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 1.63 2006/09/05 00:55:48 dillon 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 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 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)");
195 static int tcp_do_abc = 1;
196 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc, CTLFLAG_RW,
198 "TCP Appropriate Byte Counting (RFC 3465)");
201 * Define as tunable for easy testing with SACK on and off.
202 * Warning: do not change setting in the middle of an existing active TCP flow,
203 * else strange things might happen to that flow.
206 SYSCTL_INT(_net_inet_tcp, OID_AUTO, sack, CTLFLAG_RW,
207 &tcp_do_sack, 0, "Enable SACK Algorithms");
209 int tcp_do_smartsack = 1;
210 SYSCTL_INT(_net_inet_tcp, OID_AUTO, smartsack, CTLFLAG_RW,
211 &tcp_do_smartsack, 0, "Enable Smart SACK Algorithms");
213 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0,
214 "TCP Segment Reassembly Queue");
216 int tcp_reass_maxseg = 0;
217 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RD,
218 &tcp_reass_maxseg, 0,
219 "Global maximum number of TCP Segments in Reassembly Queue");
221 int tcp_reass_qsize = 0;
222 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD,
224 "Global number of TCP Segments currently in Reassembly Queue");
226 static int tcp_reass_overflows = 0;
227 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD,
228 &tcp_reass_overflows, 0,
229 "Global number of TCP Segment Reassembly Queue Overflows");
231 static void tcp_dooptions(struct tcpopt *, u_char *, int, boolean_t);
232 static void tcp_pulloutofband(struct socket *,
233 struct tcphdr *, struct mbuf *, int);
234 static int tcp_reass(struct tcpcb *, struct tcphdr *, int *,
236 static void tcp_xmit_timer(struct tcpcb *, int);
237 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *, int);
238 static void tcp_sack_rexmt(struct tcpcb *, struct tcphdr *);
240 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
242 #define ND6_HINT(tp) \
244 if ((tp) && (tp)->t_inpcb && \
245 ((tp)->t_inpcb->inp_vflag & INP_IPV6) && \
246 (tp)->t_inpcb->in6p_route.ro_rt) \
247 nd6_nud_hint((tp)->t_inpcb->in6p_route.ro_rt, NULL, 0); \
254 * Indicate whether this ack should be delayed. We can delay the ack if
255 * - delayed acks are enabled and
256 * - there is no delayed ack timer in progress and
257 * - our last ack wasn't a 0-sized window. We never want to delay
258 * the ack that opens up a 0-sized window.
260 #define DELAY_ACK(tp) \
261 (tcp_delack_enabled && !callout_pending(tp->tt_delack) && \
262 !(tp->t_flags & TF_RXWIN0SENT))
264 #define acceptable_window_update(tp, th, tiwin) \
265 (SEQ_LT(tp->snd_wl1, th->th_seq) || \
266 (tp->snd_wl1 == th->th_seq && \
267 (SEQ_LT(tp->snd_wl2, th->th_ack) || \
268 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))
271 tcp_reass(struct tcpcb *tp, struct tcphdr *th, int *tlenp, struct mbuf *m)
274 struct tseg_qent *p = NULL;
275 struct tseg_qent *te;
276 struct socket *so = tp->t_inpcb->inp_socket;
280 * Call with th == NULL after become established to
281 * force pre-ESTABLISHED data up to user socket.
287 * Limit the number of segments in the reassembly queue to prevent
288 * holding on to too many segments (and thus running out of mbufs).
289 * Make sure to let the missing segment through which caused this
290 * queue. Always keep one global queue entry spare to be able to
291 * process the missing segment.
293 if (th->th_seq != tp->rcv_nxt &&
294 tcp_reass_qsize + 1 >= tcp_reass_maxseg) {
295 tcp_reass_overflows++;
296 tcpstat.tcps_rcvmemdrop++;
298 /* no SACK block to report */
299 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
303 /* Allocate a new queue entry. */
304 MALLOC(te, struct tseg_qent *, sizeof(struct tseg_qent), M_TSEGQ,
305 M_INTWAIT | M_NULLOK);
307 tcpstat.tcps_rcvmemdrop++;
309 /* no SACK block to report */
310 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
316 * Find a segment which begins after this one does.
318 LIST_FOREACH(q, &tp->t_segq, tqe_q) {
319 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
325 * If there is a preceding segment, it may provide some of
326 * our data already. If so, drop the data from the incoming
327 * segment. If it provides all of our data, drop us.
332 /* conversion to int (in i) handles seq wraparound */
333 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
334 if (i > 0) { /* overlaps preceding segment */
335 tp->t_flags |= (TF_DUPSEG | TF_ENCLOSESEG);
336 /* enclosing block starts w/ preceding segment */
337 tp->encloseblk.rblk_start = p->tqe_th->th_seq;
339 /* preceding encloses incoming segment */
340 tp->encloseblk.rblk_end = p->tqe_th->th_seq +
342 tcpstat.tcps_rcvduppack++;
343 tcpstat.tcps_rcvdupbyte += *tlenp;
348 * Try to present any queued data
349 * at the left window edge to the user.
350 * This is needed after the 3-WHS
353 goto present; /* ??? */
358 /* incoming segment end is enclosing block end */
359 tp->encloseblk.rblk_end = th->th_seq + *tlenp +
360 ((th->th_flags & TH_FIN) != 0);
361 /* trim end of reported D-SACK block */
362 tp->reportblk.rblk_end = th->th_seq;
365 tcpstat.tcps_rcvoopack++;
366 tcpstat.tcps_rcvoobyte += *tlenp;
369 * While we overlap succeeding segments trim them or,
370 * if they are completely covered, dequeue them.
373 tcp_seq_diff_t i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
374 tcp_seq qend = q->tqe_th->th_seq + q->tqe_len;
375 struct tseg_qent *nq;
379 if (!(tp->t_flags & TF_DUPSEG)) { /* first time through */
380 tp->t_flags |= (TF_DUPSEG | TF_ENCLOSESEG);
381 tp->encloseblk = tp->reportblk;
382 /* report trailing duplicate D-SACK segment */
383 tp->reportblk.rblk_start = q->tqe_th->th_seq;
385 if ((tp->t_flags & TF_ENCLOSESEG) &&
386 SEQ_GT(qend, tp->encloseblk.rblk_end)) {
387 /* extend enclosing block if one exists */
388 tp->encloseblk.rblk_end = qend;
390 if (i < q->tqe_len) {
391 q->tqe_th->th_seq += i;
397 nq = LIST_NEXT(q, tqe_q);
398 LIST_REMOVE(q, tqe_q);
405 /* Insert the new segment queue entry into place. */
408 te->tqe_len = *tlenp;
410 /* check if can coalesce with following segment */
411 if (q != NULL && (th->th_seq + *tlenp == q->tqe_th->th_seq)) {
412 tcp_seq tend = te->tqe_th->th_seq + te->tqe_len;
414 te->tqe_len += q->tqe_len;
415 if (q->tqe_th->th_flags & TH_FIN)
416 te->tqe_th->th_flags |= TH_FIN;
417 m_cat(te->tqe_m, q->tqe_m);
418 tp->encloseblk.rblk_end = tend;
420 * When not reporting a duplicate segment, use
421 * the larger enclosing block as the SACK block.
423 if (!(tp->t_flags & TF_DUPSEG))
424 tp->reportblk.rblk_end = tend;
425 LIST_REMOVE(q, tqe_q);
431 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
433 /* check if can coalesce with preceding segment */
434 if (p->tqe_th->th_seq + p->tqe_len == th->th_seq) {
435 p->tqe_len += te->tqe_len;
436 m_cat(p->tqe_m, te->tqe_m);
437 tp->encloseblk.rblk_start = p->tqe_th->th_seq;
439 * When not reporting a duplicate segment, use
440 * the larger enclosing block as the SACK block.
442 if (!(tp->t_flags & TF_DUPSEG))
443 tp->reportblk.rblk_start = p->tqe_th->th_seq;
447 LIST_INSERT_AFTER(p, te, tqe_q);
452 * Present data to user, advancing rcv_nxt through
453 * completed sequence space.
455 if (!TCPS_HAVEESTABLISHED(tp->t_state))
457 q = LIST_FIRST(&tp->t_segq);
458 if (q == NULL || q->tqe_th->th_seq != tp->rcv_nxt)
460 tp->rcv_nxt += q->tqe_len;
461 if (!(tp->t_flags & TF_DUPSEG)) {
462 /* no SACK block to report since ACK advanced */
463 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
465 /* no enclosing block to report since ACK advanced */
466 tp->t_flags &= ~TF_ENCLOSESEG;
467 flags = q->tqe_th->th_flags & TH_FIN;
468 LIST_REMOVE(q, tqe_q);
469 KASSERT(LIST_EMPTY(&tp->t_segq) ||
470 LIST_FIRST(&tp->t_segq)->tqe_th->th_seq != tp->rcv_nxt,
471 ("segment not coalesced"));
472 if (so->so_state & SS_CANTRCVMORE)
475 sbappendstream(&so->so_rcv, q->tqe_m);
484 * TCP input routine, follows pages 65-76 of the
485 * protocol specification dated September, 1981 very closely.
489 tcp6_input(struct mbuf **mp, int *offp, int proto)
491 struct mbuf *m = *mp;
492 struct in6_ifaddr *ia6;
494 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
497 * draft-itojun-ipv6-tcp-to-anycast
498 * better place to put this in?
500 ia6 = ip6_getdstifaddr(m);
501 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
504 ip6 = mtod(m, struct ip6_hdr *);
505 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
506 offsetof(struct ip6_hdr, ip6_dst));
507 return (IPPROTO_DONE);
510 tcp_input(m, *offp, proto);
511 return (IPPROTO_DONE);
516 tcp_input(struct mbuf *m, ...)
521 struct ip *ip = NULL;
523 struct inpcb *inp = NULL;
528 struct tcpcb *tp = NULL;
530 struct socket *so = 0;
532 boolean_t ourfinisacked, needoutput = FALSE;
535 struct tcpopt to; /* options in this segment */
536 struct rmxp_tao *taop; /* pointer to our TAO cache entry */
537 struct rmxp_tao tao_noncached; /* in case there's no cached entry */
538 struct sockaddr_in *next_hop = NULL;
539 int rstreason; /* For badport_bandlim accounting purposes */
541 struct ip6_hdr *ip6 = NULL;
545 const boolean_t isipv6 = FALSE;
552 off0 = __va_arg(ap, int);
553 proto = __va_arg(ap, int);
556 tcpstat.tcps_rcvtotal++;
558 /* Grab info from and strip MT_TAG mbufs prepended to the chain. */
559 while (m->m_type == MT_TAG) {
560 if (m->_m_tag_id == PACKET_TAG_IPFORWARD)
561 next_hop = (struct sockaddr_in *)m->m_hdr.mh_data;
566 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? TRUE : FALSE;
570 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */
571 ip6 = mtod(m, struct ip6_hdr *);
572 tlen = (sizeof *ip6) + ntohs(ip6->ip6_plen) - off0;
573 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) {
574 tcpstat.tcps_rcvbadsum++;
577 th = (struct tcphdr *)((caddr_t)ip6 + off0);
580 * Be proactive about unspecified IPv6 address in source.
