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 1.52 2005/02/08 22:56:19 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_do_rfc3390 = 1;
183 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW,
185 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
187 static int tcp_do_eifel_detect = 1;
188 SYSCTL_INT(_net_inet_tcp, OID_AUTO, eifel, CTLFLAG_RW,
189 &tcp_do_eifel_detect, 0, "Eifel detection algorithm (RFC 3522)");
192 * Define as tunable for easy testing with SACK on and off.
193 * Warning: do not change setting in the middle of an existing active TCP flow,
194 * else strange things might happen to that flow.
197 SYSCTL_INT(_net_inet_tcp, OID_AUTO, sack, CTLFLAG_RW,
198 &tcp_do_sack, 0, "Enable SACK Algorithms");
200 int tcp_do_smartsack = 1;
201 SYSCTL_INT(_net_inet_tcp, OID_AUTO, smartsack, CTLFLAG_RW,
202 &tcp_do_smartsack, 0, "Enable Smart SACK Algorithms");
204 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0,
205 "TCP Segment Reassembly Queue");
207 int tcp_reass_maxseg = 0;
208 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RD,
209 &tcp_reass_maxseg, 0,
210 "Global maximum number of TCP Segments in Reassembly Queue");
212 int tcp_reass_qsize = 0;
213 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD,
215 "Global number of TCP Segments currently in Reassembly Queue");
217 static int tcp_reass_overflows = 0;
218 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD,
219 &tcp_reass_overflows, 0,
220 "Global number of TCP Segment Reassembly Queue Overflows");
222 static void tcp_dooptions(struct tcpopt *, u_char *, int, boolean_t);
223 static void tcp_pulloutofband(struct socket *,
224 struct tcphdr *, struct mbuf *, int);
225 static int tcp_reass(struct tcpcb *, struct tcphdr *, int *,
227 static void tcp_xmit_timer(struct tcpcb *, int);
228 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *, int);
229 static void tcp_sack_rexmt(struct tcpcb *, struct tcphdr *);
231 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
233 #define ND6_HINT(tp) \
235 if ((tp) && (tp)->t_inpcb && \
236 ((tp)->t_inpcb->inp_vflag & INP_IPV6) && \
237 (tp)->t_inpcb->in6p_route.ro_rt) \
238 nd6_nud_hint((tp)->t_inpcb->in6p_route.ro_rt, NULL, 0); \
245 * Indicate whether this ack should be delayed. We can delay the ack if
246 * - delayed acks are enabled and
247 * - there is no delayed ack timer in progress and
248 * - our last ack wasn't a 0-sized window. We never want to delay
249 * the ack that opens up a 0-sized window.
251 #define DELAY_ACK(tp) \
252 (tcp_delack_enabled && !callout_pending(tp->tt_delack) && \
253 !(tp->t_flags & TF_RXWIN0SENT))
255 #define acceptable_window_update(tp, th, tiwin) \
256 (SEQ_LT(tp->snd_wl1, th->th_seq) || \
257 (tp->snd_wl1 == th->th_seq && \
258 (SEQ_LT(tp->snd_wl2, th->th_ack) || \
259 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))
262 tcp_reass(struct tcpcb *tp, struct tcphdr *th, int *tlenp, struct mbuf *m)
265 struct tseg_qent *p = NULL;
266 struct tseg_qent *te;
267 struct socket *so = tp->t_inpcb->inp_socket;
271 * Call with th == NULL after become established to
272 * force pre-ESTABLISHED data up to user socket.
278 * Limit the number of segments in the reassembly queue to prevent
279 * holding on to too many segments (and thus running out of mbufs).
280 * Make sure to let the missing segment through which caused this
281 * queue. Always keep one global queue entry spare to be able to
282 * process the missing segment.
284 if (th->th_seq != tp->rcv_nxt &&
285 tcp_reass_qsize + 1 >= tcp_reass_maxseg) {
286 tcp_reass_overflows++;
287 tcpstat.tcps_rcvmemdrop++;
289 /* no SACK block to report */
290 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
294 /* Allocate a new queue entry. */
295 MALLOC(te, struct tseg_qent *, sizeof(struct tseg_qent), M_TSEGQ,
296 M_INTWAIT | M_NULLOK);
298 tcpstat.tcps_rcvmemdrop++;
300 /* no SACK block to report */
301 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
307 * Find a segment which begins after this one does.
309 LIST_FOREACH(q, &tp->t_segq, tqe_q) {
310 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
316 * If there is a preceding segment, it may provide some of
317 * our data already. If so, drop the data from the incoming
318 * segment. If it provides all of our data, drop us.
323 /* conversion to int (in i) handles seq wraparound */
324 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
325 if (i > 0) { /* overlaps preceding segment */
326 tp->t_flags |= (TF_DUPSEG | TF_ENCLOSESEG);
327 /* enclosing block starts w/ preceding segment */
328 tp->encloseblk.rblk_start = p->tqe_th->th_seq;
330 /* preceding encloses incoming segment */
331 tp->encloseblk.rblk_end = p->tqe_th->th_seq +
333 tcpstat.tcps_rcvduppack++;
334 tcpstat.tcps_rcvdupbyte += *tlenp;
339 * Try to present any queued data
340 * at the left window edge to the user.
341 * This is needed after the 3-WHS
344 goto present; /* ??? */
349 /* incoming segment end is enclosing block end */
350 tp->encloseblk.rblk_end = th->th_seq + *tlenp +
351 ((th->th_flags & TH_FIN) != 0);
352 /* trim end of reported D-SACK block */
353 tp->reportblk.rblk_end = th->th_seq;
356 tcpstat.tcps_rcvoopack++;
357 tcpstat.tcps_rcvoobyte += *tlenp;
360 * While we overlap succeeding segments trim them or,
361 * if they are completely covered, dequeue them.
364 int i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
365 tcp_seq qend = q->tqe_th->th_seq + q->tqe_len;
366 struct tseg_qent *nq;
370 if (!(tp->t_flags & TF_DUPSEG)) { /* first time through */
371 tp->t_flags |= (TF_DUPSEG | TF_ENCLOSESEG);
372 tp->encloseblk = tp->reportblk;
373 /* report trailing duplicate D-SACK segment */
374 tp->reportblk.rblk_start = q->tqe_th->th_seq;
376 if ((tp->t_flags & TF_ENCLOSESEG) &&
377 SEQ_GT(qend, tp->encloseblk.rblk_end)) {
378 /* extend enclosing block if one exists */
379 tp->encloseblk.rblk_end = qend;
381 if (i < q->tqe_len) {
382 q->tqe_th->th_seq += i;
388 nq = LIST_NEXT(q, tqe_q);
389 LIST_REMOVE(q, tqe_q);
396 /* Insert the new segment queue entry into place. */
399 te->tqe_len = *tlenp;
401 /* check if can coalesce with following segment */
402 if (q != NULL && (th->th_seq + *tlenp == q->tqe_th->th_seq)) {
403 tcp_seq tend = te->tqe_th->th_seq + te->tqe_len;
405 te->tqe_len += q->tqe_len;
406 if (q->tqe_th->th_flags & TH_FIN)
407 te->tqe_th->th_flags |= TH_FIN;
408 m_cat(te->tqe_m, q->tqe_m);
409 tp->encloseblk.rblk_end = tend;
411 * When not reporting a duplicate segment, use
412 * the larger enclosing block as the SACK block.
414 if (!(tp->t_flags & TF_DUPSEG))
415 tp->reportblk.rblk_end = tend;
416 LIST_REMOVE(q, tqe_q);
422 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
424 /* check if can coalesce with preceding segment */
425 if (p->tqe_th->th_seq + p->tqe_len == th->th_seq) {
426 p->tqe_len += te->tqe_len;
427 m_cat(p->tqe_m, te->tqe_m);
428 tp->encloseblk.rblk_start = p->tqe_th->th_seq;
430 * When not reporting a duplicate segment, use
431 * the larger enclosing block as the SACK block.
433 if (!(tp->t_flags & TF_DUPSEG))
434 tp->reportblk.rblk_start = p->tqe_th->th_seq;
438 LIST_INSERT_AFTER(p, te, tqe_q);
443 * Present data to user, advancing rcv_nxt through
444 * completed sequence space.
446 if (!TCPS_HAVEESTABLISHED(tp->t_state))
448 q = LIST_FIRST(&tp->t_segq);
449 if (q == NULL || q->tqe_th->th_seq != tp->rcv_nxt)
451 tp->rcv_nxt += q->tqe_len;
452 if (!(tp->t_flags & TF_DUPSEG)) {
453 /* no SACK block to report since ACK advanced */
454 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
456 /* no enclosing block to report since ACK advanced */
457 tp->t_flags &= ~TF_ENCLOSESEG;
458 flags = q->tqe_th->th_flags & TH_FIN;
459 LIST_REMOVE(q, tqe_q);
460 KASSERT(LIST_EMPTY(&tp->t_segq) ||
461 LIST_FIRST(&tp->t_segq)->tqe_th->th_seq != tp->rcv_nxt,
462 ("segment not coalesced"));
463 if (so->so_state & SS_CANTRCVMORE)
466 sbappendstream(&so->so_rcv, q->tqe_m);
475 * TCP input routine, follows pages 65-76 of the
476 * protocol specification dated September, 1981 very closely.
480 tcp6_input(struct mbuf **mp, int *offp, int proto)
482 struct mbuf *m = *mp;
483 struct in6_ifaddr *ia6;
485 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
488 * draft-itojun-ipv6-tcp-to-anycast
489 * better place to put this in?
491 ia6 = ip6_getdstifaddr(m);
492 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
495 ip6 = mtod(m, struct ip6_hdr *);
496 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
497 offsetof(struct ip6_hdr, ip6_dst));
498 return (IPPROTO_DONE);
501 tcp_input(m, *offp, proto);
502 return (IPPROTO_DONE);
507 tcp_input(struct mbuf *m, ...)
512 struct ip *ip = NULL;
514 struct inpcb *inp = NULL;
519 struct tcpcb *tp = NULL;
521 struct socket *so = 0;
523 boolean_t ourfinisacked, needoutput = FALSE;
526 struct tcpopt to; /* options in this segment */
527 struct rmxp_tao *taop; /* pointer to our TAO cache entry */
528 struct rmxp_tao tao_noncached; /* in case there's no cached entry */
529 struct sockaddr_in *next_hop = NULL;
530 int rstreason; /* For badport_bandlim accounting purposes */
532 struct ip6_hdr *ip6 = NULL;
536 const boolean_t isipv6 = FALSE;
543 off0 = __va_arg(ap, int);
544 proto = __va_arg(ap, int);
547 tcpstat.tcps_rcvtotal++;
549 /* Grab info from and strip MT_TAG mbufs prepended to the chain. */
550 while (m->m_type == MT_TAG) {
551 if (m->_m_tag_id == PACKET_TAG_IPFORWARD)
552 next_hop = (struct sockaddr_in *)m->m_hdr.mh_data;
557 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? TRUE : FALSE;
561 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */
562 ip6 = mtod(m, struct ip6_hdr *);
563 tlen = (sizeof *ip6) + ntohs(ip6->ip6_plen) - off0;
564 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) {
565 tcpstat.tcps_rcvbadsum++;
568 th = (struct tcphdr *)((caddr_t)ip6 + off0);
571 * Be proactive about unspecified IPv6 address in source.
