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 $
68 * $DragonFly: src/sys/netinet/tcp_input.c,v 1.68 2008/08/22 09:14:17 sephe Exp $
71 #include "opt_ipfw.h" /* for ipfw_fwd */
73 #include "opt_inet6.h"
74 #include "opt_ipsec.h"
75 #include "opt_tcpdebug.h"
76 #include "opt_tcp_input.h"
78 #include <sys/param.h>
79 #include <sys/systm.h>
80 #include <sys/kernel.h>
81 #include <sys/sysctl.h>
82 #include <sys/malloc.h>
84 #include <sys/proc.h> /* for proc0 declaration */
85 #include <sys/protosw.h>
86 #include <sys/socket.h>
87 #include <sys/socketvar.h>
88 #include <sys/syslog.h>
89 #include <sys/in_cksum.h>
91 #include <sys/socketvar2.h>
93 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
94 #include <machine/stdarg.h>
97 #include <net/route.h>
99 #include <netinet/in.h>
100 #include <netinet/in_systm.h>
101 #include <netinet/ip.h>
102 #include <netinet/ip_icmp.h> /* for ICMP_BANDLIM */
103 #include <netinet/in_var.h>
104 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */
105 #include <netinet/in_pcb.h>
106 #include <netinet/ip_var.h>
107 #include <netinet/ip6.h>
108 #include <netinet/icmp6.h>
109 #include <netinet6/nd6.h>
110 #include <netinet6/ip6_var.h>
111 #include <netinet6/in6_pcb.h>
112 #include <netinet/tcp.h>
113 #include <netinet/tcp_fsm.h>
114 #include <netinet/tcp_seq.h>
115 #include <netinet/tcp_timer.h>
116 #include <netinet/tcp_timer2.h>
117 #include <netinet/tcp_var.h>
118 #include <netinet6/tcp6_var.h>
119 #include <netinet/tcpip.h>
122 #include <netinet/tcp_debug.h>
124 u_char tcp_saveipgen[40]; /* the size must be of max ip header, now IPv6 */
125 struct tcphdr tcp_savetcp;
129 #include <netproto/ipsec/ipsec.h>
130 #include <netproto/ipsec/ipsec6.h>
134 #include <netinet6/ipsec.h>
135 #include <netinet6/ipsec6.h>
136 #include <netproto/key/key.h>
139 MALLOC_DEFINE(M_TSEGQ, "tseg_qent", "TCP segment queue entry");
141 static int log_in_vain = 0;
142 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
143 &log_in_vain, 0, "Log all incoming TCP connections");
145 static int blackhole = 0;
146 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW,
147 &blackhole, 0, "Do not send RST when dropping refused connections");
149 int tcp_delack_enabled = 1;
150 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW,
151 &tcp_delack_enabled, 0,
152 "Delay ACK to try and piggyback it onto a data packet");
154 #ifdef TCP_DROP_SYNFIN
155 static int drop_synfin = 0;
156 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW,
157 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set");
160 static int tcp_do_limitedtransmit = 1;
161 SYSCTL_INT(_net_inet_tcp, OID_AUTO, limitedtransmit, CTLFLAG_RW,
162 &tcp_do_limitedtransmit, 0, "Enable RFC 3042 (Limited Transmit)");
164 static int tcp_do_early_retransmit = 1;
165 SYSCTL_INT(_net_inet_tcp, OID_AUTO, earlyretransmit, CTLFLAG_RW,
166 &tcp_do_early_retransmit, 0, "Early retransmit");
168 int tcp_aggregate_acks = 1;
169 SYSCTL_INT(_net_inet_tcp, OID_AUTO, aggregate_acks, CTLFLAG_RW,
170 &tcp_aggregate_acks, 0, "Aggregate built-up acks into one ack");
172 int tcp_do_rfc3390 = 1;
173 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW,
175 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
177 static int tcp_do_eifel_detect = 1;
178 SYSCTL_INT(_net_inet_tcp, OID_AUTO, eifel, CTLFLAG_RW,
179 &tcp_do_eifel_detect, 0, "Eifel detection algorithm (RFC 3522)");
181 static int tcp_do_abc = 1;
182 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc, CTLFLAG_RW,
184 "TCP Appropriate Byte Counting (RFC 3465)");
187 * Define as tunable for easy testing with SACK on and off.
188 * Warning: do not change setting in the middle of an existing active TCP flow,
189 * else strange things might happen to that flow.
192 SYSCTL_INT(_net_inet_tcp, OID_AUTO, sack, CTLFLAG_RW,
193 &tcp_do_sack, 0, "Enable SACK Algorithms");
195 int tcp_do_smartsack = 1;
196 SYSCTL_INT(_net_inet_tcp, OID_AUTO, smartsack, CTLFLAG_RW,
197 &tcp_do_smartsack, 0, "Enable Smart SACK Algorithms");
199 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0,
200 "TCP Segment Reassembly Queue");
202 int tcp_reass_maxseg = 0;
203 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RD,
204 &tcp_reass_maxseg, 0,
205 "Global maximum number of TCP Segments in Reassembly Queue");
207 int tcp_reass_qsize = 0;
208 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD,
210 "Global number of TCP Segments currently in Reassembly Queue");
212 static int tcp_reass_overflows = 0;
213 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD,
214 &tcp_reass_overflows, 0,
215 "Global number of TCP Segment Reassembly Queue Overflows");
217 int tcp_do_autorcvbuf = 1;
218 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_RW,
219 &tcp_do_autorcvbuf, 0, "Enable automatic receive buffer sizing");
221 int tcp_autorcvbuf_inc = 16*1024;
222 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_RW,
223 &tcp_autorcvbuf_inc, 0,
224 "Incrementor step size of automatic receive buffer");
226 int tcp_autorcvbuf_max = 2*1024*1024;
227 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_RW,
228 &tcp_autorcvbuf_max, 0, "Max size of automatic receive buffer");
230 int tcp_sosnd_agglim = 2;
231 SYSCTL_INT(_net_inet_tcp, OID_AUTO, sosnd_agglim, CTLFLAG_RW,
232 &tcp_sosnd_agglim, 0, "TCP sosend mbuf aggregation limit");
234 static void tcp_dooptions(struct tcpopt *, u_char *, int, boolean_t);
235 static void tcp_pulloutofband(struct socket *,
236 struct tcphdr *, struct mbuf *, int);
237 static int tcp_reass(struct tcpcb *, struct tcphdr *, int *,
239 static void tcp_xmit_timer(struct tcpcb *, int);
240 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *, int);
241 static void tcp_sack_rexmt(struct tcpcb *, struct tcphdr *);
243 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
245 #define ND6_HINT(tp) \
247 if ((tp) && (tp)->t_inpcb && \
248 ((tp)->t_inpcb->inp_vflag & INP_IPV6) && \
249 (tp)->t_inpcb->in6p_route.ro_rt) \
250 nd6_nud_hint((tp)->t_inpcb->in6p_route.ro_rt, NULL, 0); \
257 * Indicate whether this ack should be delayed. We can delay the ack if
258 * - delayed acks are enabled and
259 * - there is no delayed ack timer in progress and
260 * - our last ack wasn't a 0-sized window. We never want to delay
261 * the ack that opens up a 0-sized window.
263 #define DELAY_ACK(tp) \
264 (tcp_delack_enabled && !tcp_callout_pending(tp, tp->tt_delack) && \
265 !(tp->t_flags & TF_RXWIN0SENT))
267 #define acceptable_window_update(tp, th, tiwin) \
268 (SEQ_LT(tp->snd_wl1, th->th_seq) || \
269 (tp->snd_wl1 == th->th_seq && \
270 (SEQ_LT(tp->snd_wl2, th->th_ack) || \
271 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))
274 tcp_reass(struct tcpcb *tp, struct tcphdr *th, int *tlenp, struct mbuf *m)
277 struct tseg_qent *p = NULL;
278 struct tseg_qent *te;
279 struct socket *so = tp->t_inpcb->inp_socket;
283 * Call with th == NULL after become established to
284 * force pre-ESTABLISHED data up to user socket.
290 * Limit the number of segments in the reassembly queue to prevent
291 * holding on to too many segments (and thus running out of mbufs).
292 * Make sure to let the missing segment through which caused this
293 * queue. Always keep one global queue entry spare to be able to
294 * process the missing segment.
296 if (th->th_seq != tp->rcv_nxt &&
297 tcp_reass_qsize + 1 >= tcp_reass_maxseg) {
298 tcp_reass_overflows++;
299 tcpstat.tcps_rcvmemdrop++;
301 /* no SACK block to report */
302 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
306 /* Allocate a new queue entry. */
307 MALLOC(te, struct tseg_qent *, sizeof(struct tseg_qent), M_TSEGQ,
308 M_INTWAIT | M_NULLOK);
310 tcpstat.tcps_rcvmemdrop++;
312 /* no SACK block to report */
313 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
316 atomic_add_int(&tcp_reass_qsize, 1);
319 * Find a segment which begins after this one does.
321 LIST_FOREACH(q, &tp->t_segq, tqe_q) {
322 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
328 * If there is a preceding segment, it may provide some of
329 * our data already. If so, drop the data from the incoming
330 * segment. If it provides all of our data, drop us.
335 /* conversion to int (in i) handles seq wraparound */
336 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
337 if (i > 0) { /* overlaps preceding segment */
338 tp->t_flags |= (TF_DUPSEG | TF_ENCLOSESEG);
339 /* enclosing block starts w/ preceding segment */
340 tp->encloseblk.rblk_start = p->tqe_th->th_seq;
342 /* preceding encloses incoming segment */
343 tp->encloseblk.rblk_end = p->tqe_th->th_seq +
345 tcpstat.tcps_rcvduppack++;
346 tcpstat.tcps_rcvdupbyte += *tlenp;
349 atomic_add_int(&tcp_reass_qsize, -1);
351 * Try to present any queued data
352 * at the left window edge to the user.
353 * This is needed after the 3-WHS
356 goto present; /* ??? */
361 /* incoming segment end is enclosing block end */
362 tp->encloseblk.rblk_end = th->th_seq + *tlenp +
363 ((th->th_flags & TH_FIN) != 0);
364 /* trim end of reported D-SACK block */
365 tp->reportblk.rblk_end = th->th_seq;
368 tcpstat.tcps_rcvoopack++;
369 tcpstat.tcps_rcvoobyte += *tlenp;
372 * While we overlap succeeding segments trim them or,
373 * if they are completely covered, dequeue them.
376 tcp_seq_diff_t i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
377 tcp_seq qend = q->tqe_th->th_seq + q->tqe_len;
378 struct tseg_qent *nq;
382 if (!(tp->t_flags & TF_DUPSEG)) { /* first time through */
383 tp->t_flags |= (TF_DUPSEG | TF_ENCLOSESEG);
384 tp->encloseblk = tp->reportblk;
385 /* report trailing duplicate D-SACK segment */
386 tp->reportblk.rblk_start = q->tqe_th->th_seq;
388 if ((tp->t_flags & TF_ENCLOSESEG) &&
389 SEQ_GT(qend, tp->encloseblk.rblk_end)) {
390 /* extend enclosing block if one exists */
391 tp->encloseblk.rblk_end = qend;
393 if (i < q->tqe_len) {
394 q->tqe_th->th_seq += i;
400 nq = LIST_NEXT(q, tqe_q);
401 LIST_REMOVE(q, tqe_q);
404 atomic_add_int(&tcp_reass_qsize, -1);
408 /* Insert the new segment queue entry into place. */
411 te->tqe_len = *tlenp;
413 /* check if can coalesce with following segment */
414 if (q != NULL && (th->th_seq + *tlenp == q->tqe_th->th_seq)) {
415 tcp_seq tend = te->tqe_th->th_seq + te->tqe_len;
417 te->tqe_len += q->tqe_len;
418 if (q->tqe_th->th_flags & TH_FIN)
419 te->tqe_th->th_flags |= TH_FIN;
420 m_cat(te->tqe_m, q->tqe_m);
421 tp->encloseblk.rblk_end = tend;
423 * When not reporting a duplicate segment, use
424 * the larger enclosing block as the SACK block.