581 * As we use all-zero to indicate unbounded/unconnected pcb,
582 * unspecified IPv6 address can be used to confuse us.
584 * Note that packets with unspecified IPv6 destination is
585 * already dropped in ip6_input.
587 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
593 * Get IP and TCP header together in first mbuf.
594 * Note: IP leaves IP header in first mbuf.
596 if (off0 > sizeof(struct ip)) {
598 off0 = sizeof(struct ip);
600 /* already checked and pulled up in ip_demux() */
601 KASSERT(m->m_len >= sizeof(struct tcpiphdr),
602 ("TCP header not in one mbuf: m->m_len %d", m->m_len));
603 ip = mtod(m, struct ip *);
604 ipov = (struct ipovly *)ip;
605 th = (struct tcphdr *)((caddr_t)ip + off0);
608 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
609 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
610 th->th_sum = m->m_pkthdr.csum_data;
612 th->th_sum = in_pseudo(ip->ip_src.s_addr,
614 htonl(m->m_pkthdr.csum_data +
617 th->th_sum ^= 0xffff;
620 * Checksum extended TCP header and data.
622 len = sizeof(struct ip) + tlen;
623 bzero(ipov->ih_x1, sizeof ipov->ih_x1);
624 ipov->ih_len = (u_short)tlen;
625 ipov->ih_len = htons(ipov->ih_len);
626 th->th_sum = in_cksum(m, len);
629 tcpstat.tcps_rcvbadsum++;
633 /* Re-initialization for later version check */
634 ip->ip_v = IPVERSION;
639 * Check that TCP offset makes sense,
640 * pull out TCP options and adjust length. XXX
642 off = th->th_off << 2;
643 /* already checked and pulled up in ip_demux() */
644 KASSERT(off >= sizeof(struct tcphdr) && off <= tlen,
645 ("bad TCP data offset %d (tlen %d)", off, tlen));
646 tlen -= off; /* tlen is used instead of ti->ti_len */
647 if (off > sizeof(struct tcphdr)) {
649 IP6_EXTHDR_CHECK(m, off0, off, );
650 ip6 = mtod(m, struct ip6_hdr *);
651 th = (struct tcphdr *)((caddr_t)ip6 + off0);
653 /* already pulled up in ip_demux() */
654 KASSERT(m->m_len >= sizeof(struct ip) + off,
655 ("TCP header and options not in one mbuf: "
656 "m_len %d, off %d", m->m_len, off));
658 optlen = off - sizeof(struct tcphdr);
659 optp = (u_char *)(th + 1);
661 thflags = th->th_flags;
663 #ifdef TCP_DROP_SYNFIN
665 * If the drop_synfin option is enabled, drop all packets with
666 * both the SYN and FIN bits set. This prevents e.g. nmap from
667 * identifying the TCP/IP stack.
669 * This is a violation of the TCP specification.
671 if (drop_synfin && (thflags & (TH_SYN | TH_FIN)) == (TH_SYN | TH_FIN))
676 * Convert TCP protocol specific fields to host format.
678 th->th_seq = ntohl(th->th_seq);
679 th->th_ack = ntohl(th->th_ack);
680 th->th_win = ntohs(th->th_win);
681 th->th_urp = ntohs(th->th_urp);
684 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options,
685 * until after ip6_savecontrol() is called and before other functions
686 * which don't want those proto headers.
687 * Because ip6_savecontrol() is going to parse the mbuf to
688 * search for data to be passed up to user-land, it wants mbuf
689 * parameters to be unchanged.
690 * XXX: the call of ip6_savecontrol() has been obsoleted based on
691 * latest version of the advanced API (20020110).
693 drop_hdrlen = off0 + off;
696 * Locate pcb for segment.
699 /* IPFIREWALL_FORWARD section */
700 if (next_hop != NULL && !isipv6) { /* IPv6 support is not there yet */
702 * Transparently forwarded. Pretend to be the destination.
703 * already got one like this?
705 cpu = mycpu->gd_cpuid;
706 inp = in_pcblookup_hash(&tcbinfo[cpu],
707 ip->ip_src, th->th_sport,
708 ip->ip_dst, th->th_dport,
709 0, m->m_pkthdr.rcvif);
712 * It's new. Try to find the ambushing socket.
716 * The rest of the ipfw code stores the port in
718 * (The IP address is still in network order.)
720 in_port_t dport = next_hop->sin_port ?
721 htons(next_hop->sin_port) :
724 cpu = tcp_addrcpu(ip->ip_src.s_addr, th->th_sport,
725 next_hop->sin_addr.s_addr, dport);
726 inp = in_pcblookup_hash(&tcbinfo[cpu],
727 ip->ip_src, th->th_sport,
728 next_hop->sin_addr, dport,
729 1, m->m_pkthdr.rcvif);
733 inp = in6_pcblookup_hash(&tcbinfo[0],
734 &ip6->ip6_src, th->th_sport,
735 &ip6->ip6_dst, th->th_dport,
736 1, m->m_pkthdr.rcvif);
738 cpu = mycpu->gd_cpuid;
739 inp = in_pcblookup_hash(&tcbinfo[cpu],
740 ip->ip_src, th->th_sport,
741 ip->ip_dst, th->th_dport,
742 1, m->m_pkthdr.rcvif);
747 * If the state is CLOSED (i.e., TCB does not exist) then
748 * all data in the incoming segment is discarded.
749 * If the TCB exists but is in CLOSED state, it is embryonic,
750 * but should either do a listen or a connect soon.
755 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2];
757 char dbuf[sizeof "aaa.bbb.ccc.ddd"];
758 char sbuf[sizeof "aaa.bbb.ccc.ddd"];
762 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst));
765 strcat(sbuf, ip6_sprintf(&ip6->ip6_src));
768 strcpy(dbuf, inet_ntoa(ip->ip_dst));
769 strcpy(sbuf, inet_ntoa(ip->ip_src));
771 switch (log_in_vain) {
773 if (!(thflags & TH_SYN))
777 "Connection attempt to TCP %s:%d "
778 "from %s:%d flags:0x%02x\n",
779 dbuf, ntohs(th->th_dport), sbuf,
780 ntohs(th->th_sport), thflags);
789 if (thflags & TH_SYN)
798 rstreason = BANDLIM_RST_CLOSEDPORT;
804 if (ipsec6_in_reject_so(m, inp->inp_socket)) {
805 ipsec6stat.in_polvio++;
809 if (ipsec4_in_reject_so(m, inp->inp_socket)) {
810 ipsecstat.in_polvio++;
817 if (ipsec6_in_reject(m, inp))
820 if (ipsec4_in_reject(m, inp))
827 rstreason = BANDLIM_RST_CLOSEDPORT;
830 if (tp->t_state <= TCPS_CLOSED)
833 /* Unscale the window into a 32-bit value. */
834 if (!(thflags & TH_SYN))
835 tiwin = th->th_win << tp->snd_scale;
839 so = inp->inp_socket;
842 if (so->so_options & SO_DEBUG) {
843 ostate = tp->t_state;
845 bcopy(ip6, tcp_saveipgen, sizeof(*ip6));
847 bcopy(ip, tcp_saveipgen, sizeof(*ip));
852 bzero(&to, sizeof to);
854 if (so->so_options & SO_ACCEPTCONN) {
855 struct in_conninfo inc;
858 inc.inc_isipv6 = (isipv6 == TRUE);
861 inc.inc6_faddr = ip6->ip6_src;
862 inc.inc6_laddr = ip6->ip6_dst;
863 inc.inc6_route.ro_rt = NULL; /* XXX */
865 inc.inc_faddr = ip->ip_src;
866 inc.inc_laddr = ip->ip_dst;
867 inc.inc_route.ro_rt = NULL; /* XXX */
869 inc.inc_fport = th->th_sport;
870 inc.inc_lport = th->th_dport;
873 * If the state is LISTEN then ignore segment if it contains
874 * a RST. If the segment contains an ACK then it is bad and
875 * send a RST. If it does not contain a SYN then it is not
876 * interesting; drop it.
878 * If the state is SYN_RECEIVED (syncache) and seg contains
879 * an ACK, but not for our SYN/ACK, send a RST. If the seg
880 * contains a RST, check the sequence number to see if it
881 * is a valid reset segment.
883 if ((thflags & (TH_RST | TH_ACK | TH_SYN)) != TH_SYN) {
884 if ((thflags & (TH_RST | TH_ACK | TH_SYN)) == TH_ACK) {
885 if (!syncache_expand(&inc, th, &so, m)) {
887 * No syncache entry, or ACK was not
888 * for our SYN/ACK. Send a RST.
890 tcpstat.tcps_badsyn++;
891 rstreason = BANDLIM_RST_OPENPORT;
896 * Could not complete 3-way handshake,
897 * connection is being closed down, and
898 * syncache will free mbuf.
902 * Socket is created in state SYN_RECEIVED.
903 * Continue processing segment.
908 * This is what would have happened in
909 * tcp_output() when the SYN,ACK was sent.
911 tp->snd_up = tp->snd_una;
912 tp->snd_max = tp->snd_nxt = tp->iss + 1;
913 tp->last_ack_sent = tp->rcv_nxt;
915 * XXX possible bug - it doesn't appear that tp->snd_wnd is unscaled
916 * until the _second_ ACK is received:
917 * rcv SYN (set wscale opts) --> send SYN/ACK, set snd_wnd = window.
918 * rcv ACK, calculate tiwin --> process SYN_RECEIVED, determine wscale,
919 * move to ESTAB, set snd_wnd to tiwin.
921 tp->snd_wnd = tiwin; /* unscaled */
924 if (thflags & TH_RST) {
925 syncache_chkrst(&inc, th);
928 if (thflags & TH_ACK) {
929 syncache_badack(&inc);
930 tcpstat.tcps_badsyn++;
931 rstreason = BANDLIM_RST_OPENPORT;
938 * Segment's flags are (SYN) or (SYN | FIN).