572 * As we use all-zero to indicate unbounded/unconnected pcb,
573 * unspecified IPv6 address can be used to confuse us.
575 * Note that packets with unspecified IPv6 destination is
576 * already dropped in ip6_input.
578 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
584 * Get IP and TCP header together in first mbuf.
585 * Note: IP leaves IP header in first mbuf.
587 if (off0 > sizeof(struct ip)) {
589 off0 = sizeof(struct ip);
591 /* already checked and pulled up in ip_demux() */
592 KASSERT(m->m_len >= sizeof(struct tcpiphdr),
593 ("TCP header not in one mbuf: m->m_len %d", m->m_len));
594 ip = mtod(m, struct ip *);
595 ipov = (struct ipovly *)ip;
596 th = (struct tcphdr *)((caddr_t)ip + off0);
599 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
600 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
601 th->th_sum = m->m_pkthdr.csum_data;
603 th->th_sum = in_pseudo(ip->ip_src.s_addr,
605 htonl(m->m_pkthdr.csum_data +
608 th->th_sum ^= 0xffff;
611 * Checksum extended TCP header and data.
613 len = sizeof(struct ip) + tlen;
614 bzero(ipov->ih_x1, sizeof ipov->ih_x1);
615 ipov->ih_len = (u_short)tlen;
616 ipov->ih_len = htons(ipov->ih_len);
617 th->th_sum = in_cksum(m, len);
620 tcpstat.tcps_rcvbadsum++;
624 /* Re-initialization for later version check */
625 ip->ip_v = IPVERSION;
630 * Check that TCP offset makes sense,
631 * pull out TCP options and adjust length. XXX
633 off = th->th_off << 2;
634 /* already checked and pulled up in ip_demux() */
635 KASSERT(off >= sizeof(struct tcphdr) && off <= tlen,
636 ("bad TCP data offset %d (tlen %d)", off, tlen));
637 tlen -= off; /* tlen is used instead of ti->ti_len */
638 if (off > sizeof(struct tcphdr)) {
640 IP6_EXTHDR_CHECK(m, off0, off, );
641 ip6 = mtod(m, struct ip6_hdr *);
642 th = (struct tcphdr *)((caddr_t)ip6 + off0);
644 /* already pulled up in ip_demux() */
645 KASSERT(m->m_len >= sizeof(struct ip) + off,
646 ("TCP header and options not in one mbuf: "
647 "m_len %d, off %d", m->m_len, off));
649 optlen = off - sizeof(struct tcphdr);
650 optp = (u_char *)(th + 1);
652 thflags = th->th_flags;
654 #ifdef TCP_DROP_SYNFIN
656 * If the drop_synfin option is enabled, drop all packets with
657 * both the SYN and FIN bits set. This prevents e.g. nmap from
658 * identifying the TCP/IP stack.
660 * This is a violation of the TCP specification.
662 if (drop_synfin && (thflags & (TH_SYN | TH_FIN)) == (TH_SYN | TH_FIN))
667 * Convert TCP protocol specific fields to host format.
669 th->th_seq = ntohl(th->th_seq);
670 th->th_ack = ntohl(th->th_ack);
671 th->th_win = ntohs(th->th_win);
672 th->th_urp = ntohs(th->th_urp);
675 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options,
676 * until after ip6_savecontrol() is called and before other functions
677 * which don't want those proto headers.
678 * Because ip6_savecontrol() is going to parse the mbuf to
679 * search for data to be passed up to user-land, it wants mbuf
680 * parameters to be unchanged.
681 * XXX: the call of ip6_savecontrol() has been obsoleted based on
682 * latest version of the advanced API (20020110).
684 drop_hdrlen = off0 + off;
687 * Locate pcb for segment.
690 /* IPFIREWALL_FORWARD section */
691 if (next_hop != NULL && !isipv6) { /* IPv6 support is not there yet */
693 * Transparently forwarded. Pretend to be the destination.
694 * already got one like this?
696 cpu = mycpu->gd_cpuid;
697 inp = in_pcblookup_hash(&tcbinfo[cpu],
698 ip->ip_src, th->th_sport,
699 ip->ip_dst, th->th_dport,
700 0, m->m_pkthdr.rcvif);
703 * It's new. Try to find the ambushing socket.
707 * The rest of the ipfw code stores the port in
709 * (The IP address is still in network order.)
711 in_port_t dport = next_hop->sin_port ?
712 htons(next_hop->sin_port) :
715 cpu = tcp_addrcpu(ip->ip_src.s_addr, th->th_sport,
716 next_hop->sin_addr.s_addr, dport);
717 inp = in_pcblookup_hash(&tcbinfo[cpu],
718 ip->ip_src, th->th_sport,
719 next_hop->sin_addr, dport,
720 1, m->m_pkthdr.rcvif);
724 inp = in6_pcblookup_hash(&tcbinfo[0],
725 &ip6->ip6_src, th->th_sport,
726 &ip6->ip6_dst, th->th_dport,
727 1, m->m_pkthdr.rcvif);
729 cpu = mycpu->gd_cpuid;
730 inp = in_pcblookup_hash(&tcbinfo[cpu],
731 ip->ip_src, th->th_sport,
732 ip->ip_dst, th->th_dport,
733 1, m->m_pkthdr.rcvif);
738 * If the state is CLOSED (i.e., TCB does not exist) then
739 * all data in the incoming segment is discarded.
740 * If the TCB exists but is in CLOSED state, it is embryonic,
741 * but should either do a listen or a connect soon.
746 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2];
748 char dbuf[sizeof "aaa.bbb.ccc.ddd"];
749 char sbuf[sizeof "aaa.bbb.ccc.ddd"];
753 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst));
756 strcat(sbuf, ip6_sprintf(&ip6->ip6_src));
759 strcpy(dbuf, inet_ntoa(ip->ip_dst));
760 strcpy(sbuf, inet_ntoa(ip->ip_src));
762 switch (log_in_vain) {
764 if (!(thflags & TH_SYN))
768 "Connection attempt to TCP %s:%d "
769 "from %s:%d flags:0x%02x\n",
770 dbuf, ntohs(th->th_dport), sbuf,
771 ntohs(th->th_sport), thflags);
780 if (thflags & TH_SYN)
789 rstreason = BANDLIM_RST_CLOSEDPORT;
795 if (ipsec6_in_reject_so(m, inp->inp_socket)) {
796 ipsec6stat.in_polvio++;
800 if (ipsec4_in_reject_so(m, inp->inp_socket)) {
801 ipsecstat.in_polvio++;
808 if (ipsec6_in_reject(m, inp))
811 if (ipsec4_in_reject(m, inp))
818 rstreason = BANDLIM_RST_CLOSEDPORT;
821 if (tp->t_state <= TCPS_CLOSED)
824 /* Unscale the window into a 32-bit value. */
825 if (!(thflags & TH_SYN))
826 tiwin = th->th_win << tp->snd_scale;
830 so = inp->inp_socket;
833 if (so->so_options & SO_DEBUG) {
834 ostate = tp->t_state;
836 bcopy(ip6, tcp_saveipgen, sizeof(*ip6));
838 bcopy(ip, tcp_saveipgen, sizeof(*ip));
843 bzero(&to, sizeof to);
845 if (so->so_options & SO_ACCEPTCONN) {
846 struct in_conninfo inc;
849 inc.inc_isipv6 = (isipv6 == TRUE);
852 inc.inc6_faddr = ip6->ip6_src;
853 inc.inc6_laddr = ip6->ip6_dst;
854 inc.inc6_route.ro_rt = NULL; /* XXX */
856 inc.inc_faddr = ip->ip_src;
857 inc.inc_laddr = ip->ip_dst;
858 inc.inc_route.ro_rt = NULL; /* XXX */
860 inc.inc_fport = th->th_sport;
861 inc.inc_lport = th->th_dport;
864 * If the state is LISTEN then ignore segment if it contains
865 * a RST. If the segment contains an ACK then it is bad and
866 * send a RST. If it does not contain a SYN then it is not
867 * interesting; drop it.
869 * If the state is SYN_RECEIVED (syncache) and seg contains
870 * an ACK, but not for our SYN/ACK, send a RST. If the seg
871 * contains a RST, check the sequence number to see if it
872 * is a valid reset segment.
874 if ((thflags & (TH_RST | TH_ACK | TH_SYN)) != TH_SYN) {
875 if ((thflags & (TH_RST | TH_ACK | TH_SYN)) == TH_ACK) {
876 if (!syncache_expand(&inc, th, &so, m)) {
878 * No syncache entry, or ACK was not
879 * for our SYN/ACK. Send a RST.
881 tcpstat.tcps_badsyn++;
882 rstreason = BANDLIM_RST_OPENPORT;
887 * Could not complete 3-way handshake,
888 * connection is being closed down, and
889 * syncache will free mbuf.
893 * Socket is created in state SYN_RECEIVED.
894 * Continue processing segment.
899 * This is what would have happened in
900 * tcp_output() when the SYN,ACK was sent.
902 tp->snd_up = tp->snd_una;
903 tp->snd_max = tp->snd_nxt = tp->iss + 1;
904 tp->last_ack_sent = tp->rcv_nxt;
906 * XXX possible bug - it doesn't appear that tp->snd_wnd is unscaled
907 * until the _second_ ACK is received:
908 * rcv SYN (set wscale opts) --> send SYN/ACK, set snd_wnd = window.
909 * rcv ACK, calculate tiwin --> process SYN_RECEIVED, determine wscale,
910 * move to ESTAB, set snd_wnd to tiwin.
912 tp->snd_wnd = tiwin; /* unscaled */
915 if (thflags & TH_RST) {
916 syncache_chkrst(&inc, th);
919 if (thflags & TH_ACK) {
920 syncache_badack(&inc);
921 tcpstat.tcps_badsyn++;
922 rstreason = BANDLIM_RST_OPENPORT;
929 * Segment's flags are (SYN) or (SYN | FIN).
933 * If deprecated address is forbidden,
934 * we do not accept SYN to deprecated interface
935 * address to prevent any new inbound connection from
936 * getting established.
937 * When we do not accept SYN, we send a TCP RST,
938 * with deprecated source address (instead of dropping
939 * it). We compromise it as it is much better for peer
940 * to send a RST, and RST will be the final packet
943 * If we do not forbid deprecated addresses, we accept
944 * the SYN packet. RFC2462 does not suggest dropping
946 * If we decipher RFC2462 5.5.4, it says like this:
947 * 1. use of deprecated addr with existing
948 * communication is okay - "SHOULD continue to be
950 * 2. use of it with new communication:
951 * (2a) "SHOULD NOT be used if alternate address
952 * with sufficient scope is available"
953 * (2b) nothing mentioned otherwise.
954 * Here we fall into (2b) case as we have no choice in
955 * our source address selection - we must obey the peer.
