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) 2002, 2003, 2004 Jeffrey M. Hsu. All rights reserved.
37 * License terms: all terms for the DragonFly license above plus the following:
39 * 4. All advertising materials mentioning features or use of this software
40 * must display the following acknowledgement:
42 * This product includes software developed by Jeffrey M. Hsu
43 * for the DragonFly Project.
45 * This requirement may be waived with permission from Jeffrey Hsu.
46 * This requirement will sunset and may be removed on July 8 2005,
47 * after which the standard DragonFly license (as shown above) will
52 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
53 * The Regents of the University of California. All rights reserved.
55 * Redistribution and use in source and binary forms, with or without
56 * modification, are permitted provided that the following conditions
58 * 1. Redistributions of source code must retain the above copyright
59 * notice, this list of conditions and the following disclaimer.
60 * 2. Redistributions in binary form must reproduce the above copyright
61 * notice, this list of conditions and the following disclaimer in the
62 * documentation and/or other materials provided with the distribution.
63 * 3. All advertising materials mentioning features or use of this software
64 * must display the following acknowledgement:
65 * This product includes software developed by the University of
66 * California, Berkeley and its contributors.
67 * 4. Neither the name of the University nor the names of its contributors
68 * may be used to endorse or promote products derived from this software
69 * without specific prior written permission.
71 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
72 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
73 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
74 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
75 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
76 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
77 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
78 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
79 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
80 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
83 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95
84 * $FreeBSD: src/sys/netinet/tcp_input.c,v 1.107.2.38 2003/05/21 04:46:41 cjc Exp $
85 * $DragonFly: src/sys/netinet/tcp_input.c,v 1.57 2005/03/21 01:56:22 hsu Exp $
88 #include "opt_ipfw.h" /* for ipfw_fwd */
89 #include "opt_inet6.h"
90 #include "opt_ipsec.h"
91 #include "opt_tcpdebug.h"
92 #include "opt_tcp_input.h"
94 #include <sys/param.h>
95 #include <sys/systm.h>
96 #include <sys/kernel.h>
97 #include <sys/sysctl.h>
98 #include <sys/malloc.h>
100 #include <sys/proc.h> /* for proc0 declaration */
101 #include <sys/protosw.h>
102 #include <sys/socket.h>
103 #include <sys/socketvar.h>
104 #include <sys/syslog.h>
105 #include <sys/in_cksum.h>
107 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
108 #include <machine/stdarg.h>
111 #include <net/route.h>
113 #include <netinet/in.h>
114 #include <netinet/in_systm.h>
115 #include <netinet/ip.h>
116 #include <netinet/ip_icmp.h> /* for ICMP_BANDLIM */
117 #include <netinet/in_var.h>
118 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */
119 #include <netinet/in_pcb.h>
120 #include <netinet/ip_var.h>
121 #include <netinet/ip6.h>
122 #include <netinet/icmp6.h>
123 #include <netinet6/nd6.h>
124 #include <netinet6/ip6_var.h>
125 #include <netinet6/in6_pcb.h>
126 #include <netinet/tcp.h>
127 #include <netinet/tcp_fsm.h>
128 #include <netinet/tcp_seq.h>
129 #include <netinet/tcp_timer.h>
130 #include <netinet/tcp_var.h>
131 #include <netinet6/tcp6_var.h>
132 #include <netinet/tcpip.h>
135 #include <netinet/tcp_debug.h>
137 u_char tcp_saveipgen[40]; /* the size must be of max ip header, now IPv6 */
138 struct tcphdr tcp_savetcp;
142 #include <netproto/ipsec/ipsec.h>
143 #include <netproto/ipsec/ipsec6.h>
147 #include <netinet6/ipsec.h>
148 #include <netinet6/ipsec6.h>
149 #include <netproto/key/key.h>
152 MALLOC_DEFINE(M_TSEGQ, "tseg_qent", "TCP segment queue entry");
155 static int log_in_vain = 0;
156 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
157 &log_in_vain, 0, "Log all incoming TCP connections");
159 static int blackhole = 0;
160 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW,
161 &blackhole, 0, "Do not send RST when dropping refused connections");
163 int tcp_delack_enabled = 1;
164 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW,
165 &tcp_delack_enabled, 0,
166 "Delay ACK to try and piggyback it onto a data packet");
168 #ifdef TCP_DROP_SYNFIN
169 static int drop_synfin = 0;
170 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW,
171 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set");
174 static int tcp_do_limitedtransmit = 1;
175 SYSCTL_INT(_net_inet_tcp, OID_AUTO, limitedtransmit, CTLFLAG_RW,
176 &tcp_do_limitedtransmit, 0, "Enable RFC 3042 (Limited Transmit)");
178 static int tcp_do_early_retransmit = 1;
179 SYSCTL_INT(_net_inet_tcp, OID_AUTO, earlyretransmit, CTLFLAG_RW,
180 &tcp_do_early_retransmit, 0, "Early retransmit");
182 static int tcp_aggregate_acks = 1;
183 SYSCTL_INT(_net_inet_tcp, OID_AUTO, aggregate_acks, CTLFLAG_RW,
184 &tcp_aggregate_acks, 0, "Aggregate built-up acks into one ack");
186 static int tcp_do_rfc3390 = 1;
187 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW,
189 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
191 static int tcp_do_eifel_detect = 1;
192 SYSCTL_INT(_net_inet_tcp, OID_AUTO, eifel, CTLFLAG_RW,
193 &tcp_do_eifel_detect, 0, "Eifel detection algorithm (RFC 3522)");
196 * Define as tunable for easy testing with SACK on and off.
197 * Warning: do not change setting in the middle of an existing active TCP flow,
198 * else strange things might happen to that flow.
201 SYSCTL_INT(_net_inet_tcp, OID_AUTO, sack, CTLFLAG_RW,
202 &tcp_do_sack, 0, "Enable SACK Algorithms");
204 int tcp_do_smartsack = 1;
205 SYSCTL_INT(_net_inet_tcp, OID_AUTO, smartsack, CTLFLAG_RW,
206 &tcp_do_smartsack, 0, "Enable Smart SACK Algorithms");
208 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0,
209 "TCP Segment Reassembly Queue");
211 int tcp_reass_maxseg = 0;
212 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RD,
213 &tcp_reass_maxseg, 0,
214 "Global maximum number of TCP Segments in Reassembly Queue");
216 int tcp_reass_qsize = 0;
217 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD,
219 "Global number of TCP Segments currently in Reassembly Queue");
221 static int tcp_reass_overflows = 0;
222 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD,
223 &tcp_reass_overflows, 0,
224 "Global number of TCP Segment Reassembly Queue Overflows");
226 static void tcp_dooptions(struct tcpopt *, u_char *, int, boolean_t);
227 static void tcp_pulloutofband(struct socket *,
228 struct tcphdr *, struct mbuf *, int);
229 static int tcp_reass(struct tcpcb *, struct tcphdr *, int *,
231 static void tcp_xmit_timer(struct tcpcb *, int);
232 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *, int);
233 static void tcp_sack_rexmt(struct tcpcb *, struct tcphdr *);
235 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
237 #define ND6_HINT(tp) \
239 if ((tp) && (tp)->t_inpcb && \
240 ((tp)->t_inpcb->inp_vflag & INP_IPV6) && \
241 (tp)->t_inpcb->in6p_route.ro_rt) \
242 nd6_nud_hint((tp)->t_inpcb->in6p_route.ro_rt, NULL, 0); \
249 * Indicate whether this ack should be delayed. We can delay the ack if
250 * - delayed acks are enabled and
251 * - there is no delayed ack timer in progress and
252 * - our last ack wasn't a 0-sized window. We never want to delay
253 * the ack that opens up a 0-sized window.
255 #define DELAY_ACK(tp) \
256 (tcp_delack_enabled && !callout_pending(tp->tt_delack) && \
257 !(tp->t_flags & TF_RXWIN0SENT))
259 #define acceptable_window_update(tp, th, tiwin) \
260 (SEQ_LT(tp->snd_wl1, th->th_seq) || \
261 (tp->snd_wl1 == th->th_seq && \
262 (SEQ_LT(tp->snd_wl2, th->th_ack) || \
263 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))
266 tcp_reass(struct tcpcb *tp, struct tcphdr *th, int *tlenp, struct mbuf *m)
269 struct tseg_qent *p = NULL;
270 struct tseg_qent *te;
271 struct socket *so = tp->t_inpcb->inp_socket;
275 * Call with th == NULL after become established to
276 * force pre-ESTABLISHED data up to user socket.
282 * Limit the number of segments in the reassembly queue to prevent
283 * holding on to too many segments (and thus running out of mbufs).
284 * Make sure to let the missing segment through which caused this
285 * queue. Always keep one global queue entry spare to be able to
286 * process the missing segment.
288 if (th->th_seq != tp->rcv_nxt &&
289 tcp_reass_qsize + 1 >= tcp_reass_maxseg) {
290 tcp_reass_overflows++;
291 tcpstat.tcps_rcvmemdrop++;
293 /* no SACK block to report */
294 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
298 /* Allocate a new queue entry. */
299 MALLOC(te, struct tseg_qent *, sizeof(struct tseg_qent), M_TSEGQ,
300 M_INTWAIT | M_NULLOK);
302 tcpstat.tcps_rcvmemdrop++;
304 /* no SACK block to report */
305 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
311 * Find a segment which begins after this one does.
313 LIST_FOREACH(q, &tp->t_segq, tqe_q) {
314 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
320 * If there is a preceding segment, it may provide some of
321 * our data already. If so, drop the data from the incoming
322 * segment. If it provides all of our data, drop us.
327 /* conversion to int (in i) handles seq wraparound */
328 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
329 if (i > 0) { /* overlaps preceding segment */
330 tp->t_flags |= (TF_DUPSEG | TF_ENCLOSESEG);
331 /* enclosing block starts w/ preceding segment */
332 tp->encloseblk.rblk_start = p->tqe_th->th_seq;
334 /* preceding encloses incoming segment */
335 tp->encloseblk.rblk_end = p->tqe_th->th_seq +
337 tcpstat.tcps_rcvduppack++;
338 tcpstat.tcps_rcvdupbyte += *tlenp;
343 * Try to present any queued data
344 * at the left window edge to the user.
345 * This is needed after the 3-WHS
348 goto present; /* ??? */
353 /* incoming segment end is enclosing block end */
354 tp->encloseblk.rblk_end = th->th_seq + *tlenp +
355 ((th->th_flags & TH_FIN) != 0);
356 /* trim end of reported D-SACK block */
357 tp->reportblk.rblk_end = th->th_seq;
360 tcpstat.tcps_rcvoopack++;
361 tcpstat.tcps_rcvoobyte += *tlenp;
364 * While we overlap succeeding segments trim them or,
365 * if they are completely covered, dequeue them.