942 * If deprecated address is forbidden,
943 * we do not accept SYN to deprecated interface
944 * address to prevent any new inbound connection from
945 * getting established.
946 * When we do not accept SYN, we send a TCP RST,
947 * with deprecated source address (instead of dropping
948 * it). We compromise it as it is much better for peer
949 * to send a RST, and RST will be the final packet
952 * If we do not forbid deprecated addresses, we accept
953 * the SYN packet. RFC2462 does not suggest dropping
955 * If we decipher RFC2462 5.5.4, it says like this:
956 * 1. use of deprecated addr with existing
957 * communication is okay - "SHOULD continue to be
959 * 2. use of it with new communication:
960 * (2a) "SHOULD NOT be used if alternate address
961 * with sufficient scope is available"
962 * (2b) nothing mentioned otherwise.
963 * Here we fall into (2b) case as we have no choice in
964 * our source address selection - we must obey the peer.
966 * The wording in RFC2462 is confusing, and there are
967 * multiple description text for deprecated address
968 * handling - worse, they are not exactly the same.
969 * I believe 5.5.4 is the best one, so we follow 5.5.4.
971 if (isipv6 && !ip6_use_deprecated) {
972 struct in6_ifaddr *ia6;
974 if ((ia6 = ip6_getdstifaddr(m)) &&
975 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
977 rstreason = BANDLIM_RST_OPENPORT;
983 * If it is from this socket, drop it, it must be forged.
984 * Don't bother responding if the destination was a broadcast.
986 if (th->th_dport == th->th_sport) {
988 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
992 if (ip->ip_dst.s_addr == ip->ip_src.s_addr)
997 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
999 * Note that it is quite possible to receive unicast
1000 * link-layer packets with a broadcast IP address. Use
1001 * in_broadcast() to find them.
1003 if (m->m_flags & (M_BCAST | M_MCAST))
1006 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1007 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
1010 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1011 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1012 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1013 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
1017 * SYN appears to be valid; create compressed TCP state
1018 * for syncache, or perform t/tcp connection.
1020 if (so->so_qlen <= so->so_qlimit) {
1021 tcp_dooptions(&to, optp, optlen, TRUE);
1022 if (!syncache_add(&inc, &to, th, &so, m))
1026 * Entry added to syncache, mbuf used to
1027 * send SYN,ACK packet.
1031 * Segment passed TAO tests.
1034 tp = intotcpcb(inp);
1035 tp->snd_wnd = tiwin;
1036 tp->t_starttime = ticks;
1037 tp->t_state = TCPS_ESTABLISHED;
1040 * If there is a FIN, or if there is data and the
1041 * connection is local, then delay SYN,ACK(SYN) in
1042 * the hope of piggy-backing it on a response
1043 * segment. Otherwise must send ACK now in case
1044 * the other side is slow starting.
1046 if (DELAY_ACK(tp) &&
1047 ((thflags & TH_FIN) ||
1049 ((isipv6 && in6_localaddr(&inp->in6p_faddr)) ||
1050 (!isipv6 && in_localaddr(inp->inp_faddr)))))) {
1051 callout_reset(tp->tt_delack, tcp_delacktime,
1052 tcp_timer_delack, tp);
1053 tp->t_flags |= TF_NEEDSYN;
1055 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
1057 tcpstat.tcps_connects++;
1065 /* should not happen - syncache should pick up these connections */
1066 KASSERT(tp->t_state != TCPS_LISTEN, ("tcp_input: TCPS_LISTEN state"));
1069 * Segment received on connection.
1070 * Reset idle time and keep-alive timer.
1072 tp->t_rcvtime = ticks;
1073 if (TCPS_HAVEESTABLISHED(tp->t_state))
1074 callout_reset(tp->tt_keep, tcp_keepidle, tcp_timer_keep, tp);
1078 * XXX this is tradtitional behavior, may need to be cleaned up.
1080 tcp_dooptions(&to, optp, optlen, (thflags & TH_SYN) != 0);
1081 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1082 if (to.to_flags & TOF_SCALE) {
1083 tp->t_flags |= TF_RCVD_SCALE;
1084 tp->requested_s_scale = to.to_requested_s_scale;
1086 if (to.to_flags & TOF_TS) {
1087 tp->t_flags |= TF_RCVD_TSTMP;
1088 tp->ts_recent = to.to_tsval;
1089 tp->ts_recent_age = ticks;
1091 if (to.to_flags & (TOF_CC | TOF_CCNEW))
1092 tp->t_flags |= TF_RCVD_CC;
1093 if (to.to_flags & TOF_MSS)
1094 tcp_mss(tp, to.to_mss);
1096 * Only set the TF_SACK_PERMITTED per-connection flag
1097 * if we got a SACK_PERMITTED option from the other side
1098 * and the global tcp_do_sack variable is true.
1100 if (tcp_do_sack && (to.to_flags & TOF_SACK_PERMITTED))
1101 tp->t_flags |= TF_SACK_PERMITTED;
1105 * Header prediction: check for the two common cases
1106 * of a uni-directional data xfer. If the packet has
1107 * no control flags, is in-sequence, the window didn't
1108 * change and we're not retransmitting, it's a
1109 * candidate. If the length is zero and the ack moved
1110 * forward, we're the sender side of the xfer. Just
1111 * free the data acked & wake any higher level process
1112 * that was blocked waiting for space. If the length
1113 * is non-zero and the ack didn't move, we're the
1114 * receiver side. If we're getting packets in-order
1115 * (the reassembly queue is empty), add the data to
1116 * the socket buffer and note that we need a delayed ack.
1117 * Make sure that the hidden state-flags are also off.
1118 * Since we check for TCPS_ESTABLISHED above, it can only
1121 if (tp->t_state == TCPS_ESTABLISHED &&
1122 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1123 !(tp->t_flags & (TF_NEEDSYN | TF_NEEDFIN)) &&
1124 (!(to.to_flags & TOF_TS) ||
1125 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) &&
1127 * Using the CC option is compulsory if once started:
1128 * the segment is OK if no T/TCP was negotiated or
1129 * if the segment has a CC option equal to CCrecv
1131 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) ||
1132 ((to.to_flags & TOF_CC) && to.to_cc == tp->cc_recv)) &&
1133 th->th_seq == tp->rcv_nxt &&
1134 tp->snd_nxt == tp->snd_max) {
1137 * If last ACK falls within this segment's sequence numbers,
1138 * record the timestamp.
1139 * NOTE that the test is modified according to the latest
1140 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1142 if ((to.to_flags & TOF_TS) &&
1143 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1144 tp->ts_recent_age = ticks;
1145 tp->ts_recent = to.to_tsval;
1149 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1150 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1151 tp->snd_cwnd >= tp->snd_wnd &&
1152 !IN_FASTRECOVERY(tp)) {
1154 * This is a pure ack for outstanding data.
1156 ++tcpstat.tcps_predack;
1158 * "bad retransmit" recovery
1160 * If Eifel detection applies, then
1161 * it is deterministic, so use it
1162 * unconditionally over the old heuristic.
1163 * Otherwise, fall back to the old heuristic.
1165 if (tcp_do_eifel_detect &&
1166 (to.to_flags & TOF_TS) && to.to_tsecr &&
1167 (tp->t_flags & TF_FIRSTACCACK)) {
1168 /* Eifel detection applicable. */
1169 if (to.to_tsecr < tp->t_rexmtTS) {
1170 tcp_revert_congestion_state(tp);
1171 ++tcpstat.tcps_eifeldetected;
1173 } else if (tp->t_rxtshift == 1 &&
1174 ticks < tp->t_badrxtwin) {
1175 tcp_revert_congestion_state(tp);
1176 ++tcpstat.tcps_rttdetected;
1178 tp->t_flags &= ~(TF_FIRSTACCACK |
1179 TF_FASTREXMT | TF_EARLYREXMT);
1181 * Recalculate the retransmit timer / rtt.
1183 * Some machines (certain windows boxes)
1184 * send broken timestamp replies during the
1185 * SYN+ACK phase, ignore timestamps of 0.
1187 if ((to.to_flags & TOF_TS) && to.to_tsecr) {
1189 ticks - to.to_tsecr + 1);
1190 } else if (tp->t_rtttime &&
1191 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1193 ticks - tp->t_rtttime);
1195 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1196 acked = th->th_ack - tp->snd_una;
1197 tcpstat.tcps_rcvackpack++;
1198 tcpstat.tcps_rcvackbyte += acked;
1199 sbdrop(&so->so_snd, acked);
1200 tp->snd_recover = th->th_ack - 1;
1201 tp->snd_una = th->th_ack;
1204 * Update window information.
1206 if (tiwin != tp->snd_wnd &&
1207 acceptable_window_update(tp, th, tiwin)) {
1208 /* keep track of pure window updates */
1209 if (tp->snd_wl2 == th->th_ack &&
1210 tiwin > tp->snd_wnd)
1211 tcpstat.tcps_rcvwinupd++;
1212 tp->snd_wnd = tiwin;
1213 tp->snd_wl1 = th->th_seq;
1214 tp->snd_wl2 = th->th_ack;
1215 if (tp->snd_wnd > tp->max_sndwnd)
1216 tp->max_sndwnd = tp->snd_wnd;
1219 ND6_HINT(tp); /* some progress has been done */
1221 * If all outstanding data are acked, stop
1222 * retransmit timer, otherwise restart timer
1223 * using current (possibly backed-off) value.
1224 * If process is waiting for space,
1225 * wakeup/selwakeup/signal. If data
1226 * are ready to send, let tcp_output
1227 * decide between more output or persist.
1229 if (tp->snd_una == tp->snd_max)
1230 callout_stop(tp->tt_rexmt);
1231 else if (!callout_active(tp->tt_persist))
1232 callout_reset(tp->tt_rexmt,
1234 tcp_timer_rexmt, tp);
1236 if (so->so_snd.sb_cc > 0)
1240 } else if (tiwin == tp->snd_wnd &&
1241 th->th_ack == tp->snd_una &&
1242 LIST_EMPTY(&tp->t_segq) &&
1243 tlen <= sbspace(&so->so_rcv)) {
1245 * This is a pure, in-sequence data packet
1246 * with nothing on the reassembly queue and
1247 * we have enough buffer space to take it.
1249 ++tcpstat.tcps_preddat;
1250 tp->rcv_nxt += tlen;
1251 tcpstat.tcps_rcvpack++;
1252 tcpstat.tcps_rcvbyte += tlen;
1253 ND6_HINT(tp); /* some progress has been done */
1255 * Add data to socket buffer.