426 if (!(tp->t_flags & TF_DUPSEG))
427 tp->reportblk.rblk_end = tend;
428 LIST_REMOVE(q, tqe_q);
430 atomic_add_int(&tcp_reass_qsize, -1);
434 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
436 /* check if can coalesce with preceding segment */
437 if (p->tqe_th->th_seq + p->tqe_len == th->th_seq) {
438 p->tqe_len += te->tqe_len;
439 m_cat(p->tqe_m, te->tqe_m);
440 tp->encloseblk.rblk_start = p->tqe_th->th_seq;
442 * When not reporting a duplicate segment, use
443 * the larger enclosing block as the SACK block.
445 if (!(tp->t_flags & TF_DUPSEG))
446 tp->reportblk.rblk_start = p->tqe_th->th_seq;
448 atomic_add_int(&tcp_reass_qsize, -1);
450 LIST_INSERT_AFTER(p, te, tqe_q);
456 * Present data to user, advancing rcv_nxt through
457 * completed sequence space.
459 if (!TCPS_HAVEESTABLISHED(tp->t_state))
461 q = LIST_FIRST(&tp->t_segq);
462 if (q == NULL || q->tqe_th->th_seq != tp->rcv_nxt)
464 tp->rcv_nxt += q->tqe_len;
465 if (!(tp->t_flags & TF_DUPSEG)) {
466 /* no SACK block to report since ACK advanced */
467 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
469 /* no enclosing block to report since ACK advanced */
470 tp->t_flags &= ~TF_ENCLOSESEG;
471 flags = q->tqe_th->th_flags & TH_FIN;
472 LIST_REMOVE(q, tqe_q);
473 KASSERT(LIST_EMPTY(&tp->t_segq) ||
474 LIST_FIRST(&tp->t_segq)->tqe_th->th_seq != tp->rcv_nxt,
475 ("segment not coalesced"));
476 if (so->so_state & SS_CANTRCVMORE) {
479 lwkt_gettoken(&so->so_rcv.ssb_token);
480 ssb_appendstream(&so->so_rcv, q->tqe_m);
481 lwkt_reltoken(&so->so_rcv.ssb_token);
484 atomic_add_int(&tcp_reass_qsize, -1);
491 * TCP input routine, follows pages 65-76 of the
492 * protocol specification dated September, 1981 very closely.
496 tcp6_input(struct mbuf **mp, int *offp, int proto)
498 struct mbuf *m = *mp;
499 struct in6_ifaddr *ia6;
501 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
504 * draft-itojun-ipv6-tcp-to-anycast
505 * better place to put this in?
507 ia6 = ip6_getdstifaddr(m);
508 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
511 ip6 = mtod(m, struct ip6_hdr *);
512 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
513 offsetof(struct ip6_hdr, ip6_dst));
514 return (IPPROTO_DONE);
517 tcp_input(mp, offp, proto);
518 return (IPPROTO_DONE);
523 tcp_input(struct mbuf **mp, int *offp, int proto)
527 struct ip *ip = NULL;
529 struct inpcb *inp = NULL;
535 struct tcpcb *tp = NULL;
537 struct socket *so = 0;
539 boolean_t ourfinisacked, needoutput = FALSE;
542 struct tcpopt to; /* options in this segment */
543 struct sockaddr_in *next_hop = NULL;
544 int rstreason; /* For badport_bandlim accounting purposes */
546 struct ip6_hdr *ip6 = NULL;
551 const boolean_t isipv6 = FALSE;
561 tcpstat.tcps_rcvtotal++;
563 if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
566 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
567 KKASSERT(mtag != NULL);
568 next_hop = m_tag_data(mtag);
572 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? TRUE : FALSE;
576 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */
577 ip6 = mtod(m, struct ip6_hdr *);
578 tlen = (sizeof *ip6) + ntohs(ip6->ip6_plen) - off0;
579 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) {
580 tcpstat.tcps_rcvbadsum++;
583 th = (struct tcphdr *)((caddr_t)ip6 + off0);
586 * Be proactive about unspecified IPv6 address in source.
587 * As we use all-zero to indicate unbounded/unconnected pcb,
588 * unspecified IPv6 address can be used to confuse us.
590 * Note that packets with unspecified IPv6 destination is
591 * already dropped in ip6_input.
593 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
599 * Get IP and TCP header together in first mbuf.
600 * Note: IP leaves IP header in first mbuf.
602 if (off0 > sizeof(struct ip)) {
604 off0 = sizeof(struct ip);
606 /* already checked and pulled up in ip_demux() */
607 KASSERT(m->m_len >= sizeof(struct tcpiphdr),
608 ("TCP header not in one mbuf: m->m_len %d", m->m_len));
609 ip = mtod(m, struct ip *);
610 ipov = (struct ipovly *)ip;
611 th = (struct tcphdr *)((caddr_t)ip + off0);
614 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
615 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
616 th->th_sum = m->m_pkthdr.csum_data;
618 th->th_sum = in_pseudo(ip->ip_src.s_addr,
620 htonl(m->m_pkthdr.csum_data +
623 th->th_sum ^= 0xffff;
626 * Checksum extended TCP header and data.
628 len = sizeof(struct ip) + tlen;
629 bzero(ipov->ih_x1, sizeof ipov->ih_x1);
630 ipov->ih_len = (u_short)tlen;
631 ipov->ih_len = htons(ipov->ih_len);
632 th->th_sum = in_cksum(m, len);
635 tcpstat.tcps_rcvbadsum++;
639 /* Re-initialization for later version check */
640 ip->ip_v = IPVERSION;
645 * Check that TCP offset makes sense,
646 * pull out TCP options and adjust length. XXX
648 off = th->th_off << 2;
649 /* already checked and pulled up in ip_demux() */
650 KASSERT(off >= sizeof(struct tcphdr) && off <= tlen,
651 ("bad TCP data offset %d (tlen %d)", off, tlen));
652 tlen -= off; /* tlen is used instead of ti->ti_len */
653 if (off > sizeof(struct tcphdr)) {
655 IP6_EXTHDR_CHECK(m, off0, off, IPPROTO_DONE);
656 ip6 = mtod(m, struct ip6_hdr *);
657 th = (struct tcphdr *)((caddr_t)ip6 + off0);
659 /* already pulled up in ip_demux() */
660 KASSERT(m->m_len >= sizeof(struct ip) + off,
661 ("TCP header and options not in one mbuf: "
662 "m_len %d, off %d", m->m_len, off));
664 optlen = off - sizeof(struct tcphdr);
665 optp = (u_char *)(th + 1);
667 thflags = th->th_flags;
669 #ifdef TCP_DROP_SYNFIN
671 * If the drop_synfin option is enabled, drop all packets with
672 * both the SYN and FIN bits set. This prevents e.g. nmap from
673 * identifying the TCP/IP stack.
675 * This is a violation of the TCP specification.
677 if (drop_synfin && (thflags & (TH_SYN | TH_FIN)) == (TH_SYN | TH_FIN))
682 * Convert TCP protocol specific fields to host format.
684 th->th_seq = ntohl(th->th_seq);
685 th->th_ack = ntohl(th->th_ack);
686 th->th_win = ntohs(th->th_win);
687 th->th_urp = ntohs(th->th_urp);
690 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options,
691 * until after ip6_savecontrol() is called and before other functions
692 * which don't want those proto headers.
693 * Because ip6_savecontrol() is going to parse the mbuf to
694 * search for data to be passed up to user-land, it wants mbuf
695 * parameters to be unchanged.
696 * XXX: the call of ip6_savecontrol() has been obsoleted based on
697 * latest version of the advanced API (20020110).
699 drop_hdrlen = off0 + off;
702 * Locate pcb for segment.
705 /* IPFIREWALL_FORWARD section */
706 if (next_hop != NULL && !isipv6) { /* IPv6 support is not there yet */
708 * Transparently forwarded. Pretend to be the destination.
709 * already got one like this?
711 cpu = mycpu->gd_cpuid;
712 inp = in_pcblookup_hash(&tcbinfo[cpu],
713 ip->ip_src, th->th_sport,
714 ip->ip_dst, th->th_dport,
715 0, m->m_pkthdr.rcvif);
718 * It's new. Try to find the ambushing socket.
722 * The rest of the ipfw code stores the port in
724 * (The IP address is still in network order.)
726 in_port_t dport = next_hop->sin_port ?
727 htons(next_hop->sin_port) :
730 cpu = tcp_addrcpu(ip->ip_src.s_addr, th->th_sport,
731 next_hop->sin_addr.s_addr, dport);
732 inp = in_pcblookup_hash(&tcbinfo[cpu],
733 ip->ip_src, th->th_sport,
734 next_hop->sin_addr, dport,
735 1, m->m_pkthdr.rcvif);
739 inp = in6_pcblookup_hash(&tcbinfo[0],
740 &ip6->ip6_src, th->th_sport,
741 &ip6->ip6_dst, th->th_dport,
742 1, m->m_pkthdr.rcvif);
744 cpu = mycpu->gd_cpuid;
745 inp = in_pcblookup_hash(&tcbinfo[cpu],
746 ip->ip_src, th->th_sport,
747 ip->ip_dst, th->th_dport,
748 1, m->m_pkthdr.rcvif);
753 * If the state is CLOSED (i.e., TCB does not exist) then
754 * all data in the incoming segment is discarded.
755 * If the TCB exists but is in CLOSED state, it is embryonic,
756 * but should either do a listen or a connect soon.
761 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2];
763 char dbuf[sizeof "aaa.bbb.ccc.ddd"];
764 char sbuf[sizeof "aaa.bbb.ccc.ddd"];
768 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst));
771 strcat(sbuf, ip6_sprintf(&ip6->ip6_src));
774 strcpy(dbuf, inet_ntoa(ip->ip_dst));
775 strcpy(sbuf, inet_ntoa(ip->ip_src));
777 switch (log_in_vain) {
779 if (!(thflags & TH_SYN))
783 "Connection attempt to TCP %s:%d "
784 "from %s:%d flags:0x%02x\n",
785 dbuf, ntohs(th->th_dport), sbuf,
786 ntohs(th->th_sport), thflags);
795 if (thflags & TH_SYN)
804 rstreason = BANDLIM_RST_CLOSEDPORT;
810 if (ipsec6_in_reject_so(m, inp->inp_socket)) {
811 ipsec6stat.in_polvio++;
815 if (ipsec4_in_reject_so(m, inp->inp_socket)) {
816 ipsecstat.in_polvio++;
823 if (ipsec6_in_reject(m, inp))
826 if (ipsec4_in_reject(m, inp))
830 /* Check the minimum TTL for socket. */
832 if ((isipv6 ? ip6->ip6_hlim : ip->ip_ttl) < inp->inp_ip_minttl)
838 rstreason = BANDLIM_RST_CLOSEDPORT;
841 if (tp->t_state <= TCPS_CLOSED)
844 /* Unscale the window into a 32-bit value. */
845 if (!(thflags & TH_SYN))
846 tiwin = th->th_win << tp->snd_scale;
850 so = inp->inp_socket;
853 if (so->so_options & SO_DEBUG) {
854 ostate = tp->t_state;
856 bcopy(ip6, tcp_saveipgen, sizeof(*ip6));
858 bcopy(ip, tcp_saveipgen, sizeof(*ip));
863 bzero(&to, sizeof to);
865 if (so->so_options & SO_ACCEPTCONN) {
866 struct in_conninfo inc;
869 inc.inc_isipv6 = (isipv6 == TRUE);
872 inc.inc6_faddr = ip6->ip6_src;
873 inc.inc6_laddr = ip6->ip6_dst;
874 inc.inc6_route.ro_rt = NULL; /* XXX */
876 inc.inc_faddr = ip->ip_src;
877 inc.inc_laddr = ip->ip_dst;
878 inc.inc_route.ro_rt = NULL; /* XXX */
880 inc.inc_fport = th->th_sport;
881 inc.inc_lport = th->th_dport;
884 * If the state is LISTEN then ignore segment if it contains
885 * a RST. If the segment contains an ACK then it is bad and
886 * send a RST. If it does not contain a SYN then it is not
887 * interesting; drop it.