368 tcp_seq_diff_t i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
369 tcp_seq qend = q->tqe_th->th_seq + q->tqe_len;
370 struct tseg_qent *nq;
374 if (!(tp->t_flags & TF_DUPSEG)) { /* first time through */
375 tp->t_flags |= (TF_DUPSEG | TF_ENCLOSESEG);
376 tp->encloseblk = tp->reportblk;
377 /* report trailing duplicate D-SACK segment */
378 tp->reportblk.rblk_start = q->tqe_th->th_seq;
380 if ((tp->t_flags & TF_ENCLOSESEG) &&
381 SEQ_GT(qend, tp->encloseblk.rblk_end)) {
382 /* extend enclosing block if one exists */
383 tp->encloseblk.rblk_end = qend;
385 if (i < q->tqe_len) {
386 q->tqe_th->th_seq += i;
392 nq = LIST_NEXT(q, tqe_q);
393 LIST_REMOVE(q, tqe_q);
400 /* Insert the new segment queue entry into place. */
403 te->tqe_len = *tlenp;
405 /* check if can coalesce with following segment */
406 if (q != NULL && (th->th_seq + *tlenp == q->tqe_th->th_seq)) {
407 tcp_seq tend = te->tqe_th->th_seq + te->tqe_len;
409 te->tqe_len += q->tqe_len;
410 if (q->tqe_th->th_flags & TH_FIN)
411 te->tqe_th->th_flags |= TH_FIN;
412 m_cat(te->tqe_m, q->tqe_m);
413 tp->encloseblk.rblk_end = tend;
415 * When not reporting a duplicate segment, use
416 * the larger enclosing block as the SACK block.
418 if (!(tp->t_flags & TF_DUPSEG))
419 tp->reportblk.rblk_end = tend;
420 LIST_REMOVE(q, tqe_q);
426 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
428 /* check if can coalesce with preceding segment */
429 if (p->tqe_th->th_seq + p->tqe_len == th->th_seq) {
430 p->tqe_len += te->tqe_len;
431 m_cat(p->tqe_m, te->tqe_m);
432 tp->encloseblk.rblk_start = p->tqe_th->th_seq;
434 * When not reporting a duplicate segment, use
435 * the larger enclosing block as the SACK block.
437 if (!(tp->t_flags & TF_DUPSEG))
438 tp->reportblk.rblk_start = p->tqe_th->th_seq;
442 LIST_INSERT_AFTER(p, te, tqe_q);
447 * Present data to user, advancing rcv_nxt through
448 * completed sequence space.
450 if (!TCPS_HAVEESTABLISHED(tp->t_state))
452 q = LIST_FIRST(&tp->t_segq);
453 if (q == NULL || q->tqe_th->th_seq != tp->rcv_nxt)
455 tp->rcv_nxt += q->tqe_len;
456 if (!(tp->t_flags & TF_DUPSEG)) {
457 /* no SACK block to report since ACK advanced */
458 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
460 /* no enclosing block to report since ACK advanced */
461 tp->t_flags &= ~TF_ENCLOSESEG;
462 flags = q->tqe_th->th_flags & TH_FIN;
463 LIST_REMOVE(q, tqe_q);
464 KASSERT(LIST_EMPTY(&tp->t_segq) ||
465 LIST_FIRST(&tp->t_segq)->tqe_th->th_seq != tp->rcv_nxt,
466 ("segment not coalesced"));
467 if (so->so_state & SS_CANTRCVMORE)
470 sbappendstream(&so->so_rcv, q->tqe_m);
479 * TCP input routine, follows pages 65-76 of the
480 * protocol specification dated September, 1981 very closely.
484 tcp6_input(struct mbuf **mp, int *offp, int proto)
486 struct mbuf *m = *mp;
487 struct in6_ifaddr *ia6;
489 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
492 * draft-itojun-ipv6-tcp-to-anycast
493 * better place to put this in?
495 ia6 = ip6_getdstifaddr(m);
496 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
499 ip6 = mtod(m, struct ip6_hdr *);
500 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
501 offsetof(struct ip6_hdr, ip6_dst));
502 return (IPPROTO_DONE);
505 tcp_input(m, *offp, proto);
506 return (IPPROTO_DONE);
511 tcp_input(struct mbuf *m, ...)
516 struct ip *ip = NULL;
518 struct inpcb *inp = NULL;
523 struct tcpcb *tp = NULL;
525 struct socket *so = 0;
527 boolean_t ourfinisacked, needoutput = FALSE;
530 struct tcpopt to; /* options in this segment */
531 struct rmxp_tao *taop; /* pointer to our TAO cache entry */
532 struct rmxp_tao tao_noncached; /* in case there's no cached entry */
533 struct sockaddr_in *next_hop = NULL;
534 int rstreason; /* For badport_bandlim accounting purposes */
536 struct ip6_hdr *ip6 = NULL;
540 const boolean_t isipv6 = FALSE;
547 off0 = __va_arg(ap, int);
548 proto = __va_arg(ap, int);
551 tcpstat.tcps_rcvtotal++;
553 /* Grab info from and strip MT_TAG mbufs prepended to the chain. */
554 while (m->m_type == MT_TAG) {
555 if (m->_m_tag_id == PACKET_TAG_IPFORWARD)
556 next_hop = (struct sockaddr_in *)m->m_hdr.mh_data;
561 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? TRUE : FALSE;
565 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */
566 ip6 = mtod(m, struct ip6_hdr *);
567 tlen = (sizeof *ip6) + ntohs(ip6->ip6_plen) - off0;
568 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) {
569 tcpstat.tcps_rcvbadsum++;
572 th = (struct tcphdr *)((caddr_t)ip6 + off0);
575 * Be proactive about unspecified IPv6 address in source.
576 * As we use all-zero to indicate unbounded/unconnected pcb,
577 * unspecified IPv6 address can be used to confuse us.
579 * Note that packets with unspecified IPv6 destination is
580 * already dropped in ip6_input.
582 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
588 * Get IP and TCP header together in first mbuf.
589 * Note: IP leaves IP header in first mbuf.
591 if (off0 > sizeof(struct ip)) {
593 off0 = sizeof(struct ip);
595 /* already checked and pulled up in ip_demux() */
596 KASSERT(m->m_len >= sizeof(struct tcpiphdr),
597 ("TCP header not in one mbuf: m->m_len %d", m->m_len));
598 ip = mtod(m, struct ip *);
599 ipov = (struct ipovly *)ip;
600 th = (struct tcphdr *)((caddr_t)ip + off0);
603 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
604 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
605 th->th_sum = m->m_pkthdr.csum_data;
607 th->th_sum = in_pseudo(ip->ip_src.s_addr,
609 htonl(m->m_pkthdr.csum_data +
612 th->th_sum ^= 0xffff;
615 * Checksum extended TCP header and data.
617 len = sizeof(struct ip) + tlen;
618 bzero(ipov->ih_x1, sizeof ipov->ih_x1);
619 ipov->ih_len = (u_short)tlen;
620 ipov->ih_len = htons(ipov->ih_len);
621 th->th_sum = in_cksum(m, len);
624 tcpstat.tcps_rcvbadsum++;
628 /* Re-initialization for later version check */
629 ip->ip_v = IPVERSION;
634 * Check that TCP offset makes sense,
635 * pull out TCP options and adjust length. XXX
637 off = th->th_off << 2;
638 /* already checked and pulled up in ip_demux() */
639 KASSERT(off >= sizeof(struct tcphdr) && off <= tlen,
640 ("bad TCP data offset %d (tlen %d)", off, tlen));
641 tlen -= off; /* tlen is used instead of ti->ti_len */
642 if (off > sizeof(struct tcphdr)) {
644 IP6_EXTHDR_CHECK(m, off0, off, );
645 ip6 = mtod(m, struct ip6_hdr *);
646 th = (struct tcphdr *)((caddr_t)ip6 + off0);
648 /* already pulled up in ip_demux() */
649 KASSERT(m->m_len >= sizeof(struct ip) + off,
650 ("TCP header and options not in one mbuf: "
651 "m_len %d, off %d", m->m_len, off));
653 optlen = off - sizeof(struct tcphdr);
654 optp = (u_char *)(th + 1);
656 thflags = th->th_flags;
658 #ifdef TCP_DROP_SYNFIN
660 * If the drop_synfin option is enabled, drop all packets with
661 * both the SYN and FIN bits set. This prevents e.g. nmap from
662 * identifying the TCP/IP stack.
664 * This is a violation of the TCP specification.
666 if (drop_synfin && (thflags & (TH_SYN | TH_FIN)) == (TH_SYN | TH_FIN))
671 * Convert TCP protocol specific fields to host format.
673 th->th_seq = ntohl(th->th_seq);
674 th->th_ack = ntohl(th->th_ack);
675 th->th_win = ntohs(th->th_win);
676 th->th_urp = ntohs(th->th_urp);
679 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options,
680 * until after ip6_savecontrol() is called and before other functions
681 * which don't want those proto headers.
682 * Because ip6_savecontrol() is going to parse the mbuf to
683 * search for data to be passed up to user-land, it wants mbuf
684 * parameters to be unchanged.
685 * XXX: the call of ip6_savecontrol() has been obsoleted based on
686 * latest version of the advanced API (20020110).
688 drop_hdrlen = off0 + off;
691 * Locate pcb for segment.
694 /* IPFIREWALL_FORWARD section */
695 if (next_hop != NULL && !isipv6) { /* IPv6 support is not there yet */
697 * Transparently forwarded. Pretend to be the destination.
698 * already got one like this?
700 cpu = mycpu->gd_cpuid;
701 inp = in_pcblookup_hash(&tcbinfo[cpu],
702 ip->ip_src, th->th_sport,
703 ip->ip_dst, th->th_dport,
704 0, m->m_pkthdr.rcvif);
707 * It's new. Try to find the ambushing socket.
711 * The rest of the ipfw code stores the port in
713 * (The IP address is still in network order.)
715 in_port_t dport = next_hop->sin_port ?
716 htons(next_hop->sin_port) :
719 cpu = tcp_addrcpu(ip->ip_src.s_addr, th->th_sport,
720 next_hop->sin_addr.s_addr, dport);
721 inp = in_pcblookup_hash(&tcbinfo[cpu],
722 ip->ip_src, th->th_sport,
723 next_hop->sin_addr, dport,
724 1, m->m_pkthdr.rcvif);
728 inp = in6_pcblookup_hash(&tcbinfo[0],
729 &ip6->ip6_src, th->th_sport,
730 &ip6->ip6_dst, th->th_dport,
731 1, m->m_pkthdr.rcvif);
733 cpu = mycpu->gd_cpuid;
734 inp = in_pcblookup_hash(&tcbinfo[cpu],
735 ip->ip_src, th->th_sport,
736 ip->ip_dst, th->th_dport,
737 1, m->m_pkthdr.rcvif);
742 * If the state is CLOSED (i.e., TCB does not exist) then
743 * all data in the incoming segment is discarded.
744 * If the TCB exists but is in CLOSED state, it is embryonic,
745 * but should either do a listen or a connect soon.