957 * The wording in RFC2462 is confusing, and there are
958 * multiple description text for deprecated address
959 * handling - worse, they are not exactly the same.
960 * I believe 5.5.4 is the best one, so we follow 5.5.4.
962 if (isipv6 && !ip6_use_deprecated) {
963 struct in6_ifaddr *ia6;
965 if ((ia6 = ip6_getdstifaddr(m)) &&
966 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
968 rstreason = BANDLIM_RST_OPENPORT;
974 * If it is from this socket, drop it, it must be forged.
975 * Don't bother responding if the destination was a broadcast.
977 if (th->th_dport == th->th_sport) {
979 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
983 if (ip->ip_dst.s_addr == ip->ip_src.s_addr)
988 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
990 * Note that it is quite possible to receive unicast
991 * link-layer packets with a broadcast IP address. Use
992 * in_broadcast() to find them.
994 if (m->m_flags & (M_BCAST | M_MCAST))
997 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
998 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
1001 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1002 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1003 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1004 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
1008 * SYN appears to be valid; create compressed TCP state
1009 * for syncache, or perform t/tcp connection.
1011 if (so->so_qlen <= so->so_qlimit) {
1012 tcp_dooptions(&to, optp, optlen, TRUE);
1013 if (!syncache_add(&inc, &to, th, &so, m))
1017 * Entry added to syncache, mbuf used to
1018 * send SYN,ACK packet.
1022 * Segment passed TAO tests.
1025 tp = intotcpcb(inp);
1026 tp->snd_wnd = tiwin;
1027 tp->t_starttime = ticks;
1028 tp->t_state = TCPS_ESTABLISHED;
1031 * If there is a FIN, or if there is data and the
1032 * connection is local, then delay SYN,ACK(SYN) in
1033 * the hope of piggy-backing it on a response
1034 * segment. Otherwise must send ACK now in case
1035 * the other side is slow starting.
1037 if (DELAY_ACK(tp) &&
1038 ((thflags & TH_FIN) ||
1040 ((isipv6 && in6_localaddr(&inp->in6p_faddr)) ||
1041 (!isipv6 && in_localaddr(inp->inp_faddr)))))) {
1042 callout_reset(tp->tt_delack, tcp_delacktime,
1043 tcp_timer_delack, tp);
1044 tp->t_flags |= TF_NEEDSYN;
1046 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
1048 tcpstat.tcps_connects++;
1056 /* should not happen - syncache should pick up these connections */
1057 KASSERT(tp->t_state != TCPS_LISTEN, ("tcp_input: TCPS_LISTEN state"));
1060 * Segment received on connection.
1061 * Reset idle time and keep-alive timer.
1063 tp->t_rcvtime = ticks;
1064 if (TCPS_HAVEESTABLISHED(tp->t_state))
1065 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp);
1069 * XXX this is tradtitional behavior, may need to be cleaned up.
1071 tcp_dooptions(&to, optp, optlen, (thflags & TH_SYN) != 0);
1072 if (thflags & TH_SYN) {
1073 if (to.to_flags & TOF_SCALE) {
1074 tp->t_flags |= TF_RCVD_SCALE;
1075 tp->requested_s_scale = to.to_requested_s_scale;
1077 if (to.to_flags & TOF_TS) {
1078 tp->t_flags |= TF_RCVD_TSTMP;
1079 tp->ts_recent = to.to_tsval;
1080 tp->ts_recent_age = ticks;
1082 if (to.to_flags & (TOF_CC | TOF_CCNEW))
1083 tp->t_flags |= TF_RCVD_CC;
1084 if (to.to_flags & TOF_MSS)
1085 tcp_mss(tp, to.to_mss);
1087 * Only set the TF_SACK_PERMITTED per-connection flag
1088 * if we got a SACK_PERMITTED option from the other side
1089 * and the global tcp_do_sack variable is true.
1091 if (tcp_do_sack && (to.to_flags & TOF_SACK_PERMITTED))
1092 tp->t_flags |= TF_SACK_PERMITTED;
1096 * Header prediction: check for the two common cases
1097 * of a uni-directional data xfer. If the packet has
1098 * no control flags, is in-sequence, the window didn't
1099 * change and we're not retransmitting, it's a
1100 * candidate. If the length is zero and the ack moved
1101 * forward, we're the sender side of the xfer. Just
1102 * free the data acked & wake any higher level process
1103 * that was blocked waiting for space. If the length
1104 * is non-zero and the ack didn't move, we're the
1105 * receiver side. If we're getting packets in-order
1106 * (the reassembly queue is empty), add the data to
1107 * the socket buffer and note that we need a delayed ack.
1108 * Make sure that the hidden state-flags are also off.
1109 * Since we check for TCPS_ESTABLISHED above, it can only
1112 if (tp->t_state == TCPS_ESTABLISHED &&
1113 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1114 !(tp->t_flags & (TF_NEEDSYN | TF_NEEDFIN)) &&
1115 (!(to.to_flags & TOF_TS) ||
1116 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) &&
1118 * Using the CC option is compulsory if once started:
1119 * the segment is OK if no T/TCP was negotiated or
1120 * if the segment has a CC option equal to CCrecv
1122 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) ||
1123 ((to.to_flags & TOF_CC) && to.to_cc == tp->cc_recv)) &&
1124 th->th_seq == tp->rcv_nxt &&
1125 tp->snd_nxt == tp->snd_max) {
1128 * If last ACK falls within this segment's sequence numbers,
1129 * record the timestamp.
1130 * NOTE that the test is modified according to the latest
1131 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1133 if ((to.to_flags & TOF_TS) &&
1134 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1135 tp->ts_recent_age = ticks;
1136 tp->ts_recent = to.to_tsval;
1140 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1141 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1142 tp->snd_cwnd >= tp->snd_wnd &&
1143 !IN_FASTRECOVERY(tp)) {
1145 * this is a pure ack for outstanding data.
1147 ++tcpstat.tcps_predack;
1149 * "bad retransmit" recovery
1151 * If Eifel detection applies, then
1152 * it is deterministic, so use it
1153 * unconditionally over the old heuristic.
1154 * Otherwise, fall back to the old heuristic.
1156 if (tcp_do_eifel_detect &&
1157 (to.to_flags & TOF_TS) && to.to_tsecr &&
1158 (tp->t_flags & TF_FIRSTACCACK)) {
1159 /* Eifel detection applicable. */
1160 if (to.to_tsecr < tp->t_rexmtTS) {
1161 tcp_revert_congestion_state(tp);
1162 ++tcpstat.tcps_eifeldetected;
1164 } else if (tp->t_rxtshift == 1 &&
1165 ticks < tp->t_badrxtwin) {
1166 tcp_revert_congestion_state(tp);
1167 ++tcpstat.tcps_rttdetected;
1169 tp->t_flags &= ~(TF_FIRSTACCACK |
1170 TF_FASTREXMT | TF_EARLYREXMT);
1172 * Recalculate the retransmit timer / rtt.
1174 * Some machines (certain windows boxes)
1175 * send broken timestamp replies during the
1176 * SYN+ACK phase, ignore timestamps of 0.
1178 if ((to.to_flags & TOF_TS) && to.to_tsecr) {
1180 ticks - to.to_tsecr + 1);
1181 } else if (tp->t_rtttime &&
1182 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1184 ticks - tp->t_rtttime);
1186 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1187 acked = th->th_ack - tp->snd_una;
1188 tcpstat.tcps_rcvackpack++;
1189 tcpstat.tcps_rcvackbyte += acked;
1190 sbdrop(&so->so_snd, acked);
1191 tp->snd_recover = th->th_ack - 1;
1192 tp->snd_una = th->th_ack;
1195 * Update window information.
1197 if (tiwin != tp->snd_wnd &&
1198 acceptable_window_update(tp, th, tiwin)) {
1199 /* keep track of pure window updates */
1200 if (tp->snd_wl2 == th->th_ack &&
1201 tiwin > tp->snd_wnd)
1202 tcpstat.tcps_rcvwinupd++;
1203 tp->snd_wnd = tiwin;
1204 tp->snd_wl1 = th->th_seq;
1205 tp->snd_wl2 = th->th_ack;
1206 if (tp->snd_wnd > tp->max_sndwnd)
1207 tp->max_sndwnd = tp->snd_wnd;
1210 ND6_HINT(tp); /* some progress has been done */
1212 * If all outstanding data are acked, stop
1213 * retransmit timer, otherwise restart timer
1214 * using current (possibly backed-off) value.
1215 * If process is waiting for space,
1216 * wakeup/selwakeup/signal. If data
1217 * are ready to send, let tcp_output
1218 * decide between more output or persist.
1220 if (tp->snd_una == tp->snd_max)
1221 callout_stop(tp->tt_rexmt);
1222 else if (!callout_active(tp->tt_persist))
1223 callout_reset(tp->tt_rexmt,
1225 tcp_timer_rexmt, tp);
1228 if (so->so_snd.sb_cc)
1232 } else if (tiwin == tp->snd_wnd &&
1233 th->th_ack == tp->snd_una &&
1234 LIST_EMPTY(&tp->t_segq) &&
1235 tlen <= sbspace(&so->so_rcv)) {
1237 * this is a pure, in-sequence data packet
1238 * with nothing on the reassembly queue and
1239 * we have enough buffer space to take it.
1241 ++tcpstat.tcps_preddat;
1242 tp->rcv_nxt += tlen;
1243 tcpstat.tcps_rcvpack++;
1244 tcpstat.tcps_rcvbyte += tlen;
1245 ND6_HINT(tp); /* some progress has been done */
1247 * Add data to socket buffer.
1249 if (so->so_state & SS_CANTRCVMORE) {
1252 m_adj(m, drop_hdrlen); /* delayed header drop */
1253 sbappendstream(&so->so_rcv, m);
1258 * This code is responsible for most of the ACKs
1259 * the TCP stack sends back after receiving a data
1260 * packet. Note that the DELAY_ACK check fails if
1261 * the delack timer is already running, which results
1262 * in an ack being sent every other packet (which is
1265 if (DELAY_ACK(tp)) {
1266 callout_reset(tp->tt_delack, tcp_delacktime,
1267 tcp_timer_delack, tp);
1269 tp->t_flags |= TF_ACKNOW;
1270 if (!(tp->t_flags & TF_ONOUTPUTQ)) {
1271 tp->t_flags |= TF_ONOUTPUTQ;
1272 tp->tt_cpu = mycpu->gd_cpuid;
1274 &tcpcbackq[tp->tt_cpu],
1283 * Calculate amount of space in receive window,
1284 * and then do TCP input processing.
1285 * Receive window is amount of space in rcv queue,
1286 * but not less than advertised window.
1288 recvwin = sbspace(&so->so_rcv);
1291 tp->rcv_wnd = imax(recvwin, (int)(tp->rcv_adv - tp->rcv_nxt));
1293 switch (tp->t_state) {
1295 * If the state is SYN_RECEIVED:
1296 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1298 case TCPS_SYN_RECEIVED:
1299 if ((thflags & TH_ACK) &&
1300 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1301 SEQ_GT(th->th_ack, tp->snd_max))) {
1302 rstreason = BANDLIM_RST_OPENPORT;
1308 * If the state is SYN_SENT:
1309 * if seg contains an ACK, but not for our SYN, drop the input.