889 * If the state is SYN_RECEIVED (syncache) and seg contains
890 * an ACK, but not for our SYN/ACK, send a RST. If the seg
891 * contains a RST, check the sequence number to see if it
892 * is a valid reset segment.
894 if ((thflags & (TH_RST | TH_ACK | TH_SYN)) != TH_SYN) {
895 if ((thflags & (TH_RST | TH_ACK | TH_SYN)) == TH_ACK) {
896 if (!syncache_expand(&inc, th, &so, m)) {
898 * No syncache entry, or ACK was not
899 * for our SYN/ACK. Send a RST.
901 tcpstat.tcps_badsyn++;
902 rstreason = BANDLIM_RST_OPENPORT;
907 * Could not complete 3-way handshake,
908 * connection is being closed down, and
909 * syncache will free mbuf.
912 return(IPPROTO_DONE);
915 * We must be in the correct protocol thread
916 * for this connection.
918 KKASSERT(so->so_port == &curthread->td_msgport);
921 * Socket is created in state SYN_RECEIVED.
922 * Continue processing segment.
927 * This is what would have happened in
928 * tcp_output() when the SYN,ACK was sent.
930 tp->snd_up = tp->snd_una;
931 tp->snd_max = tp->snd_nxt = tp->iss + 1;
932 tp->last_ack_sent = tp->rcv_nxt;
934 * XXX possible bug - it doesn't appear that tp->snd_wnd is unscaled
935 * until the _second_ ACK is received:
936 * rcv SYN (set wscale opts) --> send SYN/ACK, set snd_wnd = window.
937 * rcv ACK, calculate tiwin --> process SYN_RECEIVED, determine wscale,
938 * move to ESTAB, set snd_wnd to tiwin.
940 tp->snd_wnd = tiwin; /* unscaled */
943 if (thflags & TH_RST) {
944 syncache_chkrst(&inc, th);
947 if (thflags & TH_ACK) {
948 syncache_badack(&inc);
949 tcpstat.tcps_badsyn++;
950 rstreason = BANDLIM_RST_OPENPORT;
957 * Segment's flags are (SYN) or (SYN | FIN).
961 * If deprecated address is forbidden,
962 * we do not accept SYN to deprecated interface
963 * address to prevent any new inbound connection from
964 * getting established.
965 * When we do not accept SYN, we send a TCP RST,
966 * with deprecated source address (instead of dropping
967 * it). We compromise it as it is much better for peer
968 * to send a RST, and RST will be the final packet
971 * If we do not forbid deprecated addresses, we accept
972 * the SYN packet. RFC2462 does not suggest dropping
974 * If we decipher RFC2462 5.5.4, it says like this:
975 * 1. use of deprecated addr with existing
976 * communication is okay - "SHOULD continue to be
978 * 2. use of it with new communication:
979 * (2a) "SHOULD NOT be used if alternate address
980 * with sufficient scope is available"
981 * (2b) nothing mentioned otherwise.
982 * Here we fall into (2b) case as we have no choice in
983 * our source address selection - we must obey the peer.
985 * The wording in RFC2462 is confusing, and there are
986 * multiple description text for deprecated address
987 * handling - worse, they are not exactly the same.
988 * I believe 5.5.4 is the best one, so we follow 5.5.4.
990 if (isipv6 && !ip6_use_deprecated) {
991 struct in6_ifaddr *ia6;
993 if ((ia6 = ip6_getdstifaddr(m)) &&
994 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
996 rstreason = BANDLIM_RST_OPENPORT;
1002 * If it is from this socket, drop it, it must be forged.
1003 * Don't bother responding if the destination was a broadcast.
1005 if (th->th_dport == th->th_sport) {
1007 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
1011 if (ip->ip_dst.s_addr == ip->ip_src.s_addr)
1016 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
1018 * Note that it is quite possible to receive unicast
1019 * link-layer packets with a broadcast IP address. Use
1020 * in_broadcast() to find them.
1022 if (m->m_flags & (M_BCAST | M_MCAST))
1025 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1026 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
1029 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1030 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1031 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1032 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
1036 * SYN appears to be valid; create compressed TCP state
1037 * for syncache, or perform t/tcp connection.
1039 if (so->so_qlen <= so->so_qlimit) {
1040 tcp_dooptions(&to, optp, optlen, TRUE);
1041 if (!syncache_add(&inc, &to, th, &so, m))
1045 * Entry added to syncache, mbuf used to
1046 * send SYN,ACK packet.
1049 return(IPPROTO_DONE);
1052 * We must be in the correct protocol thread for
1055 KKASSERT(so->so_port == &curthread->td_msgport);
1058 tp = intotcpcb(inp);
1059 tp->snd_wnd = tiwin;
1060 tp->t_starttime = ticks;
1061 tp->t_state = TCPS_ESTABLISHED;
1064 * If there is a FIN, or if there is data and the
1065 * connection is local, then delay SYN,ACK(SYN) in
1066 * the hope of piggy-backing it on a response
1067 * segment. Otherwise must send ACK now in case
1068 * the other side is slow starting.
1070 if (DELAY_ACK(tp) &&
1071 ((thflags & TH_FIN) ||
1073 ((isipv6 && in6_localaddr(&inp->in6p_faddr)) ||
1074 (!isipv6 && in_localaddr(inp->inp_faddr)))))) {
1075 tcp_callout_reset(tp, tp->tt_delack,
1076 tcp_delacktime, tcp_timer_delack);
1077 tp->t_flags |= TF_NEEDSYN;
1079 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
1082 tcpstat.tcps_connects++;
1091 * Should not happen - syncache should pick up these connections.
1093 * Once we are past handling listen sockets we must be in the
1094 * correct protocol processing thread.
1096 KASSERT(tp->t_state != TCPS_LISTEN, ("tcp_input: TCPS_LISTEN state"));
1097 KKASSERT(so->so_port == &curthread->td_msgport);
1100 * This is the second part of the MSS DoS prevention code (after
1101 * minmss on the sending side) and it deals with too many too small
1102 * tcp packets in a too short timeframe (1 second).
1104 * XXX Removed. This code was crap. It does not scale to network
1105 * speed, and default values break NFS. Gone.
1110 * Segment received on connection.
1112 * Reset idle time and keep-alive timer. Don't waste time if less
1113 * then a second has elapsed.
1115 if ((int)(ticks - tp->t_rcvtime) > hz)
1116 tcp_timer_keep_activity(tp, thflags);
1120 * XXX this is tradtitional behavior, may need to be cleaned up.
1122 tcp_dooptions(&to, optp, optlen, (thflags & TH_SYN) != 0);
1123 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1124 if (to.to_flags & TOF_SCALE) {
1125 tp->t_flags |= TF_RCVD_SCALE;
1126 tp->requested_s_scale = to.to_requested_s_scale;
1128 if (to.to_flags & TOF_TS) {
1129 tp->t_flags |= TF_RCVD_TSTMP;
1130 tp->ts_recent = to.to_tsval;
1131 tp->ts_recent_age = ticks;
1133 if (to.to_flags & TOF_MSS)
1134 tcp_mss(tp, to.to_mss);
1136 * Only set the TF_SACK_PERMITTED per-connection flag
1137 * if we got a SACK_PERMITTED option from the other side
1138 * and the global tcp_do_sack variable is true.
1140 if (tcp_do_sack && (to.to_flags & TOF_SACK_PERMITTED))
1141 tp->t_flags |= TF_SACK_PERMITTED;
1145 * Header prediction: check for the two common cases
1146 * of a uni-directional data xfer. If the packet has
1147 * no control flags, is in-sequence, the window didn't
1148 * change and we're not retransmitting, it's a
1149 * candidate. If the length is zero and the ack moved
1150 * forward, we're the sender side of the xfer. Just
1151 * free the data acked & wake any higher level process
1152 * that was blocked waiting for space. If the length
1153 * is non-zero and the ack didn't move, we're the
1154 * receiver side. If we're getting packets in-order
1155 * (the reassembly queue is empty), add the data to
1156 * the socket buffer and note that we need a delayed ack.
1157 * Make sure that the hidden state-flags are also off.
1158 * Since we check for TCPS_ESTABLISHED above, it can only
1161 if (tp->t_state == TCPS_ESTABLISHED &&
1162 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1163 !(tp->t_flags & (TF_NEEDSYN | TF_NEEDFIN)) &&
1164 (!(to.to_flags & TOF_TS) ||
1165 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) &&
1166 th->th_seq == tp->rcv_nxt &&
1167 tp->snd_nxt == tp->snd_max) {
1170 * If last ACK falls within this segment's sequence numbers,
1171 * record the timestamp.
1172 * NOTE that the test is modified according to the latest
1173 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1175 if ((to.to_flags & TOF_TS) &&
1176 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1177 tp->ts_recent_age = ticks;
1178 tp->ts_recent = to.to_tsval;
1182 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1183 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1184 tp->snd_cwnd >= tp->snd_wnd &&
1185 !IN_FASTRECOVERY(tp)) {
1187 * This is a pure ack for outstanding data.
1189 ++tcpstat.tcps_predack;
1191 * "bad retransmit" recovery
1193 * If Eifel detection applies, then
1194 * it is deterministic, so use it
1195 * unconditionally over the old heuristic.
1196 * Otherwise, fall back to the old heuristic.
1198 if (tcp_do_eifel_detect &&
1199 (to.to_flags & TOF_TS) && to.to_tsecr &&
1200 (tp->t_flags & TF_FIRSTACCACK)) {
1201 /* Eifel detection applicable. */
1202 if (to.to_tsecr < tp->t_rexmtTS) {
1203 tcp_revert_congestion_state(tp);
1204 ++tcpstat.tcps_eifeldetected;
1206 } else if (tp->t_rxtshift == 1 &&
1207 ticks < tp->t_badrxtwin) {
1208 tcp_revert_congestion_state(tp);
1209 ++tcpstat.tcps_rttdetected;
1211 tp->t_flags &= ~(TF_FIRSTACCACK |
1212 TF_FASTREXMT | TF_EARLYREXMT);
1214 * Recalculate the retransmit timer / rtt.
1216 * Some machines (certain windows boxes)
1217 * send broken timestamp replies during the
1218 * SYN+ACK phase, ignore timestamps of 0.
1220 if ((to.to_flags & TOF_TS) && to.to_tsecr) {
1222 ticks - to.to_tsecr + 1);
1223 } else if (tp->t_rtttime &&
1224 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1226 ticks - tp->t_rtttime);
1228 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1229 acked = th->th_ack - tp->snd_una;
1230 tcpstat.tcps_rcvackpack++;
1231 tcpstat.tcps_rcvackbyte += acked;
1232 sbdrop(&so->so_snd.sb, acked);
1233 tp->snd_recover = th->th_ack - 1;
1234 tp->snd_una = th->th_ack;
1237 * Update window information.
1239 if (tiwin != tp->snd_wnd &&
1240 acceptable_window_update(tp, th, tiwin)) {
1241 /* keep track of pure window updates */
1242 if (tp->snd_wl2 == th->th_ack &&
1243 tiwin > tp->snd_wnd)
1244 tcpstat.tcps_rcvwinupd++;
1245 tp->snd_wnd = tiwin;
1246 tp->snd_wl1 = th->th_seq;
1247 tp->snd_wl2 = th->th_ack;
1248 if (tp->snd_wnd > tp->max_sndwnd)
1249 tp->max_sndwnd = tp->snd_wnd;
1252 ND6_HINT(tp); /* some progress has been done */
1254 * If all outstanding data are acked, stop
1255 * retransmit timer, otherwise restart timer
1256 * using current (possibly backed-off) value.
1257 * If process is waiting for space,
1258 * wakeup/selwakeup/signal. If data
1259 * are ready to send, let tcp_output
1260 * decide between more output or persist.