750 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2];
752 char dbuf[sizeof "aaa.bbb.ccc.ddd"];
753 char sbuf[sizeof "aaa.bbb.ccc.ddd"];
757 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst));
760 strcat(sbuf, ip6_sprintf(&ip6->ip6_src));
763 strcpy(dbuf, inet_ntoa(ip->ip_dst));
764 strcpy(sbuf, inet_ntoa(ip->ip_src));
766 switch (log_in_vain) {
768 if (!(thflags & TH_SYN))
772 "Connection attempt to TCP %s:%d "
773 "from %s:%d flags:0x%02x\n",
774 dbuf, ntohs(th->th_dport), sbuf,
775 ntohs(th->th_sport), thflags);
784 if (thflags & TH_SYN)
793 rstreason = BANDLIM_RST_CLOSEDPORT;
799 if (ipsec6_in_reject_so(m, inp->inp_socket)) {
800 ipsec6stat.in_polvio++;
804 if (ipsec4_in_reject_so(m, inp->inp_socket)) {
805 ipsecstat.in_polvio++;
812 if (ipsec6_in_reject(m, inp))
815 if (ipsec4_in_reject(m, inp))
822 rstreason = BANDLIM_RST_CLOSEDPORT;
825 if (tp->t_state <= TCPS_CLOSED)
828 /* Unscale the window into a 32-bit value. */
829 if (!(thflags & TH_SYN))
830 tiwin = th->th_win << tp->snd_scale;
834 so = inp->inp_socket;
837 if (so->so_options & SO_DEBUG) {
838 ostate = tp->t_state;
840 bcopy(ip6, tcp_saveipgen, sizeof(*ip6));
842 bcopy(ip, tcp_saveipgen, sizeof(*ip));
847 bzero(&to, sizeof to);
849 if (so->so_options & SO_ACCEPTCONN) {
850 struct in_conninfo inc;
853 inc.inc_isipv6 = (isipv6 == TRUE);
856 inc.inc6_faddr = ip6->ip6_src;
857 inc.inc6_laddr = ip6->ip6_dst;
858 inc.inc6_route.ro_rt = NULL; /* XXX */
860 inc.inc_faddr = ip->ip_src;
861 inc.inc_laddr = ip->ip_dst;
862 inc.inc_route.ro_rt = NULL; /* XXX */
864 inc.inc_fport = th->th_sport;
865 inc.inc_lport = th->th_dport;
868 * If the state is LISTEN then ignore segment if it contains
869 * a RST. If the segment contains an ACK then it is bad and
870 * send a RST. If it does not contain a SYN then it is not
871 * interesting; drop it.
873 * If the state is SYN_RECEIVED (syncache) and seg contains
874 * an ACK, but not for our SYN/ACK, send a RST. If the seg
875 * contains a RST, check the sequence number to see if it
876 * is a valid reset segment.
878 if ((thflags & (TH_RST | TH_ACK | TH_SYN)) != TH_SYN) {
879 if ((thflags & (TH_RST | TH_ACK | TH_SYN)) == TH_ACK) {
880 if (!syncache_expand(&inc, th, &so, m)) {
882 * No syncache entry, or ACK was not
883 * for our SYN/ACK. Send a RST.
885 tcpstat.tcps_badsyn++;
886 rstreason = BANDLIM_RST_OPENPORT;
891 * Could not complete 3-way handshake,
892 * connection is being closed down, and
893 * syncache will free mbuf.
897 * Socket is created in state SYN_RECEIVED.
898 * Continue processing segment.
903 * This is what would have happened in
904 * tcp_output() when the SYN,ACK was sent.
906 tp->snd_up = tp->snd_una;
907 tp->snd_max = tp->snd_nxt = tp->iss + 1;
908 tp->last_ack_sent = tp->rcv_nxt;
910 * XXX possible bug - it doesn't appear that tp->snd_wnd is unscaled
911 * until the _second_ ACK is received:
912 * rcv SYN (set wscale opts) --> send SYN/ACK, set snd_wnd = window.
913 * rcv ACK, calculate tiwin --> process SYN_RECEIVED, determine wscale,
914 * move to ESTAB, set snd_wnd to tiwin.
916 tp->snd_wnd = tiwin; /* unscaled */
919 if (thflags & TH_RST) {
920 syncache_chkrst(&inc, th);
923 if (thflags & TH_ACK) {
924 syncache_badack(&inc);
925 tcpstat.tcps_badsyn++;
926 rstreason = BANDLIM_RST_OPENPORT;
933 * Segment's flags are (SYN) or (SYN | FIN).
937 * If deprecated address is forbidden,
938 * we do not accept SYN to deprecated interface
939 * address to prevent any new inbound connection from
940 * getting established.
941 * When we do not accept SYN, we send a TCP RST,
942 * with deprecated source address (instead of dropping
943 * it). We compromise it as it is much better for peer
944 * to send a RST, and RST will be the final packet
947 * If we do not forbid deprecated addresses, we accept
948 * the SYN packet. RFC2462 does not suggest dropping
950 * If we decipher RFC2462 5.5.4, it says like this:
951 * 1. use of deprecated addr with existing
952 * communication is okay - "SHOULD continue to be
954 * 2. use of it with new communication:
955 * (2a) "SHOULD NOT be used if alternate address
956 * with sufficient scope is available"
957 * (2b) nothing mentioned otherwise.
958 * Here we fall into (2b) case as we have no choice in
959 * our source address selection - we must obey the peer.
961 * The wording in RFC2462 is confusing, and there are
962 * multiple description text for deprecated address
963 * handling - worse, they are not exactly the same.
964 * I believe 5.5.4 is the best one, so we follow 5.5.4.
966 if (isipv6 && !ip6_use_deprecated) {
967 struct in6_ifaddr *ia6;
969 if ((ia6 = ip6_getdstifaddr(m)) &&
970 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
972 rstreason = BANDLIM_RST_OPENPORT;
978 * If it is from this socket, drop it, it must be forged.
979 * Don't bother responding if the destination was a broadcast.
981 if (th->th_dport == th->th_sport) {
983 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
987 if (ip->ip_dst.s_addr == ip->ip_src.s_addr)
992 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
994 * Note that it is quite possible to receive unicast
995 * link-layer packets with a broadcast IP address. Use
996 * in_broadcast() to find them.
998 if (m->m_flags & (M_BCAST | M_MCAST))
1001 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1002 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
1005 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1006 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1007 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1008 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
1012 * SYN appears to be valid; create compressed TCP state
1013 * for syncache, or perform t/tcp connection.
1015 if (so->so_qlen <= so->so_qlimit) {
1016 tcp_dooptions(&to, optp, optlen, TRUE);
1017 if (!syncache_add(&inc, &to, th, &so, m))
1021 * Entry added to syncache, mbuf used to
1022 * send SYN,ACK packet.
1026 * Segment passed TAO tests.
1029 tp = intotcpcb(inp);
1030 tp->snd_wnd = tiwin;
1031 tp->t_starttime = ticks;
1032 tp->t_state = TCPS_ESTABLISHED;
1035 * If there is a FIN, or if there is data and the
1036 * connection is local, then delay SYN,ACK(SYN) in
1037 * the hope of piggy-backing it on a response
1038 * segment. Otherwise must send ACK now in case
1039 * the other side is slow starting.
1041 if (DELAY_ACK(tp) &&
1042 ((thflags & TH_FIN) ||
1044 ((isipv6 && in6_localaddr(&inp->in6p_faddr)) ||
1045 (!isipv6 && in_localaddr(inp->inp_faddr)))))) {
1046 callout_reset(tp->tt_delack, tcp_delacktime,
1047 tcp_timer_delack, tp);
1048 tp->t_flags |= TF_NEEDSYN;
1050 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
1052 tcpstat.tcps_connects++;
1060 /* should not happen - syncache should pick up these connections */
1061 KASSERT(tp->t_state != TCPS_LISTEN, ("tcp_input: TCPS_LISTEN state"));
1064 * Segment received on connection.
1065 * Reset idle time and keep-alive timer.
1067 tp->t_rcvtime = ticks;
1068 if (TCPS_HAVEESTABLISHED(tp->t_state))
1069 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp);
1073 * XXX this is tradtitional behavior, may need to be cleaned up.
1075 tcp_dooptions(&to, optp, optlen, (thflags & TH_SYN) != 0);
1076 if (thflags & TH_SYN) {
1077 if (to.to_flags & TOF_SCALE) {
1078 tp->t_flags |= TF_RCVD_SCALE;
1079 tp->requested_s_scale = to.to_requested_s_scale;
1081 if (to.to_flags & TOF_TS) {
1082 tp->t_flags |= TF_RCVD_TSTMP;
1083 tp->ts_recent = to.to_tsval;
1084 tp->ts_recent_age = ticks;
1086 if (to.to_flags & (TOF_CC | TOF_CCNEW))
1087 tp->t_flags |= TF_RCVD_CC;
1088 if (to.to_flags & TOF_MSS)
1089 tcp_mss(tp, to.to_mss);
1091 * Only set the TF_SACK_PERMITTED per-connection flag
1092 * if we got a SACK_PERMITTED option from the other side
1093 * and the global tcp_do_sack variable is true.
1095 if (tcp_do_sack && (to.to_flags & TOF_SACK_PERMITTED))
1096 tp->t_flags |= TF_SACK_PERMITTED;
1100 * Header prediction: check for the two common cases
1101 * of a uni-directional data xfer. If the packet has
1102 * no control flags, is in-sequence, the window didn't
1103 * change and we're not retransmitting, it's a
1104 * candidate. If the length is zero and the ack moved
1105 * forward, we're the sender side of the xfer. Just
1106 * free the data acked & wake any higher level process
1107 * that was blocked waiting for space. If the length
1108 * is non-zero and the ack didn't move, we're the
1109 * receiver side. If we're getting packets in-order
1110 * (the reassembly queue is empty), add the data to
1111 * the socket buffer and note that we need a delayed ack.
1112 * Make sure that the hidden state-flags are also off.
1113 * Since we check for TCPS_ESTABLISHED above, it can only
1116 if (tp->t_state == TCPS_ESTABLISHED &&
1117 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1118 !(tp->t_flags & (TF_NEEDSYN | TF_NEEDFIN)) &&
1119 (!(to.to_flags & TOF_TS) ||
1120 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) &&
1122 * Using the CC option is compulsory if once started:
1123 * the segment is OK if no T/TCP was negotiated or
1124 * if the segment has a CC option equal to CCrecv
1126 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) ||
1127 ((to.to_flags & TOF_CC) && to.to_cc == tp->cc_recv)) &&
1128 th->th_seq == tp->rcv_nxt &&
1129 tp->snd_nxt == tp->snd_max) {
1132 * If last ACK falls within this segment's sequence numbers,
1133 * record the timestamp.
1134 * NOTE that the test is modified according to the latest
1135 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1137 if ((to.to_flags & TOF_TS) &&
1138 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1139 tp->ts_recent_age = ticks;
1140 tp->ts_recent = to.to_tsval;
1144 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1145 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1146 tp->snd_cwnd >= tp->snd_wnd &&
1147 !IN_FASTRECOVERY(tp)) {
1149 * This is a pure ack for outstanding data.
1151 ++tcpstat.tcps_predack;
1153 * "bad retransmit" recovery
1155 * If Eifel detection applies, then
1156 * it is deterministic, so use it
1157 * unconditionally over the old heuristic.
1158 * Otherwise, fall back to the old heuristic.
1160 if (tcp_do_eifel_detect &&
1161 (to.to_flags & TOF_TS) && to.to_tsecr &&
1162 (tp->t_flags & TF_FIRSTACCACK)) {
1163 /* Eifel detection applicable. */
1164 if (to.to_tsecr < tp->t_rexmtTS) {
1165 tcp_revert_congestion_state(tp);
1166 ++tcpstat.tcps_eifeldetected;
1168 } else if (tp->t_rxtshift == 1 &&
1169 ticks < tp->t_badrxtwin) {
1170 tcp_revert_congestion_state(tp);
1171 ++tcpstat.tcps_rttdetected;
1173 tp->t_flags &= ~(TF_FIRSTACCACK |
1174 TF_FASTREXMT | TF_EARLYREXMT);
1176 * Recalculate the retransmit timer / rtt.
