
476 KASSERT(LIST_EMPTY(&tp->t_segq) ||
477 LIST_FIRST(&tp->t_segq)->tqe_th->th_seq != tp->rcv_nxt,
478 ("segment not coalesced"));
479 if (so->so_state & SS_CANTRCVMORE) {
482 lwkt_gettoken(&so->so_rcv.ssb_token);
483 ssb_appendstream(&so->so_rcv, q->tqe_m);
484 lwkt_reltoken(&so->so_rcv.ssb_token);
487 atomic_add_int(&tcp_reass_qsize, -1);
494 * TCP input routine, follows pages 65-76 of the
495 * protocol specification dated September, 1981 very closely.
499 tcp6_input(struct mbuf **mp, int *offp, int proto)
501 struct mbuf *m = *mp;
502 struct in6_ifaddr *ia6;
504 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
507 * draft-itojun-ipv6-tcp-to-anycast
508 * better place to put this in?
510 ia6 = ip6_getdstifaddr(m);
511 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
514 ip6 = mtod(m, struct ip6_hdr *);
515 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
516 offsetof(struct ip6_hdr, ip6_dst));
517 return (IPPROTO_DONE);
520 tcp_input(mp, offp, proto);
521 return (IPPROTO_DONE);
526 tcp_input(struct mbuf **mp, int *offp, int proto)
530 struct ip *ip = NULL;
532 struct inpcb *inp = NULL;
538 struct tcpcb *tp = NULL;
540 struct socket *so = NULL;
542 boolean_t ourfinisacked, needoutput = FALSE;
545 struct tcpopt to; /* options in this segment */
546 struct sockaddr_in *next_hop = NULL;
547 int rstreason; /* For badport_bandlim accounting purposes */
549 struct ip6_hdr *ip6 = NULL;
554 const boolean_t isipv6 = FALSE;
564 tcpstat.tcps_rcvtotal++;
566 if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
569 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
570 KKASSERT(mtag != NULL);
571 next_hop = m_tag_data(mtag);
575 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? TRUE : FALSE;
579 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */
580 ip6 = mtod(m, struct ip6_hdr *);
581 tlen = (sizeof *ip6) + ntohs(ip6->ip6_plen) - off0;
582 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) {
583 tcpstat.tcps_rcvbadsum++;
586 th = (struct tcphdr *)((caddr_t)ip6 + off0);
589 * Be proactive about unspecified IPv6 address in source.
590 * As we use all-zero to indicate unbounded/unconnected pcb,
591 * unspecified IPv6 address can be used to confuse us.
593 * Note that packets with unspecified IPv6 destination is
594 * already dropped in ip6_input.
596 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
602 * Get IP and TCP header together in first mbuf.
603 * Note: IP leaves IP header in first mbuf.
605 if (off0 > sizeof(struct ip)) {
607 off0 = sizeof(struct ip);
609 /* already checked and pulled up in ip_demux() */
610 KASSERT(m->m_len >= sizeof(struct tcpiphdr),
611 ("TCP header not in one mbuf: m->m_len %d", m->m_len));
612 ip = mtod(m, struct ip *);
613 ipov = (struct ipovly *)ip;
614 th = (struct tcphdr *)((caddr_t)ip + off0);
617 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
618 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
619 th->th_sum = m->m_pkthdr.csum_data;
621 th->th_sum = in_pseudo(ip->ip_src.s_addr,
623 htonl(m->m_pkthdr.csum_data +
626 th->th_sum ^= 0xffff;
629 * Checksum extended TCP header and data.
631 len = sizeof(struct ip) + tlen;
632 bzero(ipov->ih_x1, sizeof ipov->ih_x1);
633 ipov->ih_len = (u_short)tlen;
634 ipov->ih_len = htons(ipov->ih_len);
635 th->th_sum = in_cksum(m, len);
638 tcpstat.tcps_rcvbadsum++;
642 /* Re-initialization for later version check */
643 ip->ip_v = IPVERSION;
648 * Check that TCP offset makes sense,
649 * pull out TCP options and adjust length. XXX
651 off = th->th_off << 2;
652 /* already checked and pulled up in ip_demux() */
653 KASSERT(off >= sizeof(struct tcphdr) && off <= tlen,
654 ("bad TCP data offset %d (tlen %d)", off, tlen));
655 tlen -= off; /* tlen is used instead of ti->ti_len */
656 if (off > sizeof(struct tcphdr)) {
658 IP6_EXTHDR_CHECK(m, off0, off, IPPROTO_DONE);
659 ip6 = mtod(m, struct ip6_hdr *);
660 th = (struct tcphdr *)((caddr_t)ip6 + off0);
662 /* already pulled up in ip_demux() */
663 KASSERT(m->m_len >= sizeof(struct ip) + off,
664 ("TCP header and options not in one mbuf: "
665 "m_len %d, off %d", m->m_len, off));
667 optlen = off - sizeof(struct tcphdr);
668 optp = (u_char *)(th + 1);
670 thflags = th->th_flags;
672 #ifdef TCP_DROP_SYNFIN
674 * If the drop_synfin option is enabled, drop all packets with
675 * both the SYN and FIN bits set. This prevents e.g. nmap from
676 * identifying the TCP/IP stack.
678 * This is a violation of the TCP specification.
680 if (drop_synfin && (thflags & (TH_SYN | TH_FIN)) == (TH_SYN | TH_FIN))
685 * Convert TCP protocol specific fields to host format.
687 th->th_seq = ntohl(th->th_seq);
688 th->th_ack = ntohl(th->th_ack);
689 th->th_win = ntohs(th->th_win);
690 th->th_urp = ntohs(th->th_urp);
693 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options,
694 * until after ip6_savecontrol() is called and before other functions
695 * which don't want those proto headers.
696 * Because ip6_savecontrol() is going to parse the mbuf to
697 * search for data to be passed up to user-land, it wants mbuf
698 * parameters to be unchanged.
699 * XXX: the call of ip6_savecontrol() has been obsoleted based on
700 * latest version of the advanced API (20020110).
702 drop_hdrlen = off0 + off;
705 * Locate pcb for segment.
708 /* IPFIREWALL_FORWARD section */
709 if (next_hop != NULL && !isipv6) { /* IPv6 support is not there yet */
711 * Transparently forwarded. Pretend to be the destination.
712 * already got one like this?
714 cpu = mycpu->gd_cpuid;
715 inp = in_pcblookup_hash(&tcbinfo[cpu],
716 ip->ip_src, th->th_sport,
717 ip->ip_dst, th->th_dport,
718 0, m->m_pkthdr.rcvif);
721 * It's new. Try to find the ambushing socket.
725 * The rest of the ipfw code stores the port in
727 * (The IP address is still in network order.)
729 in_port_t dport = next_hop->sin_port ?
730 htons(next_hop->sin_port) :
733 cpu = tcp_addrcpu(ip->ip_src.s_addr, th->th_sport,
734 next_hop->sin_addr.s_addr, dport);
735 inp = in_pcblookup_hash(&tcbinfo[cpu],
736 ip->ip_src, th->th_sport,
737 next_hop->sin_addr, dport,
738 1, m->m_pkthdr.rcvif);
742 inp = in6_pcblookup_hash(&tcbinfo[0],
743 &ip6->ip6_src, th->th_sport,
744 &ip6->ip6_dst, th->th_dport,
745 1, m->m_pkthdr.rcvif);
747 cpu = mycpu->gd_cpuid;
748 inp = in_pcblookup_hash(&tcbinfo[cpu],
749 ip->ip_src, th->th_sport,
750 ip->ip_dst, th->th_dport,
751 1, m->m_pkthdr.rcvif);
756 * If the state is CLOSED (i.e., TCB does not exist) then
757 * all data in the incoming segment is discarded.
758 * If the TCB exists but is in CLOSED state, it is embryonic,
759 * but should either do a listen or a connect soon.
764 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2];
766 char dbuf[sizeof "aaa.bbb.ccc.ddd"];
767 char sbuf[sizeof "aaa.bbb.ccc.ddd"];
771 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst));
774 strcat(sbuf, ip6_sprintf(&ip6->ip6_src));
777 strcpy(dbuf, inet_ntoa(ip->ip_dst));
778 strcpy(sbuf, inet_ntoa(ip->ip_src));
780 switch (log_in_vain) {
782 if (!(thflags & TH_SYN))
786 "Connection attempt to TCP %s:%d "
787 "from %s:%d flags:0x%02x\n",
788 dbuf, ntohs(th->th_dport), sbuf,
789 ntohs(th->th_sport), thflags);
798 if (thflags & TH_SYN)
807 rstreason = BANDLIM_RST_CLOSEDPORT;
813 if (ipsec6_in_reject_so(m, inp->inp_socket)) {
814 ipsec6stat.in_polvio++;
818 if (ipsec4_in_reject_so(m, inp->inp_socket)) {
819 ipsecstat.in_polvio++;
826 if (ipsec6_in_reject(m, inp))
829 if (ipsec4_in_reject(m, inp))
833 /* Check the minimum TTL for socket. */
835 if ((isipv6 ? ip6->ip6_hlim : ip->ip_ttl) < inp->inp_ip_minttl)
841 rstreason = BANDLIM_RST_CLOSEDPORT;
844 if (tp->t_state <= TCPS_CLOSED)
847 /* Unscale the window into a 32-bit value. */
848 if (!(thflags & TH_SYN))
849 tiwin = th->th_win << tp->snd_scale;
853 so = inp->inp_socket;
856 if (so->so_options & SO_DEBUG) {
857 ostate = tp->t_state;
859 bcopy(ip6, tcp_saveipgen, sizeof(*ip6));
861 bcopy(ip, tcp_saveipgen, sizeof(*ip));
866 bzero(&to, sizeof to);
868 if (so->so_options & SO_ACCEPTCONN) {
869 struct in_conninfo inc;
872 inc.inc_isipv6 = (isipv6 == TRUE);
875 inc.inc6_faddr = ip6->ip6_src;
876 inc.inc6_laddr = ip6->ip6_dst;
877 inc.inc6_route.ro_rt = NULL; /* XXX */
879 inc.inc_faddr = ip->ip_src;
880 inc.inc_laddr = ip->ip_dst;
881 inc.inc_route.ro_rt = NULL; /* XXX */
883 inc.inc_fport = th->th_sport;
884 inc.inc_lport = th->th_dport;
887 * If the state is LISTEN then ignore segment if it contains
888 * a RST. If the segment contains an ACK then it is bad and
889 * send a RST. If it does not contain a SYN then it is not
890 * interesting; drop it.