1257 if (so->so_state & SS_CANTRCVMORE) {
1260 m_adj(m, drop_hdrlen); /* delayed header drop */
1261 sbappendstream(&so->so_rcv, m);
1265 * This code is responsible for most of the ACKs
1266 * the TCP stack sends back after receiving a data
1267 * packet. Note that the DELAY_ACK check fails if
1268 * the delack timer is already running, which results
1269 * in an ack being sent every other packet (which is
1272 * We then further aggregate acks by not actually
1273 * sending one until the protocol thread has completed
1274 * processing the current backlog of packets. This
1275 * does not delay the ack any further, but allows us
1276 * to take advantage of the packet aggregation that
1277 * high speed NICs do (usually blocks of 8-10 packets)
1278 * to send a single ack rather then four or five acks,
1279 * greatly reducing the ack rate, the return channel
1280 * bandwidth, and the protocol overhead on both ends.
1282 * Since this also has the effect of slowing down
1283 * the exponential slow-start ramp-up, systems with
1284 * very large bandwidth-delay products might want
1285 * to turn the feature off.
1287 if (DELAY_ACK(tp)) {
1288 callout_reset(tp->tt_delack, tcp_delacktime,
1289 tcp_timer_delack, tp);
1290 } else if (tcp_aggregate_acks) {
1291 tp->t_flags |= TF_ACKNOW;
1292 if (!(tp->t_flags & TF_ONOUTPUTQ)) {
1293 tp->t_flags |= TF_ONOUTPUTQ;
1294 tp->tt_cpu = mycpu->gd_cpuid;
1296 &tcpcbackq[tp->tt_cpu],
1300 tp->t_flags |= TF_ACKNOW;
1308 * Calculate amount of space in receive window,
1309 * and then do TCP input processing.
1310 * Receive window is amount of space in rcv queue,
1311 * but not less than advertised window.
1313 recvwin = sbspace(&so->so_rcv);
1316 tp->rcv_wnd = imax(recvwin, (int)(tp->rcv_adv - tp->rcv_nxt));
1318 switch (tp->t_state) {
1320 * If the state is SYN_RECEIVED:
1321 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1323 case TCPS_SYN_RECEIVED:
1324 if ((thflags & TH_ACK) &&
1325 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1326 SEQ_GT(th->th_ack, tp->snd_max))) {
1327 rstreason = BANDLIM_RST_OPENPORT;
1333 * If the state is SYN_SENT:
1334 * if seg contains an ACK, but not for our SYN, drop the input.
1335 * if seg contains a RST, then drop the connection.
1336 * if seg does not contain SYN, then drop it.
1337 * Otherwise this is an acceptable SYN segment
1338 * initialize tp->rcv_nxt and tp->irs
1339 * if seg contains ack then advance tp->snd_una
1340 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1341 * arrange for segment to be acked (eventually)
1342 * continue processing rest of data/controls, beginning with URG
1345 if ((taop = tcp_gettaocache(&inp->inp_inc)) == NULL) {
1346 taop = &tao_noncached;
1347 bzero(taop, sizeof *taop);
1350 if ((thflags & TH_ACK) &&
1351 (SEQ_LEQ(th->th_ack, tp->iss) ||
1352 SEQ_GT(th->th_ack, tp->snd_max))) {
1354 * If we have a cached CCsent for the remote host,
1355 * hence we haven't just crashed and restarted,
1356 * do not send a RST. This may be a retransmission
1357 * from the other side after our earlier ACK was lost.
1358 * Our new SYN, when it arrives, will serve as the
1361 if (taop->tao_ccsent != 0)
1364 rstreason = BANDLIM_UNLIMITED;
1368 if (thflags & TH_RST) {
1369 if (thflags & TH_ACK)
1370 tp = tcp_drop(tp, ECONNREFUSED);
1373 if (!(thflags & TH_SYN))
1375 tp->snd_wnd = th->th_win; /* initial send window */
1376 tp->cc_recv = to.to_cc; /* foreign CC */
1378 tp->irs = th->th_seq;
1380 if (thflags & TH_ACK) {
1382 * Our SYN was acked. If segment contains CC.ECHO
1383 * option, check it to make sure this segment really
1384 * matches our SYN. If not, just drop it as old
1385 * duplicate, but send an RST if we're still playing
1386 * by the old rules. If no CC.ECHO option, make sure
1387 * we don't get fooled into using T/TCP.
1389 if (to.to_flags & TOF_CCECHO) {
1390 if (tp->cc_send != to.to_ccecho) {
1391 if (taop->tao_ccsent != 0)
1394 rstreason = BANDLIM_UNLIMITED;
1399 tp->t_flags &= ~TF_RCVD_CC;
1400 tcpstat.tcps_connects++;
1402 /* Do window scaling on this connection? */
1403 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
1404 (TF_RCVD_SCALE | TF_REQ_SCALE)) {
1405 tp->snd_scale = tp->requested_s_scale;
1406 tp->rcv_scale = tp->request_r_scale;
1408 /* Segment is acceptable, update cache if undefined. */
1409 if (taop->tao_ccsent == 0)
1410 taop->tao_ccsent = to.to_ccecho;
1412 tp->rcv_adv += tp->rcv_wnd;
1413 tp->snd_una++; /* SYN is acked */
1414 callout_stop(tp->tt_rexmt);
1416 * If there's data, delay ACK; if there's also a FIN
1417 * ACKNOW will be turned on later.
1419 if (DELAY_ACK(tp) && tlen != 0)
1420 callout_reset(tp->tt_delack, tcp_delacktime,
1421 tcp_timer_delack, tp);
1423 tp->t_flags |= TF_ACKNOW;
1425 * Received <SYN,ACK> in SYN_SENT[*] state.
1427 * SYN_SENT --> ESTABLISHED
1428 * SYN_SENT* --> FIN_WAIT_1
1430 tp->t_starttime = ticks;
1431 if (tp->t_flags & TF_NEEDFIN) {
1432 tp->t_state = TCPS_FIN_WAIT_1;
1433 tp->t_flags &= ~TF_NEEDFIN;
1436 tp->t_state = TCPS_ESTABLISHED;
1437 callout_reset(tp->tt_keep, tcp_keepidle,
1438 tcp_timer_keep, tp);
1442 * Received initial SYN in SYN-SENT[*] state =>
1443 * simultaneous open. If segment contains CC option
1444 * and there is a cached CC, apply TAO test.
1445 * If it succeeds, connection is * half-synchronized.
1446 * Otherwise, do 3-way handshake:
1447 * SYN-SENT -> SYN-RECEIVED
1448 * SYN-SENT* -> SYN-RECEIVED*
1449 * If there was no CC option, clear cached CC value.
1451 tp->t_flags |= TF_ACKNOW;
1452 callout_stop(tp->tt_rexmt);
1453 if (to.to_flags & TOF_CC) {
1454 if (taop->tao_cc != 0 &&
1455 CC_GT(to.to_cc, taop->tao_cc)) {
1457 * update cache and make transition:
1458 * SYN-SENT -> ESTABLISHED*
1459 * SYN-SENT* -> FIN-WAIT-1*
1461 taop->tao_cc = to.to_cc;
1462 tp->t_starttime = ticks;
1463 if (tp->t_flags & TF_NEEDFIN) {
1464 tp->t_state = TCPS_FIN_WAIT_1;
1465 tp->t_flags &= ~TF_NEEDFIN;
1467 tp->t_state = TCPS_ESTABLISHED;
1468 callout_reset(tp->tt_keep,
1473 tp->t_flags |= TF_NEEDSYN;
1475 tp->t_state = TCPS_SYN_RECEIVED;
1477 /* CC.NEW or no option => invalidate cache */
1479 tp->t_state = TCPS_SYN_RECEIVED;
1485 * Advance th->th_seq to correspond to first data byte.
1486 * If data, trim to stay within window,
1487 * dropping FIN if necessary.
1490 if (tlen > tp->rcv_wnd) {
1491 todrop = tlen - tp->rcv_wnd;
1495 tcpstat.tcps_rcvpackafterwin++;
1496 tcpstat.tcps_rcvbyteafterwin += todrop;
1498 tp->snd_wl1 = th->th_seq - 1;
1499 tp->rcv_up = th->th_seq;
1501 * Client side of transaction: already sent SYN and data.
1502 * If the remote host used T/TCP to validate the SYN,
1503 * our data will be ACK'd; if so, enter normal data segment
1504 * processing in the middle of step 5, ack processing.
1505 * Otherwise, goto step 6.
1507 if (thflags & TH_ACK)
1513 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
1514 * if segment contains a SYN and CC [not CC.NEW] option:
1515 * if state == TIME_WAIT and connection duration > MSL,
1516 * drop packet and send RST;
1518 * if SEG.CC > CCrecv then is new SYN, and can implicitly
1519 * ack the FIN (and data) in retransmission queue.
1520 * Complete close and delete TCPCB. Then reprocess
1521 * segment, hoping to find new TCPCB in LISTEN state;
1523 * else must be old SYN; drop it.
1524 * else do normal processing.
1528 case TCPS_TIME_WAIT:
1529 if ((thflags & TH_SYN) &&
1530 (to.to_flags & TOF_CC) && tp->cc_recv != 0) {
1531 if (tp->t_state == TCPS_TIME_WAIT &&
1532 (ticks - tp->t_starttime) > tcp_msl) {
1533 rstreason = BANDLIM_UNLIMITED;
1536 if (CC_GT(to.to_cc, tp->cc_recv)) {
1543 break; /* continue normal processing */
1547 * States other than LISTEN or SYN_SENT.
1548 * First check the RST flag and sequence number since reset segments
1549 * are exempt from the timestamp and connection count tests. This
1550 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
1551 * below which allowed reset segments in half the sequence space
1552 * to fall though and be processed (which gives forged reset
1553 * segments with a random sequence number a 50 percent chance of
1554 * killing a connection).
1555 * Then check timestamp, if present.
1556 * Then check the connection count, if present.
1557 * Then check that at least some bytes of segment are within
1558 * receive window. If segment begins before rcv_nxt,
1559 * drop leading data (and SYN); if nothing left, just ack.
1562 * If the RST bit is set, check the sequence number to see
1563 * if this is a valid reset segment.
1565 * In all states except SYN-SENT, all reset (RST) segments
1566 * are validated by checking their SEQ-fields. A reset is
1567 * valid if its sequence number is in the window.
1568 * Note: this does not take into account delayed ACKs, so
1569 * we should test against last_ack_sent instead of rcv_nxt.
1570 * The sequence number in the reset segment is normally an
1571 * echo of our outgoing acknowledgement numbers, but some hosts
1572 * send a reset with the sequence number at the rightmost edge
1573 * of our receive window, and we have to handle this case.