1178 * Some machines (certain windows boxes)
1179 * send broken timestamp replies during the
1180 * SYN+ACK phase, ignore timestamps of 0.
1182 if ((to.to_flags & TOF_TS) && to.to_tsecr) {
1184 ticks - to.to_tsecr + 1);
1185 } else if (tp->t_rtttime &&
1186 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1188 ticks - tp->t_rtttime);
1190 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1191 acked = th->th_ack - tp->snd_una;
1192 tcpstat.tcps_rcvackpack++;
1193 tcpstat.tcps_rcvackbyte += acked;
1194 sbdrop(&so->so_snd, acked);
1195 tp->snd_recover = th->th_ack - 1;
1196 tp->snd_una = th->th_ack;
1199 * Update window information.
1201 if (tiwin != tp->snd_wnd &&
1202 acceptable_window_update(tp, th, tiwin)) {
1203 /* keep track of pure window updates */
1204 if (tp->snd_wl2 == th->th_ack &&
1205 tiwin > tp->snd_wnd)
1206 tcpstat.tcps_rcvwinupd++;
1207 tp->snd_wnd = tiwin;
1208 tp->snd_wl1 = th->th_seq;
1209 tp->snd_wl2 = th->th_ack;
1210 if (tp->snd_wnd > tp->max_sndwnd)
1211 tp->max_sndwnd = tp->snd_wnd;
1214 ND6_HINT(tp); /* some progress has been done */
1216 * If all outstanding data are acked, stop
1217 * retransmit timer, otherwise restart timer
1218 * using current (possibly backed-off) value.
1219 * If process is waiting for space,
1220 * wakeup/selwakeup/signal. If data
1221 * are ready to send, let tcp_output
1222 * decide between more output or persist.
1224 if (tp->snd_una == tp->snd_max)
1225 callout_stop(tp->tt_rexmt);
1226 else if (!callout_active(tp->tt_persist))
1227 callout_reset(tp->tt_rexmt,
1229 tcp_timer_rexmt, tp);
1231 if (so->so_snd.sb_cc > 0)
1235 } else if (tiwin == tp->snd_wnd &&
1236 th->th_ack == tp->snd_una &&
1237 LIST_EMPTY(&tp->t_segq) &&
1238 tlen <= sbspace(&so->so_rcv)) {
1240 * This is a pure, in-sequence data packet
1241 * with nothing on the reassembly queue and
1242 * we have enough buffer space to take it.
1244 ++tcpstat.tcps_preddat;
1245 tp->rcv_nxt += tlen;
1246 tcpstat.tcps_rcvpack++;
1247 tcpstat.tcps_rcvbyte += tlen;
1248 ND6_HINT(tp); /* some progress has been done */
1250 * Add data to socket buffer.
1252 if (so->so_state & SS_CANTRCVMORE) {
1255 m_adj(m, drop_hdrlen); /* delayed header drop */
1256 sbappendstream(&so->so_rcv, m);
1260 * This code is responsible for most of the ACKs
1261 * the TCP stack sends back after receiving a data
1262 * packet. Note that the DELAY_ACK check fails if
1263 * the delack timer is already running, which results
1264 * in an ack being sent every other packet (which is
1267 * We then further aggregate acks by not actually
1268 * sending one until the protocol thread has completed
1269 * processing the current backlog of packets. This
1270 * does not delay the ack any further, but allows us
1271 * to take advantage of the packet aggregation that
1272 * high speed NICs do (usually blocks of 8-10 packets)
1273 * to send a single ack rather then four or five acks,
1274 * greatly reducing the ack rate, the return channel
1275 * bandwidth, and the protocol overhead on both ends.
1277 * Since this also has the effect of slowing down
1278 * the exponential slow-start ramp-up, systems with
1279 * very large bandwidth-delay products might want
1280 * to turn the feature off.
1282 if (DELAY_ACK(tp)) {
1283 callout_reset(tp->tt_delack, tcp_delacktime,
1284 tcp_timer_delack, tp);
1285 } else if (tcp_aggregate_acks) {
1286 tp->t_flags |= TF_ACKNOW;
1287 if (!(tp->t_flags & TF_ONOUTPUTQ)) {
1288 tp->t_flags |= TF_ONOUTPUTQ;
1289 tp->tt_cpu = mycpu->gd_cpuid;
1291 &tcpcbackq[tp->tt_cpu],
1295 tp->t_flags |= TF_ACKNOW;
1303 * Calculate amount of space in receive window,
1304 * and then do TCP input processing.
1305 * Receive window is amount of space in rcv queue,
1306 * but not less than advertised window.
1308 recvwin = sbspace(&so->so_rcv);
1311 tp->rcv_wnd = imax(recvwin, (int)(tp->rcv_adv - tp->rcv_nxt));
1313 switch (tp->t_state) {
1315 * If the state is SYN_RECEIVED:
1316 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1318 case TCPS_SYN_RECEIVED:
1319 if ((thflags & TH_ACK) &&
1320 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1321 SEQ_GT(th->th_ack, tp->snd_max))) {
1322 rstreason = BANDLIM_RST_OPENPORT;
1328 * If the state is SYN_SENT:
1329 * if seg contains an ACK, but not for our SYN, drop the input.
1330 * if seg contains a RST, then drop the connection.
1331 * if seg does not contain SYN, then drop it.
1332 * Otherwise this is an acceptable SYN segment
1333 * initialize tp->rcv_nxt and tp->irs
1334 * if seg contains ack then advance tp->snd_una
1335 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1336 * arrange for segment to be acked (eventually)
1337 * continue processing rest of data/controls, beginning with URG
1340 if ((taop = tcp_gettaocache(&inp->inp_inc)) == NULL) {
1341 taop = &tao_noncached;
1342 bzero(taop, sizeof *taop);
1345 if ((thflags & TH_ACK) &&
1346 (SEQ_LEQ(th->th_ack, tp->iss) ||
1347 SEQ_GT(th->th_ack, tp->snd_max))) {
1349 * If we have a cached CCsent for the remote host,
1350 * hence we haven't just crashed and restarted,
1351 * do not send a RST. This may be a retransmission
1352 * from the other side after our earlier ACK was lost.
1353 * Our new SYN, when it arrives, will serve as the
1356 if (taop->tao_ccsent != 0)
1359 rstreason = BANDLIM_UNLIMITED;
1363 if (thflags & TH_RST) {
1364 if (thflags & TH_ACK)
1365 tp = tcp_drop(tp, ECONNREFUSED);
1368 if (!(thflags & TH_SYN))
1370 tp->snd_wnd = th->th_win; /* initial send window */
1371 tp->cc_recv = to.to_cc; /* foreign CC */
1373 tp->irs = th->th_seq;
1375 if (thflags & TH_ACK) {
1377 * Our SYN was acked. If segment contains CC.ECHO
1378 * option, check it to make sure this segment really
1379 * matches our SYN. If not, just drop it as old
1380 * duplicate, but send an RST if we're still playing
1381 * by the old rules. If no CC.ECHO option, make sure
1382 * we don't get fooled into using T/TCP.
1384 if (to.to_flags & TOF_CCECHO) {
1385 if (tp->cc_send != to.to_ccecho) {
1386 if (taop->tao_ccsent != 0)
1389 rstreason = BANDLIM_UNLIMITED;
1394 tp->t_flags &= ~TF_RCVD_CC;
1395 tcpstat.tcps_connects++;
1397 /* Do window scaling on this connection? */
1398 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
1399 (TF_RCVD_SCALE | TF_REQ_SCALE)) {
1400 tp->snd_scale = tp->requested_s_scale;
1401 tp->rcv_scale = tp->request_r_scale;
1403 /* Segment is acceptable, update cache if undefined. */
1404 if (taop->tao_ccsent == 0)
1405 taop->tao_ccsent = to.to_ccecho;
1407 tp->rcv_adv += tp->rcv_wnd;
1408 tp->snd_una++; /* SYN is acked */
1409 callout_stop(tp->tt_rexmt);
1411 * If there's data, delay ACK; if there's also a FIN
1412 * ACKNOW will be turned on later.
1414 if (DELAY_ACK(tp) && tlen != 0)
1415 callout_reset(tp->tt_delack, tcp_delacktime,
1416 tcp_timer_delack, tp);
1418 tp->t_flags |= TF_ACKNOW;
1420 * Received <SYN,ACK> in SYN_SENT[*] state.
1422 * SYN_SENT --> ESTABLISHED
1423 * SYN_SENT* --> FIN_WAIT_1
1425 tp->t_starttime = ticks;
1426 if (tp->t_flags & TF_NEEDFIN) {
1427 tp->t_state = TCPS_FIN_WAIT_1;
1428 tp->t_flags &= ~TF_NEEDFIN;
1431 tp->t_state = TCPS_ESTABLISHED;
1432 callout_reset(tp->tt_keep, tcp_keepidle,
1433 tcp_timer_keep, tp);
1437 * Received initial SYN in SYN-SENT[*] state =>
1438 * simultaneous open. If segment contains CC option
1439 * and there is a cached CC, apply TAO test.
1440 * If it succeeds, connection is * half-synchronized.
1441 * Otherwise, do 3-way handshake:
1442 * SYN-SENT -> SYN-RECEIVED
1443 * SYN-SENT* -> SYN-RECEIVED*
1444 * If there was no CC option, clear cached CC value.
1446 tp->t_flags |= TF_ACKNOW;
1447 callout_stop(tp->tt_rexmt);
1448 if (to.to_flags & TOF_CC) {
1449 if (taop->tao_cc != 0 &&
1450 CC_GT(to.to_cc, taop->tao_cc)) {
1452 * update cache and make transition:
1453 * SYN-SENT -> ESTABLISHED*
1454 * SYN-SENT* -> FIN-WAIT-1*
1456 taop->tao_cc = to.to_cc;
1457 tp->t_starttime = ticks;
1458 if (tp->t_flags & TF_NEEDFIN) {
1459 tp->t_state = TCPS_FIN_WAIT_1;
1460 tp->t_flags &= ~TF_NEEDFIN;
1462 tp->t_state = TCPS_ESTABLISHED;
1463 callout_reset(tp->tt_keep,
1468 tp->t_flags |= TF_NEEDSYN;
1470 tp->t_state = TCPS_SYN_RECEIVED;
1472 /* CC.NEW or no option => invalidate cache */
1474 tp->t_state = TCPS_SYN_RECEIVED;
1480 * Advance th->th_seq to correspond to first data byte.
1481 * If data, trim to stay within window,
1482 * dropping FIN if necessary.
1485 if (tlen > tp->rcv_wnd) {
1486 todrop = tlen - tp->rcv_wnd;
1490 tcpstat.tcps_rcvpackafterwin++;
1491 tcpstat.tcps_rcvbyteafterwin += todrop;
1493 tp->snd_wl1 = th->th_seq - 1;
1494 tp->rcv_up = th->th_seq;
1496 * Client side of transaction: already sent SYN and data.
1497 * If the remote host used T/TCP to validate the SYN,
1498 * our data will be ACK'd; if so, enter normal data segment
1499 * processing in the middle of step 5, ack processing.
1500 * Otherwise, goto step 6.
1502 if (thflags & TH_ACK)
1508 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
1509 * if segment contains a SYN and CC [not CC.NEW] option:
1510 * if state == TIME_WAIT and connection duration > MSL,
1511 * drop packet and send RST;
1513 * if SEG.CC > CCrecv then is new SYN, and can implicitly
1514 * ack the FIN (and data) in retransmission queue.