1310 * if seg contains a RST, then drop the connection.
1311 * if seg does not contain SYN, then drop it.
1312 * Otherwise this is an acceptable SYN segment
1313 * initialize tp->rcv_nxt and tp->irs
1314 * if seg contains ack then advance tp->snd_una
1315 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1316 * arrange for segment to be acked (eventually)
1317 * continue processing rest of data/controls, beginning with URG
1320 if ((taop = tcp_gettaocache(&inp->inp_inc)) == NULL) {
1321 taop = &tao_noncached;
1322 bzero(taop, sizeof *taop);
1325 if ((thflags & TH_ACK) &&
1326 (SEQ_LEQ(th->th_ack, tp->iss) ||
1327 SEQ_GT(th->th_ack, tp->snd_max))) {
1329 * If we have a cached CCsent for the remote host,
1330 * hence we haven't just crashed and restarted,
1331 * do not send a RST. This may be a retransmission
1332 * from the other side after our earlier ACK was lost.
1333 * Our new SYN, when it arrives, will serve as the
1336 if (taop->tao_ccsent != 0)
1339 rstreason = BANDLIM_UNLIMITED;
1343 if (thflags & TH_RST) {
1344 if (thflags & TH_ACK)
1345 tp = tcp_drop(tp, ECONNREFUSED);
1348 if (!(thflags & TH_SYN))
1350 tp->snd_wnd = th->th_win; /* initial send window */
1351 tp->cc_recv = to.to_cc; /* foreign CC */
1353 tp->irs = th->th_seq;
1355 if (thflags & TH_ACK) {
1357 * Our SYN was acked. If segment contains CC.ECHO
1358 * option, check it to make sure this segment really
1359 * matches our SYN. If not, just drop it as old
1360 * duplicate, but send an RST if we're still playing
1361 * by the old rules. If no CC.ECHO option, make sure
1362 * we don't get fooled into using T/TCP.
1364 if (to.to_flags & TOF_CCECHO) {
1365 if (tp->cc_send != to.to_ccecho) {
1366 if (taop->tao_ccsent != 0)
1369 rstreason = BANDLIM_UNLIMITED;
1374 tp->t_flags &= ~TF_RCVD_CC;
1375 tcpstat.tcps_connects++;
1377 /* Do window scaling on this connection? */
1378 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
1379 (TF_RCVD_SCALE | TF_REQ_SCALE)) {
1380 tp->snd_scale = tp->requested_s_scale;
1381 tp->rcv_scale = tp->request_r_scale;
1383 /* Segment is acceptable, update cache if undefined. */
1384 if (taop->tao_ccsent == 0)
1385 taop->tao_ccsent = to.to_ccecho;
1387 tp->rcv_adv += tp->rcv_wnd;
1388 tp->snd_una++; /* SYN is acked */
1389 callout_stop(tp->tt_rexmt);
1391 * If there's data, delay ACK; if there's also a FIN
1392 * ACKNOW will be turned on later.
1394 if (DELAY_ACK(tp) && tlen != 0)
1395 callout_reset(tp->tt_delack, tcp_delacktime,
1396 tcp_timer_delack, tp);
1398 tp->t_flags |= TF_ACKNOW;
1400 * Received <SYN,ACK> in SYN_SENT[*] state.
1402 * SYN_SENT --> ESTABLISHED
1403 * SYN_SENT* --> FIN_WAIT_1
1405 tp->t_starttime = ticks;
1406 if (tp->t_flags & TF_NEEDFIN) {
1407 tp->t_state = TCPS_FIN_WAIT_1;
1408 tp->t_flags &= ~TF_NEEDFIN;
1411 tp->t_state = TCPS_ESTABLISHED;
1412 callout_reset(tp->tt_keep, tcp_keepidle,
1413 tcp_timer_keep, tp);
1417 * Received initial SYN in SYN-SENT[*] state =>
1418 * simultaneous open. If segment contains CC option
1419 * and there is a cached CC, apply TAO test.
1420 * If it succeeds, connection is * half-synchronized.
1421 * Otherwise, do 3-way handshake:
1422 * SYN-SENT -> SYN-RECEIVED
1423 * SYN-SENT* -> SYN-RECEIVED*
1424 * If there was no CC option, clear cached CC value.
1426 tp->t_flags |= TF_ACKNOW;
1427 callout_stop(tp->tt_rexmt);
1428 if (to.to_flags & TOF_CC) {
1429 if (taop->tao_cc != 0 &&
1430 CC_GT(to.to_cc, taop->tao_cc)) {
1432 * update cache and make transition:
1433 * SYN-SENT -> ESTABLISHED*
1434 * SYN-SENT* -> FIN-WAIT-1*
1436 taop->tao_cc = to.to_cc;
1437 tp->t_starttime = ticks;
1438 if (tp->t_flags & TF_NEEDFIN) {
1439 tp->t_state = TCPS_FIN_WAIT_1;
1440 tp->t_flags &= ~TF_NEEDFIN;
1442 tp->t_state = TCPS_ESTABLISHED;
1443 callout_reset(tp->tt_keep,
1448 tp->t_flags |= TF_NEEDSYN;
1450 tp->t_state = TCPS_SYN_RECEIVED;
1452 /* CC.NEW or no option => invalidate cache */
1454 tp->t_state = TCPS_SYN_RECEIVED;
1460 * Advance th->th_seq to correspond to first data byte.
1461 * If data, trim to stay within window,
1462 * dropping FIN if necessary.
1465 if (tlen > tp->rcv_wnd) {
1466 todrop = tlen - tp->rcv_wnd;
1470 tcpstat.tcps_rcvpackafterwin++;
1471 tcpstat.tcps_rcvbyteafterwin += todrop;
1473 tp->snd_wl1 = th->th_seq - 1;
1474 tp->rcv_up = th->th_seq;
1476 * Client side of transaction: already sent SYN and data.
1477 * If the remote host used T/TCP to validate the SYN,
1478 * our data will be ACK'd; if so, enter normal data segment
1479 * processing in the middle of step 5, ack processing.
1480 * Otherwise, goto step 6.
1482 if (thflags & TH_ACK)
1488 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
1489 * if segment contains a SYN and CC [not CC.NEW] option:
1490 * if state == TIME_WAIT and connection duration > MSL,
1491 * drop packet and send RST;
1493 * if SEG.CC > CCrecv then is new SYN, and can implicitly
1494 * ack the FIN (and data) in retransmission queue.
1495 * Complete close and delete TCPCB. Then reprocess
1496 * segment, hoping to find new TCPCB in LISTEN state;
1498 * else must be old SYN; drop it.
1499 * else do normal processing.
1503 case TCPS_TIME_WAIT:
1504 if ((thflags & TH_SYN) &&
1505 (to.to_flags & TOF_CC) && tp->cc_recv != 0) {
1506 if (tp->t_state == TCPS_TIME_WAIT &&
1507 (ticks - tp->t_starttime) > tcp_msl) {
1508 rstreason = BANDLIM_UNLIMITED;
1511 if (CC_GT(to.to_cc, tp->cc_recv)) {
1518 break; /* continue normal processing */
1522 * States other than LISTEN or SYN_SENT.
1523 * First check the RST flag and sequence number since reset segments
1524 * are exempt from the timestamp and connection count tests. This
1525 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
1526 * below which allowed reset segments in half the sequence space
1527 * to fall though and be processed (which gives forged reset
1528 * segments with a random sequence number a 50 percent chance of
1529 * killing a connection).
1530 * Then check timestamp, if present.
1531 * Then check the connection count, if present.
1532 * Then check that at least some bytes of segment are within
1533 * receive window. If segment begins before rcv_nxt,
1534 * drop leading data (and SYN); if nothing left, just ack.
1537 * If the RST bit is set, check the sequence number to see
1538 * if this is a valid reset segment.
1540 * In all states except SYN-SENT, all reset (RST) segments
1541 * are validated by checking their SEQ-fields. A reset is
1542 * valid if its sequence number is in the window.
1543 * Note: this does not take into account delayed ACKs, so
1544 * we should test against last_ack_sent instead of rcv_nxt.
1545 * The sequence number in the reset segment is normally an
1546 * echo of our outgoing acknowledgement numbers, but some hosts
1547 * send a reset with the sequence number at the rightmost edge
1548 * of our receive window, and we have to handle this case.
1549 * If we have multiple segments in flight, the intial reset
1550 * segment sequence numbers will be to the left of last_ack_sent,
1551 * but they will eventually catch up.
1552 * In any case, it never made sense to trim reset segments to
1553 * fit the receive window since RFC 1122 says:
1554 * 4.2.2.12 RST Segment: RFC-793 Section 3.4
1556 * A TCP SHOULD allow a received RST segment to include data.
1559 * It has been suggested that a RST segment could contain
1560 * ASCII text that encoded and explained the cause of the
1561 * RST. No standard has yet been established for such
1564 * If the reset segment passes the sequence number test examine
1566 * SYN_RECEIVED STATE:
1567 * If passive open, return to LISTEN state.
1568 * If active open, inform user that connection was refused.
1569 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
1570 * Inform user that connection was reset, and close tcb.
1571 * CLOSING, LAST_ACK STATES:
1574 * Drop the segment - see Stevens, vol. 2, p. 964 and
1577 if (thflags & TH_RST) {
1578 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
1579 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
1580 switch (tp->t_state) {
1582 case TCPS_SYN_RECEIVED:
1583 so->so_error = ECONNREFUSED;
1586 case TCPS_ESTABLISHED:
1587 case TCPS_FIN_WAIT_1:
1588 case TCPS_FIN_WAIT_2:
1589 case TCPS_CLOSE_WAIT:
1590 so->so_error = ECONNRESET;
1592 tp->t_state = TCPS_CLOSED;
1593 tcpstat.tcps_drops++;
1602 case TCPS_TIME_WAIT:
1610 * RFC 1323 PAWS: If we have a timestamp reply on this segment
1611 * and it's less than ts_recent, drop it.
1613 if ((to.to_flags & TOF_TS) && tp->ts_recent != 0 &&
1614 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
1616 /* Check to see if ts_recent is over 24 days old. */
1617 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1619 * Invalidate ts_recent. If this segment updates
1620 * ts_recent, the age will be reset later and ts_recent
1621 * will get a valid value. If it does not, setting
1622 * ts_recent to zero will at least satisfy the
1623 * requirement that zero be placed in the timestamp
1624 * echo reply when ts_recent isn't valid. The
1625 * age isn't reset until we get a valid ts_recent
1626 * because we don't want out-of-order segments to be
1627 * dropped when ts_recent is old.
1631 tcpstat.tcps_rcvduppack++;
1632 tcpstat.tcps_rcvdupbyte += tlen;
1633 tcpstat.tcps_pawsdrop++;
1642 * If T/TCP was negotiated and the segment doesn't have CC,
1643 * or if its CC is wrong then drop the segment.
1644 * RST segments do not have to comply with this.