1262 if (tp->snd_una == tp->snd_max) {
1263 tcp_callout_stop(tp, tp->tt_rexmt);
1264 } else if (!tcp_callout_active(tp,
1266 tcp_callout_reset(tp, tp->tt_rexmt,
1267 tp->t_rxtcur, tcp_timer_rexmt);
1270 if (so->so_snd.ssb_cc > 0)
1272 return(IPPROTO_DONE);
1274 } else if (tiwin == tp->snd_wnd &&
1275 th->th_ack == tp->snd_una &&
1276 LIST_EMPTY(&tp->t_segq) &&
1277 tlen <= ssb_space(&so->so_rcv)) {
1278 u_long newsize = 0; /* automatic sockbuf scaling */
1280 * This is a pure, in-sequence data packet
1281 * with nothing on the reassembly queue and
1282 * we have enough buffer space to take it.
1284 ++tcpstat.tcps_preddat;
1285 tp->rcv_nxt += tlen;
1286 tcpstat.tcps_rcvpack++;
1287 tcpstat.tcps_rcvbyte += tlen;
1288 ND6_HINT(tp); /* some progress has been done */
1290 * Automatic sizing of receive socket buffer. Often the send
1291 * buffer size is not optimally adjusted to the actual network
1292 * conditions at hand (delay bandwidth product). Setting the
1293 * buffer size too small limits throughput on links with high
1294 * bandwidth and high delay (eg. trans-continental/oceanic links).
1296 * On the receive side the socket buffer memory is only rarely
1297 * used to any significant extent. This allows us to be much
1298 * more aggressive in scaling the receive socket buffer. For
1299 * the case that the buffer space is actually used to a large
1300 * extent and we run out of kernel memory we can simply drop
1301 * the new segments; TCP on the sender will just retransmit it
1302 * later. Setting the buffer size too big may only consume too
1303 * much kernel memory if the application doesn't read() from
1304 * the socket or packet loss or reordering makes use of the
1307 * The criteria to step up the receive buffer one notch are:
1308 * 1. the number of bytes received during the time it takes
1309 * one timestamp to be reflected back to us (the RTT);
1310 * 2. received bytes per RTT is within seven eighth of the
1311 * current socket buffer size;
1312 * 3. receive buffer size has not hit maximal automatic size;
1314 * This algorithm does one step per RTT at most and only if
1315 * we receive a bulk stream w/o packet losses or reorderings.
1316 * Shrinking the buffer during idle times is not necessary as
1317 * it doesn't consume any memory when idle.
1319 * TODO: Only step up if the application is actually serving
1320 * the buffer to better manage the socket buffer resources.
1322 if (tcp_do_autorcvbuf &&
1324 (so->so_rcv.ssb_flags & SSB_AUTOSIZE)) {
1325 if (to.to_tsecr > tp->rfbuf_ts &&
1326 to.to_tsecr - tp->rfbuf_ts < hz) {
1328 (so->so_rcv.ssb_hiwat / 8 * 7) &&
1329 so->so_rcv.ssb_hiwat <
1330 tcp_autorcvbuf_max) {
1332 ulmin(so->so_rcv.ssb_hiwat +
1334 tcp_autorcvbuf_max);
1336 /* Start over with next RTT. */
1340 tp->rfbuf_cnt += tlen; /* add up */
1343 * Add data to socket buffer.
1345 if (so->so_state & SS_CANTRCVMORE) {
1349 * Set new socket buffer size, give up when
1352 * Adjusting the size can mess up ACK
1353 * sequencing when pure window updates are
1354 * being avoided (which is the default),
1357 lwkt_gettoken(&so->so_rcv.ssb_token);
1359 tp->t_flags |= TF_RXRESIZED;
1360 if (!ssb_reserve(&so->so_rcv, newsize,
1362 atomic_clear_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE);
1365 (TCP_MAXWIN << tp->rcv_scale)) {
1366 atomic_clear_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE);
1369 m_adj(m, drop_hdrlen); /* delayed header drop */
1370 ssb_appendstream(&so->so_rcv, m);
1371 lwkt_reltoken(&so->so_rcv.ssb_token);
1375 * This code is responsible for most of the ACKs
1376 * the TCP stack sends back after receiving a data
1377 * packet. Note that the DELAY_ACK check fails if
1378 * the delack timer is already running, which results
1379 * in an ack being sent every other packet (which is
1382 * We then further aggregate acks by not actually
1383 * sending one until the protocol thread has completed
1384 * processing the current backlog of packets. This
1385 * does not delay the ack any further, but allows us
1386 * to take advantage of the packet aggregation that
1387 * high speed NICs do (usually blocks of 8-10 packets)
1388 * to send a single ack rather then four or five acks,
1389 * greatly reducing the ack rate, the return channel
1390 * bandwidth, and the protocol overhead on both ends.
1392 * Since this also has the effect of slowing down
1393 * the exponential slow-start ramp-up, systems with
1394 * very large bandwidth-delay products might want
1395 * to turn the feature off.
1397 if (DELAY_ACK(tp)) {
1398 tcp_callout_reset(tp, tp->tt_delack,
1399 tcp_delacktime, tcp_timer_delack);
1400 } else if (tcp_aggregate_acks) {
1401 tp->t_flags |= TF_ACKNOW;
1402 if (!(tp->t_flags & TF_ONOUTPUTQ)) {
1403 tp->t_flags |= TF_ONOUTPUTQ;
1404 tp->tt_cpu = mycpu->gd_cpuid;
1406 &tcpcbackq[tp->tt_cpu],
1410 tp->t_flags |= TF_ACKNOW;
1413 return(IPPROTO_DONE);
1418 * Calculate amount of space in receive window,
1419 * and then do TCP input processing.
1420 * Receive window is amount of space in rcv queue,
1421 * but not less than advertised window.
1423 recvwin = ssb_space(&so->so_rcv);
1426 tp->rcv_wnd = imax(recvwin, (int)(tp->rcv_adv - tp->rcv_nxt));
1428 /* Reset receive buffer auto scaling when not in bulk receive mode. */
1432 switch (tp->t_state) {
1434 * If the state is SYN_RECEIVED:
1435 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1437 case TCPS_SYN_RECEIVED:
1438 if ((thflags & TH_ACK) &&
1439 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1440 SEQ_GT(th->th_ack, tp->snd_max))) {
1441 rstreason = BANDLIM_RST_OPENPORT;
1447 * If the state is SYN_SENT:
1448 * if seg contains an ACK, but not for our SYN, drop the input.
1449 * if seg contains a RST, then drop the connection.
1450 * if seg does not contain SYN, then drop it.
1451 * Otherwise this is an acceptable SYN segment
1452 * initialize tp->rcv_nxt and tp->irs
1453 * if seg contains ack then advance tp->snd_una
1454 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1455 * arrange for segment to be acked (eventually)
1456 * continue processing rest of data/controls, beginning with URG
1459 if ((thflags & TH_ACK) &&
1460 (SEQ_LEQ(th->th_ack, tp->iss) ||
1461 SEQ_GT(th->th_ack, tp->snd_max))) {
1462 rstreason = BANDLIM_UNLIMITED;
1465 if (thflags & TH_RST) {
1466 if (thflags & TH_ACK)
1467 tp = tcp_drop(tp, ECONNREFUSED);
1470 if (!(thflags & TH_SYN))
1472 tp->snd_wnd = th->th_win; /* initial send window */
1474 tp->irs = th->th_seq;
1476 if (thflags & TH_ACK) {
1477 /* Our SYN was acked. */
1478 tcpstat.tcps_connects++;
1480 /* Do window scaling on this connection? */
1481 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
1482 (TF_RCVD_SCALE | TF_REQ_SCALE)) {
1483 tp->snd_scale = tp->requested_s_scale;
1484 tp->rcv_scale = tp->request_r_scale;
1486 tp->rcv_adv += tp->rcv_wnd;
1487 tp->snd_una++; /* SYN is acked */
1488 tcp_callout_stop(tp, tp->tt_rexmt);
1490 * If there's data, delay ACK; if there's also a FIN
1491 * ACKNOW will be turned on later.
1493 if (DELAY_ACK(tp) && tlen != 0) {
1494 tcp_callout_reset(tp, tp->tt_delack,
1495 tcp_delacktime, tcp_timer_delack);
1497 tp->t_flags |= TF_ACKNOW;
1500 * Received <SYN,ACK> in SYN_SENT[*] state.
1502 * SYN_SENT --> ESTABLISHED
1503 * SYN_SENT* --> FIN_WAIT_1
1505 tp->t_starttime = ticks;
1506 if (tp->t_flags & TF_NEEDFIN) {
1507 tp->t_state = TCPS_FIN_WAIT_1;
1508 tp->t_flags &= ~TF_NEEDFIN;
1511 tp->t_state = TCPS_ESTABLISHED;
1512 tcp_callout_reset(tp, tp->tt_keep,
1513 tcp_getkeepidle(tp),
1518 * Received initial SYN in SYN-SENT[*] state =>
1519 * simultaneous open.
1520 * Do 3-way handshake:
1521 * SYN-SENT -> SYN-RECEIVED
1522 * SYN-SENT* -> SYN-RECEIVED*
1524 tp->t_flags |= TF_ACKNOW;
1525 tcp_callout_stop(tp, tp->tt_rexmt);
1526 tp->t_state = TCPS_SYN_RECEIVED;
1531 * Advance th->th_seq to correspond to first data byte.
1532 * If data, trim to stay within window,
1533 * dropping FIN if necessary.
1536 if (tlen > tp->rcv_wnd) {
1537 todrop = tlen - tp->rcv_wnd;
1541 tcpstat.tcps_rcvpackafterwin++;
1542 tcpstat.tcps_rcvbyteafterwin += todrop;
1544 tp->snd_wl1 = th->th_seq - 1;
1545 tp->rcv_up = th->th_seq;
1547 * Client side of transaction: already sent SYN and data.
1548 * If the remote host used T/TCP to validate the SYN,
1549 * our data will be ACK'd; if so, enter normal data segment
1550 * processing in the middle of step 5, ack processing.
1551 * Otherwise, goto step 6.
1553 if (thflags & TH_ACK)
1559 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
1560 * do normal processing (we no longer bother with T/TCP).
1564 case TCPS_TIME_WAIT:
1565 break; /* continue normal processing */
1569 * States other than LISTEN or SYN_SENT.
1570 * First check the RST flag and sequence number since reset segments
1571 * are exempt from the timestamp and connection count tests. This
1572 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
1573 * below which allowed reset segments in half the sequence space
1574 * to fall though and be processed (which gives forged reset
1575 * segments with a random sequence number a 50 percent chance of
1576 * killing a connection).
1577 * Then check timestamp, if present.
1578 * Then check the connection count, if present.
1579 * Then check that at least some bytes of segment are within
1580 * receive window. If segment begins before rcv_nxt,
1581 * drop leading data (and SYN); if nothing left, just ack.
1584 * If the RST bit is set, check the sequence number to see
1585 * if this is a valid reset segment.
1587 * In all states except SYN-SENT, all reset (RST) segments
1588 * are validated by checking their SEQ-fields. A reset is
1589 * valid if its sequence number is in the window.
1590 * Note: this does not take into account delayed ACKs, so
1591 * we should test against last_ack_sent instead of rcv_nxt.
1592 * The sequence number in the reset segment is normally an
1593 * echo of our outgoing acknowledgement numbers, but some hosts
1594 * send a reset with the sequence number at the rightmost edge
1595 * of our receive window, and we have to handle this case.
1596 * If we have multiple segments in flight, the intial reset
1597 * segment sequence numbers will be to the left of last_ack_sent,
1598 * but they will eventually catch up.
1599 * In any case, it never made sense to trim reset segments to
1600 * fit the receive window since RFC 1122 says:
1601 * 4.2.2.12 RST Segment: RFC-793 Section 3.4
1603 * A TCP SHOULD allow a received RST segment to include data.
1606 * It has been suggested that a RST segment could contain
1607 * ASCII text that encoded and explained the cause of the
1608 * RST. No standard has yet been established for such
1611 * If the reset segment passes the sequence number test examine
1613 * SYN_RECEIVED STATE:
1614 * If passive open, return to LISTEN state.
1615 * If active open, inform user that connection was refused.