892 * If the state is SYN_RECEIVED (syncache) and seg contains
893 * an ACK, but not for our SYN/ACK, send a RST. If the seg
894 * contains a RST, check the sequence number to see if it
895 * is a valid reset segment.
897 if ((thflags & (TH_RST | TH_ACK | TH_SYN)) != TH_SYN) {
898 if ((thflags & (TH_RST | TH_ACK | TH_SYN)) == TH_ACK) {
899 if (!syncache_expand(&inc, th, &so, m)) {
901 * No syncache entry, or ACK was not
902 * for our SYN/ACK. Send a RST.
904 tcpstat.tcps_badsyn++;
905 rstreason = BANDLIM_RST_OPENPORT;
910 * Could not complete 3-way handshake,
911 * connection is being closed down, and
912 * syncache will free mbuf.
915 return(IPPROTO_DONE);
918 * We must be in the correct protocol thread
919 * for this connection.
921 KKASSERT(so->so_port == &curthread->td_msgport);
924 * Socket is created in state SYN_RECEIVED.
925 * Continue processing segment.
930 * This is what would have happened in
931 * tcp_output() when the SYN,ACK was sent.
933 tp->snd_up = tp->snd_una;
934 tp->snd_max = tp->snd_nxt = tp->iss + 1;
935 tp->last_ack_sent = tp->rcv_nxt;
937 * XXX possible bug - it doesn't appear that tp->snd_wnd is unscaled
938 * until the _second_ ACK is received:
939 * rcv SYN (set wscale opts) --> send SYN/ACK, set snd_wnd = window.
940 * rcv ACK, calculate tiwin --> process SYN_RECEIVED, determine wscale,
941 * move to ESTAB, set snd_wnd to tiwin.
943 tp->snd_wnd = tiwin; /* unscaled */
946 if (thflags & TH_RST) {
947 syncache_chkrst(&inc, th);
950 if (thflags & TH_ACK) {
951 syncache_badack(&inc);
952 tcpstat.tcps_badsyn++;
953 rstreason = BANDLIM_RST_OPENPORT;
960 * Segment's flags are (SYN) or (SYN | FIN).
964 * If deprecated address is forbidden,
965 * we do not accept SYN to deprecated interface
966 * address to prevent any new inbound connection from
967 * getting established.
968 * When we do not accept SYN, we send a TCP RST,
969 * with deprecated source address (instead of dropping
970 * it). We compromise it as it is much better for peer
971 * to send a RST, and RST will be the final packet
974 * If we do not forbid deprecated addresses, we accept
975 * the SYN packet. RFC2462 does not suggest dropping
977 * If we decipher RFC2462 5.5.4, it says like this:
978 * 1. use of deprecated addr with existing
979 * communication is okay - "SHOULD continue to be
981 * 2. use of it with new communication:
982 * (2a) "SHOULD NOT be used if alternate address
983 * with sufficient scope is available"
984 * (2b) nothing mentioned otherwise.
985 * Here we fall into (2b) case as we have no choice in
986 * our source address selection - we must obey the peer.
988 * The wording in RFC2462 is confusing, and there are
989 * multiple description text for deprecated address
990 * handling - worse, they are not exactly the same.
991 * I believe 5.5.4 is the best one, so we follow 5.5.4.
993 if (isipv6 && !ip6_use_deprecated) {
994 struct in6_ifaddr *ia6;
996 if ((ia6 = ip6_getdstifaddr(m)) &&
997 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
999 rstreason = BANDLIM_RST_OPENPORT;
1005 * If it is from this socket, drop it, it must be forged.
1006 * Don't bother responding if the destination was a broadcast.
1008 if (th->th_dport == th->th_sport) {
1010 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
1014 if (ip->ip_dst.s_addr == ip->ip_src.s_addr)
1019 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
1021 * Note that it is quite possible to receive unicast
1022 * link-layer packets with a broadcast IP address. Use
1023 * in_broadcast() to find them.
1025 if (m->m_flags & (M_BCAST | M_MCAST))
1028 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1029 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
1032 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1033 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1034 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1035 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
1039 * SYN appears to be valid; create compressed TCP state
1040 * for syncache, or perform t/tcp connection.
1042 if (so->so_qlen <= so->so_qlimit) {
1043 tcp_dooptions(&to, optp, optlen, TRUE);
1044 if (!syncache_add(&inc, &to, th, so, m))
1048 * Entry added to syncache, mbuf used to
1049 * send SYN,ACK packet.
1051 return(IPPROTO_DONE);
1058 * Should not happen - syncache should pick up these connections.
1060 * Once we are past handling listen sockets we must be in the
1061 * correct protocol processing thread.
1063 KASSERT(tp->t_state != TCPS_LISTEN, ("tcp_input: TCPS_LISTEN state"));
1064 KKASSERT(so->so_port == &curthread->td_msgport);
1067 * This is the second part of the MSS DoS prevention code (after
1068 * minmss on the sending side) and it deals with too many too small
1069 * tcp packets in a too short timeframe (1 second).
1071 * XXX Removed. This code was crap. It does not scale to network
1072 * speed, and default values break NFS. Gone.
1077 * Segment received on connection.
1079 * Reset idle time and keep-alive timer. Don't waste time if less
1080 * then a second has elapsed.
1082 if ((int)(ticks - tp->t_rcvtime) > hz)
1083 tcp_timer_keep_activity(tp, thflags);
1087 * XXX this is tradtitional behavior, may need to be cleaned up.
1089 tcp_dooptions(&to, optp, optlen, (thflags & TH_SYN) != 0);
1090 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1091 if (to.to_flags & TOF_SCALE) {
1092 tp->t_flags |= TF_RCVD_SCALE;
1093 tp->requested_s_scale = to.to_requested_s_scale;
1095 if (to.to_flags & TOF_TS) {
1096 tp->t_flags |= TF_RCVD_TSTMP;
1097 tp->ts_recent = to.to_tsval;
1098 tp->ts_recent_age = ticks;
1100 if (!(to.to_flags & TOF_MSS))
1102 tcp_mss(tp, to.to_mss);
1104 * Only set the TF_SACK_PERMITTED per-connection flag
1105 * if we got a SACK_PERMITTED option from the other side
1106 * and the global tcp_do_sack variable is true.
1108 if (tcp_do_sack && (to.to_flags & TOF_SACK_PERMITTED))
1109 tp->t_flags |= TF_SACK_PERMITTED;
1113 * Header prediction: check for the two common cases
1114 * of a uni-directional data xfer. If the packet has
1115 * no control flags, is in-sequence, the window didn't
1116 * change and we're not retransmitting, it's a
1117 * candidate. If the length is zero and the ack moved
1118 * forward, we're the sender side of the xfer. Just
1119 * free the data acked & wake any higher level process
1120 * that was blocked waiting for space. If the length
1121 * is non-zero and the ack didn't move, we're the
1122 * receiver side. If we're getting packets in-order
1123 * (the reassembly queue is empty), add the data to
1124 * the socket buffer and note that we need a delayed ack.
1125 * Make sure that the hidden state-flags are also off.
1126 * Since we check for TCPS_ESTABLISHED above, it can only
1129 if (tp->t_state == TCPS_ESTABLISHED &&
1130 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1131 !(tp->t_flags & (TF_NEEDSYN | TF_NEEDFIN)) &&
1132 (!(to.to_flags & TOF_TS) ||
1133 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) &&
1134 th->th_seq == tp->rcv_nxt &&
1135 tp->snd_nxt == tp->snd_max) {
1138 * If last ACK falls within this segment's sequence numbers,
1139 * record the timestamp.
1140 * NOTE that the test is modified according to the latest
1141 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1143 if ((to.to_flags & TOF_TS) &&
1144 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1145 tp->ts_recent_age = ticks;
1146 tp->ts_recent = to.to_tsval;
1150 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1151 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1152 tp->snd_cwnd >= tp->snd_wnd &&
1153 !IN_FASTRECOVERY(tp)) {
1155 * This is a pure ack for outstanding data.
1157 ++tcpstat.tcps_predack;
1159 * "bad retransmit" recovery
1161 * If Eifel detection applies, then
1162 * it is deterministic, so use it
1163 * unconditionally over the old heuristic.
1164 * Otherwise, fall back to the old heuristic.
1166 if (tcp_do_eifel_detect &&
1167 (to.to_flags & TOF_TS) && to.to_tsecr &&
1168 (tp->t_flags & TF_FIRSTACCACK)) {
1169 /* Eifel detection applicable. */
1170 if (to.to_tsecr < tp->t_rexmtTS) {
1171 tcp_revert_congestion_state(tp);
1172 ++tcpstat.tcps_eifeldetected;
1173 if (tp->t_rxtshift != 1 ||
1174 ticks >= tp->t_badrxtwin)
1175 ++tcpstat.tcps_rttcantdetect;
1177 } else if (tp->t_rxtshift == 1 &&
1178 ticks < tp->t_badrxtwin) {
1179 tcp_revert_congestion_state(tp);
1180 ++tcpstat.tcps_rttdetected;
1182 tp->t_flags &= ~(TF_FIRSTACCACK |
1183 TF_FASTREXMT | TF_EARLYREXMT);
1185 * Recalculate the retransmit timer / rtt.
1187 * Some machines (certain windows boxes)
1188 * send broken timestamp replies during the
1189 * SYN+ACK phase, ignore timestamps of 0.
1191 if ((to.to_flags & TOF_TS) && to.to_tsecr) {
1193 ticks - to.to_tsecr + 1);
1194 } else if (tp->t_rtttime &&
1195 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1197 ticks - tp->t_rtttime);
1199 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1200 acked = th->th_ack - tp->snd_una;
1201 tcpstat.tcps_rcvackpack++;
1202 tcpstat.tcps_rcvackbyte += acked;
1203 sbdrop(&so->so_snd.sb, acked);
1204 tp->snd_recover = th->th_ack - 1;
1205 tp->snd_una = th->th_ack;
1208 * Update window information.
1210 if (tiwin != tp->snd_wnd &&
1211 acceptable_window_update(tp, th, tiwin)) {
1212 /* keep track of pure window updates */
1213 if (tp->snd_wl2 == th->th_ack &&
1214 tiwin > tp->snd_wnd)
1215 tcpstat.tcps_rcvwinupd++;
1216 tp->snd_wnd = tiwin;
1217 tp->snd_wl1 = th->th_seq;
1218 tp->snd_wl2 = th->th_ack;
1219 if (tp->snd_wnd > tp->max_sndwnd)
1220 tp->max_sndwnd = tp->snd_wnd;
1223 ND6_HINT(tp); /* some progress has been done */
1225 * If all outstanding data are acked, stop
1226 * retransmit timer, otherwise restart timer
1227 * using current (possibly backed-off) value.