1646 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) &&
1647 (!(to.to_flags & TOF_CC) || tp->cc_recv != to.to_cc))
1651 * In the SYN-RECEIVED state, validate that the packet belongs to
1652 * this connection before trimming the data to fit the receive
1653 * window. Check the sequence number versus IRS since we know
1654 * the sequence numbers haven't wrapped. This is a partial fix
1655 * for the "LAND" DoS attack.
1657 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
1658 rstreason = BANDLIM_RST_OPENPORT;
1662 todrop = tp->rcv_nxt - th->th_seq;
1664 if (TCP_DO_SACK(tp)) {
1665 /* Report duplicate segment at head of packet. */
1666 tp->reportblk.rblk_start = th->th_seq;
1667 tp->reportblk.rblk_end = th->th_seq + tlen;
1668 if (thflags & TH_FIN)
1669 ++tp->reportblk.rblk_end;
1670 if (SEQ_GT(tp->reportblk.rblk_end, tp->rcv_nxt))
1671 tp->reportblk.rblk_end = tp->rcv_nxt;
1672 tp->t_flags |= (TF_DUPSEG | TF_SACKLEFT | TF_ACKNOW);
1674 if (thflags & TH_SYN) {
1684 * Following if statement from Stevens, vol. 2, p. 960.
1686 if (todrop > tlen ||
1687 (todrop == tlen && !(thflags & TH_FIN))) {
1689 * Any valid FIN must be to the left of the window.
1690 * At this point the FIN must be a duplicate or out
1691 * of sequence; drop it.
1696 * Send an ACK to resynchronize and drop any data.
1697 * But keep on processing for RST or ACK.
1699 tp->t_flags |= TF_ACKNOW;
1701 tcpstat.tcps_rcvduppack++;
1702 tcpstat.tcps_rcvdupbyte += todrop;
1704 tcpstat.tcps_rcvpartduppack++;
1705 tcpstat.tcps_rcvpartdupbyte += todrop;
1707 drop_hdrlen += todrop; /* drop from the top afterwards */
1708 th->th_seq += todrop;
1710 if (th->th_urp > todrop)
1711 th->th_urp -= todrop;
1719 * If new data are received on a connection after the
1720 * user processes are gone, then RST the other end.
1722 if ((so->so_state & SS_NOFDREF) &&
1723 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1725 tcpstat.tcps_rcvafterclose++;
1726 rstreason = BANDLIM_UNLIMITED;
1731 * If segment ends after window, drop trailing data
1732 * (and PUSH and FIN); if nothing left, just ACK.
1734 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
1736 tcpstat.tcps_rcvpackafterwin++;
1737 if (todrop >= tlen) {
1738 tcpstat.tcps_rcvbyteafterwin += tlen;
1740 * If a new connection request is received
1741 * while in TIME_WAIT, drop the old connection
1742 * and start over if the sequence numbers
1743 * are above the previous ones.
1745 if (thflags & TH_SYN &&
1746 tp->t_state == TCPS_TIME_WAIT &&
1747 SEQ_GT(th->th_seq, tp->rcv_nxt)) {
1752 * If window is closed can only take segments at
1753 * window edge, and have to drop data and PUSH from
1754 * incoming segments. Continue processing, but
1755 * remember to ack. Otherwise, drop segment
1758 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1759 tp->t_flags |= TF_ACKNOW;
1760 tcpstat.tcps_rcvwinprobe++;
1764 tcpstat.tcps_rcvbyteafterwin += todrop;
1767 thflags &= ~(TH_PUSH | TH_FIN);
1771 * If last ACK falls within this segment's sequence numbers,
1772 * record its timestamp.
1773 * NOTE that the test is modified according to the latest
1774 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1776 if ((to.to_flags & TOF_TS) && SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1777 tp->ts_recent_age = ticks;
1778 tp->ts_recent = to.to_tsval;
1782 * If a SYN is in the window, then this is an
1783 * error and we send an RST and drop the connection.
1785 if (thflags & TH_SYN) {
1786 tp = tcp_drop(tp, ECONNRESET);
1787 rstreason = BANDLIM_UNLIMITED;
1792 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
1793 * flag is on (half-synchronized state), then queue data for
1794 * later processing; else drop segment and return.
1796 if (!(thflags & TH_ACK)) {
1797 if (tp->t_state == TCPS_SYN_RECEIVED ||
1798 (tp->t_flags & TF_NEEDSYN))
1807 switch (tp->t_state) {
1809 * In SYN_RECEIVED state, the ACK acknowledges our SYN, so enter
1810 * ESTABLISHED state and continue processing.
1811 * The ACK was checked above.
1813 case TCPS_SYN_RECEIVED:
1815 tcpstat.tcps_connects++;
1817 /* Do window scaling? */
1818 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
1819 (TF_RCVD_SCALE | TF_REQ_SCALE)) {
1820 tp->snd_scale = tp->requested_s_scale;
1821 tp->rcv_scale = tp->request_r_scale;
1824 * Upon successful completion of 3-way handshake,
1825 * update cache.CC if it was undefined, pass any queued
1826 * data to the user, and advance state appropriately.
1828 if ((taop = tcp_gettaocache(&inp->inp_inc)) != NULL &&
1830 taop->tao_cc = tp->cc_recv;
1834 * SYN-RECEIVED -> ESTABLISHED
1835 * SYN-RECEIVED* -> FIN-WAIT-1
1837 tp->t_starttime = ticks;
1838 if (tp->t_flags & TF_NEEDFIN) {
1839 tp->t_state = TCPS_FIN_WAIT_1;
1840 tp->t_flags &= ~TF_NEEDFIN;
1842 tp->t_state = TCPS_ESTABLISHED;
1843 callout_reset(tp->tt_keep, tcp_keepidle,
1844 tcp_timer_keep, tp);
1847 * If segment contains data or ACK, will call tcp_reass()
1848 * later; if not, do so now to pass queued data to user.
1850 if (tlen == 0 && !(thflags & TH_FIN))
1851 tcp_reass(tp, NULL, NULL, NULL);
1855 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1856 * ACKs. If the ack is in the range
1857 * tp->snd_una < th->th_ack <= tp->snd_max
1858 * then advance tp->snd_una to th->th_ack and drop
1859 * data from the retransmission queue. If this ACK reflects
1860 * more up to date window information we update our window information.
1862 case TCPS_ESTABLISHED:
1863 case TCPS_FIN_WAIT_1:
1864 case TCPS_FIN_WAIT_2:
1865 case TCPS_CLOSE_WAIT:
1868 case TCPS_TIME_WAIT:
1870 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
1871 if (TCP_DO_SACK(tp))
1872 tcp_sack_update_scoreboard(tp, &to);
1873 if (tlen != 0 || tiwin != tp->snd_wnd) {
1877 tcpstat.tcps_rcvdupack++;
1878 if (!callout_active(tp->tt_rexmt) ||
1879 th->th_ack != tp->snd_una) {
1884 * We have outstanding data (other than
1885 * a window probe), this is a completely
1886 * duplicate ack (ie, window info didn't
1887 * change), the ack is the biggest we've
1888 * seen and we've seen exactly our rexmt
1889 * threshhold of them, so assume a packet
1890 * has been dropped and retransmit it.
1891 * Kludge snd_nxt & the congestion
1892 * window so we send only this one
1895 if (IN_FASTRECOVERY(tp)) {
1896 if (TCP_DO_SACK(tp)) {
1897 /* No artifical cwnd inflation. */
1898 tcp_sack_rexmt(tp, th);
1901 * Dup acks mean that packets
1902 * have left the network
1903 * (they're now cached at the
1904 * receiver) so bump cwnd by
1905 * the amount in the receiver
1906 * to keep a constant cwnd
1907 * packets in the network.
1909 tp->snd_cwnd += tp->t_maxseg;
1912 } else if (SEQ_LT(th->th_ack, tp->snd_recover)) {
1915 } else if (++tp->t_dupacks == tcprexmtthresh) {
1916 tcp_seq old_snd_nxt;
1920 if (tcp_do_eifel_detect &&
1921 (tp->t_flags & TF_RCVD_TSTMP)) {
1922 tcp_save_congestion_state(tp);
1923 tp->t_flags |= TF_FASTREXMT;
1926 * We know we're losing at the current
1927 * window size, so do congestion avoidance:
1928 * set ssthresh to half the current window
1929 * and pull our congestion window back to the
1932 win = min(tp->snd_wnd, tp->snd_cwnd) / 2 /
1936 tp->snd_ssthresh = win * tp->t_maxseg;
1937 ENTER_FASTRECOVERY(tp);
1938 tp->snd_recover = tp->snd_max;
1939 callout_stop(tp->tt_rexmt);
1941 old_snd_nxt = tp->snd_nxt;
1942 tp->snd_nxt = th->th_ack;
1943 tp->snd_cwnd = tp->t_maxseg;
1945 ++tcpstat.tcps_sndfastrexmit;
1946 tp->snd_cwnd = tp->snd_ssthresh;
1947 tp->rexmt_high = tp->snd_nxt;
1948 if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
1949 tp->snd_nxt = old_snd_nxt;
1950 KASSERT(tp->snd_limited <= 2,
1951 ("tp->snd_limited too big"));
1952 if (TCP_DO_SACK(tp))
1953 tcp_sack_rexmt(tp, th);
1955 tp->snd_cwnd += tp->t_maxseg *
1956 (tp->t_dupacks - tp->snd_limited);
1957 } else if (tcp_do_limitedtransmit) {
1958 u_long oldcwnd = tp->snd_cwnd;
1959 tcp_seq oldsndmax = tp->snd_max;
1960 /* outstanding data */
1961 uint32_t ownd = tp->snd_max - tp->snd_una;
1964 #define iceildiv(n, d) (((n)+(d)-1) / (d))
1966 KASSERT(tp->t_dupacks == 1 ||
1968 ("dupacks not 1 or 2"));
1969 if (tp->t_dupacks == 1)
1970 tp->snd_limited = 0;
1971 tp->snd_cwnd = ownd +
1972 (tp->t_dupacks - tp->snd_limited) *
1975 tp->snd_cwnd = oldcwnd;
1976 sent = tp->snd_max - oldsndmax;
1977 if (sent > tp->t_maxseg) {
1978 KASSERT((tp->t_dupacks == 2 &&
1979 tp->snd_limited == 0) ||
1980 (sent == tp->t_maxseg + 1 &&
1981 tp->t_flags & TF_SENTFIN),
1983 KASSERT(sent <= tp->t_maxseg * 2,
1984 ("sent too many segments"));
1985 tp->snd_limited = 2;
1986 tcpstat.tcps_sndlimited += 2;
1987 } else if (sent > 0) {
1989 ++tcpstat.tcps_sndlimited;
1990 } else if (tcp_do_early_retransmit &&
1991 (tcp_do_eifel_detect &&
1992 (tp->t_flags & TF_RCVD_TSTMP)) &&
1993 ownd < 4 * tp->t_maxseg &&
1994 tp->t_dupacks + 1 >=
1995 iceildiv(ownd, tp->t_maxseg) &&
1996 (!TCP_DO_SACK(tp) ||
1997 ownd <= tp->t_maxseg ||
1998 tcp_sack_has_sacked(&tp->scb,
1999 ownd - tp->t_maxseg))) {
2000 ++tcpstat.tcps_sndearlyrexmit;
2001 tp->t_flags |= TF_EARLYREXMT;
2002 goto fastretransmit;
2008 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una"));
2010 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2012 * Detected optimistic ACK attack.