1616 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
1617 * Inform user that connection was reset, and close tcb.
1618 * CLOSING, LAST_ACK STATES:
1621 * Drop the segment - see Stevens, vol. 2, p. 964 and
1624 if (thflags & TH_RST) {
1625 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
1626 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
1627 switch (tp->t_state) {
1629 case TCPS_SYN_RECEIVED:
1630 so->so_error = ECONNREFUSED;
1633 case TCPS_ESTABLISHED:
1634 case TCPS_FIN_WAIT_1:
1635 case TCPS_FIN_WAIT_2:
1636 case TCPS_CLOSE_WAIT:
1637 so->so_error = ECONNRESET;
1639 tp->t_state = TCPS_CLOSED;
1640 tcpstat.tcps_drops++;
1649 case TCPS_TIME_WAIT:
1657 * RFC 1323 PAWS: If we have a timestamp reply on this segment
1658 * and it's less than ts_recent, drop it.
1660 if ((to.to_flags & TOF_TS) && tp->ts_recent != 0 &&
1661 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
1663 /* Check to see if ts_recent is over 24 days old. */
1664 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1666 * Invalidate ts_recent. If this segment updates
1667 * ts_recent, the age will be reset later and ts_recent
1668 * will get a valid value. If it does not, setting
1669 * ts_recent to zero will at least satisfy the
1670 * requirement that zero be placed in the timestamp
1671 * echo reply when ts_recent isn't valid. The
1672 * age isn't reset until we get a valid ts_recent
1673 * because we don't want out-of-order segments to be
1674 * dropped when ts_recent is old.
1678 tcpstat.tcps_rcvduppack++;
1679 tcpstat.tcps_rcvdupbyte += tlen;
1680 tcpstat.tcps_pawsdrop++;
1688 * In the SYN-RECEIVED state, validate that the packet belongs to
1689 * this connection before trimming the data to fit the receive
1690 * window. Check the sequence number versus IRS since we know
1691 * the sequence numbers haven't wrapped. This is a partial fix
1692 * for the "LAND" DoS attack.
1694 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
1695 rstreason = BANDLIM_RST_OPENPORT;
1699 todrop = tp->rcv_nxt - th->th_seq;
1701 if (TCP_DO_SACK(tp)) {
1702 /* Report duplicate segment at head of packet. */
1703 tp->reportblk.rblk_start = th->th_seq;
1704 tp->reportblk.rblk_end = th->th_seq + tlen;
1705 if (thflags & TH_FIN)
1706 ++tp->reportblk.rblk_end;
1707 if (SEQ_GT(tp->reportblk.rblk_end, tp->rcv_nxt))
1708 tp->reportblk.rblk_end = tp->rcv_nxt;
1709 tp->t_flags |= (TF_DUPSEG | TF_SACKLEFT | TF_ACKNOW);
1711 if (thflags & TH_SYN) {
1721 * Following if statement from Stevens, vol. 2, p. 960.
1723 if (todrop > tlen ||
1724 (todrop == tlen && !(thflags & TH_FIN))) {
1726 * Any valid FIN must be to the left of the window.
1727 * At this point the FIN must be a duplicate or out
1728 * of sequence; drop it.
1733 * Send an ACK to resynchronize and drop any data.
1734 * But keep on processing for RST or ACK.
1736 tp->t_flags |= TF_ACKNOW;
1738 tcpstat.tcps_rcvduppack++;
1739 tcpstat.tcps_rcvdupbyte += todrop;
1741 tcpstat.tcps_rcvpartduppack++;
1742 tcpstat.tcps_rcvpartdupbyte += todrop;
1744 drop_hdrlen += todrop; /* drop from the top afterwards */
1745 th->th_seq += todrop;
1747 if (th->th_urp > todrop)
1748 th->th_urp -= todrop;
1756 * If new data are received on a connection after the
1757 * user processes are gone, then RST the other end.
1759 if ((so->so_state & SS_NOFDREF) &&
1760 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1762 tcpstat.tcps_rcvafterclose++;
1763 rstreason = BANDLIM_UNLIMITED;
1768 * If segment ends after window, drop trailing data
1769 * (and PUSH and FIN); if nothing left, just ACK.
1771 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
1773 tcpstat.tcps_rcvpackafterwin++;
1774 if (todrop >= tlen) {
1775 tcpstat.tcps_rcvbyteafterwin += tlen;
1777 * If a new connection request is received
1778 * while in TIME_WAIT, drop the old connection
1779 * and start over if the sequence numbers
1780 * are above the previous ones.
1782 if (thflags & TH_SYN &&
1783 tp->t_state == TCPS_TIME_WAIT &&
1784 SEQ_GT(th->th_seq, tp->rcv_nxt)) {
1789 * If window is closed can only take segments at
1790 * window edge, and have to drop data and PUSH from
1791 * incoming segments. Continue processing, but
1792 * remember to ack. Otherwise, drop segment
1795 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1796 tp->t_flags |= TF_ACKNOW;
1797 tcpstat.tcps_rcvwinprobe++;
1801 tcpstat.tcps_rcvbyteafterwin += todrop;
1804 thflags &= ~(TH_PUSH | TH_FIN);
1808 * If last ACK falls within this segment's sequence numbers,
1809 * record its timestamp.
1811 * 1) That the test incorporates suggestions from the latest
1812 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1813 * 2) That updating only on newer timestamps interferes with
1814 * our earlier PAWS tests, so this check should be solely
1815 * predicated on the sequence space of this segment.
1816 * 3) That we modify the segment boundary check to be
1817 * Last.ACK.Sent <= SEG.SEQ + SEG.LEN
1818 * instead of RFC1323's
1819 * Last.ACK.Sent < SEG.SEQ + SEG.LEN,
1820 * This modified check allows us to overcome RFC1323's
1821 * limitations as described in Stevens TCP/IP Illustrated
1822 * Vol. 2 p.869. In such cases, we can still calculate the
1823 * RTT correctly when RCV.NXT == Last.ACK.Sent.
1825 if ((to.to_flags & TOF_TS) && SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
1826 SEQ_LEQ(tp->last_ack_sent, (th->th_seq + tlen
1827 + ((thflags & TH_SYN) != 0)
1828 + ((thflags & TH_FIN) != 0)))) {
1829 tp->ts_recent_age = ticks;
1830 tp->ts_recent = to.to_tsval;
1834 * If a SYN is in the window, then this is an
1835 * error and we send an RST and drop the connection.
1837 if (thflags & TH_SYN) {
1838 tp = tcp_drop(tp, ECONNRESET);
1839 rstreason = BANDLIM_UNLIMITED;
1844 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
1845 * flag is on (half-synchronized state), then queue data for
1846 * later processing; else drop segment and return.
1848 if (!(thflags & TH_ACK)) {
1849 if (tp->t_state == TCPS_SYN_RECEIVED ||
1850 (tp->t_flags & TF_NEEDSYN))
1859 switch (tp->t_state) {
1861 * In SYN_RECEIVED state, the ACK acknowledges our SYN, so enter
1862 * ESTABLISHED state and continue processing.
1863 * The ACK was checked above.
1865 case TCPS_SYN_RECEIVED:
1867 tcpstat.tcps_connects++;
1869 /* Do window scaling? */
1870 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
1871 (TF_RCVD_SCALE | TF_REQ_SCALE)) {
1872 tp->snd_scale = tp->requested_s_scale;
1873 tp->rcv_scale = tp->request_r_scale;
1877 * SYN-RECEIVED -> ESTABLISHED
1878 * SYN-RECEIVED* -> FIN-WAIT-1
1880 tp->t_starttime = ticks;
1881 if (tp->t_flags & TF_NEEDFIN) {
1882 tp->t_state = TCPS_FIN_WAIT_1;
1883 tp->t_flags &= ~TF_NEEDFIN;
1885 tp->t_state = TCPS_ESTABLISHED;
1886 tcp_callout_reset(tp, tp->tt_keep,
1887 tcp_getkeepidle(tp),
1891 * If segment contains data or ACK, will call tcp_reass()
1892 * later; if not, do so now to pass queued data to user.
1894 if (tlen == 0 && !(thflags & TH_FIN))
1895 tcp_reass(tp, NULL, NULL, NULL);
1899 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1900 * ACKs. If the ack is in the range
1901 * tp->snd_una < th->th_ack <= tp->snd_max
1902 * then advance tp->snd_una to th->th_ack and drop
1903 * data from the retransmission queue. If this ACK reflects
1904 * more up to date window information we update our window information.
1906 case TCPS_ESTABLISHED:
1907 case TCPS_FIN_WAIT_1:
1908 case TCPS_FIN_WAIT_2:
1909 case TCPS_CLOSE_WAIT:
1912 case TCPS_TIME_WAIT:
1914 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
1915 if (TCP_DO_SACK(tp))
1916 tcp_sack_update_scoreboard(tp, &to);
1917 if (tlen != 0 || tiwin != tp->snd_wnd) {
1921 tcpstat.tcps_rcvdupack++;
1922 if (!tcp_callout_active(tp, tp->tt_rexmt) ||
1923 th->th_ack != tp->snd_una) {
1928 * We have outstanding data (other than
1929 * a window probe), this is a completely
1930 * duplicate ack (ie, window info didn't
1931 * change), the ack is the biggest we've
1932 * seen and we've seen exactly our rexmt
1933 * threshhold of them, so assume a packet
1934 * has been dropped and retransmit it.
1935 * Kludge snd_nxt & the congestion
1936 * window so we send only this one
1939 if (IN_FASTRECOVERY(tp)) {
1940 if (TCP_DO_SACK(tp)) {
1941 /* No artifical cwnd inflation. */
1942 tcp_sack_rexmt(tp, th);
1945 * Dup acks mean that packets
1946 * have left the network
1947 * (they're now cached at the
1948 * receiver) so bump cwnd by
1949 * the amount in the receiver
1950 * to keep a constant cwnd
1951 * packets in the network.
1953 tp->snd_cwnd += tp->t_maxseg;
1956 } else if (SEQ_LT(th->th_ack, tp->snd_recover)) {
1959 } else if (++tp->t_dupacks == tcprexmtthresh) {
1960 tcp_seq old_snd_nxt;
1964 if (tcp_do_eifel_detect &&
1965 (tp->t_flags & TF_RCVD_TSTMP)) {
1966 tcp_save_congestion_state(tp);
1967 tp->t_flags |= TF_FASTREXMT;
1970 * We know we're losing at the current
1971 * window size, so do congestion avoidance:
1972 * set ssthresh to half the current window
1973 * and pull our congestion window back to the
1976 win = min(tp->snd_wnd, tp->snd_cwnd) / 2 /
1980 tp->snd_ssthresh = win * tp->t_maxseg;
1981 ENTER_FASTRECOVERY(tp);
1982 tp->snd_recover = tp->snd_max;
1983 tcp_callout_stop(tp, tp->tt_rexmt);
1985 old_snd_nxt = tp->snd_nxt;
1986 tp->snd_nxt = th->th_ack;
1987 tp->snd_cwnd = tp->t_maxseg;
1989 ++tcpstat.tcps_sndfastrexmit;
1990 tp->snd_cwnd = tp->snd_ssthresh;
1991 tp->rexmt_high = tp->snd_nxt;
1992 if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
1993 tp->snd_nxt = old_snd_nxt;
1994 KASSERT(tp->snd_limited <= 2,
1995 ("tp->snd_limited too big"));
1996 if (TCP_DO_SACK(tp))
1997 tcp_sack_rexmt(tp, th);
1999 tp->snd_cwnd += tp->t_maxseg *
2000 (tp->t_dupacks - tp->snd_limited);
2001 } else if (tcp_do_limitedtransmit) {
2002 u_long oldcwnd = tp->snd_cwnd;
2003 tcp_seq oldsndmax = tp->snd_max;
2004 tcp_seq oldsndnxt = tp->snd_nxt;
2005 /* outstanding data */
2006 uint32_t ownd = tp->snd_max - tp->snd_una;
2009 #define iceildiv(n, d) (((n)+(d)-1) / (d))
2011 KASSERT(tp->t_dupacks == 1 ||
2013 ("dupacks not 1 or 2"));
2014 if (tp->t_dupacks == 1)
2015 tp->snd_limited = 0;
2016 tp->snd_nxt = tp->snd_max;
2017 tp->snd_cwnd = ownd +
2018 (tp->t_dupacks - tp->snd_limited) *
2023 * Other acks may have been processed,
2024 * snd_nxt cannot be reset to a value less
2027 if (SEQ_LT(oldsndnxt, oldsndmax)) {
2028 if (SEQ_GT(oldsndnxt, tp->snd_una))
2029 tp->snd_nxt = oldsndnxt;
2031 tp->snd_nxt = tp->snd_una;
2033 tp->snd_cwnd = oldcwnd;
2034 sent = tp->snd_max - oldsndmax;
2035 if (sent > tp->t_maxseg) {
2036 KASSERT((tp->t_dupacks == 2 &&
2037 tp->snd_limited == 0) ||
2038 (sent == tp->t_maxseg + 1 &&
2039 tp->t_flags & TF_SENTFIN),
2041 KASSERT(sent <= tp->t_maxseg * 2,
2042 ("sent too many segments"));
2043 tp->snd_limited = 2;
2044 tcpstat.tcps_sndlimited += 2;
2045 } else if (sent > 0) {
2047 ++tcpstat.tcps_sndlimited;
2048 } else if (tcp_do_early_retransmit &&
2049 (tcp_do_eifel_detect &&
2050 (tp->t_flags & TF_RCVD_TSTMP)) &&
2051 ownd < 4 * tp->t_maxseg &&
2052 tp->t_dupacks + 1 >=
2053 iceildiv(ownd, tp->t_maxseg) &&
2054 (!TCP_DO_SACK(tp) ||
2055 ownd <= tp->t_maxseg ||
2056 tcp_sack_has_sacked(&tp->scb,
2057 ownd - tp->t_maxseg))) {
2058 ++tcpstat.tcps_sndearlyrexmit;
2059 tp->t_flags |= TF_EARLYREXMT;
2060 goto fastretransmit;
2066 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una"));
2068 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2070 * Detected optimistic ACK attack.