1515 * Complete close and delete TCPCB. Then reprocess
1516 * segment, hoping to find new TCPCB in LISTEN state;
1518 * else must be old SYN; drop it.
1519 * else do normal processing.
1523 case TCPS_TIME_WAIT:
1524 if ((thflags & TH_SYN) &&
1525 (to.to_flags & TOF_CC) && tp->cc_recv != 0) {
1526 if (tp->t_state == TCPS_TIME_WAIT &&
1527 (ticks - tp->t_starttime) > tcp_msl) {
1528 rstreason = BANDLIM_UNLIMITED;
1531 if (CC_GT(to.to_cc, tp->cc_recv)) {
1538 break; /* continue normal processing */
1542 * States other than LISTEN or SYN_SENT.
1543 * First check the RST flag and sequence number since reset segments
1544 * are exempt from the timestamp and connection count tests. This
1545 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
1546 * below which allowed reset segments in half the sequence space
1547 * to fall though and be processed (which gives forged reset
1548 * segments with a random sequence number a 50 percent chance of
1549 * killing a connection).
1550 * Then check timestamp, if present.
1551 * Then check the connection count, if present.
1552 * Then check that at least some bytes of segment are within
1553 * receive window. If segment begins before rcv_nxt,
1554 * drop leading data (and SYN); if nothing left, just ack.
1557 * If the RST bit is set, check the sequence number to see
1558 * if this is a valid reset segment.
1560 * In all states except SYN-SENT, all reset (RST) segments
1561 * are validated by checking their SEQ-fields. A reset is
1562 * valid if its sequence number is in the window.
1563 * Note: this does not take into account delayed ACKs, so
1564 * we should test against last_ack_sent instead of rcv_nxt.
1565 * The sequence number in the reset segment is normally an
1566 * echo of our outgoing acknowledgement numbers, but some hosts
1567 * send a reset with the sequence number at the rightmost edge
1568 * of our receive window, and we have to handle this case.
1569 * If we have multiple segments in flight, the intial reset
1570 * segment sequence numbers will be to the left of last_ack_sent,
1571 * but they will eventually catch up.
1572 * In any case, it never made sense to trim reset segments to
1573 * fit the receive window since RFC 1122 says:
1574 * 4.2.2.12 RST Segment: RFC-793 Section 3.4
1576 * A TCP SHOULD allow a received RST segment to include data.
1579 * It has been suggested that a RST segment could contain
1580 * ASCII text that encoded and explained the cause of the
1581 * RST. No standard has yet been established for such
1584 * If the reset segment passes the sequence number test examine
1586 * SYN_RECEIVED STATE:
1587 * If passive open, return to LISTEN state.
1588 * If active open, inform user that connection was refused.
1589 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
1590 * Inform user that connection was reset, and close tcb.
1591 * CLOSING, LAST_ACK STATES:
1594 * Drop the segment - see Stevens, vol. 2, p. 964 and
1597 if (thflags & TH_RST) {
1598 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
1599 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
1600 switch (tp->t_state) {
1602 case TCPS_SYN_RECEIVED:
1603 so->so_error = ECONNREFUSED;
1606 case TCPS_ESTABLISHED:
1607 case TCPS_FIN_WAIT_1:
1608 case TCPS_FIN_WAIT_2:
1609 case TCPS_CLOSE_WAIT:
1610 so->so_error = ECONNRESET;
1612 tp->t_state = TCPS_CLOSED;
1613 tcpstat.tcps_drops++;
1622 case TCPS_TIME_WAIT:
1630 * RFC 1323 PAWS: If we have a timestamp reply on this segment
1631 * and it's less than ts_recent, drop it.
1633 if ((to.to_flags & TOF_TS) && tp->ts_recent != 0 &&
1634 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
1636 /* Check to see if ts_recent is over 24 days old. */
1637 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1639 * Invalidate ts_recent. If this segment updates
1640 * ts_recent, the age will be reset later and ts_recent
1641 * will get a valid value. If it does not, setting
1642 * ts_recent to zero will at least satisfy the
1643 * requirement that zero be placed in the timestamp
1644 * echo reply when ts_recent isn't valid. The
1645 * age isn't reset until we get a valid ts_recent
1646 * because we don't want out-of-order segments to be
1647 * dropped when ts_recent is old.
1651 tcpstat.tcps_rcvduppack++;
1652 tcpstat.tcps_rcvdupbyte += tlen;
1653 tcpstat.tcps_pawsdrop++;
1662 * If T/TCP was negotiated and the segment doesn't have CC,
1663 * or if its CC is wrong then drop the segment.
1664 * RST segments do not have to comply with this.
1666 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) &&
1667 (!(to.to_flags & TOF_CC) || tp->cc_recv != to.to_cc))
1671 * In the SYN-RECEIVED state, validate that the packet belongs to
1672 * this connection before trimming the data to fit the receive
1673 * window. Check the sequence number versus IRS since we know
1674 * the sequence numbers haven't wrapped. This is a partial fix
1675 * for the "LAND" DoS attack.
1677 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
1678 rstreason = BANDLIM_RST_OPENPORT;
1682 todrop = tp->rcv_nxt - th->th_seq;
1684 if (TCP_DO_SACK(tp)) {
1685 /* Report duplicate segment at head of packet. */
1686 tp->reportblk.rblk_start = th->th_seq;
1687 tp->reportblk.rblk_end = th->th_seq + tlen;
1688 if (thflags & TH_FIN)
1689 ++tp->reportblk.rblk_end;
1690 if (SEQ_GT(tp->reportblk.rblk_end, tp->rcv_nxt))
1691 tp->reportblk.rblk_end = tp->rcv_nxt;
1692 tp->t_flags |= (TF_DUPSEG | TF_SACKLEFT | TF_ACKNOW);
1694 if (thflags & TH_SYN) {
1704 * Following if statement from Stevens, vol. 2, p. 960.
1706 if (todrop > tlen ||
1707 (todrop == tlen && !(thflags & TH_FIN))) {
1709 * Any valid FIN must be to the left of the window.
1710 * At this point the FIN must be a duplicate or out
1711 * of sequence; drop it.
1716 * Send an ACK to resynchronize and drop any data.
1717 * But keep on processing for RST or ACK.
1719 tp->t_flags |= TF_ACKNOW;
1721 tcpstat.tcps_rcvduppack++;
1722 tcpstat.tcps_rcvdupbyte += todrop;
1724 tcpstat.tcps_rcvpartduppack++;
1725 tcpstat.tcps_rcvpartdupbyte += todrop;
1727 drop_hdrlen += todrop; /* drop from the top afterwards */
1728 th->th_seq += todrop;
1730 if (th->th_urp > todrop)
1731 th->th_urp -= todrop;
1739 * If new data are received on a connection after the
1740 * user processes are gone, then RST the other end.
1742 if ((so->so_state & SS_NOFDREF) &&
1743 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1745 tcpstat.tcps_rcvafterclose++;
1746 rstreason = BANDLIM_UNLIMITED;
1751 * If segment ends after window, drop trailing data
1752 * (and PUSH and FIN); if nothing left, just ACK.
1754 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
1756 tcpstat.tcps_rcvpackafterwin++;
1757 if (todrop >= tlen) {
1758 tcpstat.tcps_rcvbyteafterwin += tlen;
1760 * If a new connection request is received
1761 * while in TIME_WAIT, drop the old connection
1762 * and start over if the sequence numbers
1763 * are above the previous ones.
1765 if (thflags & TH_SYN &&
1766 tp->t_state == TCPS_TIME_WAIT &&
1767 SEQ_GT(th->th_seq, tp->rcv_nxt)) {
1772 * If window is closed can only take segments at
1773 * window edge, and have to drop data and PUSH from
1774 * incoming segments. Continue processing, but
1775 * remember to ack. Otherwise, drop segment
1778 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1779 tp->t_flags |= TF_ACKNOW;
1780 tcpstat.tcps_rcvwinprobe++;
1784 tcpstat.tcps_rcvbyteafterwin += todrop;
1787 thflags &= ~(TH_PUSH | TH_FIN);
1791 * If last ACK falls within this segment's sequence numbers,
1792 * record its timestamp.
1793 * NOTE that the test is modified according to the latest
1794 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1796 if ((to.to_flags & TOF_TS) && SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1797 tp->ts_recent_age = ticks;
1798 tp->ts_recent = to.to_tsval;
1802 * If a SYN is in the window, then this is an
1803 * error and we send an RST and drop the connection.
1805 if (thflags & TH_SYN) {
1806 tp = tcp_drop(tp, ECONNRESET);
1807 rstreason = BANDLIM_UNLIMITED;
1812 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
1813 * flag is on (half-synchronized state), then queue data for
1814 * later processing; else drop segment and return.
1816 if (!(thflags & TH_ACK)) {
1817 if (tp->t_state == TCPS_SYN_RECEIVED ||
1818 (tp->t_flags & TF_NEEDSYN))
1827 switch (tp->t_state) {
1829 * In SYN_RECEIVED state, the ACK acknowledges our SYN, so enter
1830 * ESTABLISHED state and continue processing.
1831 * The ACK was checked above.
1833 case TCPS_SYN_RECEIVED:
1835 tcpstat.tcps_connects++;
1837 /* Do window scaling? */
1838 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
1839 (TF_RCVD_SCALE | TF_REQ_SCALE)) {
1840 tp->snd_scale = tp->requested_s_scale;
1841 tp->rcv_scale = tp->request_r_scale;
1844 * Upon successful completion of 3-way handshake,
1845 * update cache.CC if it was undefined, pass any queued
1846 * data to the user, and advance state appropriately.
1848 if ((taop = tcp_gettaocache(&inp->inp_inc)) != NULL &&
1850 taop->tao_cc = tp->cc_recv;
1854 * SYN-RECEIVED -> ESTABLISHED
1855 * SYN-RECEIVED* -> FIN-WAIT-1
1857 tp->t_starttime = ticks;
1858 if (tp->t_flags & TF_NEEDFIN) {
1859 tp->t_state = TCPS_FIN_WAIT_1;
1860 tp->t_flags &= ~TF_NEEDFIN;
1862 tp->t_state = TCPS_ESTABLISHED;
1863 callout_reset(tp->tt_keep, tcp_keepidle,
1864 tcp_timer_keep, tp);
1867 * If segment contains data or ACK, will call tcp_reass()
1868 * later; if not, do so now to pass queued data to user.
1870 if (tlen == 0 && !(thflags & TH_FIN))
1871 tcp_reass(tp, NULL, NULL, NULL);
1875 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1876 * ACKs. If the ack is in the range
1877 * tp->snd_una < th->th_ack <= tp->snd_max
1878 * then advance tp->snd_una to th->th_ack and drop
1879 * data from the retransmission queue. If this ACK reflects
1880 * more up to date window information we update our window information.
1882 case TCPS_ESTABLISHED:
1883 case TCPS_FIN_WAIT_1:
1884 case TCPS_FIN_WAIT_2:
1885 case TCPS_CLOSE_WAIT:
1888 case TCPS_TIME_WAIT:
1890 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
1891 if (TCP_DO_SACK(tp))
1892 tcp_sack_update_scoreboard(tp, &to);
1893 if (tlen != 0 || tiwin != tp->snd_wnd) {
1897 tcpstat.tcps_rcvdupack++;
1898 if (!callout_active(tp->tt_rexmt) ||
1899 th->th_ack != tp->snd_una) {
1904 * We have outstanding data (other than
1905 * a window probe), this is a completely
1906 * duplicate ack (ie, window info didn't
1907 * change), the ack is the biggest we've
1908 * seen and we've seen exactly our rexmt
1909 * threshhold of them, so assume a packet
1910 * has been dropped and retransmit it.