2013 * Force slow-start to de-synchronize attack.
2015 tp->snd_cwnd = tp->t_maxseg;
2017 tcpstat.tcps_rcvacktoomuch++;
2021 * If we reach this point, ACK is not a duplicate,
2022 * i.e., it ACKs something we sent.
2024 if (tp->t_flags & TF_NEEDSYN) {
2026 * T/TCP: Connection was half-synchronized, and our
2027 * SYN has been ACK'd (so connection is now fully
2028 * synchronized). Go to non-starred state,
2029 * increment snd_una for ACK of SYN, and check if
2030 * we can do window scaling.
2032 tp->t_flags &= ~TF_NEEDSYN;
2034 /* Do window scaling? */
2035 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
2036 (TF_RCVD_SCALE | TF_REQ_SCALE)) {
2037 tp->snd_scale = tp->requested_s_scale;
2038 tp->rcv_scale = tp->request_r_scale;
2043 acked = th->th_ack - tp->snd_una;
2044 tcpstat.tcps_rcvackpack++;
2045 tcpstat.tcps_rcvackbyte += acked;
2047 if (tcp_do_eifel_detect && acked > 0 &&
2048 (to.to_flags & TOF_TS) && (to.to_tsecr != 0) &&
2049 (tp->t_flags & TF_FIRSTACCACK)) {
2050 /* Eifel detection applicable. */
2051 if (to.to_tsecr < tp->t_rexmtTS) {
2052 ++tcpstat.tcps_eifeldetected;
2053 tcp_revert_congestion_state(tp);
2054 if (tp->t_rxtshift == 1 &&
2055 ticks >= tp->t_badrxtwin)
2056 ++tcpstat.tcps_rttcantdetect;
2058 } else if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) {
2060 * If we just performed our first retransmit,
2061 * and the ACK arrives within our recovery window,
2062 * then it was a mistake to do the retransmit
2063 * in the first place. Recover our original cwnd
2064 * and ssthresh, and proceed to transmit where we
2067 tcp_revert_congestion_state(tp);
2068 ++tcpstat.tcps_rttdetected;
2072 * If we have a timestamp reply, update smoothed
2073 * round trip time. If no timestamp is present but
2074 * transmit timer is running and timed sequence
2075 * number was acked, update smoothed round trip time.
2076 * Since we now have an rtt measurement, cancel the
2077 * timer backoff (cf., Phil Karn's retransmit alg.).
2078 * Recompute the initial retransmit timer.
2080 * Some machines (certain windows boxes) send broken
2081 * timestamp replies during the SYN+ACK phase, ignore
2084 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0))
2085 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1);
2086 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq))
2087 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2088 tcp_xmit_bandwidth_limit(tp, th->th_ack);
2091 * If no data (only SYN) was ACK'd,
2092 * skip rest of ACK processing.
2097 /* Stop looking for an acceptable ACK since one was received. */
2098 tp->t_flags &= ~(TF_FIRSTACCACK | TF_FASTREXMT | TF_EARLYREXMT);
2100 if (acked > so->so_snd.sb_cc) {
2101 tp->snd_wnd -= so->so_snd.sb_cc;
2102 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc);
2103 ourfinisacked = TRUE;
2105 sbdrop(&so->so_snd, acked);
2106 tp->snd_wnd -= acked;
2107 ourfinisacked = FALSE;
2112 * Update window information.
2113 * Don't look at window if no ACK:
2114 * TAC's send garbage on first SYN.
2116 if (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2117 (tp->snd_wl1 == th->th_seq &&
2118 (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2119 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)))) {
2120 /* keep track of pure window updates */
2121 if (tlen == 0 && tp->snd_wl2 == th->th_ack &&
2122 tiwin > tp->snd_wnd)
2123 tcpstat.tcps_rcvwinupd++;
2124 tp->snd_wnd = tiwin;
2125 tp->snd_wl1 = th->th_seq;
2126 tp->snd_wl2 = th->th_ack;
2127 if (tp->snd_wnd > tp->max_sndwnd)
2128 tp->max_sndwnd = tp->snd_wnd;
2132 tp->snd_una = th->th_ack;
2133 if (TCP_DO_SACK(tp))
2134 tcp_sack_update_scoreboard(tp, &to);
2135 if (IN_FASTRECOVERY(tp)) {
2136 if (SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2137 EXIT_FASTRECOVERY(tp);
2140 * If the congestion window was inflated
2141 * to account for the other side's
2142 * cached packets, retract it.
2144 * Window inflation should have left us
2145 * with approximately snd_ssthresh outstanding
2146 * data. But, in case we would be inclined
2147 * to send a burst, better do it using
2150 if (!TCP_DO_SACK(tp))
2151 tp->snd_cwnd = tp->snd_ssthresh;
2153 if (SEQ_GT(th->th_ack + tp->snd_cwnd,
2154 tp->snd_max + 2 * tp->t_maxseg))
2156 (tp->snd_max - tp->snd_una) +
2159 if (TCP_DO_SACK(tp)) {
2160 tp->snd_max_rexmt = tp->snd_max;
2161 tcp_sack_rexmt(tp, th);
2163 tcp_newreno_partial_ack(tp, th, acked);
2169 * When new data is acked, open the congestion window.
2170 * If the window gives us less than ssthresh packets
2171 * in flight, open exponentially (maxseg per packet).
2172 * Otherwise open linearly: maxseg per window
2173 * (maxseg^2 / cwnd per packet).
2175 u_int cw = tp->snd_cwnd;
2178 if (cw > tp->snd_ssthresh)
2179 incr = tp->t_maxseg * tp->t_maxseg / cw;
2181 incr = tp->t_maxseg;
2182 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale);
2183 tp->snd_recover = th->th_ack - 1;
2185 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2186 tp->snd_nxt = tp->snd_una;
2189 * If all outstanding data is acked, stop retransmit
2190 * timer and remember to restart (more output or persist).
2191 * If there is more data to be acked, restart retransmit
2192 * timer, using current (possibly backed-off) value.
2194 if (th->th_ack == tp->snd_max) {
2195 callout_stop(tp->tt_rexmt);
2197 } else if (!callout_active(tp->tt_persist))
2198 callout_reset(tp->tt_rexmt, tp->t_rxtcur,
2199 tcp_timer_rexmt, tp);
2201 switch (tp->t_state) {
2203 * In FIN_WAIT_1 STATE in addition to the processing
2204 * for the ESTABLISHED state if our FIN is now acknowledged
2205 * then enter FIN_WAIT_2.
2207 case TCPS_FIN_WAIT_1:
2208 if (ourfinisacked) {
2210 * If we can't receive any more
2211 * data, then closing user can proceed.
2212 * Starting the timer is contrary to the
2213 * specification, but if we don't get a FIN
2214 * we'll hang forever.
2216 if (so->so_state & SS_CANTRCVMORE) {
2217 soisdisconnected(so);
2218 callout_reset(tp->tt_2msl, tcp_maxidle,
2219 tcp_timer_2msl, tp);
2221 tp->t_state = TCPS_FIN_WAIT_2;
2226 * In CLOSING STATE in addition to the processing for
2227 * the ESTABLISHED state if the ACK acknowledges our FIN
2228 * then enter the TIME-WAIT state, otherwise ignore
2232 if (ourfinisacked) {
2233 tp->t_state = TCPS_TIME_WAIT;
2234 tcp_canceltimers(tp);
2235 /* Shorten TIME_WAIT [RFC-1644, p.28] */
2236 if (tp->cc_recv != 0 &&
2237 (ticks - tp->t_starttime) < tcp_msl)
2238 callout_reset(tp->tt_2msl,
2239 tp->t_rxtcur * TCPTV_TWTRUNC,
2240 tcp_timer_2msl, tp);
2242 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2243 tcp_timer_2msl, tp);
2244 soisdisconnected(so);
2249 * In LAST_ACK, we may still be waiting for data to drain
2250 * and/or to be acked, as well as for the ack of our FIN.
2251 * If our FIN is now acknowledged, delete the TCB,
2252 * enter the closed state and return.
2255 if (ourfinisacked) {
2262 * In TIME_WAIT state the only thing that should arrive
2263 * is a retransmission of the remote FIN. Acknowledge
2264 * it and restart the finack timer.
2266 case TCPS_TIME_WAIT:
2267 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2268 tcp_timer_2msl, tp);
2275 * Update window information.
2276 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2278 if ((thflags & TH_ACK) &&
2279 acceptable_window_update(tp, th, tiwin)) {
2280 /* keep track of pure window updates */
2281 if (tlen == 0 && tp->snd_wl2 == th->th_ack &&
2282 tiwin > tp->snd_wnd)
2283 tcpstat.tcps_rcvwinupd++;
2284 tp->snd_wnd = tiwin;
2285 tp->snd_wl1 = th->th_seq;
2286 tp->snd_wl2 = th->th_ack;
2287 if (tp->snd_wnd > tp->max_sndwnd)
2288 tp->max_sndwnd = tp->snd_wnd;
2293 * Process segments with URG.
2295 if ((thflags & TH_URG) && th->th_urp &&
2296 !TCPS_HAVERCVDFIN(tp->t_state)) {
2298 * This is a kludge, but if we receive and accept
2299 * random urgent pointers, we'll crash in
2300 * soreceive. It's hard to imagine someone
2301 * actually wanting to send this much urgent data.
2303 if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2304 th->th_urp = 0; /* XXX */
2305 thflags &= ~TH_URG; /* XXX */
2306 goto dodata; /* XXX */
2309 * If this segment advances the known urgent pointer,
2310 * then mark the data stream. This should not happen
2311 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2312 * a FIN has been received from the remote side.
2313 * In these states we ignore the URG.
2315 * According to RFC961 (Assigned Protocols),
2316 * the urgent pointer points to the last octet
2317 * of urgent data. We continue, however,
2318 * to consider it to indicate the first octet
2319 * of data past the urgent section as the original
2320 * spec states (in one of two places).
2322 if (SEQ_GT(th->th_seq + th->th_urp, tp->rcv_up)) {
2323 tp->rcv_up = th->th_seq + th->th_urp;
2324 so->so_oobmark = so->so_rcv.sb_cc +
2325 (tp->rcv_up - tp->rcv_nxt) - 1;
2326 if (so->so_oobmark == 0)
2327 so->so_state |= SS_RCVATMARK;
2329 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2332 * Remove out of band data so doesn't get presented to user.
2333 * This can happen independent of advancing the URG pointer,
2334 * but if two URG's are pending at once, some out-of-band
2335 * data may creep in... ick.
2337 if (th->th_urp <= (u_long)tlen &&
2338 !(so->so_options & SO_OOBINLINE)) {
2339 /* hdr drop is delayed */
2340 tcp_pulloutofband(so, th, m, drop_hdrlen);
2344 * If no out of band data is expected,
2345 * pull receive urgent pointer along
2346 * with the receive window.