1574 * If we have multiple segments in flight, the intial reset
1575 * segment sequence numbers will be to the left of last_ack_sent,
1576 * but they will eventually catch up.
1577 * In any case, it never made sense to trim reset segments to
1578 * fit the receive window since RFC 1122 says:
1579 * 4.2.2.12 RST Segment: RFC-793 Section 3.4
1581 * A TCP SHOULD allow a received RST segment to include data.
1584 * It has been suggested that a RST segment could contain
1585 * ASCII text that encoded and explained the cause of the
1586 * RST. No standard has yet been established for such
1589 * If the reset segment passes the sequence number test examine
1591 * SYN_RECEIVED STATE:
1592 * If passive open, return to LISTEN state.
1593 * If active open, inform user that connection was refused.
1594 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
1595 * Inform user that connection was reset, and close tcb.
1596 * CLOSING, LAST_ACK STATES:
1599 * Drop the segment - see Stevens, vol. 2, p. 964 and
1602 if (thflags & TH_RST) {
1603 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
1604 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
1605 switch (tp->t_state) {
1607 case TCPS_SYN_RECEIVED:
1608 so->so_error = ECONNREFUSED;
1611 case TCPS_ESTABLISHED:
1612 case TCPS_FIN_WAIT_1:
1613 case TCPS_FIN_WAIT_2:
1614 case TCPS_CLOSE_WAIT:
1615 so->so_error = ECONNRESET;
1617 tp->t_state = TCPS_CLOSED;
1618 tcpstat.tcps_drops++;
1627 case TCPS_TIME_WAIT:
1635 * RFC 1323 PAWS: If we have a timestamp reply on this segment
1636 * and it's less than ts_recent, drop it.
1638 if ((to.to_flags & TOF_TS) && tp->ts_recent != 0 &&
1639 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
1641 /* Check to see if ts_recent is over 24 days old. */
1642 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1644 * Invalidate ts_recent. If this segment updates
1645 * ts_recent, the age will be reset later and ts_recent
1646 * will get a valid value. If it does not, setting
1647 * ts_recent to zero will at least satisfy the
1648 * requirement that zero be placed in the timestamp
1649 * echo reply when ts_recent isn't valid. The
1650 * age isn't reset until we get a valid ts_recent
1651 * because we don't want out-of-order segments to be
1652 * dropped when ts_recent is old.
1656 tcpstat.tcps_rcvduppack++;
1657 tcpstat.tcps_rcvdupbyte += tlen;
1658 tcpstat.tcps_pawsdrop++;
1667 * If T/TCP was negotiated and the segment doesn't have CC,
1668 * or if its CC is wrong then drop the segment.
1669 * RST segments do not have to comply with this.
1671 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) &&
1672 (!(to.to_flags & TOF_CC) || tp->cc_recv != to.to_cc))
1676 * In the SYN-RECEIVED state, validate that the packet belongs to
1677 * this connection before trimming the data to fit the receive
1678 * window. Check the sequence number versus IRS since we know
1679 * the sequence numbers haven't wrapped. This is a partial fix
1680 * for the "LAND" DoS attack.
1682 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
1683 rstreason = BANDLIM_RST_OPENPORT;
1687 todrop = tp->rcv_nxt - th->th_seq;
1689 if (TCP_DO_SACK(tp)) {
1690 /* Report duplicate segment at head of packet. */
1691 tp->reportblk.rblk_start = th->th_seq;
1692 tp->reportblk.rblk_end = th->th_seq + tlen;
1693 if (thflags & TH_FIN)
1694 ++tp->reportblk.rblk_end;
1695 if (SEQ_GT(tp->reportblk.rblk_end, tp->rcv_nxt))
1696 tp->reportblk.rblk_end = tp->rcv_nxt;
1697 tp->t_flags |= (TF_DUPSEG | TF_SACKLEFT | TF_ACKNOW);
1699 if (thflags & TH_SYN) {
1709 * Following if statement from Stevens, vol. 2, p. 960.
1711 if (todrop > tlen ||
1712 (todrop == tlen && !(thflags & TH_FIN))) {
1714 * Any valid FIN must be to the left of the window.
1715 * At this point the FIN must be a duplicate or out
1716 * of sequence; drop it.
1721 * Send an ACK to resynchronize and drop any data.
1722 * But keep on processing for RST or ACK.
1724 tp->t_flags |= TF_ACKNOW;
1726 tcpstat.tcps_rcvduppack++;
1727 tcpstat.tcps_rcvdupbyte += todrop;
1729 tcpstat.tcps_rcvpartduppack++;
1730 tcpstat.tcps_rcvpartdupbyte += todrop;
1732 drop_hdrlen += todrop; /* drop from the top afterwards */
1733 th->th_seq += todrop;
1735 if (th->th_urp > todrop)
1736 th->th_urp -= todrop;
1744 * If new data are received on a connection after the
1745 * user processes are gone, then RST the other end.
1747 if ((so->so_state & SS_NOFDREF) &&
1748 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1750 tcpstat.tcps_rcvafterclose++;
1751 rstreason = BANDLIM_UNLIMITED;
1756 * If segment ends after window, drop trailing data
1757 * (and PUSH and FIN); if nothing left, just ACK.
1759 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
1761 tcpstat.tcps_rcvpackafterwin++;
1762 if (todrop >= tlen) {
1763 tcpstat.tcps_rcvbyteafterwin += tlen;
1765 * If a new connection request is received
1766 * while in TIME_WAIT, drop the old connection
1767 * and start over if the sequence numbers
1768 * are above the previous ones.
1770 if (thflags & TH_SYN &&
1771 tp->t_state == TCPS_TIME_WAIT &&
1772 SEQ_GT(th->th_seq, tp->rcv_nxt)) {
1777 * If window is closed can only take segments at
1778 * window edge, and have to drop data and PUSH from
1779 * incoming segments. Continue processing, but
1780 * remember to ack. Otherwise, drop segment
1783 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1784 tp->t_flags |= TF_ACKNOW;
1785 tcpstat.tcps_rcvwinprobe++;
1789 tcpstat.tcps_rcvbyteafterwin += todrop;
1792 thflags &= ~(TH_PUSH | TH_FIN);
1796 * If last ACK falls within this segment's sequence numbers,
1797 * record its timestamp.
1799 * 1) That the test incorporates suggestions from the latest
1800 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1801 * 2) That updating only on newer timestamps interferes with
1802 * our earlier PAWS tests, so this check should be solely
1803 * predicated on the sequence space of this segment.
1804 * 3) That we modify the segment boundary check to be
1805 * Last.ACK.Sent <= SEG.SEQ + SEG.LEN
1806 * instead of RFC1323's
1807 * Last.ACK.Sent < SEG.SEQ + SEG.LEN,
1808 * This modified check allows us to overcome RFC1323's
1809 * limitations as described in Stevens TCP/IP Illustrated
1810 * Vol. 2 p.869. In such cases, we can still calculate the
1811 * RTT correctly when RCV.NXT == Last.ACK.Sent.
1813 if ((to.to_flags & TOF_TS) && SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
1814 SEQ_LEQ(tp->last_ack_sent, (th->th_seq + tlen
1815 + ((thflags & TH_SYN) != 0)
1816 + ((thflags & TH_FIN) != 0)))) {
1817 tp->ts_recent_age = ticks;
1818 tp->ts_recent = to.to_tsval;
1822 * If a SYN is in the window, then this is an
1823 * error and we send an RST and drop the connection.
1825 if (thflags & TH_SYN) {
1826 tp = tcp_drop(tp, ECONNRESET);
1827 rstreason = BANDLIM_UNLIMITED;
1832 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
1833 * flag is on (half-synchronized state), then queue data for
1834 * later processing; else drop segment and return.
1836 if (!(thflags & TH_ACK)) {
1837 if (tp->t_state == TCPS_SYN_RECEIVED ||
1838 (tp->t_flags & TF_NEEDSYN))
1847 switch (tp->t_state) {
1849 * In SYN_RECEIVED state, the ACK acknowledges our SYN, so enter
1850 * ESTABLISHED state and continue processing.
1851 * The ACK was checked above.
1853 case TCPS_SYN_RECEIVED:
1855 tcpstat.tcps_connects++;
1857 /* Do window scaling? */
1858 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
1859 (TF_RCVD_SCALE | TF_REQ_SCALE)) {
1860 tp->snd_scale = tp->requested_s_scale;
1861 tp->rcv_scale = tp->request_r_scale;
1864 * Upon successful completion of 3-way handshake,
1865 * update cache.CC if it was undefined, pass any queued
1866 * data to the user, and advance state appropriately.
1868 if ((taop = tcp_gettaocache(&inp->inp_inc)) != NULL &&
1870 taop->tao_cc = tp->cc_recv;
1874 * SYN-RECEIVED -> ESTABLISHED
1875 * SYN-RECEIVED* -> FIN-WAIT-1
1877 tp->t_starttime = ticks;
1878 if (tp->t_flags & TF_NEEDFIN) {
1879 tp->t_state = TCPS_FIN_WAIT_1;
1880 tp->t_flags &= ~TF_NEEDFIN;
1882 tp->t_state = TCPS_ESTABLISHED;
1883 callout_reset(tp->tt_keep, tcp_keepidle,
1884 tcp_timer_keep, tp);
1887 * If segment contains data or ACK, will call tcp_reass()
1888 * later; if not, do so now to pass queued data to user.
1890 if (tlen == 0 && !(thflags & TH_FIN))
1891 tcp_reass(tp, NULL, NULL, NULL);
1895 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1896 * ACKs. If the ack is in the range
1897 * tp->snd_una < th->th_ack <= tp->snd_max
1898 * then advance tp->snd_una to th->th_ack and drop
1899 * data from the retransmission queue. If this ACK reflects
1900 * more up to date window information we update our window information.
1902 case TCPS_ESTABLISHED:
1903 case TCPS_FIN_WAIT_1:
1904 case TCPS_FIN_WAIT_2:
1905 case TCPS_CLOSE_WAIT:
1908 case TCPS_TIME_WAIT:
1910 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
1911 if (TCP_DO_SACK(tp))
1912 tcp_sack_update_scoreboard(tp, &to);
1913 if (tlen != 0 || tiwin != tp->snd_wnd) {
1917 tcpstat.tcps_rcvdupack++;
1918 if (!callout_active(tp->tt_rexmt) ||
1919 th->th_ack != tp->snd_una) {
1924 * We have outstanding data (other than
1925 * a window probe), this is a completely
1926 * duplicate ack (ie, window info didn't
1927 * change), the ack is the biggest we've
1928 * seen and we've seen exactly our rexmt
1929 * threshhold of them, so assume a packet
1930 * has been dropped and retransmit it.