2071 * Force slow-start to de-synchronize attack.
2073 tp->snd_cwnd = tp->t_maxseg;
2076 tcpstat.tcps_rcvacktoomuch++;
2080 * If we reach this point, ACK is not a duplicate,
2081 * i.e., it ACKs something we sent.
2083 if (tp->t_flags & TF_NEEDSYN) {
2085 * T/TCP: Connection was half-synchronized, and our
2086 * SYN has been ACK'd (so connection is now fully
2087 * synchronized). Go to non-starred state,
2088 * increment snd_una for ACK of SYN, and check if
2089 * we can do window scaling.
2091 tp->t_flags &= ~TF_NEEDSYN;
2093 /* Do window scaling? */
2094 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
2095 (TF_RCVD_SCALE | TF_REQ_SCALE)) {
2096 tp->snd_scale = tp->requested_s_scale;
2097 tp->rcv_scale = tp->request_r_scale;
2102 acked = th->th_ack - tp->snd_una;
2103 tcpstat.tcps_rcvackpack++;
2104 tcpstat.tcps_rcvackbyte += acked;
2106 if (tcp_do_eifel_detect && acked > 0 &&
2107 (to.to_flags & TOF_TS) && (to.to_tsecr != 0) &&
2108 (tp->t_flags & TF_FIRSTACCACK)) {
2109 /* Eifel detection applicable. */
2110 if (to.to_tsecr < tp->t_rexmtTS) {
2111 ++tcpstat.tcps_eifeldetected;
2112 tcp_revert_congestion_state(tp);
2113 if (tp->t_rxtshift == 1 &&
2114 ticks >= tp->t_badrxtwin)
2115 ++tcpstat.tcps_rttcantdetect;
2117 } else if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) {
2119 * If we just performed our first retransmit,
2120 * and the ACK arrives within our recovery window,
2121 * then it was a mistake to do the retransmit
2122 * in the first place. Recover our original cwnd
2123 * and ssthresh, and proceed to transmit where we
2126 tcp_revert_congestion_state(tp);
2127 ++tcpstat.tcps_rttdetected;
2131 * If we have a timestamp reply, update smoothed
2132 * round trip time. If no timestamp is present but
2133 * transmit timer is running and timed sequence
2134 * number was acked, update smoothed round trip time.
2135 * Since we now have an rtt measurement, cancel the
2136 * timer backoff (cf., Phil Karn's retransmit alg.).
2137 * Recompute the initial retransmit timer.
2139 * Some machines (certain windows boxes) send broken
2140 * timestamp replies during the SYN+ACK phase, ignore
2143 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0))
2144 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1);
2145 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq))
2146 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2147 tcp_xmit_bandwidth_limit(tp, th->th_ack);
2150 * If no data (only SYN) was ACK'd,
2151 * skip rest of ACK processing.
2156 /* Stop looking for an acceptable ACK since one was received. */
2157 tp->t_flags &= ~(TF_FIRSTACCACK | TF_FASTREXMT | TF_EARLYREXMT);
2159 if (acked > so->so_snd.ssb_cc) {
2160 tp->snd_wnd -= so->so_snd.ssb_cc;
2161 sbdrop(&so->so_snd.sb, (int)so->so_snd.ssb_cc);
2162 ourfinisacked = TRUE;
2164 sbdrop(&so->so_snd.sb, acked);
2165 tp->snd_wnd -= acked;
2166 ourfinisacked = FALSE;
2171 * Update window information.
2172 * Don't look at window if no ACK:
2173 * TAC's send garbage on first SYN.
2175 if (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2176 (tp->snd_wl1 == th->th_seq &&
2177 (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2178 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)))) {
2179 /* keep track of pure window updates */
2180 if (tlen == 0 && tp->snd_wl2 == th->th_ack &&
2181 tiwin > tp->snd_wnd)
2182 tcpstat.tcps_rcvwinupd++;
2183 tp->snd_wnd = tiwin;
2184 tp->snd_wl1 = th->th_seq;
2185 tp->snd_wl2 = th->th_ack;
2186 if (tp->snd_wnd > tp->max_sndwnd)
2187 tp->max_sndwnd = tp->snd_wnd;
2191 tp->snd_una = th->th_ack;
2192 if (TCP_DO_SACK(tp))
2193 tcp_sack_update_scoreboard(tp, &to);
2194 if (IN_FASTRECOVERY(tp)) {
2195 if (SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2196 EXIT_FASTRECOVERY(tp);
2199 * If the congestion window was inflated
2200 * to account for the other side's
2201 * cached packets, retract it.
2203 if (!TCP_DO_SACK(tp))
2204 tp->snd_cwnd = tp->snd_ssthresh;
2207 * Window inflation should have left us
2208 * with approximately snd_ssthresh outstanding
2209 * data. But, in case we would be inclined
2210 * to send a burst, better do it using
2213 if (SEQ_GT(th->th_ack + tp->snd_cwnd,
2214 tp->snd_max + 2 * tp->t_maxseg))
2216 (tp->snd_max - tp->snd_una) +
2221 if (TCP_DO_SACK(tp)) {
2222 tp->snd_max_rexmt = tp->snd_max;
2223 tcp_sack_rexmt(tp, th);
2225 tcp_newreno_partial_ack(tp, th, acked);
2231 * Open the congestion window. When in slow-start,
2232 * open exponentially: maxseg per packet. Otherwise,
2233 * open linearly: maxseg per window.
2235 if (tp->snd_cwnd <= tp->snd_ssthresh) {
2237 (SEQ_LT(tp->snd_nxt, tp->snd_max) ?
2238 tp->t_maxseg : 2 * tp->t_maxseg);
2241 tp->snd_cwnd += tcp_do_abc ?
2242 min(acked, abc_sslimit) : tp->t_maxseg;
2244 /* linear increase */
2245 tp->snd_wacked += tcp_do_abc ? acked :
2247 if (tp->snd_wacked >= tp->snd_cwnd) {
2248 tp->snd_wacked -= tp->snd_cwnd;
2249 tp->snd_cwnd += tp->t_maxseg;
2252 tp->snd_cwnd = min(tp->snd_cwnd,
2253 TCP_MAXWIN << tp->snd_scale);
2254 tp->snd_recover = th->th_ack - 1;
2256 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2257 tp->snd_nxt = tp->snd_una;
2260 * If all outstanding data is acked, stop retransmit
2261 * timer and remember to restart (more output or persist).
2262 * If there is more data to be acked, restart retransmit
2263 * timer, using current (possibly backed-off) value.
2265 if (th->th_ack == tp->snd_max) {
2266 tcp_callout_stop(tp, tp->tt_rexmt);
2268 } else if (!tcp_callout_active(tp, tp->tt_persist)) {
2269 tcp_callout_reset(tp, tp->tt_rexmt, tp->t_rxtcur,
2273 switch (tp->t_state) {
2275 * In FIN_WAIT_1 STATE in addition to the processing
2276 * for the ESTABLISHED state if our FIN is now acknowledged
2277 * then enter FIN_WAIT_2.
2279 case TCPS_FIN_WAIT_1:
2280 if (ourfinisacked) {
2282 * If we can't receive any more
2283 * data, then closing user can proceed.
2284 * Starting the timer is contrary to the
2285 * specification, but if we don't get a FIN
2286 * we'll hang forever.
2288 if (so->so_state & SS_CANTRCVMORE) {
2289 soisdisconnected(so);
2290 tcp_callout_reset(tp, tp->tt_2msl,
2291 tp->t_maxidle, tcp_timer_2msl);
2293 tp->t_state = TCPS_FIN_WAIT_2;
2298 * In CLOSING STATE in addition to the processing for
2299 * the ESTABLISHED state if the ACK acknowledges our FIN
2300 * then enter the TIME-WAIT state, otherwise ignore
2304 if (ourfinisacked) {
2305 tp->t_state = TCPS_TIME_WAIT;
2306 tcp_canceltimers(tp);
2307 tcp_callout_reset(tp, tp->tt_2msl,
2308 2 * tcp_msl, tcp_timer_2msl);
2309 soisdisconnected(so);
2314 * In LAST_ACK, we may still be waiting for data to drain
2315 * and/or to be acked, as well as for the ack of our FIN.
2316 * If our FIN is now acknowledged, delete the TCB,
2317 * enter the closed state and return.
2320 if (ourfinisacked) {
2327 * In TIME_WAIT state the only thing that should arrive
2328 * is a retransmission of the remote FIN. Acknowledge
2329 * it and restart the finack timer.
2331 case TCPS_TIME_WAIT:
2332 tcp_callout_reset(tp, tp->tt_2msl, 2 * tcp_msl,
2340 * Update window information.
2341 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2343 if ((thflags & TH_ACK) &&
2344 acceptable_window_update(tp, th, tiwin)) {
2345 /* keep track of pure window updates */
2346 if (tlen == 0 && tp->snd_wl2 == th->th_ack &&
2347 tiwin > tp->snd_wnd)
2348 tcpstat.tcps_rcvwinupd++;
2349 tp->snd_wnd = tiwin;
2350 tp->snd_wl1 = th->th_seq;
2351 tp->snd_wl2 = th->th_ack;
2352 if (tp->snd_wnd > tp->max_sndwnd)
2353 tp->max_sndwnd = tp->snd_wnd;
2358 * Process segments with URG.
2360 if ((thflags & TH_URG) && th->th_urp &&
2361 !TCPS_HAVERCVDFIN(tp->t_state)) {
2363 * This is a kludge, but if we receive and accept
2364 * random urgent pointers, we'll crash in
2365 * soreceive. It's hard to imagine someone
2366 * actually wanting to send this much urgent data.
2368 if (th->th_urp + so->so_rcv.ssb_cc > sb_max) {
2369 th->th_urp = 0; /* XXX */
2370 thflags &= ~TH_URG; /* XXX */
2371 goto dodata; /* XXX */
2374 * If this segment advances the known urgent pointer,
2375 * then mark the data stream. This should not happen
2376 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2377 * a FIN has been received from the remote side.