1228 * If process is waiting for space,
1229 * wakeup/selwakeup/signal. If data
1230 * are ready to send, let tcp_output
1231 * decide between more output or persist.
1233 if (tp->snd_una == tp->snd_max) {
1234 tcp_callout_stop(tp, tp->tt_rexmt);
1235 } else if (!tcp_callout_active(tp,
1237 tcp_callout_reset(tp, tp->tt_rexmt,
1238 tp->t_rxtcur, tcp_timer_rexmt);
1241 if (so->so_snd.ssb_cc > 0)
1243 return(IPPROTO_DONE);
1245 } else if (tiwin == tp->snd_wnd &&
1246 th->th_ack == tp->snd_una &&
1247 LIST_EMPTY(&tp->t_segq) &&
1248 tlen <= ssb_space(&so->so_rcv)) {
1249 u_long newsize = 0; /* automatic sockbuf scaling */
1251 * This is a pure, in-sequence data packet
1252 * with nothing on the reassembly queue and
1253 * we have enough buffer space to take it.
1255 ++tcpstat.tcps_preddat;
1256 tp->rcv_nxt += tlen;
1257 tcpstat.tcps_rcvpack++;
1258 tcpstat.tcps_rcvbyte += tlen;
1259 ND6_HINT(tp); /* some progress has been done */
1261 * Automatic sizing of receive socket buffer. Often the send
1262 * buffer size is not optimally adjusted to the actual network
1263 * conditions at hand (delay bandwidth product). Setting the
1264 * buffer size too small limits throughput on links with high
1265 * bandwidth and high delay (eg. trans-continental/oceanic links).
1267 * On the receive side the socket buffer memory is only rarely
1268 * used to any significant extent. This allows us to be much
1269 * more aggressive in scaling the receive socket buffer. For
1270 * the case that the buffer space is actually used to a large
1271 * extent and we run out of kernel memory we can simply drop
1272 * the new segments; TCP on the sender will just retransmit it
1273 * later. Setting the buffer size too big may only consume too
1274 * much kernel memory if the application doesn't read() from
1275 * the socket or packet loss or reordering makes use of the
1278 * The criteria to step up the receive buffer one notch are:
1279 * 1. the number of bytes received during the time it takes
1280 * one timestamp to be reflected back to us (the RTT);
1281 * 2. received bytes per RTT is within seven eighth of the
1282 * current socket buffer size;
1283 * 3. receive buffer size has not hit maximal automatic size;
1285 * This algorithm does one step per RTT at most and only if
1286 * we receive a bulk stream w/o packet losses or reorderings.
1287 * Shrinking the buffer during idle times is not necessary as
1288 * it doesn't consume any memory when idle.
1290 * TODO: Only step up if the application is actually serving
1291 * the buffer to better manage the socket buffer resources.
1293 if (tcp_do_autorcvbuf &&
1295 (so->so_rcv.ssb_flags & SSB_AUTOSIZE)) {
1296 if (to.to_tsecr > tp->rfbuf_ts &&
1297 to.to_tsecr - tp->rfbuf_ts < hz) {
1299 (so->so_rcv.ssb_hiwat / 8 * 7) &&
1300 so->so_rcv.ssb_hiwat <
1301 tcp_autorcvbuf_max) {
1303 ulmin(so->so_rcv.ssb_hiwat +
1305 tcp_autorcvbuf_max);
1307 /* Start over with next RTT. */
1311 tp->rfbuf_cnt += tlen; /* add up */
1314 * Add data to socket buffer.
1316 if (so->so_state & SS_CANTRCVMORE) {
1320 * Set new socket buffer size, give up when
1323 * Adjusting the size can mess up ACK
1324 * sequencing when pure window updates are
1325 * being avoided (which is the default),
1328 lwkt_gettoken(&so->so_rcv.ssb_token);
1330 tp->t_flags |= TF_RXRESIZED;
1331 if (!ssb_reserve(&so->so_rcv, newsize,
1333 atomic_clear_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE);
1336 (TCP_MAXWIN << tp->rcv_scale)) {
1337 atomic_clear_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE);
1340 m_adj(m, drop_hdrlen); /* delayed header drop */
1341 ssb_appendstream(&so->so_rcv, m);
1342 lwkt_reltoken(&so->so_rcv.ssb_token);
1346 * This code is responsible for most of the ACKs
1347 * the TCP stack sends back after receiving a data
1348 * packet. Note that the DELAY_ACK check fails if
1349 * the delack timer is already running, which results
1350 * in an ack being sent every other packet (which is
1353 * We then further aggregate acks by not actually
1354 * sending one until the protocol thread has completed
1355 * processing the current backlog of packets. This
1356 * does not delay the ack any further, but allows us
1357 * to take advantage of the packet aggregation that
1358 * high speed NICs do (usually blocks of 8-10 packets)
1359 * to send a single ack rather then four or five acks,
1360 * greatly reducing the ack rate, the return channel
1361 * bandwidth, and the protocol overhead on both ends.
1363 * Since this also has the effect of slowing down
1364 * the exponential slow-start ramp-up, systems with
1365 * very large bandwidth-delay products might want
1366 * to turn the feature off.
1368 if (DELAY_ACK(tp)) {
1369 tcp_callout_reset(tp, tp->tt_delack,
1370 tcp_delacktime, tcp_timer_delack);
1371 } else if (tcp_aggregate_acks) {
1372 tp->t_flags |= TF_ACKNOW;
1373 if (!(tp->t_flags & TF_ONOUTPUTQ)) {
1374 tp->t_flags |= TF_ONOUTPUTQ;
1375 tp->tt_cpu = mycpu->gd_cpuid;
1377 &tcpcbackq[tp->tt_cpu],
1381 tp->t_flags |= TF_ACKNOW;
1384 return(IPPROTO_DONE);
1389 * Calculate amount of space in receive window,
1390 * and then do TCP input processing.
1391 * Receive window is amount of space in rcv queue,
1392 * but not less than advertised window.
1394 recvwin = ssb_space(&so->so_rcv);
1397 tp->rcv_wnd = imax(recvwin, (int)(tp->rcv_adv - tp->rcv_nxt));
1399 /* Reset receive buffer auto scaling when not in bulk receive mode. */
1403 switch (tp->t_state) {
1405 * If the state is SYN_RECEIVED:
1406 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1408 case TCPS_SYN_RECEIVED:
1409 if ((thflags & TH_ACK) &&
1410 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1411 SEQ_GT(th->th_ack, tp->snd_max))) {
1412 rstreason = BANDLIM_RST_OPENPORT;
1418 * If the state is SYN_SENT:
1419 * if seg contains an ACK, but not for our SYN, drop the input.
1420 * if seg contains a RST, then drop the connection.
1421 * if seg does not contain SYN, then drop it.
1422 * Otherwise this is an acceptable SYN segment
1423 * initialize tp->rcv_nxt and tp->irs
1424 * if seg contains ack then advance tp->snd_una
1425 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1426 * arrange for segment to be acked (eventually)
1427 * continue processing rest of data/controls, beginning with URG
1430 if ((thflags & TH_ACK) &&
1431 (SEQ_LEQ(th->th_ack, tp->iss) ||
1432 SEQ_GT(th->th_ack, tp->snd_max))) {
1433 rstreason = BANDLIM_UNLIMITED;
1436 if (thflags & TH_RST) {
1437 if (thflags & TH_ACK)
1438 tp = tcp_drop(tp, ECONNREFUSED);
1441 if (!(thflags & TH_SYN))
1443 tp->snd_wnd = th->th_win; /* initial send window */
1445 tp->irs = th->th_seq;
1447 if (thflags & TH_ACK) {
1448 /* Our SYN was acked. */
1449 tcpstat.tcps_connects++;
1451 /* Do window scaling on this connection? */
1452 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
1453 (TF_RCVD_SCALE | TF_REQ_SCALE)) {
1454 tp->snd_scale = tp->requested_s_scale;
1455 tp->rcv_scale = tp->request_r_scale;
1457 tp->rcv_adv += tp->rcv_wnd;
1458 tp->snd_una++; /* SYN is acked */
1459 tcp_callout_stop(tp, tp->tt_rexmt);
1461 * If there's data, delay ACK; if there's also a FIN
1462 * ACKNOW will be turned on later.
1464 if (DELAY_ACK(tp) && tlen != 0) {
1465 tcp_callout_reset(tp, tp->tt_delack,
1466 tcp_delacktime, tcp_timer_delack);
1468 tp->t_flags |= TF_ACKNOW;
1471 * Received <SYN,ACK> in SYN_SENT[*] state.
1473 * SYN_SENT --> ESTABLISHED
1474 * SYN_SENT* --> FIN_WAIT_1
1476 tp->t_starttime = ticks;
1477 if (tp->t_flags & TF_NEEDFIN) {
1478 tp->t_state = TCPS_FIN_WAIT_1;
1479 tp->t_flags &= ~TF_NEEDFIN;
1482 tcp_established(tp);
1486 * Received initial SYN in SYN-SENT[*] state =>
1487 * simultaneous open.
1488 * Do 3-way handshake:
1489 * SYN-SENT -> SYN-RECEIVED
1490 * SYN-SENT* -> SYN-RECEIVED*
1492 tp->t_flags |= TF_ACKNOW;
1493 tcp_callout_stop(tp, tp->tt_rexmt);
1494 tp->t_state = TCPS_SYN_RECEIVED;
1498 * Advance th->th_seq to correspond to first data byte.
1499 * If data, trim to stay within window,
1500 * dropping FIN if necessary.
1503 if (tlen > tp->rcv_wnd) {
1504 todrop = tlen - tp->rcv_wnd;
1508 tcpstat.tcps_rcvpackafterwin++;
1509 tcpstat.tcps_rcvbyteafterwin += todrop;
1511 tp->snd_wl1 = th->th_seq - 1;
1512 tp->rcv_up = th->th_seq;
1514 * Client side of transaction: already sent SYN and data.
1515 * If the remote host used T/TCP to validate the SYN,
1516 * our data will be ACK'd; if so, enter normal data segment
1517 * processing in the middle of step 5, ack processing.