1911 * Kludge snd_nxt & the congestion
1912 * window so we send only this one
1915 if (IN_FASTRECOVERY(tp)) {
1916 if (TCP_DO_SACK(tp)) {
1917 /* No artifical cwnd inflation. */
1918 tcp_sack_rexmt(tp, th);
1921 * Dup acks mean that packets
1922 * have left the network
1923 * (they're now cached at the
1924 * receiver) so bump cwnd by
1925 * the amount in the receiver
1926 * to keep a constant cwnd
1927 * packets in the network.
1929 tp->snd_cwnd += tp->t_maxseg;
1932 } else if (SEQ_LT(th->th_ack, tp->snd_recover)) {
1935 } else if (++tp->t_dupacks == tcprexmtthresh) {
1936 tcp_seq old_snd_nxt;
1940 if (tcp_do_eifel_detect &&
1941 (tp->t_flags & TF_RCVD_TSTMP)) {
1942 tcp_save_congestion_state(tp);
1943 tp->t_flags |= TF_FASTREXMT;
1946 * We know we're losing at the current
1947 * window size, so do congestion avoidance:
1948 * set ssthresh to half the current window
1949 * and pull our congestion window back to the
1952 win = min(tp->snd_wnd, tp->snd_cwnd) / 2 /
1956 tp->snd_ssthresh = win * tp->t_maxseg;
1957 ENTER_FASTRECOVERY(tp);
1958 tp->snd_recover = tp->snd_max;
1959 callout_stop(tp->tt_rexmt);
1961 old_snd_nxt = tp->snd_nxt;
1962 tp->snd_nxt = th->th_ack;
1963 tp->snd_cwnd = tp->t_maxseg;
1965 ++tcpstat.tcps_sndfastrexmit;
1966 tp->snd_cwnd = tp->snd_ssthresh;
1967 tp->rexmt_high = tp->snd_nxt;
1968 if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
1969 tp->snd_nxt = old_snd_nxt;
1970 KASSERT(tp->snd_limited <= 2,
1971 ("tp->snd_limited too big"));
1972 if (TCP_DO_SACK(tp))
1973 tcp_sack_rexmt(tp, th);
1975 tp->snd_cwnd += tp->t_maxseg *
1976 (tp->t_dupacks - tp->snd_limited);
1977 } else if (tcp_do_limitedtransmit) {
1978 u_long oldcwnd = tp->snd_cwnd;
1979 tcp_seq oldsndmax = tp->snd_max;
1980 tcp_seq oldsndnxt = tp->snd_nxt;
1981 /* outstanding data */
1982 uint32_t ownd = tp->snd_max - tp->snd_una;
1985 #define iceildiv(n, d) (((n)+(d)-1) / (d))
1987 KASSERT(tp->t_dupacks == 1 ||
1989 ("dupacks not 1 or 2"));
1990 if (tp->t_dupacks == 1)
1991 tp->snd_limited = 0;
1992 oldsndnxt = tp->snd_nxt;
1993 tp->snd_nxt = tp->snd_max;
1994 tp->snd_cwnd = ownd +
1995 (tp->t_dupacks - tp->snd_limited) *
1998 if (SEQ_LT(oldsndnxt, oldsndmax))
1999 tp->snd_nxt = oldsndnxt;
2000 tp->snd_cwnd = oldcwnd;
2001 sent = tp->snd_max - oldsndmax;
2002 if (sent > tp->t_maxseg) {
2003 KASSERT((tp->t_dupacks == 2 &&
2004 tp->snd_limited == 0) ||
2005 (sent == tp->t_maxseg + 1 &&
2006 tp->t_flags & TF_SENTFIN),
2008 KASSERT(sent <= tp->t_maxseg * 2,
2009 ("sent too many segments"));
2010 tp->snd_limited = 2;
2011 tcpstat.tcps_sndlimited += 2;
2012 } else if (sent > 0) {
2014 ++tcpstat.tcps_sndlimited;
2015 } else if (tcp_do_early_retransmit &&
2016 (tcp_do_eifel_detect &&
2017 (tp->t_flags & TF_RCVD_TSTMP)) &&
2018 ownd < 4 * tp->t_maxseg &&
2019 tp->t_dupacks + 1 >=
2020 iceildiv(ownd, tp->t_maxseg) &&
2021 (!TCP_DO_SACK(tp) ||
2022 ownd <= tp->t_maxseg ||
2023 tcp_sack_has_sacked(&tp->scb,
2024 ownd - tp->t_maxseg))) {
2025 ++tcpstat.tcps_sndearlyrexmit;
2026 tp->t_flags |= TF_EARLYREXMT;
2027 goto fastretransmit;
2033 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una"));
2035 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2037 * Detected optimistic ACK attack.
2038 * Force slow-start to de-synchronize attack.
2040 tp->snd_cwnd = tp->t_maxseg;
2042 tcpstat.tcps_rcvacktoomuch++;
2046 * If we reach this point, ACK is not a duplicate,
2047 * i.e., it ACKs something we sent.
2049 if (tp->t_flags & TF_NEEDSYN) {
2051 * T/TCP: Connection was half-synchronized, and our
2052 * SYN has been ACK'd (so connection is now fully
2053 * synchronized). Go to non-starred state,
2054 * increment snd_una for ACK of SYN, and check if
2055 * we can do window scaling.
2057 tp->t_flags &= ~TF_NEEDSYN;
2059 /* Do window scaling? */
2060 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
2061 (TF_RCVD_SCALE | TF_REQ_SCALE)) {
2062 tp->snd_scale = tp->requested_s_scale;
2063 tp->rcv_scale = tp->request_r_scale;
2068 acked = th->th_ack - tp->snd_una;
2069 tcpstat.tcps_rcvackpack++;
2070 tcpstat.tcps_rcvackbyte += acked;
2072 if (tcp_do_eifel_detect && acked > 0 &&
2073 (to.to_flags & TOF_TS) && (to.to_tsecr != 0) &&
2074 (tp->t_flags & TF_FIRSTACCACK)) {
2075 /* Eifel detection applicable. */
2076 if (to.to_tsecr < tp->t_rexmtTS) {
2077 ++tcpstat.tcps_eifeldetected;
2078 tcp_revert_congestion_state(tp);
2079 if (tp->t_rxtshift == 1 &&
2080 ticks >= tp->t_badrxtwin)
2081 ++tcpstat.tcps_rttcantdetect;
2083 } else if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) {
2085 * If we just performed our first retransmit,
2086 * and the ACK arrives within our recovery window,
2087 * then it was a mistake to do the retransmit
2088 * in the first place. Recover our original cwnd
2089 * and ssthresh, and proceed to transmit where we
2092 tcp_revert_congestion_state(tp);
2093 ++tcpstat.tcps_rttdetected;
2097 * If we have a timestamp reply, update smoothed
2098 * round trip time. If no timestamp is present but
2099 * transmit timer is running and timed sequence
2100 * number was acked, update smoothed round trip time.
2101 * Since we now have an rtt measurement, cancel the
2102 * timer backoff (cf., Phil Karn's retransmit alg.).
2103 * Recompute the initial retransmit timer.
2105 * Some machines (certain windows boxes) send broken
2106 * timestamp replies during the SYN+ACK phase, ignore
2109 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0))
2110 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1);
2111 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq))
2112 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2113 tcp_xmit_bandwidth_limit(tp, th->th_ack);
2116 * If no data (only SYN) was ACK'd,
2117 * skip rest of ACK processing.
2122 /* Stop looking for an acceptable ACK since one was received. */
2123 tp->t_flags &= ~(TF_FIRSTACCACK | TF_FASTREXMT | TF_EARLYREXMT);
2125 if (acked > so->so_snd.sb_cc) {
2126 tp->snd_wnd -= so->so_snd.sb_cc;
2127 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc);
2128 ourfinisacked = TRUE;
2130 sbdrop(&so->so_snd, acked);
2131 tp->snd_wnd -= acked;
2132 ourfinisacked = FALSE;
2137 * Update window information.
2138 * Don't look at window if no ACK:
2139 * TAC's send garbage on first SYN.
2141 if (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2142 (tp->snd_wl1 == th->th_seq &&
2143 (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2144 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)))) {
2145 /* keep track of pure window updates */
2146 if (tlen == 0 && tp->snd_wl2 == th->th_ack &&
2147 tiwin > tp->snd_wnd)
2148 tcpstat.tcps_rcvwinupd++;
2149 tp->snd_wnd = tiwin;
2150 tp->snd_wl1 = th->th_seq;
2151 tp->snd_wl2 = th->th_ack;
2152 if (tp->snd_wnd > tp->max_sndwnd)
2153 tp->max_sndwnd = tp->snd_wnd;
2157 tp->snd_una = th->th_ack;
2158 if (TCP_DO_SACK(tp))
2159 tcp_sack_update_scoreboard(tp, &to);
2160 if (IN_FASTRECOVERY(tp)) {
2161 if (SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2162 EXIT_FASTRECOVERY(tp);
2165 * If the congestion window was inflated
2166 * to account for the other side's
2167 * cached packets, retract it.
2169 * Window inflation should have left us
2170 * with approximately snd_ssthresh outstanding
2171 * data. But, in case we would be inclined
2172 * to send a burst, better do it using
2175 if (!TCP_DO_SACK(tp))
2176 tp->snd_cwnd = tp->snd_ssthresh;
2178 if (SEQ_GT(th->th_ack + tp->snd_cwnd,
2179 tp->snd_max + 2 * tp->t_maxseg))
2181 (tp->snd_max - tp->snd_una) +
2184 if (TCP_DO_SACK(tp)) {
2185 tp->snd_max_rexmt = tp->snd_max;
2186 tcp_sack_rexmt(tp, th);
2188 tcp_newreno_partial_ack(tp, th, acked);
2194 * When new data is acked, open the congestion window.
2195 * If the window gives us less than ssthresh packets
2196 * in flight, open exponentially (maxseg per packet).
2197 * Otherwise open linearly: maxseg per window
2198 * (maxseg^2 / cwnd per packet).
2200 u_int cw = tp->snd_cwnd;
2203 if (cw > tp->snd_ssthresh)
2204 incr = tp->t_maxseg * tp->t_maxseg / cw;
2206 incr = tp->t_maxseg;
2207 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale);
2208 tp->snd_recover = th->th_ack - 1;
2210 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2211 tp->snd_nxt = tp->snd_una;
2214 * If all outstanding data is acked, stop retransmit
2215 * timer and remember to restart (more output or persist).
2216 * If there is more data to be acked, restart retransmit
2217 * timer, using current (possibly backed-off) value.
2219 if (th->th_ack == tp->snd_max) {
2220 callout_stop(tp->tt_rexmt);
2222 } else if (!callout_active(tp->tt_persist))
2223 callout_reset(tp->tt_rexmt, tp->t_rxtcur,
2224 tcp_timer_rexmt, tp);
2226 switch (tp->t_state) {
2228 * In FIN_WAIT_1 STATE in addition to the processing
2229 * for the ESTABLISHED state if our FIN is now acknowledged
2230 * then enter FIN_WAIT_2.