1931 * Kludge snd_nxt & the congestion
1932 * window so we send only this one
1935 if (IN_FASTRECOVERY(tp)) {
1936 if (TCP_DO_SACK(tp)) {
1937 /* No artifical cwnd inflation. */
1938 tcp_sack_rexmt(tp, th);
1941 * Dup acks mean that packets
1942 * have left the network
1943 * (they're now cached at the
1944 * receiver) so bump cwnd by
1945 * the amount in the receiver
1946 * to keep a constant cwnd
1947 * packets in the network.
1949 tp->snd_cwnd += tp->t_maxseg;
1952 } else if (SEQ_LT(th->th_ack, tp->snd_recover)) {
1955 } else if (++tp->t_dupacks == tcprexmtthresh) {
1956 tcp_seq old_snd_nxt;
1960 if (tcp_do_eifel_detect &&
1961 (tp->t_flags & TF_RCVD_TSTMP)) {
1962 tcp_save_congestion_state(tp);
1963 tp->t_flags |= TF_FASTREXMT;
1966 * We know we're losing at the current
1967 * window size, so do congestion avoidance:
1968 * set ssthresh to half the current window
1969 * and pull our congestion window back to the
1972 win = min(tp->snd_wnd, tp->snd_cwnd) / 2 /
1976 tp->snd_ssthresh = win * tp->t_maxseg;
1977 ENTER_FASTRECOVERY(tp);
1978 tp->snd_recover = tp->snd_max;
1979 callout_stop(tp->tt_rexmt);
1981 old_snd_nxt = tp->snd_nxt;
1982 tp->snd_nxt = th->th_ack;
1983 tp->snd_cwnd = tp->t_maxseg;
1985 ++tcpstat.tcps_sndfastrexmit;
1986 tp->snd_cwnd = tp->snd_ssthresh;
1987 tp->rexmt_high = tp->snd_nxt;
1988 if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
1989 tp->snd_nxt = old_snd_nxt;
1990 KASSERT(tp->snd_limited <= 2,
1991 ("tp->snd_limited too big"));
1992 if (TCP_DO_SACK(tp))
1993 tcp_sack_rexmt(tp, th);
1995 tp->snd_cwnd += tp->t_maxseg *
1996 (tp->t_dupacks - tp->snd_limited);
1997 } else if (tcp_do_limitedtransmit) {
1998 u_long oldcwnd = tp->snd_cwnd;
1999 tcp_seq oldsndmax = tp->snd_max;
2000 tcp_seq oldsndnxt = tp->snd_nxt;
2001 /* outstanding data */
2002 uint32_t ownd = tp->snd_max - tp->snd_una;
2005 #define iceildiv(n, d) (((n)+(d)-1) / (d))
2007 KASSERT(tp->t_dupacks == 1 ||
2009 ("dupacks not 1 or 2"));
2010 if (tp->t_dupacks == 1)
2011 tp->snd_limited = 0;
2012 tp->snd_nxt = tp->snd_max;
2013 tp->snd_cwnd = ownd +
2014 (tp->t_dupacks - tp->snd_limited) *
2017 if (SEQ_LT(oldsndnxt, oldsndmax))
2018 tp->snd_nxt = oldsndnxt;
2019 tp->snd_cwnd = oldcwnd;
2020 sent = tp->snd_max - oldsndmax;
2021 if (sent > tp->t_maxseg) {
2022 KASSERT((tp->t_dupacks == 2 &&
2023 tp->snd_limited == 0) ||
2024 (sent == tp->t_maxseg + 1 &&
2025 tp->t_flags & TF_SENTFIN),
2027 KASSERT(sent <= tp->t_maxseg * 2,
2028 ("sent too many segments"));
2029 tp->snd_limited = 2;
2030 tcpstat.tcps_sndlimited += 2;
2031 } else if (sent > 0) {
2033 ++tcpstat.tcps_sndlimited;
2034 } else if (tcp_do_early_retransmit &&
2035 (tcp_do_eifel_detect &&
2036 (tp->t_flags & TF_RCVD_TSTMP)) &&
2037 ownd < 4 * tp->t_maxseg &&
2038 tp->t_dupacks + 1 >=
2039 iceildiv(ownd, tp->t_maxseg) &&
2040 (!TCP_DO_SACK(tp) ||
2041 ownd <= tp->t_maxseg ||
2042 tcp_sack_has_sacked(&tp->scb,
2043 ownd - tp->t_maxseg))) {
2044 ++tcpstat.tcps_sndearlyrexmit;
2045 tp->t_flags |= TF_EARLYREXMT;
2046 goto fastretransmit;
2052 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una"));
2054 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2056 * Detected optimistic ACK attack.
2057 * Force slow-start to de-synchronize attack.
2059 tp->snd_cwnd = tp->t_maxseg;
2062 tcpstat.tcps_rcvacktoomuch++;
2066 * If we reach this point, ACK is not a duplicate,
2067 * i.e., it ACKs something we sent.
2069 if (tp->t_flags & TF_NEEDSYN) {
2071 * T/TCP: Connection was half-synchronized, and our
2072 * SYN has been ACK'd (so connection is now fully
2073 * synchronized). Go to non-starred state,
2074 * increment snd_una for ACK of SYN, and check if
2075 * we can do window scaling.
2077 tp->t_flags &= ~TF_NEEDSYN;
2079 /* Do window scaling? */
2080 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
2081 (TF_RCVD_SCALE | TF_REQ_SCALE)) {
2082 tp->snd_scale = tp->requested_s_scale;
2083 tp->rcv_scale = tp->request_r_scale;
2088 acked = th->th_ack - tp->snd_una;
2089 tcpstat.tcps_rcvackpack++;
2090 tcpstat.tcps_rcvackbyte += acked;
2092 if (tcp_do_eifel_detect && acked > 0 &&
2093 (to.to_flags & TOF_TS) && (to.to_tsecr != 0) &&
2094 (tp->t_flags & TF_FIRSTACCACK)) {
2095 /* Eifel detection applicable. */
2096 if (to.to_tsecr < tp->t_rexmtTS) {
2097 ++tcpstat.tcps_eifeldetected;
2098 tcp_revert_congestion_state(tp);
2099 if (tp->t_rxtshift == 1 &&
2100 ticks >= tp->t_badrxtwin)
2101 ++tcpstat.tcps_rttcantdetect;
2103 } else if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) {
2105 * If we just performed our first retransmit,
2106 * and the ACK arrives within our recovery window,
2107 * then it was a mistake to do the retransmit
2108 * in the first place. Recover our original cwnd
2109 * and ssthresh, and proceed to transmit where we
2112 tcp_revert_congestion_state(tp);
2113 ++tcpstat.tcps_rttdetected;
2117 * If we have a timestamp reply, update smoothed
2118 * round trip time. If no timestamp is present but
2119 * transmit timer is running and timed sequence
2120 * number was acked, update smoothed round trip time.
2121 * Since we now have an rtt measurement, cancel the
2122 * timer backoff (cf., Phil Karn's retransmit alg.).
2123 * Recompute the initial retransmit timer.
2125 * Some machines (certain windows boxes) send broken
2126 * timestamp replies during the SYN+ACK phase, ignore
2129 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0))
2130 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1);
2131 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq))
2132 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2133 tcp_xmit_bandwidth_limit(tp, th->th_ack);
2136 * If no data (only SYN) was ACK'd,
2137 * skip rest of ACK processing.
2142 /* Stop looking for an acceptable ACK since one was received. */
2143 tp->t_flags &= ~(TF_FIRSTACCACK | TF_FASTREXMT | TF_EARLYREXMT);
2145 if (acked > so->so_snd.sb_cc) {
2146 tp->snd_wnd -= so->so_snd.sb_cc;
2147 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc);
2148 ourfinisacked = TRUE;
2150 sbdrop(&so->so_snd, acked);
2151 tp->snd_wnd -= acked;
2152 ourfinisacked = FALSE;
2157 * Update window information.
2158 * Don't look at window if no ACK:
2159 * TAC's send garbage on first SYN.
2161 if (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2162 (tp->snd_wl1 == th->th_seq &&
2163 (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2164 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)))) {
2165 /* keep track of pure window updates */
2166 if (tlen == 0 && tp->snd_wl2 == th->th_ack &&
2167 tiwin > tp->snd_wnd)
2168 tcpstat.tcps_rcvwinupd++;
2169 tp->snd_wnd = tiwin;
2170 tp->snd_wl1 = th->th_seq;
2171 tp->snd_wl2 = th->th_ack;
2172 if (tp->snd_wnd > tp->max_sndwnd)
2173 tp->max_sndwnd = tp->snd_wnd;
2177 tp->snd_una = th->th_ack;
2178 if (TCP_DO_SACK(tp))
2179 tcp_sack_update_scoreboard(tp, &to);
2180 if (IN_FASTRECOVERY(tp)) {
2181 if (SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2182 EXIT_FASTRECOVERY(tp);
2185 * If the congestion window was inflated
2186 * to account for the other side's
2187 * cached packets, retract it.
2189 if (!TCP_DO_SACK(tp))
2190 tp->snd_cwnd = tp->snd_ssthresh;
2193 * Window inflation should have left us
2194 * with approximately snd_ssthresh outstanding
2195 * data. But, in case we would be inclined
2196 * to send a burst, better do it using
2199 if (SEQ_GT(th->th_ack + tp->snd_cwnd,
2200 tp->snd_max + 2 * tp->t_maxseg))
2202 (tp->snd_max - tp->snd_una) +
2207 if (TCP_DO_SACK(tp)) {
2208 tp->snd_max_rexmt = tp->snd_max;
2209 tcp_sack_rexmt(tp, th);
2211 tcp_newreno_partial_ack(tp, th, acked);
2217 * Open the congestion window. When in slow-start,
2218 * open exponentially: maxseg per packet. Otherwise,
2219 * open linearly: maxseg per window.
2221 if (tp->snd_cwnd <= tp->snd_ssthresh) {
2223 (SEQ_LT(tp->snd_nxt, tp->snd_max) ?
2224 tp->t_maxseg : 2 * tp->t_maxseg);
2227 tp->snd_cwnd += tcp_do_abc ?