1518 * Otherwise, goto step 6.
1520 if (thflags & TH_ACK)
1526 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
1527 * do normal processing (we no longer bother with T/TCP).
1531 case TCPS_TIME_WAIT:
1532 break; /* continue normal processing */
1536 * States other than LISTEN or SYN_SENT.
1537 * First check the RST flag and sequence number since reset segments
1538 * are exempt from the timestamp and connection count tests. This
1539 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
1540 * below which allowed reset segments in half the sequence space
1541 * to fall though and be processed (which gives forged reset
1542 * segments with a random sequence number a 50 percent chance of
1543 * killing a connection).
1544 * Then check timestamp, if present.
1545 * Then check the connection count, if present.
1546 * Then check that at least some bytes of segment are within
1547 * receive window. If segment begins before rcv_nxt,
1548 * drop leading data (and SYN); if nothing left, just ack.
1551 * If the RST bit is set, check the sequence number to see
1552 * if this is a valid reset segment.
1554 * In all states except SYN-SENT, all reset (RST) segments
1555 * are validated by checking their SEQ-fields. A reset is
1556 * valid if its sequence number is in the window.
1557 * Note: this does not take into account delayed ACKs, so
1558 * we should test against last_ack_sent instead of rcv_nxt.
1559 * The sequence number in the reset segment is normally an
1560 * echo of our outgoing acknowledgement numbers, but some hosts
1561 * send a reset with the sequence number at the rightmost edge
1562 * of our receive window, and we have to handle this case.
1563 * If we have multiple segments in flight, the intial reset
1564 * segment sequence numbers will be to the left of last_ack_sent,
1565 * but they will eventually catch up.
1566 * In any case, it never made sense to trim reset segments to
1567 * fit the receive window since RFC 1122 says:
1568 * 4.2.2.12 RST Segment: RFC-793 Section 3.4
1570 * A TCP SHOULD allow a received RST segment to include data.
1573 * It has been suggested that a RST segment could contain
1574 * ASCII text that encoded and explained the cause of the
1575 * RST. No standard has yet been established for such
1578 * If the reset segment passes the sequence number test examine
1580 * SYN_RECEIVED STATE:
1581 * If passive open, return to LISTEN state.
1582 * If active open, inform user that connection was refused.
1583 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
1584 * Inform user that connection was reset, and close tcb.
1585 * CLOSING, LAST_ACK STATES:
1588 * Drop the segment - see Stevens, vol. 2, p. 964 and
1591 if (thflags & TH_RST) {
1592 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
1593 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
1594 switch (tp->t_state) {
1596 case TCPS_SYN_RECEIVED:
1597 so->so_error = ECONNREFUSED;
1600 case TCPS_ESTABLISHED:
1601 case TCPS_FIN_WAIT_1:
1602 case TCPS_FIN_WAIT_2:
1603 case TCPS_CLOSE_WAIT:
1604 so->so_error = ECONNRESET;
1606 tp->t_state = TCPS_CLOSED;
1607 tcpstat.tcps_drops++;
1616 case TCPS_TIME_WAIT:
1624 * RFC 1323 PAWS: If we have a timestamp reply on this segment
1625 * and it's less than ts_recent, drop it.
1627 if ((to.to_flags & TOF_TS) && tp->ts_recent != 0 &&
1628 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
1630 /* Check to see if ts_recent is over 24 days old. */
1631 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1633 * Invalidate ts_recent. If this segment updates
1634 * ts_recent, the age will be reset later and ts_recent
1635 * will get a valid value. If it does not, setting
1636 * ts_recent to zero will at least satisfy the
1637 * requirement that zero be placed in the timestamp
1638 * echo reply when ts_recent isn't valid. The
1639 * age isn't reset until we get a valid ts_recent
1640 * because we don't want out-of-order segments to be
1641 * dropped when ts_recent is old.
1645 tcpstat.tcps_rcvduppack++;
1646 tcpstat.tcps_rcvdupbyte += tlen;
1647 tcpstat.tcps_pawsdrop++;
1655 * In the SYN-RECEIVED state, validate that the packet belongs to
1656 * this connection before trimming the data to fit the receive
1657 * window. Check the sequence number versus IRS since we know
1658 * the sequence numbers haven't wrapped. This is a partial fix
1659 * for the "LAND" DoS attack.
1661 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
1662 rstreason = BANDLIM_RST_OPENPORT;
1666 todrop = tp->rcv_nxt - th->th_seq;
1668 if (TCP_DO_SACK(tp)) {
1669 /* Report duplicate segment at head of packet. */
1670 tp->reportblk.rblk_start = th->th_seq;
1671 tp->reportblk.rblk_end = th->th_seq + tlen;
1672 if (thflags & TH_FIN)
1673 ++tp->reportblk.rblk_end;
1674 if (SEQ_GT(tp->reportblk.rblk_end, tp->rcv_nxt))
1675 tp->reportblk.rblk_end = tp->rcv_nxt;
1676 tp->t_flags |= (TF_DUPSEG | TF_SACKLEFT | TF_ACKNOW);
1678 if (thflags & TH_SYN) {
1688 * Following if statement from Stevens, vol. 2, p. 960.
1690 if (todrop > tlen ||
1691 (todrop == tlen && !(thflags & TH_FIN))) {
1693 * Any valid FIN must be to the left of the window.
1694 * At this point the FIN must be a duplicate or out
1695 * of sequence; drop it.
1700 * Send an ACK to resynchronize and drop any data.
1701 * But keep on processing for RST or ACK.
1703 tp->t_flags |= TF_ACKNOW;
1705 tcpstat.tcps_rcvduppack++;
1706 tcpstat.tcps_rcvdupbyte += todrop;
1708 tcpstat.tcps_rcvpartduppack++;
1709 tcpstat.tcps_rcvpartdupbyte += todrop;
1711 drop_hdrlen += todrop; /* drop from the top afterwards */
1712 th->th_seq += todrop;
1714 if (th->th_urp > todrop)
1715 th->th_urp -= todrop;
1723 * If new data are received on a connection after the
1724 * user processes are gone, then RST the other end.
1726 if ((so->so_state & SS_NOFDREF) &&
1727 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1729 tcpstat.tcps_rcvafterclose++;
1730 rstreason = BANDLIM_UNLIMITED;
1735 * If segment ends after window, drop trailing data
1736 * (and PUSH and FIN); if nothing left, just ACK.
1738 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
1740 tcpstat.tcps_rcvpackafterwin++;
1741 if (todrop >= tlen) {
1742 tcpstat.tcps_rcvbyteafterwin += tlen;
1744 * If a new connection request is received
1745 * while in TIME_WAIT, drop the old connection
1746 * and start over if the sequence numbers
1747 * are above the previous ones.
1749 if (thflags & TH_SYN &&
1750 tp->t_state == TCPS_TIME_WAIT &&
1751 SEQ_GT(th->th_seq, tp->rcv_nxt)) {
1756 * If window is closed can only take segments at
1757 * window edge, and have to drop data and PUSH from
1758 * incoming segments. Continue processing, but
1759 * remember to ack. Otherwise, drop segment
1762 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1763 tp->t_flags |= TF_ACKNOW;
1764 tcpstat.tcps_rcvwinprobe++;
1768 tcpstat.tcps_rcvbyteafterwin += todrop;
1771 thflags &= ~(TH_PUSH | TH_FIN);
1775 * If last ACK falls within this segment's sequence numbers,
1776 * record its timestamp.
1778 * 1) That the test incorporates suggestions from the latest
1779 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1780 * 2) That updating only on newer timestamps interferes with
1781 * our earlier PAWS tests, so this check should be solely
1782 * predicated on the sequence space of this segment.
1783 * 3) That we modify the segment boundary check to be
1784 * Last.ACK.Sent <= SEG.SEQ + SEG.LEN
1785 * instead of RFC1323's
1786 * Last.ACK.Sent < SEG.SEQ + SEG.LEN,
1787 * This modified check allows us to overcome RFC1323's
1788 * limitations as described in Stevens TCP/IP Illustrated
1789 * Vol. 2 p.869. In such cases, we can still calculate the
1790 * RTT correctly when RCV.NXT == Last.ACK.Sent.
1792 if ((to.to_flags & TOF_TS) && SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
1793 SEQ_LEQ(tp->last_ack_sent, (th->th_seq + tlen
1794 + ((thflags & TH_SYN) != 0)
1795 + ((thflags & TH_FIN) != 0)))) {
1796 tp->ts_recent_age = ticks;
1797 tp->ts_recent = to.to_tsval;
1801 * If a SYN is in the window, then this is an
1802 * error and we send an RST and drop the connection.
1804 if (thflags & TH_SYN) {
1805 tp = tcp_drop(tp, ECONNRESET);
1806 rstreason = BANDLIM_UNLIMITED;
1811 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
1812 * flag is on (half-synchronized state), then queue data for
1813 * later processing; else drop segment and return.
1815 if (!(thflags & TH_ACK)) {
1816 if (tp->t_state == TCPS_SYN_RECEIVED ||
1817 (tp->t_flags & TF_NEEDSYN))
1826 switch (tp->t_state) {
1828 * In SYN_RECEIVED state, the ACK acknowledges our SYN, so enter
1829 * ESTABLISHED state and continue processing.
1830 * The ACK was checked above.
1832 case TCPS_SYN_RECEIVED:
1834 tcpstat.tcps_connects++;
1836 /* Do window scaling? */
1837 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
1838 (TF_RCVD_SCALE | TF_REQ_SCALE)) {
1839 tp->snd_scale = tp->requested_s_scale;
1840 tp->rcv_scale = tp->request_r_scale;
1844 * SYN-RECEIVED -> ESTABLISHED
1845 * SYN-RECEIVED* -> FIN-WAIT-1
1847 tp->t_starttime = ticks;
1848 if (tp->t_flags & TF_NEEDFIN) {
1849 tp->t_state = TCPS_FIN_WAIT_1;
1850 tp->t_flags &= ~TF_NEEDFIN;
1852 tcp_established(tp);
1855 * If segment contains data or ACK, will call tcp_reass()
1856 * later; if not, do so now to pass queued data to user.
1858 if (tlen == 0 && !(thflags & TH_FIN))
1859 tcp_reass(tp, NULL, NULL, NULL);
1863 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1864 * ACKs. If the ack is in the range
1865 * tp->snd_una < th->th_ack <= tp->snd_max
1866 * then advance tp->snd_una to th->th_ack and drop
1867 * data from the retransmission queue. If this ACK reflects
1868 * more up to date window information we update our window information.