2232 case TCPS_FIN_WAIT_1:
2233 if (ourfinisacked) {
2235 * If we can't receive any more
2236 * data, then closing user can proceed.
2237 * Starting the timer is contrary to the
2238 * specification, but if we don't get a FIN
2239 * we'll hang forever.
2241 if (so->so_state & SS_CANTRCVMORE) {
2242 soisdisconnected(so);
2243 callout_reset(tp->tt_2msl, tcp_maxidle,
2244 tcp_timer_2msl, tp);
2246 tp->t_state = TCPS_FIN_WAIT_2;
2251 * In CLOSING STATE in addition to the processing for
2252 * the ESTABLISHED state if the ACK acknowledges our FIN
2253 * then enter the TIME-WAIT state, otherwise ignore
2257 if (ourfinisacked) {
2258 tp->t_state = TCPS_TIME_WAIT;
2259 tcp_canceltimers(tp);
2260 /* Shorten TIME_WAIT [RFC-1644, p.28] */
2261 if (tp->cc_recv != 0 &&
2262 (ticks - tp->t_starttime) < tcp_msl)
2263 callout_reset(tp->tt_2msl,
2264 tp->t_rxtcur * TCPTV_TWTRUNC,
2265 tcp_timer_2msl, tp);
2267 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2268 tcp_timer_2msl, tp);
2269 soisdisconnected(so);
2274 * In LAST_ACK, we may still be waiting for data to drain
2275 * and/or to be acked, as well as for the ack of our FIN.
2276 * If our FIN is now acknowledged, delete the TCB,
2277 * enter the closed state and return.
2280 if (ourfinisacked) {
2287 * In TIME_WAIT state the only thing that should arrive
2288 * is a retransmission of the remote FIN. Acknowledge
2289 * it and restart the finack timer.
2291 case TCPS_TIME_WAIT:
2292 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2293 tcp_timer_2msl, tp);
2300 * Update window information.
2301 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2303 if ((thflags & TH_ACK) &&
2304 acceptable_window_update(tp, th, tiwin)) {
2305 /* keep track of pure window updates */
2306 if (tlen == 0 && tp->snd_wl2 == th->th_ack &&
2307 tiwin > tp->snd_wnd)
2308 tcpstat.tcps_rcvwinupd++;
2309 tp->snd_wnd = tiwin;
2310 tp->snd_wl1 = th->th_seq;
2311 tp->snd_wl2 = th->th_ack;
2312 if (tp->snd_wnd > tp->max_sndwnd)
2313 tp->max_sndwnd = tp->snd_wnd;
2318 * Process segments with URG.
2320 if ((thflags & TH_URG) && th->th_urp &&
2321 !TCPS_HAVERCVDFIN(tp->t_state)) {
2323 * This is a kludge, but if we receive and accept
2324 * random urgent pointers, we'll crash in
2325 * soreceive. It's hard to imagine someone
2326 * actually wanting to send this much urgent data.
2328 if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2329 th->th_urp = 0; /* XXX */
2330 thflags &= ~TH_URG; /* XXX */
2331 goto dodata; /* XXX */
2334 * If this segment advances the known urgent pointer,
2335 * then mark the data stream. This should not happen
2336 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2337 * a FIN has been received from the remote side.
2338 * In these states we ignore the URG.
2340 * According to RFC961 (Assigned Protocols),
2341 * the urgent pointer points to the last octet
2342 * of urgent data. We continue, however,
2343 * to consider it to indicate the first octet
2344 * of data past the urgent section as the original
2345 * spec states (in one of two places).
2347 if (SEQ_GT(th->th_seq + th->th_urp, tp->rcv_up)) {
2348 tp->rcv_up = th->th_seq + th->th_urp;
2349 so->so_oobmark = so->so_rcv.sb_cc +
2350 (tp->rcv_up - tp->rcv_nxt) - 1;
2351 if (so->so_oobmark == 0)
2352 so->so_state |= SS_RCVATMARK;
2354 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2357 * Remove out of band data so doesn't get presented to user.
2358 * This can happen independent of advancing the URG pointer,
2359 * but if two URG's are pending at once, some out-of-band
2360 * data may creep in... ick.
2362 if (th->th_urp <= (u_long)tlen &&
2363 !(so->so_options & SO_OOBINLINE)) {
2364 /* hdr drop is delayed */
2365 tcp_pulloutofband(so, th, m, drop_hdrlen);
2369 * If no out of band data is expected,
2370 * pull receive urgent pointer along
2371 * with the receive window.
2373 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2374 tp->rcv_up = tp->rcv_nxt;
2379 * Process the segment text, merging it into the TCP sequencing queue,
2380 * and arranging for acknowledgment of receipt if necessary.
2381 * This process logically involves adjusting tp->rcv_wnd as data
2382 * is presented to the user (this happens in tcp_usrreq.c,
2383 * case PRU_RCVD). If a FIN has already been received on this
2384 * connection then we just ignore the text.
2386 if ((tlen || (thflags & TH_FIN)) && !TCPS_HAVERCVDFIN(tp->t_state)) {
2387 m_adj(m, drop_hdrlen); /* delayed header drop */
2389 * Insert segment which includes th into TCP reassembly queue
2390 * with control block tp. Set thflags to whether reassembly now
2391 * includes a segment with FIN. This handles the common case
2392 * inline (segment is the next to be received on an established
2393 * connection, and the queue is empty), avoiding linkage into
2394 * and removal from the queue and repetition of various
2396 * Set DELACK for segments received in order, but ack
2397 * immediately when segments are out of order (so
2398 * fast retransmit can work).
2400 if (th->th_seq == tp->rcv_nxt &&
2401 LIST_EMPTY(&tp->t_segq) &&
2402 TCPS_HAVEESTABLISHED(tp->t_state)) {
2404 callout_reset(tp->tt_delack, tcp_delacktime,
2405 tcp_timer_delack, tp);
2407 tp->t_flags |= TF_ACKNOW;
2408 tp->rcv_nxt += tlen;
2409 thflags = th->th_flags & TH_FIN;
2410 tcpstat.tcps_rcvpack++;
2411 tcpstat.tcps_rcvbyte += tlen;
2413 if (so->so_state & SS_CANTRCVMORE)
2416 sbappendstream(&so->so_rcv, m);
2419 if (!(tp->t_flags & TF_DUPSEG)) {
2420 /* Initialize SACK report block. */
2421 tp->reportblk.rblk_start = th->th_seq;
2422 tp->reportblk.rblk_end = th->th_seq + tlen +
2423 ((thflags & TH_FIN) != 0);
2425 thflags = tcp_reass(tp, th, &tlen, m);
2426 tp->t_flags |= TF_ACKNOW;
2430 * Note the amount of data that peer has sent into
2431 * our window, in order to estimate the sender's
2434 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2441 * If FIN is received ACK the FIN and let the user know
2442 * that the connection is closing.
2444 if (thflags & TH_FIN) {
2445 if (!TCPS_HAVERCVDFIN(tp->t_state)) {
2448 * If connection is half-synchronized
2449 * (ie NEEDSYN flag on) then delay ACK,
2450 * so it may be piggybacked when SYN is sent.
2451 * Otherwise, since we received a FIN then no
2452 * more input can be expected, send ACK now.
2454 if (DELAY_ACK(tp) && (tp->t_flags & TF_NEEDSYN))
2455 callout_reset(tp->tt_delack, tcp_delacktime,
2456 tcp_timer_delack, tp);
2458 tp->t_flags |= TF_ACKNOW;
2462 switch (tp->t_state) {
2464 * In SYN_RECEIVED and ESTABLISHED STATES
2465 * enter the CLOSE_WAIT state.
2467 case TCPS_SYN_RECEIVED:
2468 tp->t_starttime = ticks;
2470 case TCPS_ESTABLISHED:
2471 tp->t_state = TCPS_CLOSE_WAIT;
2475 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2476 * enter the CLOSING state.
2478 case TCPS_FIN_WAIT_1:
2479 tp->t_state = TCPS_CLOSING;
2483 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2484 * starting the time-wait timer, turning off the other
2487 case TCPS_FIN_WAIT_2:
2488 tp->t_state = TCPS_TIME_WAIT;
2489 tcp_canceltimers(tp);
2490 /* Shorten TIME_WAIT [RFC-1644, p.28] */
2491 if (tp->cc_recv != 0 &&
2492 (ticks - tp->t_starttime) < tcp_msl) {
2493 callout_reset(tp->tt_2msl,
2494 tp->t_rxtcur * TCPTV_TWTRUNC,
2495 tcp_timer_2msl, tp);
2496 /* For transaction client, force ACK now. */
2497 tp->t_flags |= TF_ACKNOW;
2500 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2501 tcp_timer_2msl, tp);
2502 soisdisconnected(so);
2506 * In TIME_WAIT state restart the 2 MSL time_wait timer.
2508 case TCPS_TIME_WAIT:
2509 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2510 tcp_timer_2msl, 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))
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;
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);
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);
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);
2666 if (optlen != TCPOLEN_CC)
2668 to->to_flags |= TOF_CC;
2669 bcopy(cp + 2, &to->to_cc, sizeof to->to_cc);
2670 to->to_cc = ntohl(to->to_cc);
2673 if (optlen != TCPOLEN_CC)
2677 to->to_flags |= TOF_CCNEW;
2678 bcopy(cp + 2, &to->to_cc, sizeof to->to_cc);
2679 to->to_cc = ntohl(to->to_cc);
2682 if (optlen != TCPOLEN_CC)
2686 to->to_flags |= TOF_CCECHO;
2687 bcopy(cp + 2, &to->to_ccecho, sizeof to->to_ccecho);
2688 to->to_ccecho = ntohl(to->to_ccecho);
2690 case TCPOPT_SACK_PERMITTED:
2691 if (optlen != TCPOLEN_SACK_PERMITTED)
2695 to->to_flags |= TOF_SACK_PERMITTED;
2698 if ((optlen - 2) & 0x07) /* not multiple of 8 */
2700 to->to_nsackblocks = (optlen - 2) / 8;
2701 to->to_sackblocks = (struct raw_sackblock *) (cp + 2);
2702 to->to_flags |= TOF_SACK;
2703 for (i = 0; i < to->to_nsackblocks; i++) {
2704 struct raw_sackblock *r = &to->to_sackblocks[i];
2706 r->rblk_start = ntohl(r->rblk_start);
2707 r->rblk_end = ntohl(r->rblk_end);
2717 * Pull out of band byte out of a segment so
2718 * it doesn't appear in the user's data queue.
2719 * It is still reflected in the segment length for
2720 * sequencing purposes.
2721 * "off" is the delayed to be dropped hdrlen.
2724 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, int off)
2726 int cnt = off + th->th_urp - 1;
2729 if (m->m_len > cnt) {
2730 char *cp = mtod(m, caddr_t) + cnt;
2731 struct tcpcb *tp = sototcpcb(so);
2734 tp->t_oobflags |= TCPOOB_HAVEDATA;
2735 bcopy(cp + 1, cp, m->m_len - cnt - 1);
2737 if (m->m_flags & M_PKTHDR)
2746 panic("tcp_pulloutofband");
2750 * Collect new round-trip time estimate
2751 * and update averages and current timeout.
2754 tcp_xmit_timer(struct tcpcb *tp, int rtt)
2758 tcpstat.tcps_rttupdated++;
2760 if (tp->t_srtt != 0) {
2762 * srtt is stored as fixed point with 5 bits after the
2763 * binary point (i.e., scaled by 8). The following magic
2764 * is equivalent to the smoothing algorithm in rfc793 with
2765 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2766 * point). Adjust rtt to origin 0.