2228 min(acked, abc_sslimit) : tp->t_maxseg;
2230 /* linear increase */
2231 tp->snd_wacked += tcp_do_abc ? acked :
2233 if (tp->snd_wacked >= tp->snd_cwnd) {
2234 tp->snd_wacked -= tp->snd_cwnd;
2235 tp->snd_cwnd += tp->t_maxseg;
2238 tp->snd_cwnd = min(tp->snd_cwnd,
2239 TCP_MAXWIN << tp->snd_scale);
2240 tp->snd_recover = th->th_ack - 1;
2242 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2243 tp->snd_nxt = tp->snd_una;
2246 * If all outstanding data is acked, stop retransmit
2247 * timer and remember to restart (more output or persist).
2248 * If there is more data to be acked, restart retransmit
2249 * timer, using current (possibly backed-off) value.
2251 if (th->th_ack == tp->snd_max) {
2252 callout_stop(tp->tt_rexmt);
2254 } else if (!callout_active(tp->tt_persist))
2255 callout_reset(tp->tt_rexmt, tp->t_rxtcur,
2256 tcp_timer_rexmt, tp);
2258 switch (tp->t_state) {
2260 * In FIN_WAIT_1 STATE in addition to the processing
2261 * for the ESTABLISHED state if our FIN is now acknowledged
2262 * then enter FIN_WAIT_2.
2264 case TCPS_FIN_WAIT_1:
2265 if (ourfinisacked) {
2267 * If we can't receive any more
2268 * data, then closing user can proceed.
2269 * Starting the timer is contrary to the
2270 * specification, but if we don't get a FIN
2271 * we'll hang forever.
2273 if (so->so_state & SS_CANTRCVMORE) {
2274 soisdisconnected(so);
2275 callout_reset(tp->tt_2msl, tcp_maxidle,
2276 tcp_timer_2msl, tp);
2278 tp->t_state = TCPS_FIN_WAIT_2;
2283 * In CLOSING STATE in addition to the processing for
2284 * the ESTABLISHED state if the ACK acknowledges our FIN
2285 * then enter the TIME-WAIT state, otherwise ignore
2289 if (ourfinisacked) {
2290 tp->t_state = TCPS_TIME_WAIT;
2291 tcp_canceltimers(tp);
2292 /* Shorten TIME_WAIT [RFC-1644, p.28] */
2293 if (tp->cc_recv != 0 &&
2294 (ticks - tp->t_starttime) < tcp_msl)
2295 callout_reset(tp->tt_2msl,
2296 tp->t_rxtcur * TCPTV_TWTRUNC,
2297 tcp_timer_2msl, tp);
2299 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2300 tcp_timer_2msl, tp);
2301 soisdisconnected(so);
2306 * In LAST_ACK, we may still be waiting for data to drain
2307 * and/or to be acked, as well as for the ack of our FIN.
2308 * If our FIN is now acknowledged, delete the TCB,
2309 * enter the closed state and return.
2312 if (ourfinisacked) {
2319 * In TIME_WAIT state the only thing that should arrive
2320 * is a retransmission of the remote FIN. Acknowledge
2321 * it and restart the finack timer.
2323 case TCPS_TIME_WAIT:
2324 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2325 tcp_timer_2msl, tp);
2332 * Update window information.
2333 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2335 if ((thflags & TH_ACK) &&
2336 acceptable_window_update(tp, th, tiwin)) {
2337 /* keep track of pure window updates */
2338 if (tlen == 0 && tp->snd_wl2 == th->th_ack &&
2339 tiwin > tp->snd_wnd)
2340 tcpstat.tcps_rcvwinupd++;
2341 tp->snd_wnd = tiwin;
2342 tp->snd_wl1 = th->th_seq;
2343 tp->snd_wl2 = th->th_ack;
2344 if (tp->snd_wnd > tp->max_sndwnd)
2345 tp->max_sndwnd = tp->snd_wnd;
2350 * Process segments with URG.
2352 if ((thflags & TH_URG) && th->th_urp &&
2353 !TCPS_HAVERCVDFIN(tp->t_state)) {
2355 * This is a kludge, but if we receive and accept
2356 * random urgent pointers, we'll crash in
2357 * soreceive. It's hard to imagine someone
2358 * actually wanting to send this much urgent data.
2360 if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2361 th->th_urp = 0; /* XXX */
2362 thflags &= ~TH_URG; /* XXX */
2363 goto dodata; /* XXX */
2366 * If this segment advances the known urgent pointer,
2367 * then mark the data stream. This should not happen
2368 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2369 * a FIN has been received from the remote side.
2370 * In these states we ignore the URG.
2372 * According to RFC961 (Assigned Protocols),
2373 * the urgent pointer points to the last octet
2374 * of urgent data. We continue, however,
2375 * to consider it to indicate the first octet
2376 * of data past the urgent section as the original
2377 * spec states (in one of two places).
2379 if (SEQ_GT(th->th_seq + th->th_urp, tp->rcv_up)) {
2380 tp->rcv_up = th->th_seq + th->th_urp;
2381 so->so_oobmark = so->so_rcv.sb_cc +
2382 (tp->rcv_up - tp->rcv_nxt) - 1;
2383 if (so->so_oobmark == 0)
2384 so->so_state |= SS_RCVATMARK;
2386 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2389 * Remove out of band data so doesn't get presented to user.
2390 * This can happen independent of advancing the URG pointer,
2391 * but if two URG's are pending at once, some out-of-band
2392 * data may creep in... ick.
2394 if (th->th_urp <= (u_long)tlen &&
2395 !(so->so_options & SO_OOBINLINE)) {
2396 /* hdr drop is delayed */
2397 tcp_pulloutofband(so, th, m, drop_hdrlen);
2401 * If no out of band data is expected,
2402 * pull receive urgent pointer along
2403 * with the receive window.
2405 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2406 tp->rcv_up = tp->rcv_nxt;
2411 * Process the segment text, merging it into the TCP sequencing queue,
2412 * and arranging for acknowledgment of receipt if necessary.
2413 * This process logically involves adjusting tp->rcv_wnd as data
2414 * is presented to the user (this happens in tcp_usrreq.c,
2415 * case PRU_RCVD). If a FIN has already been received on this
2416 * connection then we just ignore the text.
2418 if ((tlen || (thflags & TH_FIN)) && !TCPS_HAVERCVDFIN(tp->t_state)) {
2419 m_adj(m, drop_hdrlen); /* delayed header drop */
2421 * Insert segment which includes th into TCP reassembly queue
2422 * with control block tp. Set thflags to whether reassembly now
2423 * includes a segment with FIN. This handles the common case
2424 * inline (segment is the next to be received on an established
2425 * connection, and the queue is empty), avoiding linkage into
2426 * and removal from the queue and repetition of various
2428 * Set DELACK for segments received in order, but ack
2429 * immediately when segments are out of order (so
2430 * fast retransmit can work).
2432 if (th->th_seq == tp->rcv_nxt &&
2433 LIST_EMPTY(&tp->t_segq) &&
2434 TCPS_HAVEESTABLISHED(tp->t_state)) {
2436 callout_reset(tp->tt_delack, tcp_delacktime,
2437 tcp_timer_delack, tp);
2439 tp->t_flags |= TF_ACKNOW;
2440 tp->rcv_nxt += tlen;
2441 thflags = th->th_flags & TH_FIN;
2442 tcpstat.tcps_rcvpack++;
2443 tcpstat.tcps_rcvbyte += tlen;
2445 if (so->so_state & SS_CANTRCVMORE)
2448 sbappendstream(&so->so_rcv, m);
2451 if (!(tp->t_flags & TF_DUPSEG)) {
2452 /* Initialize SACK report block. */
2453 tp->reportblk.rblk_start = th->th_seq;
2454 tp->reportblk.rblk_end = th->th_seq + tlen +
2455 ((thflags & TH_FIN) != 0);
2457 thflags = tcp_reass(tp, th, &tlen, m);
2458 tp->t_flags |= TF_ACKNOW;
2462 * Note the amount of data that peer has sent into
2463 * our window, in order to estimate the sender's
2466 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2473 * If FIN is received ACK the FIN and let the user know
2474 * that the connection is closing.
2476 if (thflags & TH_FIN) {
2477 if (!TCPS_HAVERCVDFIN(tp->t_state)) {
2480 * If connection is half-synchronized
2481 * (ie NEEDSYN flag on) then delay ACK,
2482 * so it may be piggybacked when SYN is sent.
2483 * Otherwise, since we received a FIN then no
2484 * more input can be expected, send ACK now.
2486 if (DELAY_ACK(tp) && (tp->t_flags & TF_NEEDSYN))
2487 callout_reset(tp->tt_delack, tcp_delacktime,
2488 tcp_timer_delack, tp);
2490 tp->t_flags |= TF_ACKNOW;
2494 switch (tp->t_state) {
2496 * In SYN_RECEIVED and ESTABLISHED STATES
2497 * enter the CLOSE_WAIT state.
2499 case TCPS_SYN_RECEIVED:
2500 tp->t_starttime = ticks;
2502 case TCPS_ESTABLISHED:
2503 tp->t_state = TCPS_CLOSE_WAIT;
2507 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2508 * enter the CLOSING state.
2510 case TCPS_FIN_WAIT_1:
2511 tp->t_state = TCPS_CLOSING;
2515 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2516 * starting the time-wait timer, turning off the other
2519 case TCPS_FIN_WAIT_2:
2520 tp->t_state = TCPS_TIME_WAIT;
2521 tcp_canceltimers(tp);
2522 /* Shorten TIME_WAIT [RFC-1644, p.28] */
2523 if (tp->cc_recv != 0 &&
2524 (ticks - tp->t_starttime) < tcp_msl) {
2525 callout_reset(tp->tt_2msl,
2526 tp->t_rxtcur * TCPTV_TWTRUNC,
2527 tcp_timer_2msl, tp);
2528 /* For transaction client, force ACK now. */
2529 tp->t_flags |= TF_ACKNOW;
2532 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2533 tcp_timer_2msl, tp);
2534 soisdisconnected(so);
2538 * In TIME_WAIT state restart the 2 MSL time_wait timer.
2540 case TCPS_TIME_WAIT:
2541 callout_reset(tp->tt_2msl, 2 * tcp_msl,
2542 tcp_timer_2msl, tp);
2548 if (so->so_options & SO_DEBUG)
2549 tcp_trace(TA_INPUT, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2553 * Return any desired output.
2555 if (needoutput || (tp->t_flags & TF_ACKNOW))
2561 * Generate an ACK dropping incoming segment if it occupies
2562 * sequence space, where the ACK reflects our state.
2564 * We can now skip the test for the RST flag since all
2565 * paths to this code happen after packets containing
2566 * RST have been dropped.