2378 * In these states we ignore the URG.
2380 * According to RFC961 (Assigned Protocols),
2381 * the urgent pointer points to the last octet
2382 * of urgent data. We continue, however,
2383 * to consider it to indicate the first octet
2384 * of data past the urgent section as the original
2385 * spec states (in one of two places).
2387 if (SEQ_GT(th->th_seq + th->th_urp, tp->rcv_up)) {
2388 tp->rcv_up = th->th_seq + th->th_urp;
2389 so->so_oobmark = so->so_rcv.ssb_cc +
2390 (tp->rcv_up - tp->rcv_nxt) - 1;
2391 if (so->so_oobmark == 0)
2392 sosetstate(so, SS_RCVATMARK);
2394 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2397 * Remove out of band data so doesn't get presented to user.
2398 * This can happen independent of advancing the URG pointer,
2399 * but if two URG's are pending at once, some out-of-band
2400 * data may creep in... ick.
2402 if (th->th_urp <= (u_long)tlen &&
2403 !(so->so_options & SO_OOBINLINE)) {
2404 /* hdr drop is delayed */
2405 tcp_pulloutofband(so, th, m, drop_hdrlen);
2409 * If no out of band data is expected,
2410 * pull receive urgent pointer along
2411 * with the receive window.
2413 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2414 tp->rcv_up = tp->rcv_nxt;
2419 * Process the segment text, merging it into the TCP sequencing queue,
2420 * and arranging for acknowledgment of receipt if necessary.
2421 * This process logically involves adjusting tp->rcv_wnd as data
2422 * is presented to the user (this happens in tcp_usrreq.c,
2423 * case PRU_RCVD). If a FIN has already been received on this
2424 * connection then we just ignore the text.
2426 if ((tlen || (thflags & TH_FIN)) && !TCPS_HAVERCVDFIN(tp->t_state)) {
2427 m_adj(m, drop_hdrlen); /* delayed header drop */
2429 * Insert segment which includes th into TCP reassembly queue
2430 * with control block tp. Set thflags to whether reassembly now
2431 * includes a segment with FIN. This handles the common case
2432 * inline (segment is the next to be received on an established
2433 * connection, and the queue is empty), avoiding linkage into
2434 * and removal from the queue and repetition of various
2436 * Set DELACK for segments received in order, but ack
2437 * immediately when segments are out of order (so
2438 * fast retransmit can work).
2440 if (th->th_seq == tp->rcv_nxt &&
2441 LIST_EMPTY(&tp->t_segq) &&
2442 TCPS_HAVEESTABLISHED(tp->t_state)) {
2443 if (DELAY_ACK(tp)) {
2444 tcp_callout_reset(tp, tp->tt_delack,
2445 tcp_delacktime, tcp_timer_delack);
2447 tp->t_flags |= TF_ACKNOW;
2449 tp->rcv_nxt += tlen;
2450 thflags = th->th_flags & TH_FIN;
2451 tcpstat.tcps_rcvpack++;
2452 tcpstat.tcps_rcvbyte += tlen;
2454 if (so->so_state & SS_CANTRCVMORE) {
2457 lwkt_gettoken(&so->so_rcv.ssb_token);
2458 ssb_appendstream(&so->so_rcv, m);
2459 lwkt_reltoken(&so->so_rcv.ssb_token);
2463 if (!(tp->t_flags & TF_DUPSEG)) {
2464 /* Initialize SACK report block. */
2465 tp->reportblk.rblk_start = th->th_seq;
2466 tp->reportblk.rblk_end = th->th_seq + tlen +
2467 ((thflags & TH_FIN) != 0);
2469 thflags = tcp_reass(tp, th, &tlen, m);
2470 tp->t_flags |= TF_ACKNOW;
2474 * Note the amount of data that peer has sent into
2475 * our window, in order to estimate the sender's
2478 len = so->so_rcv.ssb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2485 * If FIN is received ACK the FIN and let the user know
2486 * that the connection is closing.
2488 if (thflags & TH_FIN) {
2489 if (!TCPS_HAVERCVDFIN(tp->t_state)) {
2492 * If connection is half-synchronized
2493 * (ie NEEDSYN flag on) then delay ACK,
2494 * so it may be piggybacked when SYN is sent.
2495 * Otherwise, since we received a FIN then no
2496 * more input can be expected, send ACK now.
2498 if (DELAY_ACK(tp) && (tp->t_flags & TF_NEEDSYN)) {
2499 tcp_callout_reset(tp, tp->tt_delack,
2500 tcp_delacktime, tcp_timer_delack);
2502 tp->t_flags |= TF_ACKNOW;
2507 switch (tp->t_state) {
2509 * In SYN_RECEIVED and ESTABLISHED STATES
2510 * enter the CLOSE_WAIT state.
2512 case TCPS_SYN_RECEIVED:
2513 tp->t_starttime = ticks;
2515 case TCPS_ESTABLISHED:
2516 tp->t_state = TCPS_CLOSE_WAIT;
2520 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2521 * enter the CLOSING state.
2523 case TCPS_FIN_WAIT_1:
2524 tp->t_state = TCPS_CLOSING;
2528 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2529 * starting the time-wait timer, turning off the other
2532 case TCPS_FIN_WAIT_2:
2533 tp->t_state = TCPS_TIME_WAIT;
2534 tcp_canceltimers(tp);
2535 tcp_callout_reset(tp, tp->tt_2msl, 2 * tcp_msl,
2537 soisdisconnected(so);
2541 * In TIME_WAIT state restart the 2 MSL time_wait timer.
2543 case TCPS_TIME_WAIT:
2544 tcp_callout_reset(tp, tp->tt_2msl, 2 * tcp_msl,
2551 if (so->so_options & SO_DEBUG)
2552 tcp_trace(TA_INPUT, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2556 * Return any desired output.
2558 if (needoutput || (tp->t_flags & TF_ACKNOW))
2560 return(IPPROTO_DONE);
2564 * Generate an ACK dropping incoming segment if it occupies
2565 * sequence space, where the ACK reflects our state.
2567 * We can now skip the test for the RST flag since all
2568 * paths to this code happen after packets containing
2569 * RST have been dropped.
2571 * In the SYN-RECEIVED state, don't send an ACK unless the
2572 * segment we received passes the SYN-RECEIVED ACK test.
2573 * If it fails send a RST. This breaks the loop in the
2574 * "LAND" DoS attack, and also prevents an ACK storm
2575 * between two listening ports that have been sent forged
2576 * SYN segments, each with the source address of the other.
2578 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
2579 (SEQ_GT(tp->snd_una, th->th_ack) ||
2580 SEQ_GT(th->th_ack, tp->snd_max)) ) {
2581 rstreason = BANDLIM_RST_OPENPORT;
2585 if (so->so_options & SO_DEBUG)
2586 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2589 tp->t_flags |= TF_ACKNOW;
2591 return(IPPROTO_DONE);
2595 * Generate a RST, dropping incoming segment.
2596 * Make ACK acceptable to originator of segment.
2597 * Don't bother to respond if destination was broadcast/multicast.
2599 if ((thflags & TH_RST) || m->m_flags & (M_BCAST | M_MCAST))
2602 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
2603 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
2606 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
2607 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
2608 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
2609 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2612 /* IPv6 anycast check is done at tcp6_input() */
2615 * Perform bandwidth limiting.
2618 if (badport_bandlim(rstreason) < 0)
2623 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2624 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2626 if (thflags & TH_ACK)
2627 /* mtod() below is safe as long as hdr dropping is delayed */
2628 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack,
2631 if (thflags & TH_SYN)
2633 /* mtod() below is safe as long as hdr dropping is delayed */
2634 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq + tlen,
2635 (tcp_seq)0, TH_RST | TH_ACK);
2637 return(IPPROTO_DONE);
2641 * Drop space held by incoming segment and return.
2644 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2645 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2648 return(IPPROTO_DONE);
2652 * Parse TCP options and place in tcpopt.
2655 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, boolean_t is_syn)
2660 for (; cnt > 0; cnt -= optlen, cp += optlen) {
2662 if (opt == TCPOPT_EOL)
2664 if (opt == TCPOPT_NOP)
2670 if (optlen < 2 || optlen > cnt)
2675 if (optlen != TCPOLEN_MAXSEG)
2679 to->to_flags |= TOF_MSS;
2680 bcopy(cp + 2, &to->to_mss, sizeof to->to_mss);
2681 to->to_mss = ntohs(to->to_mss);
2684 if (optlen != TCPOLEN_WINDOW)
2688 to->to_flags |= TOF_SCALE;
2689 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
2691 case TCPOPT_TIMESTAMP:
2692 if (optlen != TCPOLEN_TIMESTAMP)
2694 to->to_flags |= TOF_TS;
2695 bcopy(cp + 2, &to->to_tsval, sizeof to->to_tsval);
2696 to->to_tsval = ntohl(to->to_tsval);
2697 bcopy(cp + 6, &to->to_tsecr, sizeof to->to_tsecr);
2698 to->to_tsecr = ntohl(to->to_tsecr);
2700 * If echoed timestamp is later than the current time,
2701 * fall back to non RFC1323 RTT calculation.
2703 if (to->to_tsecr != 0 && TSTMP_GT(to->to_tsecr, ticks))
2706 case TCPOPT_SACK_PERMITTED:
2707 if (optlen != TCPOLEN_SACK_PERMITTED)
2711 to->to_flags |= TOF_SACK_PERMITTED;
2714 if ((optlen - 2) & 0x07) /* not multiple of 8 */
2716 to->to_nsackblocks = (optlen - 2) / 8;
2717 to->to_sackblocks = (struct raw_sackblock *) (cp + 2);
2718 to->to_flags |= TOF_SACK;
2719 for (i = 0; i < to->to_nsackblocks; i++) {
2720 struct raw_sackblock *r = &to->to_sackblocks[i];
2722 r->rblk_start = ntohl(r->rblk_start);
2723 r->rblk_end = ntohl(r->rblk_end);
2726 #ifdef TCP_SIGNATURE
2728 * XXX In order to reply to a host which has set the
2729 * TCP_SIGNATURE option in its initial SYN, we have to
2730 * record the fact that the option was observed here
2731 * for the syncache code to perform the correct response.
2733 case TCPOPT_SIGNATURE:
2734 if (optlen != TCPOLEN_SIGNATURE)
2736 to->to_flags |= (TOF_SIGNATURE | TOF_SIGLEN);
2738 #endif /* TCP_SIGNATURE */
2746 * Pull out of band byte out of a segment so
2747 * it doesn't appear in the user's data queue.
2748 * It is still reflected in the segment length for
2749 * sequencing purposes.
2750 * "off" is the delayed to be dropped hdrlen.
2753 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, int off)
2755 int cnt = off + th->th_urp - 1;
2758 if (m->m_len > cnt) {
2759 char *cp = mtod(m, caddr_t) + cnt;
2760 struct tcpcb *tp = sototcpcb(so);
2763 tp->t_oobflags |= TCPOOB_HAVEDATA;
2764 bcopy(cp + 1, cp, m->m_len - cnt - 1);
2766 if (m->m_flags & M_PKTHDR)
2775 panic("tcp_pulloutofband");
2779 * Collect new round-trip time estimate
2780 * and update averages and current timeout.
2783 tcp_xmit_timer(struct tcpcb *tp, int rtt)
2787 tcpstat.tcps_rttupdated++;
2789 if (tp->t_srtt != 0) {
2791 * srtt is stored as fixed point with 5 bits after the
2792 * binary point (i.e., scaled by 8). The following magic
2793 * is equivalent to the smoothing algorithm in rfc793 with
2794 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2795 * point). Adjust rtt to origin 0.
2797 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
2798 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
2800 if ((tp->t_srtt += delta) <= 0)
2804 * We accumulate a smoothed rtt variance (actually, a
2805 * smoothed mean difference), then set the retransmit
2806 * timer to smoothed rtt + 4 times the smoothed variance.
2807 * rttvar is stored as fixed point with 4 bits after the
2808 * binary point (scaled by 16). The following is
2809 * equivalent to rfc793 smoothing with an alpha of .75
2810 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
2811 * rfc793's wired-in beta.