1870 case TCPS_ESTABLISHED:
1871 case TCPS_FIN_WAIT_1:
1872 case TCPS_FIN_WAIT_2:
1873 case TCPS_CLOSE_WAIT:
1876 case TCPS_TIME_WAIT:
1878 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
1879 if (TCP_DO_SACK(tp))
1880 tcp_sack_update_scoreboard(tp, &to);
1881 if (tlen != 0 || tiwin != tp->snd_wnd) {
1885 tcpstat.tcps_rcvdupack++;
1886 if (!tcp_callout_active(tp, tp->tt_rexmt) ||
1887 th->th_ack != tp->snd_una) {
1892 * We have outstanding data (other than
1893 * a window probe), this is a completely
1894 * duplicate ack (ie, window info didn't
1895 * change), the ack is the biggest we've
1896 * seen and we've seen exactly our rexmt
1897 * threshhold of them, so assume a packet
1898 * has been dropped and retransmit it.
1899 * Kludge snd_nxt & the congestion
1900 * window so we send only this one
1903 if (IN_FASTRECOVERY(tp)) {
1904 if (TCP_DO_SACK(tp)) {
1905 /* No artifical cwnd inflation. */
1906 tcp_sack_rexmt(tp, th);
1909 * Dup acks mean that packets
1910 * have left the network
1911 * (they're now cached at the
1912 * receiver) so bump cwnd by
1913 * the amount in the receiver
1914 * to keep a constant cwnd
1915 * packets in the network.
1917 tp->snd_cwnd += tp->t_maxseg;
1920 } else if (SEQ_LT(th->th_ack, tp->snd_recover)) {
1923 } else if (++tp->t_dupacks == tcprexmtthresh) {
1924 tcp_seq old_snd_nxt;
1928 if (tcp_do_eifel_detect &&
1929 (tp->t_flags & TF_RCVD_TSTMP)) {
1930 tcp_save_congestion_state(tp);
1931 tp->t_flags |= TF_FASTREXMT;
1934 * We know we're losing at the current
1935 * window size, so do congestion avoidance:
1936 * set ssthresh to half the current window
1937 * and pull our congestion window back to the
1940 win = min(tp->snd_wnd, tp->snd_cwnd) / 2 /
1944 tp->snd_ssthresh = win * tp->t_maxseg;
1945 ENTER_FASTRECOVERY(tp);
1946 tp->snd_recover = tp->snd_max;
1947 tcp_callout_stop(tp, tp->tt_rexmt);
1949 old_snd_nxt = tp->snd_nxt;
1950 tp->snd_nxt = th->th_ack;
1951 tp->snd_cwnd = tp->t_maxseg;
1953 ++tcpstat.tcps_sndfastrexmit;
1954 tp->snd_cwnd = tp->snd_ssthresh;
1955 tp->rexmt_high = tp->snd_nxt;
1956 if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
1957 tp->snd_nxt = old_snd_nxt;
1958 KASSERT(tp->snd_limited <= 2,
1959 ("tp->snd_limited too big"));
1960 if (TCP_DO_SACK(tp))
1961 tcp_sack_rexmt(tp, th);
1963 tp->snd_cwnd += tp->t_maxseg *
1964 (tp->t_dupacks - tp->snd_limited);
1965 } else if (tcp_do_limitedtransmit) {
1966 u_long oldcwnd = tp->snd_cwnd;
1967 tcp_seq oldsndmax = tp->snd_max;
1968 tcp_seq oldsndnxt = tp->snd_nxt;
1969 /* outstanding data */
1970 uint32_t ownd = tp->snd_max - tp->snd_una;
1973 #define iceildiv(n, d) (((n)+(d)-1) / (d))
1975 KASSERT(tp->t_dupacks == 1 ||
1977 ("dupacks not 1 or 2"));
1978 if (tp->t_dupacks == 1)
1979 tp->snd_limited = 0;
1980 tp->snd_nxt = tp->snd_max;
1981 tp->snd_cwnd = ownd +
1982 (tp->t_dupacks - tp->snd_limited) *
1987 * Other acks may have been processed,
1988 * snd_nxt cannot be reset to a value less
1991 if (SEQ_LT(oldsndnxt, oldsndmax)) {
1992 if (SEQ_GT(oldsndnxt, tp->snd_una))
1993 tp->snd_nxt = oldsndnxt;
1995 tp->snd_nxt = tp->snd_una;
1997 tp->snd_cwnd = oldcwnd;
1998 sent = tp->snd_max - oldsndmax;
1999 if (sent > tp->t_maxseg) {
2000 KASSERT((tp->t_dupacks == 2 &&
2001 tp->snd_limited == 0) ||
2002 (sent == tp->t_maxseg + 1 &&
2003 tp->t_flags & TF_SENTFIN),
2005 KASSERT(sent <= tp->t_maxseg * 2,
2006 ("sent too many segments"));
2007 tp->snd_limited = 2;
2008 tcpstat.tcps_sndlimited += 2;
2009 } else if (sent > 0) {
2011 ++tcpstat.tcps_sndlimited;
2012 } else if (tcp_do_early_retransmit &&
2013 (tcp_do_eifel_detect &&
2014 (tp->t_flags & TF_RCVD_TSTMP)) &&
2015 ownd < 4 * tp->t_maxseg &&
2016 tp->t_dupacks + 1 >=
2017 iceildiv(ownd, tp->t_maxseg) &&
2018 (!TCP_DO_SACK(tp) ||
2019 ownd <= tp->t_maxseg ||
2020 tcp_sack_has_sacked(&tp->scb,
2021 ownd - tp->t_maxseg))) {
2022 ++tcpstat.tcps_sndearlyrexmit;
2023 tp->t_flags |= TF_EARLYREXMT;
2024 goto fastretransmit;
2030 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una"));
2032 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2034 * Detected optimistic ACK attack.
2035 * Force slow-start to de-synchronize attack.
2037 tp->snd_cwnd = tp->t_maxseg;
2040 tcpstat.tcps_rcvacktoomuch++;
2044 * If we reach this point, ACK is not a duplicate,
2045 * i.e., it ACKs something we sent.
2047 if (tp->t_flags & TF_NEEDSYN) {
2049 * T/TCP: Connection was half-synchronized, and our
2050 * SYN has been ACK'd (so connection is now fully
2051 * synchronized). Go to non-starred state,
2052 * increment snd_una for ACK of SYN, and check if
2053 * we can do window scaling.
2055 tp->t_flags &= ~TF_NEEDSYN;
2057 /* Do window scaling? */
2058 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
2059 (TF_RCVD_SCALE | TF_REQ_SCALE)) {
2060 tp->snd_scale = tp->requested_s_scale;
2061 tp->rcv_scale = tp->request_r_scale;
2066 acked = th->th_ack - tp->snd_una;
2067 tcpstat.tcps_rcvackpack++;
2068 tcpstat.tcps_rcvackbyte += acked;
2070 if (tcp_do_eifel_detect && acked > 0 &&
2071 (to.to_flags & TOF_TS) && (to.to_tsecr != 0) &&
2072 (tp->t_flags & TF_FIRSTACCACK)) {
2073 /* Eifel detection applicable. */
2074 if (to.to_tsecr < tp->t_rexmtTS) {
2075 ++tcpstat.tcps_eifeldetected;
2076 tcp_revert_congestion_state(tp);
2077 if (tp->t_rxtshift != 1 ||
2078 ticks >= tp->t_badrxtwin)
2079 ++tcpstat.tcps_rttcantdetect;
2081 } else if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) {
2083 * If we just performed our first retransmit,
2084 * and the ACK arrives within our recovery window,
2085 * then it was a mistake to do the retransmit
2086 * in the first place. Recover our original cwnd
2087 * and ssthresh, and proceed to transmit where we
2090 tcp_revert_congestion_state(tp);
2091 ++tcpstat.tcps_rttdetected;
2095 * If we have a timestamp reply, update smoothed
2096 * round trip time. If no timestamp is present but
2097 * transmit timer is running and timed sequence
2098 * number was acked, update smoothed round trip time.
2099 * Since we now have an rtt measurement, cancel the
2100 * timer backoff (cf., Phil Karn's retransmit alg.).
2101 * Recompute the initial retransmit timer.
2103 * Some machines (certain windows boxes) send broken
2104 * timestamp replies during the SYN+ACK phase, ignore
2107 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0))
2108 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1);
2109 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq))
2110 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2111 tcp_xmit_bandwidth_limit(tp, th->th_ack);
2114 * If no data (only SYN) was ACK'd,
2115 * skip rest of ACK processing.
2120 /* Stop looking for an acceptable ACK since one was received. */
2121 tp->t_flags &= ~(TF_FIRSTACCACK | TF_FASTREXMT | TF_EARLYREXMT);
2123 if (acked > so->so_snd.ssb_cc) {
2124 tp->snd_wnd -= so->so_snd.ssb_cc;
2125 sbdrop(&so->so_snd.sb, (int)so->so_snd.ssb_cc);
2126 ourfinisacked = TRUE;
2128 sbdrop(&so->so_snd.sb, acked);
2129 tp->snd_wnd -= acked;
2130 ourfinisacked = FALSE;
2135 * Update window information.
2136 * Don't look at window if no ACK:
2137 * TAC's send garbage on first SYN.
2139 if (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2140 (tp->snd_wl1 == th->th_seq &&
2141 (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2142 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)))) {
2143 /* keep track of pure window updates */
2144 if (tlen == 0 && tp->snd_wl2 == th->th_ack &&
2145 tiwin > tp->snd_wnd)
2146 tcpstat.tcps_rcvwinupd++;
2147 tp->snd_wnd = tiwin;
2148 tp->snd_wl1 = th->th_seq;
2149 tp->snd_wl2 = th->th_ack;
2150 if (tp->snd_wnd > tp->max_sndwnd)
2151 tp->max_sndwnd = tp->snd_wnd;
2155 tp->snd_una = th->th_ack;
2156 if (TCP_DO_SACK(tp))
2157 tcp_sack_update_scoreboard(tp, &to);
2158 if (IN_FASTRECOVERY(tp)) {
2159 if (SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2160 EXIT_FASTRECOVERY(tp);
2163 * If the congestion window was inflated
2164 * to account for the other side's
2165 * cached packets, retract it.