2768 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
2769 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
2771 if ((tp->t_srtt += delta) <= 0)
2775 * We accumulate a smoothed rtt variance (actually, a
2776 * smoothed mean difference), then set the retransmit
2777 * timer to smoothed rtt + 4 times the smoothed variance.
2778 * rttvar is stored as fixed point with 4 bits after the
2779 * binary point (scaled by 16). The following is
2780 * equivalent to rfc793 smoothing with an alpha of .75
2781 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
2782 * rfc793's wired-in beta.
2786 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
2787 if ((tp->t_rttvar += delta) <= 0)
2789 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
2790 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2793 * No rtt measurement yet - use the unsmoothed rtt.
2794 * Set the variance to half the rtt (so our first
2795 * retransmit happens at 3*rtt).
2797 tp->t_srtt = rtt << TCP_RTT_SHIFT;
2798 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
2799 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2805 * the retransmit should happen at rtt + 4 * rttvar.
2806 * Because of the way we do the smoothing, srtt and rttvar
2807 * will each average +1/2 tick of bias. When we compute
2808 * the retransmit timer, we want 1/2 tick of rounding and
2809 * 1 extra tick because of +-1/2 tick uncertainty in the
2810 * firing of the timer. The bias will give us exactly the
2811 * 1.5 tick we need. But, because the bias is
2812 * statistical, we have to test that we don't drop below
2813 * the minimum feasible timer (which is 2 ticks).
2815 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
2816 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
2819 * We received an ack for a packet that wasn't retransmitted;
2820 * it is probably safe to discard any error indications we've
2821 * received recently. This isn't quite right, but close enough
2822 * for now (a route might have failed after we sent a segment,
2823 * and the return path might not be symmetrical).
2825 tp->t_softerror = 0;
2829 * Determine a reasonable value for maxseg size.
2830 * If the route is known, check route for mtu.
2831 * If none, use an mss that can be handled on the outgoing
2832 * interface without forcing IP to fragment; if bigger than
2833 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2834 * to utilize large mbufs. If no route is found, route has no mtu,
2835 * or the destination isn't local, use a default, hopefully conservative
2836 * size (usually 512 or the default IP max size, but no more than the mtu
2837 * of the interface), as we can't discover anything about intervening
2838 * gateways or networks. We also initialize the congestion/slow start
2839 * window to be a single segment if the destination isn't local.
2840 * While looking at the routing entry, we also initialize other path-dependent
2841 * parameters from pre-set or cached values in the routing entry.
2843 * Also take into account the space needed for options that we
2844 * send regularly. Make maxseg shorter by that amount to assure
2845 * that we can send maxseg amount of data even when the options
2846 * are present. Store the upper limit of the length of options plus
2849 * NOTE that this routine is only called when we process an incoming
2850 * segment, for outgoing segments only tcp_mssopt is called.
2852 * In case of T/TCP, we call this routine during implicit connection
2853 * setup as well (offer = -1), to initialize maxseg from the cached
2857 tcp_mss(struct tcpcb *tp, int offer)
2863 struct inpcb *inp = tp->t_inpcb;
2865 struct rmxp_tao *taop;
2866 int origoffer = offer;
2868 boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) ? TRUE : FALSE);
2869 size_t min_protoh = isipv6 ?
2870 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
2871 sizeof(struct tcpiphdr);
2873 const boolean_t isipv6 = FALSE;
2874 const size_t min_protoh = sizeof(struct tcpiphdr);
2878 rt = tcp_rtlookup6(&inp->inp_inc);
2880 rt = tcp_rtlookup(&inp->inp_inc);
2882 tp->t_maxopd = tp->t_maxseg =
2883 (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
2887 so = inp->inp_socket;
2889 taop = rmx_taop(rt->rt_rmx);
2891 * Offer == -1 means that we didn't receive SYN yet,
2892 * use cached value in that case;
2895 offer = taop->tao_mssopt;
2897 * Offer == 0 means that there was no MSS on the SYN segment,
2898 * in this case we use tcp_mssdflt.
2901 offer = (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
2904 * Sanity check: make sure that maxopd will be large
2905 * enough to allow some data on segments even is the
2906 * all the option space is used (40bytes). Otherwise
2907 * funny things may happen in tcp_output.
2909 offer = max(offer, 64);
2910 taop->tao_mssopt = offer;
2913 * While we're here, check if there's an initial rtt
2914 * or rttvar. Convert from the route-table units
2915 * to scaled multiples of the slow timeout timer.
2917 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
2919 * XXX the lock bit for RTT indicates that the value
2920 * is also a minimum value; this is subject to time.
2922 if (rt->rt_rmx.rmx_locks & RTV_RTT)
2923 tp->t_rttmin = rtt / (RTM_RTTUNIT / hz);
2924 tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE));
2925 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
2926 tcpstat.tcps_usedrtt++;
2927 if (rt->rt_rmx.rmx_rttvar) {
2928 tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
2929 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE));
2930 tcpstat.tcps_usedrttvar++;
2932 /* default variation is +- 1 rtt */
2934 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
2936 TCPT_RANGESET(tp->t_rxtcur,
2937 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
2938 tp->t_rttmin, TCPTV_REXMTMAX);
2941 * if there's an mtu associated with the route, use it
2942 * else, use the link mtu.
2944 if (rt->rt_rmx.rmx_mtu)
2945 mss = rt->rt_rmx.rmx_mtu - min_protoh;
2948 mss = ND_IFINFO(rt->rt_ifp)->linkmtu - min_protoh;
2949 if (!in6_localaddr(&inp->in6p_faddr))
2950 mss = min(mss, tcp_v6mssdflt);
2952 mss = ifp->if_mtu - min_protoh;
2953 if (!in_localaddr(inp->inp_faddr))
2954 mss = min(mss, tcp_mssdflt);
2957 mss = min(mss, offer);
2959 * maxopd stores the maximum length of data AND options
2960 * in a segment; maxseg is the amount of data in a normal
2961 * segment. We need to store this value (maxopd) apart
2962 * from maxseg, because now every segment carries options
2963 * and thus we normally have somewhat less data in segments.
2968 * In case of T/TCP, origoffer==-1 indicates, that no segments
2969 * were received yet. In this case we just guess, otherwise
2970 * we do the same as before T/TCP.
2972 if ((tp->t_flags & (TF_REQ_TSTMP | TF_NOOPT)) == TF_REQ_TSTMP &&
2974 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
2975 mss -= TCPOLEN_TSTAMP_APPA;
2976 if ((tp->t_flags & (TF_REQ_CC | TF_NOOPT)) == TF_REQ_CC &&
2978 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC))
2979 mss -= TCPOLEN_CC_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.sb_hiwat;
3001 bufsize = roundup(bufsize, mss);
3002 if (bufsize > sb_max)
3004 if (bufsize > so->so_snd.sb_hiwat)
3005 sbreserve(&so->so_snd, bufsize, so, NULL);
3010 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0)
3012 bufsize = so->so_rcv.sb_hiwat;
3013 if (bufsize > mss) {
3014 bufsize = roundup(bufsize, mss);
3015 if (bufsize > sb_max)
3017 if (bufsize > so->so_rcv.sb_hiwat)
3018 sbreserve(&so->so_rcv, bufsize, so, NULL);
3022 * Set the slow-start flight size depending on whether this
3023 * is a local network or not.
3026 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380));
3030 if (rt->rt_rmx.rmx_ssthresh) {
3032 * There's some sort of gateway or interface
3033 * buffer limit on the path. Use this to set
3034 * the slow start threshhold, but set the
3035 * threshold to no less than 2*mss.
3037 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
3038 tcpstat.tcps_usedssthresh++;
3043 * Determine the MSS option to send on an outgoing SYN.
3046 tcp_mssopt(struct tcpcb *tp)
3051 ((tp->t_inpcb->inp_vflag & INP_IPV6) ? TRUE : FALSE);
3052 int min_protoh = isipv6 ?
3053 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
3054 sizeof(struct tcpiphdr);
3056 const boolean_t isipv6 = FALSE;
3057 const size_t min_protoh = sizeof(struct tcpiphdr);
3061 rt = tcp_rtlookup6(&tp->t_inpcb->inp_inc);
3063 rt = tcp_rtlookup(&tp->t_inpcb->inp_inc);
3065 return (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
3067 return (rt->rt_ifp->if_mtu - min_protoh);
3071 * When a partial ack arrives, force the retransmission of the
3072 * next unacknowledged segment. Do not exit Fast Recovery.
3074 * Implement the Slow-but-Steady variant of NewReno by restarting the
3075 * the retransmission timer. Turn it off here so it can be restarted
3076 * later in tcp_output().
3079 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th, int acked)
3081 tcp_seq old_snd_nxt = tp->snd_nxt;
3082 u_long ocwnd = tp->snd_cwnd;
3084 callout_stop(tp->tt_rexmt);
3086 tp->snd_nxt = th->th_ack;
3087 /* Set snd_cwnd to one segment beyond acknowledged offset. */
3088 tp->snd_cwnd = tp->t_maxseg;
3089 tp->t_flags |= TF_ACKNOW;
3091 if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
3092 tp->snd_nxt = old_snd_nxt;
3093 /* partial window deflation */
3094 tp->snd_cwnd = ocwnd - acked + tp->t_maxseg;
3098 * In contrast to the Slow-but-Steady NewReno variant,
3099 * we do not reset the retransmission timer for SACK retransmissions,
3100 * except when retransmitting snd_una.
3103 tcp_sack_rexmt(struct tcpcb *tp, struct tcphdr *th)
3105 uint32_t pipe, seglen;
3108 tcp_seq old_snd_nxt = tp->snd_nxt;
3109 u_long ocwnd = tp->snd_cwnd;
3110 int nseg = 0; /* consecutive new segments */
3111 #define MAXBURST 4 /* limit burst of new packets on partial ack */
3114 pipe = tcp_sack_compute_pipe(tp);
3115 while ((tcp_seq_diff_t)(ocwnd - pipe) >= (tcp_seq_diff_t)tp->t_maxseg &&
3116 (!tcp_do_smartsack || nseg < MAXBURST) &&
3117 tcp_sack_nextseg(tp, &nextrexmt, &seglen, &lostdup)) {
3119 tcp_seq old_snd_max;
3122 if (nextrexmt == tp->snd_max) ++nseg;
3123 tp->snd_nxt = nextrexmt;
3124 tp->snd_cwnd = nextrexmt - tp->snd_una + seglen;
3125 old_snd_max = tp->snd_max;
3126 if (nextrexmt == tp->snd_una)
3127 callout_stop(tp->tt_rexmt);
3128 error = tcp_output(tp);
3131 sent = tp->snd_nxt - nextrexmt;
3136 tcpstat.tcps_sndsackpack++;
3137 tcpstat.tcps_sndsackbyte += sent;
3138 if (SEQ_LT(nextrexmt, old_snd_max) &&
3139 SEQ_LT(tp->rexmt_high, tp->snd_nxt))
3140 tp->rexmt_high = seq_min(tp->snd_nxt, old_snd_max);
3142 if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
3143 tp->snd_nxt = old_snd_nxt;
3144 tp->snd_cwnd = ocwnd;
tuxillo's projects / dragonfly.git / blob