2815 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
2816 if ((tp->t_rttvar += delta) <= 0)
2818 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
2819 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2822 * No rtt measurement yet - use the unsmoothed rtt.
2823 * Set the variance to half the rtt (so our first
2824 * retransmit happens at 3*rtt).
2826 tp->t_srtt = rtt << TCP_RTT_SHIFT;
2827 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
2828 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2834 * the retransmit should happen at rtt + 4 * rttvar.
2835 * Because of the way we do the smoothing, srtt and rttvar
2836 * will each average +1/2 tick of bias. When we compute
2837 * the retransmit timer, we want 1/2 tick of rounding and
2838 * 1 extra tick because of +-1/2 tick uncertainty in the
2839 * firing of the timer. The bias will give us exactly the
2840 * 1.5 tick we need. But, because the bias is
2841 * statistical, we have to test that we don't drop below
2842 * the minimum feasible timer (which is 2 ticks).
2844 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
2845 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
2848 * We received an ack for a packet that wasn't retransmitted;
2849 * it is probably safe to discard any error indications we've
2850 * received recently. This isn't quite right, but close enough
2851 * for now (a route might have failed after we sent a segment,
2852 * and the return path might not be symmetrical).
2854 tp->t_softerror = 0;
2858 * Determine a reasonable value for maxseg size.
2859 * If the route is known, check route for mtu.
2860 * If none, use an mss that can be handled on the outgoing
2861 * interface without forcing IP to fragment; if bigger than
2862 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2863 * to utilize large mbufs. If no route is found, route has no mtu,
2864 * or the destination isn't local, use a default, hopefully conservative
2865 * size (usually 512 or the default IP max size, but no more than the mtu
2866 * of the interface), as we can't discover anything about intervening
2867 * gateways or networks. We also initialize the congestion/slow start
2868 * window to be a single segment if the destination isn't local.
2869 * While looking at the routing entry, we also initialize other path-dependent
2870 * parameters from pre-set or cached values in the routing entry.
2872 * Also take into account the space needed for options that we
2873 * send regularly. Make maxseg shorter by that amount to assure
2874 * that we can send maxseg amount of data even when the options
2875 * are present. Store the upper limit of the length of options plus
2878 * NOTE that this routine is only called when we process an incoming
2879 * segment, for outgoing segments only tcp_mssopt is called.
2882 tcp_mss(struct tcpcb *tp, int offer)
2888 struct inpcb *inp = tp->t_inpcb;
2891 boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) ? TRUE : FALSE);
2892 size_t min_protoh = isipv6 ?
2893 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
2894 sizeof(struct tcpiphdr);
2896 const boolean_t isipv6 = FALSE;
2897 const size_t min_protoh = sizeof(struct tcpiphdr);
2901 rt = tcp_rtlookup6(&inp->inp_inc);
2903 rt = tcp_rtlookup(&inp->inp_inc);
2905 tp->t_maxopd = tp->t_maxseg =
2906 (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
2910 so = inp->inp_socket;
2913 * Offer == 0 means that there was no MSS on the SYN segment,
2914 * in this case we use either the interface mtu or tcp_mssdflt.
2916 * An offer which is too large will be cut down later.
2920 if (in6_localaddr(&inp->in6p_faddr)) {
2921 offer = ND_IFINFO(rt->rt_ifp)->linkmtu -
2924 offer = tcp_v6mssdflt;
2927 if (in_localaddr(inp->inp_faddr))
2928 offer = ifp->if_mtu - min_protoh;
2930 offer = tcp_mssdflt;
2935 * Prevent DoS attack with too small MSS. Round up
2936 * to at least minmss.
2938 * Sanity check: make sure that maxopd will be large
2939 * enough to allow some data on segments even is the
2940 * all the option space is used (40bytes). Otherwise
2941 * funny things may happen in tcp_output.
2943 offer = max(offer, tcp_minmss);
2944 offer = max(offer, 64);
2946 rt->rt_rmx.rmx_mssopt = offer;
2949 * While we're here, check if there's an initial rtt
2950 * or rttvar. Convert from the route-table units
2951 * to scaled multiples of the slow timeout timer.
2953 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
2955 * XXX the lock bit for RTT indicates that the value
2956 * is also a minimum value; this is subject to time.
2958 if (rt->rt_rmx.rmx_locks & RTV_RTT)
2959 tp->t_rttmin = rtt / (RTM_RTTUNIT / hz);
2960 tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE));
2961 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
2962 tcpstat.tcps_usedrtt++;
2963 if (rt->rt_rmx.rmx_rttvar) {
2964 tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
2965 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE));
2966 tcpstat.tcps_usedrttvar++;
2968 /* default variation is +- 1 rtt */
2970 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
2972 TCPT_RANGESET(tp->t_rxtcur,
2973 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
2974 tp->t_rttmin, TCPTV_REXMTMAX);
2978 * if there's an mtu associated with the route, use it
2979 * else, use the link mtu. Take the smaller of mss or offer
2982 if (rt->rt_rmx.rmx_mtu) {
2983 mss = rt->rt_rmx.rmx_mtu - min_protoh;
2986 mss = ND_IFINFO(rt->rt_ifp)->linkmtu - min_protoh;
2988 mss = ifp->if_mtu - min_protoh;
2990 mss = min(mss, offer);
2993 * maxopd stores the maximum length of data AND options
2994 * in a segment; maxseg is the amount of data in a normal
2995 * segment. We need to store this value (maxopd) apart
2996 * from maxseg, because now every segment carries options
2997 * and thus we normally have somewhat less data in segments.
3001 if ((tp->t_flags & (TF_REQ_TSTMP | TF_NOOPT)) == TF_REQ_TSTMP &&
3002 ((tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
3003 mss -= TCPOLEN_TSTAMP_APPA;
3005 #if (MCLBYTES & (MCLBYTES - 1)) == 0
3007 mss &= ~(MCLBYTES-1);
3010 mss = mss / MCLBYTES * MCLBYTES;
3013 * If there's a pipesize, change the socket buffer
3014 * to that size. Make the socket buffers an integral
3015 * number of mss units; if the mss is larger than
3016 * the socket buffer, decrease the mss.
3019 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0)
3021 bufsize = so->so_snd.ssb_hiwat;
3025 bufsize = roundup(bufsize, mss);
3026 if (bufsize > sb_max)
3028 if (bufsize > so->so_snd.ssb_hiwat)
3029 ssb_reserve(&so->so_snd, bufsize, so, NULL);
3034 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0)
3036 bufsize = so->so_rcv.ssb_hiwat;
3037 if (bufsize > mss) {
3038 bufsize = roundup(bufsize, mss);
3039 if (bufsize > sb_max)
3041 if (bufsize > so->so_rcv.ssb_hiwat) {
3042 lwkt_gettoken(&so->so_rcv.ssb_token);
3043 ssb_reserve(&so->so_rcv, bufsize, so, NULL);
3044 lwkt_reltoken(&so->so_rcv.ssb_token);
3049 * Set the slow-start flight size depending on whether this
3050 * is a local network or not.
3053 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380));
3057 if (rt->rt_rmx.rmx_ssthresh) {
3059 * There's some sort of gateway or interface
3060 * buffer limit on the path. Use this to set
3061 * the slow start threshhold, but set the
3062 * threshold to no less than 2*mss.
3064 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
3065 tcpstat.tcps_usedssthresh++;
3070 * Determine the MSS option to send on an outgoing SYN.
3073 tcp_mssopt(struct tcpcb *tp)
3078 ((tp->t_inpcb->inp_vflag & INP_IPV6) ? TRUE : FALSE);
3079 int min_protoh = isipv6 ?
3080 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
3081 sizeof(struct tcpiphdr);
3083 const boolean_t isipv6 = FALSE;
3084 const size_t min_protoh = sizeof(struct tcpiphdr);
3088 rt = tcp_rtlookup6(&tp->t_inpcb->inp_inc);
3090 rt = tcp_rtlookup(&tp->t_inpcb->inp_inc);
3092 return (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
3094 return (rt->rt_ifp->if_mtu - min_protoh);
3098 * When a partial ack arrives, force the retransmission of the
3099 * next unacknowledged segment. Do not exit Fast Recovery.
3101 * Implement the Slow-but-Steady variant of NewReno by restarting the
3102 * the retransmission timer. Turn it off here so it can be restarted
3103 * later in tcp_output().
3106 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th, int acked)
3108 tcp_seq old_snd_nxt = tp->snd_nxt;
3109 u_long ocwnd = tp->snd_cwnd;
3111 tcp_callout_stop(tp, tp->tt_rexmt);
3113 tp->snd_nxt = th->th_ack;
3114 /* Set snd_cwnd to one segment beyond acknowledged offset. */
3115 tp->snd_cwnd = tp->t_maxseg;
3116 tp->t_flags |= TF_ACKNOW;
3118 if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
3119 tp->snd_nxt = old_snd_nxt;
3120 /* partial window deflation */
3122 tp->snd_cwnd = ocwnd - acked + tp->t_maxseg;
3124 tp->snd_cwnd = tp->t_maxseg;
3128 * In contrast to the Slow-but-Steady NewReno variant,
3129 * we do not reset the retransmission timer for SACK retransmissions,
3130 * except when retransmitting snd_una.
3133 tcp_sack_rexmt(struct tcpcb *tp, struct tcphdr *th)
3135 uint32_t pipe, seglen;
3138 tcp_seq old_snd_nxt = tp->snd_nxt;
3139 u_long ocwnd = tp->snd_cwnd;
3140 int nseg = 0; /* consecutive new segments */
3141 #define MAXBURST 4 /* limit burst of new packets on partial ack */
3144 pipe = tcp_sack_compute_pipe(tp);
3145 while ((tcp_seq_diff_t)(ocwnd - pipe) >= (tcp_seq_diff_t)tp->t_maxseg &&
3146 (!tcp_do_smartsack || nseg < MAXBURST) &&
3147 tcp_sack_nextseg(tp, &nextrexmt, &seglen, &lostdup)) {
3149 tcp_seq old_snd_max;
3152 if (nextrexmt == tp->snd_max)
3154 tp->snd_nxt = nextrexmt;
3155 tp->snd_cwnd = nextrexmt - tp->snd_una + seglen;
3156 old_snd_max = tp->snd_max;
3157 if (nextrexmt == tp->snd_una)
3158 tcp_callout_stop(tp, tp->tt_rexmt);
3159 error = tcp_output(tp);
3162 sent = tp->snd_nxt - nextrexmt;
3167 tcpstat.tcps_sndsackpack++;
3168 tcpstat.tcps_sndsackbyte += sent;
3169 if (SEQ_LT(nextrexmt, old_snd_max) &&
3170 SEQ_LT(tp->rexmt_high, tp->snd_nxt))
3171 tp->rexmt_high = seq_min(tp->snd_nxt, old_snd_max);
3173 if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
3174 tp->snd_nxt = old_snd_nxt;
3175 tp->snd_cwnd = ocwnd;
3179 * Reset idle time and keep-alive timer, typically called when a valid
3180 * tcp packet is received but may also be called when FASTKEEP is set
3181 * to prevent the previous long-timeout from calculating to a drop.
3183 * Only update t_rcvtime for non-SYN packets.
3185 * Handle the case where one side thinks the connection is established
3186 * but the other side has, say, rebooted without cleaning out the
3187 * connection. The SYNs could be construed as an attack and wind
3188 * up ignored, but in case it isn't an attack we can validate the
3189 * connection by forcing a keepalive.
3192 tcp_timer_keep_activity(struct tcpcb *tp, int thflags)
3194 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
3195 if ((thflags & (TH_SYN | TH_ACK)) == TH_SYN) {
3196 tp->t_flags |= TF_KEEPALIVE;
3197 tcp_callout_reset(tp, tp->tt_keep, hz / 2,
3200 tp->t_rcvtime = ticks;
3201 tp->t_flags &= ~TF_KEEPALIVE;
3202 tcp_callout_reset(tp, tp->tt_keep,
3203 tcp_getkeepidle(tp),