2167 if (!TCP_DO_SACK(tp))
2168 tp->snd_cwnd = tp->snd_ssthresh;
2171 * Window inflation should have left us
2172 * with approximately snd_ssthresh outstanding
2173 * data. But, in case we would be inclined
2174 * to send a burst, better do it using
2177 if (SEQ_GT(th->th_ack + tp->snd_cwnd,
2178 tp->snd_max + 2 * tp->t_maxseg))
2180 (tp->snd_max - tp->snd_una) +
2185 if (TCP_DO_SACK(tp)) {
2186 tp->snd_max_rexmt = tp->snd_max;
2187 tcp_sack_rexmt(tp, th);
2189 tcp_newreno_partial_ack(tp, th, acked);
2195 * Open the congestion window. When in slow-start,
2196 * open exponentially: maxseg per packet. Otherwise,
2197 * open linearly: maxseg per window.
2199 if (tp->snd_cwnd <= tp->snd_ssthresh) {
2201 (SEQ_LT(tp->snd_nxt, tp->snd_max) ?
2202 tp->t_maxseg : 2 * tp->t_maxseg);
2205 tp->snd_cwnd += tcp_do_abc ?
2206 min(acked, abc_sslimit) : tp->t_maxseg;
2208 /* linear increase */
2209 tp->snd_wacked += tcp_do_abc ? acked :
2211 if (tp->snd_wacked >= tp->snd_cwnd) {
2212 tp->snd_wacked -= tp->snd_cwnd;
2213 tp->snd_cwnd += tp->t_maxseg;
2216 tp->snd_cwnd = min(tp->snd_cwnd,
2217 TCP_MAXWIN << tp->snd_scale);
2218 tp->snd_recover = th->th_ack - 1;
2220 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2221 tp->snd_nxt = tp->snd_una;
2224 * If all outstanding data is acked, stop retransmit
2225 * timer and remember to restart (more output or persist).
2226 * If there is more data to be acked, restart retransmit
2227 * timer, using current (possibly backed-off) value.
2229 if (th->th_ack == tp->snd_max) {
2230 tcp_callout_stop(tp, tp->tt_rexmt);
2232 } else if (!tcp_callout_active(tp, tp->tt_persist)) {
2233 tcp_callout_reset(tp, tp->tt_rexmt, tp->t_rxtcur,
2237 switch (tp->t_state) {
2239 * In FIN_WAIT_1 STATE in addition to the processing
2240 * for the ESTABLISHED state if our FIN is now acknowledged
2241 * then enter FIN_WAIT_2.
2243 case TCPS_FIN_WAIT_1:
2244 if (ourfinisacked) {
2246 * If we can't receive any more
2247 * data, then closing user can proceed.
2248 * Starting the timer is contrary to the
2249 * specification, but if we don't get a FIN
2250 * we'll hang forever.
2252 if (so->so_state & SS_CANTRCVMORE) {
2253 soisdisconnected(so);
2254 tcp_callout_reset(tp, tp->tt_2msl,
2255 tp->t_maxidle, tcp_timer_2msl);
2257 tp->t_state = TCPS_FIN_WAIT_2;
2262 * In CLOSING STATE in addition to the processing for
2263 * the ESTABLISHED state if the ACK acknowledges our FIN
2264 * then enter the TIME-WAIT state, otherwise ignore
2268 if (ourfinisacked) {
2269 tp->t_state = TCPS_TIME_WAIT;
2270 tcp_canceltimers(tp);
2271 tcp_callout_reset(tp, tp->tt_2msl,
2272 2 * tcp_rmx_msl(tp),
2274 soisdisconnected(so);
2279 * In LAST_ACK, we may still be waiting for data to drain
2280 * and/or to be acked, as well as for the ack of our FIN.
2281 * If our FIN is now acknowledged, delete the TCB,
2282 * enter the closed state and return.
2285 if (ourfinisacked) {
2292 * In TIME_WAIT state the only thing that should arrive
2293 * is a retransmission of the remote FIN. Acknowledge
2294 * it and restart the finack timer.
2296 case TCPS_TIME_WAIT:
2297 tcp_callout_reset(tp, tp->tt_2msl, 2 * tcp_rmx_msl(tp),
2305 * Update window information.
2306 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2308 if ((thflags & TH_ACK) &&
2309 acceptable_window_update(tp, th, tiwin)) {
2310 /* keep track of pure window updates */
2311 if (tlen == 0 && tp->snd_wl2 == th->th_ack &&
2312 tiwin > tp->snd_wnd)
2313 tcpstat.tcps_rcvwinupd++;
2314 tp->snd_wnd = tiwin;
2315 tp->snd_wl1 = th->th_seq;
2316 tp->snd_wl2 = th->th_ack;
2317 if (tp->snd_wnd > tp->max_sndwnd)
2318 tp->max_sndwnd = tp->snd_wnd;
2323 * Process segments with URG.
2325 if ((thflags & TH_URG) && th->th_urp &&
2326 !TCPS_HAVERCVDFIN(tp->t_state)) {
2328 * This is a kludge, but if we receive and accept
2329 * random urgent pointers, we'll crash in
2330 * soreceive. It's hard to imagine someone
2331 * actually wanting to send this much urgent data.
2333 if (th->th_urp + so->so_rcv.ssb_cc > sb_max) {
2334 th->th_urp = 0; /* XXX */
2335 thflags &= ~TH_URG; /* XXX */
2336 goto dodata; /* XXX */
2339 * If this segment advances the known urgent pointer,
2340 * then mark the data stream. This should not happen
2341 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2342 * a FIN has been received from the remote side.
2343 * In these states we ignore the URG.
2345 * According to RFC961 (Assigned Protocols),
2346 * the urgent pointer points to the last octet
2347 * of urgent data. We continue, however,
2348 * to consider it to indicate the first octet
2349 * of data past the urgent section as the original
2350 * spec states (in one of two places).
2352 if (SEQ_GT(th->th_seq + th->th_urp, tp->rcv_up)) {
2353 tp->rcv_up = th->th_seq + th->th_urp;
2354 so->so_oobmark = so->so_rcv.ssb_cc +
2355 (tp->rcv_up - tp->rcv_nxt) - 1;
2356 if (so->so_oobmark == 0)
2357 sosetstate(so, SS_RCVATMARK);
2359 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2362 * Remove out of band data so doesn't get presented to user.
2363 * This can happen independent of advancing the URG pointer,
2364 * but if two URG's are pending at once, some out-of-band
2365 * data may creep in... ick.
2367 if (th->th_urp <= (u_long)tlen &&
2368 !(so->so_options & SO_OOBINLINE)) {
2369 /* hdr drop is delayed */
2370 tcp_pulloutofband(so, th, m, drop_hdrlen);
2374 * If no out of band data is expected,
2375 * pull receive urgent pointer along
2376 * with the receive window.
2378 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2379 tp->rcv_up = tp->rcv_nxt;
2384 * Process the segment text, merging it into the TCP sequencing queue,
2385 * and arranging for acknowledgment of receipt if necessary.
2386 * This process logically involves adjusting tp->rcv_wnd as data
2387 * is presented to the user (this happens in tcp_usrreq.c,
2388 * case PRU_RCVD). If a FIN has already been received on this
2389 * connection then we just ignore the text.
2391 if ((tlen || (thflags & TH_FIN)) && !TCPS_HAVERCVDFIN(tp->t_state)) {
2392 m_adj(m, drop_hdrlen); /* delayed header drop */
2394 * Insert segment which includes th into TCP reassembly queue
2395 * with control block tp. Set thflags to whether reassembly now
2396 * includes a segment with FIN. This handles the common case
2397 * inline (segment is the next to be received on an established
2398 * connection, and the queue is empty), avoiding linkage into
2399 * and removal from the queue and repetition of various
2401 * Set DELACK for segments received in order, but ack
2402 * immediately when segments are out of order (so
2403 * fast retransmit can work).
2405 if (th->th_seq == tp->rcv_nxt &&
2406 LIST_EMPTY(&tp->t_segq) &&
2407 TCPS_HAVEESTABLISHED(tp->t_state)) {
2408 if (DELAY_ACK(tp)) {
2409 tcp_callout_reset(tp, tp->tt_delack,
2410 tcp_delacktime, tcp_timer_delack);
2412 tp->t_flags |= TF_ACKNOW;
2414 tp->rcv_nxt += tlen;
2415 thflags = th->th_flags & TH_FIN;
2416 tcpstat.tcps_rcvpack++;
2417 tcpstat.tcps_rcvbyte += tlen;
2419 if (so->so_state & SS_CANTRCVMORE) {
2422 lwkt_gettoken(&so->so_rcv.ssb_token);
2423 ssb_appendstream(&so->so_rcv, m);
2424 lwkt_reltoken(&so->so_rcv.ssb_token);
2428 if (!(tp->t_flags & TF_DUPSEG)) {
2429 /* Initialize SACK report block. */
2430 tp->reportblk.rblk_start = th->th_seq;
2431 tp->reportblk.rblk_end = th->th_seq + tlen +
2432 ((thflags & TH_FIN) != 0);
2434 thflags = tcp_reass(tp, th, &tlen, m);
2435 tp->t_flags |= TF_ACKNOW;
2439 * Note the amount of data that peer has sent into
2440 * our window, in order to estimate the sender's
2443 len = so->so_rcv.ssb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2450 * If FIN is received ACK the FIN and let the user know
2451 * that the connection is closing.
2453 if (thflags & TH_FIN) {
2454 if (!TCPS_HAVERCVDFIN(tp->t_state)) {
2457 * If connection is half-synchronized
2458 * (ie NEEDSYN flag on) then delay ACK,
2459 * so it may be piggybacked when SYN is sent.
2460 * Otherwise, since we received a FIN then no
2461 * more input can be expected, send ACK now.
2463 if (DELAY_ACK(tp) && (tp->t_flags & TF_NEEDSYN)) {
2464 tcp_callout_reset(tp, tp->tt_delack,
2465 tcp_delacktime, tcp_timer_delack);
2467 tp->t_flags |= TF_ACKNOW;
2472 switch (tp->t_state) {
2474 * In SYN_RECEIVED and ESTABLISHED STATES
2475 * enter the CLOSE_WAIT state.
2477 case TCPS_SYN_RECEIVED:
2478 tp->t_starttime = ticks;
2480 case TCPS_ESTABLISHED:
2481 tp->t_state = TCPS_CLOSE_WAIT;
2485 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2486 * enter the CLOSING state.