2348 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2349 tp->rcv_up = tp->rcv_nxt;
2354 * Process the segment text, merging it into the TCP sequencing queue,
2355 * and arranging for acknowledgment of receipt if necessary.
2356 * This process logically involves adjusting tp->rcv_wnd as data
2357 * is presented to the user (this happens in tcp_usrreq.c,
2358 * case PRU_RCVD). If a FIN has already been received on this
2359 * connection then we just ignore the text.
2361 if ((tlen || (thflags & TH_FIN)) && !TCPS_HAVERCVDFIN(tp->t_state)) {
2362 m_adj(m, drop_hdrlen); /* delayed header drop */
2364 * Insert segment which includes th into TCP reassembly queue
2365 * with control block tp. Set thflags to whether reassembly now
2366 * includes a segment with FIN. This handles the common case
2367 * inline (segment is the next to be received on an established
2368 * connection, and the queue is empty), avoiding linkage into
2369 * and removal from the queue and repetition of various
2371 * Set DELACK for segments received in order, but ack
2372 * immediately when segments are out of order (so
2373 * fast retransmit can work).
2375 if (th->th_seq == tp->rcv_nxt &&
2376 LIST_EMPTY(&tp->t_segq) &&
2377 TCPS_HAVEESTABLISHED(tp->t_state)) {
2379 callout_reset(tp->tt_delack, tcp_delacktime,
2380 tcp_timer_delack, tp);
2382 tp->t_flags |= TF_ACKNOW;
2383 tp->rcv_nxt += tlen;
2384 thflags = th->th_flags & TH_FIN;
2385 tcpstat.tcps_rcvpack++;
2386 tcpstat.tcps_rcvbyte += tlen;
2388 if (so->so_state & SS_CANTRCVMORE)
2391 sbappendstream(&so->so_rcv, m);
2394 if (!(tp->t_flags & TF_DUPSEG)) {
2395 /* Initialize SACK report block. */
2396 tp->reportblk.rblk_start = th->th_seq;
2397 tp->reportblk.rblk_end = th->th_seq + tlen +
2398 ((thflags & TH_FIN) != 0);
2400 thflags = tcp_reass(tp, th, &tlen, m);
2401 tp->t_flags |= TF_ACKNOW;
2405 * Note the amount of data that peer has sent into
2406 * our window, in order to estimate the sender's
2409 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2416 * If FIN is received ACK the FIN and let the user know
2417 * that the connection is closing.
2419 if (thflags & TH_FIN) {
2420 if (!TCPS_HAVERCVDFIN(tp->t_state)) {
2423 * If connection is half-synchronized
2424 * (ie NEEDSYN flag on) then delay ACK,
2425 * so it may be piggybacked when SYN is sent.
2426 * Otherwise, since we received a FIN then no
2427 * more input can be expected, send ACK now.
2429 if (DELAY_ACK(tp) && (tp->t_flags & TF_NEEDSYN))
2430 callout_reset(tp->tt_delack, tcp_delacktime,
2431 tcp_timer_delack, tp);
2433 tp->t_flags |= TF_ACKNOW;
2437 switch (tp->t_state) {
2439 * In SYN_RECEIVED and ESTABLISHED STATES
2440 * enter the CLOSE_WAIT state.
2442 case TCPS_SYN_RECEIVED:
2443 tp->t_starttime = ticks;
2445 case TCPS_ESTABLISHED:
2446 tp->t_state = TCPS_CLOSE_WAIT;
2450 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2451 * enter the CLOSING state.
2453 case TCPS_FIN_WAIT_1:
2454 tp->t_state = TCPS_CLOSING;
2458 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2459 * starting the time-wait timer, turning off the other
2462 case TCPS_FIN_WAIT_2:
2463 tp->t_state = TCPS_TIME_WAIT;
2464 tcp_canceltimers(tp);
2465 /* Shorten TIME_WAIT [RFC-1644, p.28] */
2466 if (tp->cc_recv != 0 &&
2467 (ticks - tp->t_starttime) < tcp_msl) {
2468 callout_reset(tp->tt_2msl,
2469 tp->t_rxtcur * TCPTV_TWTRUNC,
2470 tcp_timer_2msl, tp);
2471 /* For transaction client, force ACK now. */
2472 tp->t_flags |= TF_ACKNOW;
2475 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2476 tcp_timer_2msl, tp);
2477 soisdisconnected(so);
2481 * In TIME_WAIT state restart the 2 MSL time_wait timer.
2483 case TCPS_TIME_WAIT:
2484 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2485 tcp_timer_2msl, tp);
2491 if (so->so_options & SO_DEBUG)
2492 tcp_trace(TA_INPUT, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2496 * Return any desired output.
2498 if (needoutput || (tp->t_flags & TF_ACKNOW))
2504 * Generate an ACK dropping incoming segment if it occupies
2505 * sequence space, where the ACK reflects our state.
2507 * We can now skip the test for the RST flag since all
2508 * paths to this code happen after packets containing
2509 * RST have been dropped.
2511 * In the SYN-RECEIVED state, don't send an ACK unless the
2512 * segment we received passes the SYN-RECEIVED ACK test.
2513 * If it fails send a RST. This breaks the loop in the
2514 * "LAND" DoS attack, and also prevents an ACK storm
2515 * between two listening ports that have been sent forged
2516 * SYN segments, each with the source address of the other.
2518 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
2519 (SEQ_GT(tp->snd_una, th->th_ack) ||
2520 SEQ_GT(th->th_ack, tp->snd_max)) ) {
2521 rstreason = BANDLIM_RST_OPENPORT;
2525 if (so->so_options & SO_DEBUG)
2526 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2529 tp->t_flags |= TF_ACKNOW;
2535 * Generate a RST, dropping incoming segment.
2536 * Make ACK acceptable to originator of segment.
2537 * Don't bother to respond if destination was broadcast/multicast.
2539 if ((thflags & TH_RST) || m->m_flags & (M_BCAST | M_MCAST))
2542 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
2543 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
2546 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
2547 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
2548 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
2549 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2552 /* IPv6 anycast check is done at tcp6_input() */
2555 * Perform bandwidth limiting.
2558 if (badport_bandlim(rstreason) < 0)
2563 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2564 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2566 if (thflags & TH_ACK)
2567 /* mtod() below is safe as long as hdr dropping is delayed */
2568 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack,
2571 if (thflags & TH_SYN)
2573 /* mtod() below is safe as long as hdr dropping is delayed */
2574 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq + tlen,
2575 (tcp_seq)0, TH_RST | TH_ACK);
2581 * Drop space held by incoming segment and return.
2584 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2585 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2592 * Parse TCP options and place in tcpopt.
2595 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, boolean_t is_syn)
2600 for (; cnt > 0; cnt -= optlen, cp += optlen) {
2602 if (opt == TCPOPT_EOL)
2604 if (opt == TCPOPT_NOP)
2610 if (optlen < 2 || optlen > cnt)
2615 if (optlen != TCPOLEN_MAXSEG)
2619 to->to_flags |= TOF_MSS;
2620 bcopy(cp + 2, &to->to_mss, sizeof to->to_mss);
2621 to->to_mss = ntohs(to->to_mss);
2624 if (optlen != TCPOLEN_WINDOW)
2628 to->to_flags |= TOF_SCALE;
2629 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
2631 case TCPOPT_TIMESTAMP:
2632 if (optlen != TCPOLEN_TIMESTAMP)
2634 to->to_flags |= TOF_TS;
2635 bcopy(cp + 2, &to->to_tsval, sizeof to->to_tsval);
2636 to->to_tsval = ntohl(to->to_tsval);
2637 bcopy(cp + 6, &to->to_tsecr, sizeof to->to_tsecr);
2638 to->to_tsecr = ntohl(to->to_tsecr);
2641 if (optlen != TCPOLEN_CC)
2643 to->to_flags |= TOF_CC;
2644 bcopy(cp + 2, &to->to_cc, sizeof to->to_cc);
2645 to->to_cc = ntohl(to->to_cc);
2648 if (optlen != TCPOLEN_CC)
2652 to->to_flags |= TOF_CCNEW;
2653 bcopy(cp + 2, &to->to_cc, sizeof to->to_cc);
2654 to->to_cc = ntohl(to->to_cc);
2657 if (optlen != TCPOLEN_CC)
2661 to->to_flags |= TOF_CCECHO;
2662 bcopy(cp + 2, &to->to_ccecho, sizeof to->to_ccecho);
2663 to->to_ccecho = ntohl(to->to_ccecho);
2665 case TCPOPT_SACK_PERMITTED:
2666 if (optlen != TCPOLEN_SACK_PERMITTED)
2670 to->to_flags |= TOF_SACK_PERMITTED;
2673 if ((optlen - 2) & 0x07) /* not multiple of 8 */
2675 to->to_nsackblocks = (optlen - 2) / 8;
2676 to->to_sackblocks = (struct raw_sackblock *) (cp + 2);
2677 to->to_flags |= TOF_SACK;
2678 for (i = 0; i < to->to_nsackblocks; i++) {
2679 struct raw_sackblock *r = &to->to_sackblocks[i];
2681 r->rblk_start = ntohl(r->rblk_start);
2682 r->rblk_end = ntohl(r->rblk_end);
2692 * Pull out of band byte out of a segment so
2693 * it doesn't appear in the user's data queue.
2694 * It is still reflected in the segment length for
2695 * sequencing purposes.
2696 * "off" is the delayed to be dropped hdrlen.
2699 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, int off)
2701 int cnt = off + th->th_urp - 1;
2704 if (m->m_len > cnt) {
2705 char *cp = mtod(m, caddr_t) + cnt;
2706 struct tcpcb *tp = sototcpcb(so);
2709 tp->t_oobflags |= TCPOOB_HAVEDATA;
2710 bcopy(cp + 1, cp, m->m_len - cnt - 1);
2712 if (m->m_flags & M_PKTHDR)
2721 panic("tcp_pulloutofband");
2725 * Collect new round-trip time estimate
2726 * and update averages and current timeout.
2729 tcp_xmit_timer(struct tcpcb *tp, int rtt)
2733 tcpstat.tcps_rttupdated++;
2735 if (tp->t_srtt != 0) {
2737 * srtt is stored as fixed point with 5 bits after the
2738 * binary point (i.e., scaled by 8). The following magic
2739 * is equivalent to the smoothing algorithm in rfc793 with
2740 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2741 * point). Adjust rtt to origin 0.
2743 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
2744 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
2746 if ((tp->t_srtt += delta) <= 0)
2750 * We accumulate a smoothed rtt variance (actually, a
2751 * smoothed mean difference), then set the retransmit
2752 * timer to smoothed rtt + 4 times the smoothed variance.
2753 * rttvar is stored as fixed point with 4 bits after the
2754 * binary point (scaled by 16). The following is
2755 * equivalent to rfc793 smoothing with an alpha of .75
2756 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
2757 * rfc793's wired-in beta.