2488 case TCPS_FIN_WAIT_1:
2489 tp->t_state = TCPS_CLOSING;
2493 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2494 * starting the time-wait timer, turning off the other
2497 case TCPS_FIN_WAIT_2:
2498 tp->t_state = TCPS_TIME_WAIT;
2499 tcp_canceltimers(tp);
2500 tcp_callout_reset(tp, tp->tt_2msl, 2 * tcp_rmx_msl(tp),
2502 soisdisconnected(so);
2506 * In TIME_WAIT state restart the 2 MSL time_wait timer.
2508 case TCPS_TIME_WAIT:
2509 tcp_callout_reset(tp, tp->tt_2msl, 2 * tcp_rmx_msl(tp),
2516 if (so->so_options & SO_DEBUG)
2517 tcp_trace(TA_INPUT, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2521 * Return any desired output.
2523 if (needoutput || (tp->t_flags & TF_ACKNOW))
2525 return(IPPROTO_DONE);
2529 * Generate an ACK dropping incoming segment if it occupies
2530 * sequence space, where the ACK reflects our state.
2532 * We can now skip the test for the RST flag since all
2533 * paths to this code happen after packets containing
2534 * RST have been dropped.
2536 * In the SYN-RECEIVED state, don't send an ACK unless the
2537 * segment we received passes the SYN-RECEIVED ACK test.
2538 * If it fails send a RST. This breaks the loop in the
2539 * "LAND" DoS attack, and also prevents an ACK storm
2540 * between two listening ports that have been sent forged
2541 * SYN segments, each with the source address of the other.
2543 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
2544 (SEQ_GT(tp->snd_una, th->th_ack) ||
2545 SEQ_GT(th->th_ack, tp->snd_max)) ) {
2546 rstreason = BANDLIM_RST_OPENPORT;
2550 if (so->so_options & SO_DEBUG)
2551 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2554 tp->t_flags |= TF_ACKNOW;
2556 return(IPPROTO_DONE);
2560 * Generate a RST, dropping incoming segment.
2561 * Make ACK acceptable to originator of segment.
2562 * Don't bother to respond if destination was broadcast/multicast.
2564 if ((thflags & TH_RST) || m->m_flags & (M_BCAST | M_MCAST))
2567 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
2568 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
2571 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
2572 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
2573 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
2574 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2577 /* IPv6 anycast check is done at tcp6_input() */
2580 * Perform bandwidth limiting.
2583 if (badport_bandlim(rstreason) < 0)
2588 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2589 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2591 if (thflags & TH_ACK)
2592 /* mtod() below is safe as long as hdr dropping is delayed */
2593 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack,
2596 if (thflags & TH_SYN)
2598 /* mtod() below is safe as long as hdr dropping is delayed */
2599 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq + tlen,
2600 (tcp_seq)0, TH_RST | TH_ACK);
2602 return(IPPROTO_DONE);
2606 * Drop space held by incoming segment and return.
2609 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2610 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2613 return(IPPROTO_DONE);
2617 * Parse TCP options and place in tcpopt.
2620 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, boolean_t is_syn)
2625 for (; cnt > 0; cnt -= optlen, cp += optlen) {
2627 if (opt == TCPOPT_EOL)
2629 if (opt == TCPOPT_NOP)
2635 if (optlen < 2 || optlen > cnt)
2640 if (optlen != TCPOLEN_MAXSEG)
2644 to->to_flags |= TOF_MSS;
2645 bcopy(cp + 2, &to->to_mss, sizeof to->to_mss);
2646 to->to_mss = ntohs(to->to_mss);
2649 if (optlen != TCPOLEN_WINDOW)
2653 to->to_flags |= TOF_SCALE;
2654 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
2656 case TCPOPT_TIMESTAMP:
2657 if (optlen != TCPOLEN_TIMESTAMP)
2659 to->to_flags |= TOF_TS;
2660 bcopy(cp + 2, &to->to_tsval, sizeof to->to_tsval);
2661 to->to_tsval = ntohl(to->to_tsval);
2662 bcopy(cp + 6, &to->to_tsecr, sizeof to->to_tsecr);
2663 to->to_tsecr = ntohl(to->to_tsecr);
2665 * If echoed timestamp is later than the current time,
2666 * fall back to non RFC1323 RTT calculation.
2668 if (to->to_tsecr != 0 && TSTMP_GT(to->to_tsecr, ticks))
2671 case TCPOPT_SACK_PERMITTED:
2672 if (optlen != TCPOLEN_SACK_PERMITTED)
2676 to->to_flags |= TOF_SACK_PERMITTED;
2679 if ((optlen - 2) & 0x07) /* not multiple of 8 */
2681 to->to_nsackblocks = (optlen - 2) / 8;
2682 to->to_sackblocks = (struct raw_sackblock *) (cp + 2);
2683 to->to_flags |= TOF_SACK;
2684 for (i = 0; i < to->to_nsackblocks; i++) {
2685 struct raw_sackblock *r = &to->to_sackblocks[i];
2687 r->rblk_start = ntohl(r->rblk_start);
2688 r->rblk_end = ntohl(r->rblk_end);
2690 if (SEQ_LEQ(r->rblk_end, r->rblk_start)) {
2692 * Invalid SACK block; discard all
2695 to->to_nsackblocks = 0;
2696 to->to_sackblocks = NULL;
2697 to->to_flags &= ~TOF_SACK;
2702 #ifdef TCP_SIGNATURE
2704 * XXX In order to reply to a host which has set the
2705 * TCP_SIGNATURE option in its initial SYN, we have to
2706 * record the fact that the option was observed here
2707 * for the syncache code to perform the correct response.
2709 case TCPOPT_SIGNATURE:
2710 if (optlen != TCPOLEN_SIGNATURE)
2712 to->to_flags |= (TOF_SIGNATURE | TOF_SIGLEN);
2714 #endif /* TCP_SIGNATURE */
2722 * Pull out of band byte out of a segment so
2723 * it doesn't appear in the user's data queue.
2724 * It is still reflected in the segment length for
2725 * sequencing purposes.
2726 * "off" is the delayed to be dropped hdrlen.
2729 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, int off)
2731 int cnt = off + th->th_urp - 1;
2734 if (m->m_len > cnt) {
2735 char *cp = mtod(m, caddr_t) + cnt;
2736 struct tcpcb *tp = sototcpcb(so);
2739 tp->t_oobflags |= TCPOOB_HAVEDATA;
2740 bcopy(cp + 1, cp, m->m_len - cnt - 1);
2742 if (m->m_flags & M_PKTHDR)
2751 panic("tcp_pulloutofband");
2755 * Collect new round-trip time estimate
2756 * and update averages and current timeout.
2759 tcp_xmit_timer(struct tcpcb *tp, int rtt)
2763 tcpstat.tcps_rttupdated++;
2765 if (tp->t_srtt != 0) {
2767 * srtt is stored as fixed point with 5 bits after the
2768 * binary point (i.e., scaled by 8). The following magic
2769 * is equivalent to the smoothing algorithm in rfc793 with
2770 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2771 * point). Adjust rtt to origin 0.
2773 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
2774 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
2776 if ((tp->t_srtt += delta) <= 0)
2780 * We accumulate a smoothed rtt variance (actually, a
2781 * smoothed mean difference), then set the retransmit
2782 * timer to smoothed rtt + 4 times the smoothed variance.
2783 * rttvar is stored as fixed point with 4 bits after the
2784 * binary point (scaled by 16). The following is
2785 * equivalent to rfc793 smoothing with an alpha of .75
2786 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
2787 * rfc793's wired-in beta.
2791 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
2792 if ((tp->t_rttvar += delta) <= 0)
2794 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
2795 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2798 * No rtt measurement yet - use the unsmoothed rtt.
2799 * Set the variance to half the rtt (so our first
2800 * retransmit happens at 3*rtt).
2802 tp->t_srtt = rtt << TCP_RTT_SHIFT;
2803 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
2804 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2810 * the retransmit should happen at rtt + 4 * rttvar.
2811 * Because of the way we do the smoothing, srtt and rttvar
2812 * will each average +1/2 tick of bias. When we compute
2813 * the retransmit timer, we want 1/2 tick of rounding and
2814 * 1 extra tick because of +-1/2 tick uncertainty in the
2815 * firing of the timer. The bias will give us exactly the
2816 * 1.5 tick we need. But, because the bias is
2817 * statistical, we have to test that we don't drop below
2818 * the minimum feasible timer (which is 2 ticks).
2820 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
2821 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
2824 * We received an ack for a packet that wasn't retransmitted;
2825 * it is probably safe to discard any error indications we've
2826 * received recently. This isn't quite right, but close enough
2827 * for now (a route might have failed after we sent a segment,
2828 * and the return path might not be symmetrical).
2830 tp->t_softerror = 0;
2834 * Determine a reasonable value for maxseg size.
2835 * If the route is known, check route for mtu.
2836 * If none, use an mss that can be handled on the outgoing
2837 * interface without forcing IP to fragment; if bigger than
2838 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2839 * to utilize large mbufs. If no route is found, route has no mtu,
2840 * or the destination isn't local, use a default, hopefully conservative
2841 * size (usually 512 or the default IP max size, but no more than the mtu
2842 * of the interface), as we can't discover anything about intervening
2843 * gateways or networks. We also initialize the congestion/slow start
2844 * window to be a single segment if the destination isn't local.
2845 * While looking at the routing entry, we also initialize other path-dependent
2846 * parameters from pre-set or cached values in the routing entry.
2848 * Also take into account the space needed for options that we
2849 * send regularly. Make maxseg shorter by that amount to assure
2850 * that we can send maxseg amount of data even when the options
2851 * are present. Store the upper limit of the length of options plus
2854 * NOTE that this routine is only called when we process an incoming
2855 * segment, for outgoing segments only tcp_mssopt is called.
2858 tcp_mss(struct tcpcb *tp, int offer)
2864 struct inpcb *inp = tp->t_inpcb;
2867 boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) ? TRUE : FALSE);
2868 size_t min_protoh = isipv6 ?
2869 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
2870 sizeof(struct tcpiphdr);
2872 const boolean_t isipv6 = FALSE;
2873 const size_t min_protoh = sizeof(struct tcpiphdr);
2877 rt = tcp_rtlookup6(&inp->inp_inc);
2879 rt = tcp_rtlookup(&inp->inp_inc);
2881 tp->t_maxopd = tp->t_maxseg =
2882 (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
2886 so = inp->inp_socket;
2889 * Offer == 0 means that there was no MSS on the SYN segment,
2890 * in this case we use either the interface mtu or tcp_mssdflt.