2761 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
2762 if ((tp->t_rttvar += delta) <= 0)
2764 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
2765 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2768 * No rtt measurement yet - use the unsmoothed rtt.
2769 * Set the variance to half the rtt (so our first
2770 * retransmit happens at 3*rtt).
2772 tp->t_srtt = rtt << TCP_RTT_SHIFT;
2773 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
2774 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2780 * the retransmit should happen at rtt + 4 * rttvar.
2781 * Because of the way we do the smoothing, srtt and rttvar
2782 * will each average +1/2 tick of bias. When we compute
2783 * the retransmit timer, we want 1/2 tick of rounding and
2784 * 1 extra tick because of +-1/2 tick uncertainty in the
2785 * firing of the timer. The bias will give us exactly the
2786 * 1.5 tick we need. But, because the bias is
2787 * statistical, we have to test that we don't drop below
2788 * the minimum feasible timer (which is 2 ticks).
2790 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
2791 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
2794 * We received an ack for a packet that wasn't retransmitted;
2795 * it is probably safe to discard any error indications we've
2796 * received recently. This isn't quite right, but close enough
2797 * for now (a route might have failed after we sent a segment,
2798 * and the return path might not be symmetrical).
2800 tp->t_softerror = 0;
2804 * Determine a reasonable value for maxseg size.
2805 * If the route is known, check route for mtu.
2806 * If none, use an mss that can be handled on the outgoing
2807 * interface without forcing IP to fragment; if bigger than
2808 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2809 * to utilize large mbufs. If no route is found, route has no mtu,
2810 * or the destination isn't local, use a default, hopefully conservative
2811 * size (usually 512 or the default IP max size, but no more than the mtu
2812 * of the interface), as we can't discover anything about intervening
2813 * gateways or networks. We also initialize the congestion/slow start
2814 * window to be a single segment if the destination isn't local.
2815 * While looking at the routing entry, we also initialize other path-dependent
2816 * parameters from pre-set or cached values in the routing entry.
2818 * Also take into account the space needed for options that we
2819 * send regularly. Make maxseg shorter by that amount to assure
2820 * that we can send maxseg amount of data even when the options
2821 * are present. Store the upper limit of the length of options plus
2824 * NOTE that this routine is only called when we process an incoming
2825 * segment, for outgoing segments only tcp_mssopt is called.
2827 * In case of T/TCP, we call this routine during implicit connection
2828 * setup as well (offer = -1), to initialize maxseg from the cached
2832 tcp_mss(struct tcpcb *tp, int offer)
2838 struct inpcb *inp = tp->t_inpcb;
2840 struct rmxp_tao *taop;
2841 int origoffer = offer;
2843 boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) ? TRUE : FALSE);
2844 size_t min_protoh = isipv6 ?
2845 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
2846 sizeof(struct tcpiphdr);
2848 const boolean_t isipv6 = FALSE;
2849 const size_t min_protoh = sizeof(struct tcpiphdr);
2853 rt = tcp_rtlookup6(&inp->inp_inc);
2855 rt = tcp_rtlookup(&inp->inp_inc);
2857 tp->t_maxopd = tp->t_maxseg =
2858 (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
2862 so = inp->inp_socket;
2864 taop = rmx_taop(rt->rt_rmx);
2866 * Offer == -1 means that we didn't receive SYN yet,
2867 * use cached value in that case;
2870 offer = taop->tao_mssopt;
2872 * Offer == 0 means that there was no MSS on the SYN segment,
2873 * in this case we use tcp_mssdflt.
2876 offer = (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
2879 * Sanity check: make sure that maxopd will be large
2880 * enough to allow some data on segments even is the
2881 * all the option space is used (40bytes). Otherwise
2882 * funny things may happen in tcp_output.
2884 offer = max(offer, 64);
2885 taop->tao_mssopt = offer;
2888 * While we're here, check if there's an initial rtt
2889 * or rttvar. Convert from the route-table units
2890 * to scaled multiples of the slow timeout timer.
2892 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
2894 * XXX the lock bit for RTT indicates that the value
2895 * is also a minimum value; this is subject to time.
2897 if (rt->rt_rmx.rmx_locks & RTV_RTT)
2898 tp->t_rttmin = rtt / (RTM_RTTUNIT / hz);
2899 tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE));
2900 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
2901 tcpstat.tcps_usedrtt++;
2902 if (rt->rt_rmx.rmx_rttvar) {
2903 tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
2904 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE));
2905 tcpstat.tcps_usedrttvar++;
2907 /* default variation is +- 1 rtt */
2909 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
2911 TCPT_RANGESET(tp->t_rxtcur,
2912 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
2913 tp->t_rttmin, TCPTV_REXMTMAX);
2916 * if there's an mtu associated with the route, use it
2917 * else, use the link mtu.
2919 if (rt->rt_rmx.rmx_mtu)
2920 mss = rt->rt_rmx.rmx_mtu - min_protoh;
2923 mss = ND_IFINFO(rt->rt_ifp)->linkmtu - min_protoh;
2924 if (!in6_localaddr(&inp->in6p_faddr))
2925 mss = min(mss, tcp_v6mssdflt);
2927 mss = ifp->if_mtu - min_protoh;
2928 if (!in_localaddr(inp->inp_faddr))
2929 mss = min(mss, tcp_mssdflt);
2932 mss = min(mss, offer);
2934 * maxopd stores the maximum length of data AND options
2935 * in a segment; maxseg is the amount of data in a normal
2936 * segment. We need to store this value (maxopd) apart
2937 * from maxseg, because now every segment carries options
2938 * and thus we normally have somewhat less data in segments.
2943 * In case of T/TCP, origoffer==-1 indicates, that no segments
2944 * were received yet. In this case we just guess, otherwise
2945 * we do the same as before T/TCP.
2947 if ((tp->t_flags & (TF_REQ_TSTMP | TF_NOOPT)) == TF_REQ_TSTMP &&
2949 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
2950 mss -= TCPOLEN_TSTAMP_APPA;
2951 if ((tp->t_flags & (TF_REQ_CC | TF_NOOPT)) == TF_REQ_CC &&
2953 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC))
2954 mss -= TCPOLEN_CC_APPA;
2956 #if (MCLBYTES & (MCLBYTES - 1)) == 0
2958 mss &= ~(MCLBYTES-1);
2961 mss = mss / MCLBYTES * MCLBYTES;
2964 * If there's a pipesize, change the socket buffer
2965 * to that size. Make the socket buffers an integral
2966 * number of mss units; if the mss is larger than
2967 * the socket buffer, decrease the mss.
2970 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0)
2972 bufsize = so->so_snd.sb_hiwat;
2976 bufsize = roundup(bufsize, mss);
2977 if (bufsize > sb_max)
2979 if (bufsize > so->so_snd.sb_hiwat)
2980 sbreserve(&so->so_snd, bufsize, so, NULL);
2985 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0)
2987 bufsize = so->so_rcv.sb_hiwat;
2988 if (bufsize > mss) {
2989 bufsize = roundup(bufsize, mss);
2990 if (bufsize > sb_max)
2992 if (bufsize > so->so_rcv.sb_hiwat)
2993 sbreserve(&so->so_rcv, bufsize, so, NULL);
2997 * Set the slow-start flight size depending on whether this
2998 * is a local network or not.
3001 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380));
3005 if (rt->rt_rmx.rmx_ssthresh) {
3007 * There's some sort of gateway or interface
3008 * buffer limit on the path. Use this to set
3009 * the slow start threshhold, but set the
3010 * threshold to no less than 2*mss.
3012 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
3013 tcpstat.tcps_usedssthresh++;
3018 * Determine the MSS option to send on an outgoing SYN.
3021 tcp_mssopt(struct tcpcb *tp)
3026 ((tp->t_inpcb->inp_vflag & INP_IPV6) ? TRUE : FALSE);
3027 int min_protoh = isipv6 ?
3028 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
3029 sizeof(struct tcpiphdr);
3031 const boolean_t isipv6 = FALSE;
3032 const size_t min_protoh = sizeof(struct tcpiphdr);
3036 rt = tcp_rtlookup6(&tp->t_inpcb->inp_inc);
3038 rt = tcp_rtlookup(&tp->t_inpcb->inp_inc);
3040 return (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
3042 return (rt->rt_ifp->if_mtu - min_protoh);
3046 * When a partial ack arrives, force the retransmission of the
3047 * next unacknowledged segment. Do not exit Fast Recovery.
3049 * Implement the Slow-but-Steady variant of NewReno by restarting the
3050 * the retransmission timer. Turn it off here so it can be restarted
3051 * later in tcp_output().
3054 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th, int acked)
3056 tcp_seq old_snd_nxt = tp->snd_nxt;
3057 u_long ocwnd = tp->snd_cwnd;
3059 callout_stop(tp->tt_rexmt);
3061 tp->snd_nxt = th->th_ack;
3062 /* Set snd_cwnd to one segment beyond acknowledged offset. */
3063 tp->snd_cwnd = tp->t_maxseg;
3064 tp->t_flags |= TF_ACKNOW;
3066 if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
3067 tp->snd_nxt = old_snd_nxt;
3068 /* partial window deflation */
3069 tp->snd_cwnd = ocwnd - acked + tp->t_maxseg;
3073 * In contrast to the Slow-but-Steady NewReno variant,
3074 * we do not reset the retransmission timer for SACK retransmissions,
3075 * except when retransmitting snd_una.
3078 tcp_sack_rexmt(struct tcpcb *tp, struct tcphdr *th)
3080 uint32_t pipe, seglen;
3083 tcp_seq old_snd_nxt = tp->snd_nxt;
3084 u_long ocwnd = tp->snd_cwnd;
3085 int nseg = 0; /* consecutive new segments */
3086 #define MAXBURST 4 /* limit burst of new packets on partial ack */
3089 pipe = tcp_sack_compute_pipe(tp);
3090 while ((tcp_seq_diff_t)(ocwnd - pipe) >= (tcp_seq_diff_t)tp->t_maxseg &&
3091 (!tcp_do_smartsack || nseg < MAXBURST) &&
3092 tcp_sack_nextseg(tp, &nextrexmt, &seglen, &lostdup)) {
3094 tcp_seq old_snd_max;
3097 if (nextrexmt == tp->snd_max) ++nseg;
3098 tp->snd_nxt = nextrexmt;
3099 tp->snd_cwnd = nextrexmt - tp->snd_una + seglen;
3100 old_snd_max = tp->snd_max;
3101 if (nextrexmt == tp->snd_una)
3102 callout_stop(tp->tt_rexmt);
3103 error = tcp_output(tp);
3106 sent = tp->snd_nxt - nextrexmt;
3111 tcpstat.tcps_sndsackpack++;
3112 tcpstat.tcps_sndsackbyte += sent;
3113 if (SEQ_LT(nextrexmt, old_snd_max) &&
3114 SEQ_LT(tp->rexmt_high, tp->snd_nxt))
3115 tp->rexmt_high = seq_min(tp->snd_nxt, old_snd_max);
3117 if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
3118 tp->snd_nxt = old_snd_nxt;
3119 tp->snd_cwnd = ocwnd;