2892 * An offer which is too large will be cut down later.
2896 if (in6_localaddr(&inp->in6p_faddr)) {
2897 offer = ND_IFINFO(rt->rt_ifp)->linkmtu -
2900 offer = tcp_v6mssdflt;
2903 if (in_localaddr(inp->inp_faddr))
2904 offer = ifp->if_mtu - min_protoh;
2906 offer = tcp_mssdflt;
2911 * Prevent DoS attack with too small MSS. Round up
2912 * to at least minmss.
2914 * Sanity check: make sure that maxopd will be large
2915 * enough to allow some data on segments even is the
2916 * all the option space is used (40bytes). Otherwise
2917 * funny things may happen in tcp_output.
2919 offer = max(offer, tcp_minmss);
2920 offer = max(offer, 64);
2922 rt->rt_rmx.rmx_mssopt = offer;
2925 * While we're here, check if there's an initial rtt
2926 * or rttvar. Convert from the route-table units
2927 * to scaled multiples of the slow timeout timer.
2929 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
2931 * XXX the lock bit for RTT indicates that the value
2932 * is also a minimum value; this is subject to time.
2934 if (rt->rt_rmx.rmx_locks & RTV_RTT)
2935 tp->t_rttmin = rtt / (RTM_RTTUNIT / hz);
2936 tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE));
2937 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
2938 tcpstat.tcps_usedrtt++;
2939 if (rt->rt_rmx.rmx_rttvar) {
2940 tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
2941 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE));
2942 tcpstat.tcps_usedrttvar++;
2944 /* default variation is +- 1 rtt */
2946 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
2948 TCPT_RANGESET(tp->t_rxtcur,
2949 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
2950 tp->t_rttmin, TCPTV_REXMTMAX);
2954 * if there's an mtu associated with the route, use it
2955 * else, use the link mtu. Take the smaller of mss or offer
2958 if (rt->rt_rmx.rmx_mtu) {
2959 mss = rt->rt_rmx.rmx_mtu - min_protoh;
2962 mss = ND_IFINFO(rt->rt_ifp)->linkmtu - min_protoh;
2964 mss = ifp->if_mtu - min_protoh;
2966 mss = min(mss, offer);
2969 * maxopd stores the maximum length of data AND options
2970 * in a segment; maxseg is the amount of data in a normal
2971 * segment. We need to store this value (maxopd) apart
2972 * from maxseg, because now every segment carries options
2973 * and thus we normally have somewhat less data in segments.
2977 if ((tp->t_flags & (TF_REQ_TSTMP | TF_NOOPT)) == TF_REQ_TSTMP &&
2978 ((tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
2979 mss -= TCPOLEN_TSTAMP_APPA;
2981 #if (MCLBYTES & (MCLBYTES - 1)) == 0
2983 mss &= ~(MCLBYTES-1);
2986 mss = mss / MCLBYTES * MCLBYTES;
2989 * If there's a pipesize, change the socket buffer
2990 * to that size. Make the socket buffers an integral
2991 * number of mss units; if the mss is larger than
2992 * the socket buffer, decrease the mss.
2995 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0)
2997 bufsize = so->so_snd.ssb_hiwat;
3001 bufsize = roundup(bufsize, mss);
3002 if (bufsize > sb_max)
3004 if (bufsize > so->so_snd.ssb_hiwat)
3005 ssb_reserve(&so->so_snd, bufsize, so, NULL);
3010 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0)
3012 bufsize = so->so_rcv.ssb_hiwat;
3013 if (bufsize > mss) {
3014 bufsize = roundup(bufsize, mss);
3015 if (bufsize > sb_max)
3017 if (bufsize > so->so_rcv.ssb_hiwat) {
3018 lwkt_gettoken(&so->so_rcv.ssb_token);
3019 ssb_reserve(&so->so_rcv, bufsize, so, NULL);
3020 lwkt_reltoken(&so->so_rcv.ssb_token);
3025 * Set the slow-start flight size depending on whether this
3026 * is a local network or not.
3029 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380));
3033 if (rt->rt_rmx.rmx_ssthresh) {
3035 * There's some sort of gateway or interface
3036 * buffer limit on the path. Use this to set
3037 * the slow start threshhold, but set the
3038 * threshold to no less than 2*mss.
3040 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
3041 tcpstat.tcps_usedssthresh++;
3046 * Determine the MSS option to send on an outgoing SYN.
3049 tcp_mssopt(struct tcpcb *tp)
3054 ((tp->t_inpcb->inp_vflag & INP_IPV6) ? TRUE : FALSE);
3055 int min_protoh = isipv6 ?
3056 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
3057 sizeof(struct tcpiphdr);
3059 const boolean_t isipv6 = FALSE;
3060 const size_t min_protoh = sizeof(struct tcpiphdr);
3064 rt = tcp_rtlookup6(&tp->t_inpcb->inp_inc);
3066 rt = tcp_rtlookup(&tp->t_inpcb->inp_inc);
3068 return (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
3070 return (rt->rt_ifp->if_mtu - min_protoh);
3074 * When a partial ack arrives, force the retransmission of the
3075 * next unacknowledged segment. Do not exit Fast Recovery.
3077 * Implement the Slow-but-Steady variant of NewReno by restarting the
3078 * the retransmission timer. Turn it off here so it can be restarted
3079 * later in tcp_output().
3082 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th, int acked)
3084 tcp_seq old_snd_nxt = tp->snd_nxt;
3085 u_long ocwnd = tp->snd_cwnd;
3087 tcp_callout_stop(tp, tp->tt_rexmt);
3089 tp->snd_nxt = th->th_ack;
3090 /* Set snd_cwnd to one segment beyond acknowledged offset. */
3091 tp->snd_cwnd = tp->t_maxseg;
3092 tp->t_flags |= TF_ACKNOW;
3094 if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
3095 tp->snd_nxt = old_snd_nxt;
3096 /* partial window deflation */
3098 tp->snd_cwnd = ocwnd - acked + tp->t_maxseg;
3100 tp->snd_cwnd = tp->t_maxseg;
3104 * In contrast to the Slow-but-Steady NewReno variant,
3105 * we do not reset the retransmission timer for SACK retransmissions,
3106 * except when retransmitting snd_una.
3109 tcp_sack_rexmt(struct tcpcb *tp, struct tcphdr *th)
3111 uint32_t pipe, seglen;
3114 tcp_seq old_snd_nxt = tp->snd_nxt;
3115 u_long ocwnd = tp->snd_cwnd;
3116 int nseg = 0; /* consecutive new segments */
3117 #define MAXBURST 4 /* limit burst of new packets on partial ack */
3120 pipe = tcp_sack_compute_pipe(tp);
3121 while ((tcp_seq_diff_t)(ocwnd - pipe) >= (tcp_seq_diff_t)tp->t_maxseg &&
3122 (!tcp_do_smartsack || nseg < MAXBURST) &&
3123 tcp_sack_nextseg(tp, &nextrexmt, &seglen, &lostdup)) {
3125 tcp_seq old_snd_max;
3128 if (nextrexmt == tp->snd_max)
3130 tp->snd_nxt = nextrexmt;
3131 tp->snd_cwnd = nextrexmt - tp->snd_una + seglen;
3132 old_snd_max = tp->snd_max;
3133 if (nextrexmt == tp->snd_una)
3134 tcp_callout_stop(tp, tp->tt_rexmt);
3135 error = tcp_output(tp);
3138 sent = tp->snd_nxt - nextrexmt;
3143 tcpstat.tcps_sndsackpack++;
3144 tcpstat.tcps_sndsackbyte += sent;
3145 if (SEQ_LT(nextrexmt, old_snd_max) &&
3146 SEQ_LT(tp->rexmt_high, tp->snd_nxt))
3147 tp->rexmt_high = seq_min(tp->snd_nxt, old_snd_max);
3149 if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
3150 tp->snd_nxt = old_snd_nxt;
3151 tp->snd_cwnd = ocwnd;
3155 * Reset idle time and keep-alive timer, typically called when a valid
3156 * tcp packet is received but may also be called when FASTKEEP is set
3157 * to prevent the previous long-timeout from calculating to a drop.
3159 * Only update t_rcvtime for non-SYN packets.
3161 * Handle the case where one side thinks the connection is established
3162 * but the other side has, say, rebooted without cleaning out the
3163 * connection. The SYNs could be construed as an attack and wind
3164 * up ignored, but in case it isn't an attack we can validate the
3165 * connection by forcing a keepalive.
3168 tcp_timer_keep_activity(struct tcpcb *tp, int thflags)
3170 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
3171 if ((thflags & (TH_SYN | TH_ACK)) == TH_SYN) {
3172 tp->t_flags |= TF_KEEPALIVE;
3173 tcp_callout_reset(tp, tp->tt_keep, hz / 2,
3176 tp->t_rcvtime = ticks;
3177 tp->t_flags &= ~TF_KEEPALIVE;
3178 tcp_callout_reset(tp, tp->tt_keep,
3179 tcp_getkeepidle(tp),
3186 tcp_rmx_msl(const struct tcpcb *tp)
3189 struct inpcb *inp = tp->t_inpcb;
3192 boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) ? TRUE : FALSE);
3194 const boolean_t isipv6 = FALSE;
3198 rt = tcp_rtlookup6(&inp->inp_inc);
3200 rt = tcp_rtlookup(&inp->inp_inc);
3201 if (rt == NULL || rt->rt_rmx.rmx_msl == 0)
3204 msl = (rt->rt_rmx.rmx_msl * hz) / 1000;
3212 tcp_established(struct tcpcb *tp)
3214 tp->t_state = TCPS_ESTABLISHED;
3215 tcp_callout_reset(tp, tp->tt_keep, tcp_getkeepidle(tp), tcp_timer_keep);
3217 if (tp->t_flags & TF_SYN_WASLOST) {
3220 * "If the SYN or SYN/ACK is lost, the initial window used by
3221 * a sender after a correctly transmitted SYN MUST be one
3222 * segment consisting of MSS bytes."
3224 tp->snd_cwnd = tp->t_maxseg;
3228 * "If the timer expires awaiting the ACK of a SYN segment
3229 * and the TCP implementation is using an RTO less than 3
3230 * seconds, the RTO MUST be re-initialized to 3 seconds
3231 * when data transmission begins"
3233 if (tp->t_rxtcur < TCPTV_RTOBASE3)
3234 tp->t_rxtcur = TCPTV_RTOBASE3;