2568 * In the SYN-RECEIVED state, don't send an ACK unless the
2569 * segment we received passes the SYN-RECEIVED ACK test.
2570 * If it fails send a RST. This breaks the loop in the
2571 * "LAND" DoS attack, and also prevents an ACK storm
2572 * between two listening ports that have been sent forged
2573 * SYN segments, each with the source address of the other.
2575 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
2576 (SEQ_GT(tp->snd_una, th->th_ack) ||
2577 SEQ_GT(th->th_ack, tp->snd_max)) ) {
2578 rstreason = BANDLIM_RST_OPENPORT;
2582 if (so->so_options & SO_DEBUG)
2583 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2586 tp->t_flags |= TF_ACKNOW;
2592 * Generate a RST, dropping incoming segment.
2593 * Make ACK acceptable to originator of segment.
2594 * Don't bother to respond if destination was broadcast/multicast.
2596 if ((thflags & TH_RST) || m->m_flags & (M_BCAST | M_MCAST))
2599 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
2600 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
2603 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
2604 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
2605 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
2606 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2609 /* IPv6 anycast check is done at tcp6_input() */
2612 * Perform bandwidth limiting.
2615 if (badport_bandlim(rstreason) < 0)
2620 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2621 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2623 if (thflags & TH_ACK)
2624 /* mtod() below is safe as long as hdr dropping is delayed */
2625 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack,
2628 if (thflags & TH_SYN)
2630 /* mtod() below is safe as long as hdr dropping is delayed */
2631 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq + tlen,
2632 (tcp_seq)0, TH_RST | TH_ACK);
2638 * Drop space held by incoming segment and return.
2641 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2642 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2649 * Parse TCP options and place in tcpopt.
2652 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, boolean_t is_syn)
2657 for (; cnt > 0; cnt -= optlen, cp += optlen) {
2659 if (opt == TCPOPT_EOL)
2661 if (opt == TCPOPT_NOP)
2667 if (optlen < 2 || optlen > cnt)
2672 if (optlen != TCPOLEN_MAXSEG)
2676 to->to_flags |= TOF_MSS;
2677 bcopy(cp + 2, &to->to_mss, sizeof to->to_mss);
2678 to->to_mss = ntohs(to->to_mss);
2681 if (optlen != TCPOLEN_WINDOW)
2685 to->to_flags |= TOF_SCALE;
2686 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
2688 case TCPOPT_TIMESTAMP:
2689 if (optlen != TCPOLEN_TIMESTAMP)
2691 to->to_flags |= TOF_TS;
2692 bcopy(cp + 2, &to->to_tsval, sizeof to->to_tsval);
2693 to->to_tsval = ntohl(to->to_tsval);
2694 bcopy(cp + 6, &to->to_tsecr, sizeof to->to_tsecr);
2695 to->to_tsecr = ntohl(to->to_tsecr);
2697 * If echoed timestamp is later than the current time,
2698 * fall back to non RFC1323 RTT calculation.
2700 if (to->to_tsecr != 0 && TSTMP_GT(to->to_tsecr, ticks))
2704 if (optlen != TCPOLEN_CC)
2706 to->to_flags |= TOF_CC;
2707 bcopy(cp + 2, &to->to_cc, sizeof to->to_cc);
2708 to->to_cc = ntohl(to->to_cc);
2711 if (optlen != TCPOLEN_CC)
2715 to->to_flags |= TOF_CCNEW;
2716 bcopy(cp + 2, &to->to_cc, sizeof to->to_cc);
2717 to->to_cc = ntohl(to->to_cc);
2720 if (optlen != TCPOLEN_CC)
2724 to->to_flags |= TOF_CCECHO;
2725 bcopy(cp + 2, &to->to_ccecho, sizeof to->to_ccecho);
2726 to->to_ccecho = ntohl(to->to_ccecho);
2728 case TCPOPT_SACK_PERMITTED:
2729 if (optlen != TCPOLEN_SACK_PERMITTED)
2733 to->to_flags |= TOF_SACK_PERMITTED;
2736 if ((optlen - 2) & 0x07) /* not multiple of 8 */
2738 to->to_nsackblocks = (optlen - 2) / 8;
2739 to->to_sackblocks = (struct raw_sackblock *) (cp + 2);
2740 to->to_flags |= TOF_SACK;
2741 for (i = 0; i < to->to_nsackblocks; i++) {
2742 struct raw_sackblock *r = &to->to_sackblocks[i];
2744 r->rblk_start = ntohl(r->rblk_start);
2745 r->rblk_end = ntohl(r->rblk_end);
2755 * Pull out of band byte out of a segment so
2756 * it doesn't appear in the user's data queue.
2757 * It is still reflected in the segment length for
2758 * sequencing purposes.
2759 * "off" is the delayed to be dropped hdrlen.
2762 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, int off)
2764 int cnt = off + th->th_urp - 1;
2767 if (m->m_len > cnt) {
2768 char *cp = mtod(m, caddr_t) + cnt;
2769 struct tcpcb *tp = sototcpcb(so);
2772 tp->t_oobflags |= TCPOOB_HAVEDATA;
2773 bcopy(cp + 1, cp, m->m_len - cnt - 1);
2775 if (m->m_flags & M_PKTHDR)
2784 panic("tcp_pulloutofband");
2788 * Collect new round-trip time estimate
2789 * and update averages and current timeout.
2792 tcp_xmit_timer(struct tcpcb *tp, int rtt)
2796 tcpstat.tcps_rttupdated++;
2798 if (tp->t_srtt != 0) {
2800 * srtt is stored as fixed point with 5 bits after the
2801 * binary point (i.e., scaled by 8). The following magic
2802 * is equivalent to the smoothing algorithm in rfc793 with
2803 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2804 * point). Adjust rtt to origin 0.
2806 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
2807 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
2809 if ((tp->t_srtt += delta) <= 0)
2813 * We accumulate a smoothed rtt variance (actually, a
2814 * smoothed mean difference), then set the retransmit
2815 * timer to smoothed rtt + 4 times the smoothed variance.
2816 * rttvar is stored as fixed point with 4 bits after the
2817 * binary point (scaled by 16). The following is
2818 * equivalent to rfc793 smoothing with an alpha of .75
2819 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
2820 * rfc793's wired-in beta.
2824 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
2825 if ((tp->t_rttvar += delta) <= 0)
2827 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
2828 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2831 * No rtt measurement yet - use the unsmoothed rtt.
2832 * Set the variance to half the rtt (so our first
2833 * retransmit happens at 3*rtt).
2835 tp->t_srtt = rtt << TCP_RTT_SHIFT;
2836 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
2837 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2843 * the retransmit should happen at rtt + 4 * rttvar.
2844 * Because of the way we do the smoothing, srtt and rttvar
2845 * will each average +1/2 tick of bias. When we compute
2846 * the retransmit timer, we want 1/2 tick of rounding and
2847 * 1 extra tick because of +-1/2 tick uncertainty in the
2848 * firing of the timer. The bias will give us exactly the
2849 * 1.5 tick we need. But, because the bias is
2850 * statistical, we have to test that we don't drop below
2851 * the minimum feasible timer (which is 2 ticks).
2853 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
2854 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
2857 * We received an ack for a packet that wasn't retransmitted;
2858 * it is probably safe to discard any error indications we've
2859 * received recently. This isn't quite right, but close enough
2860 * for now (a route might have failed after we sent a segment,
2861 * and the return path might not be symmetrical).
2863 tp->t_softerror = 0;
2867 * Determine a reasonable value for maxseg size.
2868 * If the route is known, check route for mtu.
2869 * If none, use an mss that can be handled on the outgoing
2870 * interface without forcing IP to fragment; if bigger than
2871 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2872 * to utilize large mbufs. If no route is found, route has no mtu,
2873 * or the destination isn't local, use a default, hopefully conservative
2874 * size (usually 512 or the default IP max size, but no more than the mtu
2875 * of the interface), as we can't discover anything about intervening
2876 * gateways or networks. We also initialize the congestion/slow start
2877 * window to be a single segment if the destination isn't local.
2878 * While looking at the routing entry, we also initialize other path-dependent
2879 * parameters from pre-set or cached values in the routing entry.
2881 * Also take into account the space needed for options that we
2882 * send regularly. Make maxseg shorter by that amount to assure
2883 * that we can send maxseg amount of data even when the options
2884 * are present. Store the upper limit of the length of options plus
2887 * NOTE that this routine is only called when we process an incoming
2888 * segment, for outgoing segments only tcp_mssopt is called.
2890 * In case of T/TCP, we call this routine during implicit connection
2891 * setup as well (offer = -1), to initialize maxseg from the cached
2895 tcp_mss(struct tcpcb *tp, int offer)
2901 struct inpcb *inp = tp->t_inpcb;
2903 struct rmxp_tao *taop;
2904 int origoffer = offer;
2906 boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) ? TRUE : FALSE);
2907 size_t min_protoh = isipv6 ?
2908 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
2909 sizeof(struct tcpiphdr);
2911 const boolean_t isipv6 = FALSE;
2912 const size_t min_protoh = sizeof(struct tcpiphdr);
2916 rt = tcp_rtlookup6(&inp->inp_inc);
2918 rt = tcp_rtlookup(&inp->inp_inc);
2920 tp->t_maxopd = tp->t_maxseg =
2921 (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
2925 so = inp->inp_socket;
2927 taop = rmx_taop(rt->rt_rmx);
2929 * Offer == -1 means that we didn't receive SYN yet,
2930 * use cached value in that case;
2933 offer = taop->tao_mssopt;
2935 * Offer == 0 means that there was no MSS on the SYN segment,
2936 * in this case we use tcp_mssdflt.
2939 offer = (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
2942 * Sanity check: make sure that maxopd will be large
2943 * enough to allow some data on segments even is the
2944 * all the option space is used (40bytes). Otherwise
2945 * funny things may happen in tcp_output.
2947 offer = max(offer, 64);
2948 taop->tao_mssopt = offer;
2951 * While we're here, check if there's an initial rtt
2952 * or rttvar. Convert from the route-table units
2953 * to scaled multiples of the slow timeout timer.
2955 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
2957 * XXX the lock bit for RTT indicates that the value
2958 * is also a minimum value; this is subject to time.
2960 if (rt->rt_rmx.rmx_locks & RTV_RTT)
2961 tp->t_rttmin = rtt / (RTM_RTTUNIT / hz);
2962 tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE));
2963 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
2964 tcpstat.tcps_usedrtt++;
2965 if (rt->rt_rmx.rmx_rttvar) {
2966 tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
2967 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE));
2968 tcpstat.tcps_usedrttvar++;
2970 /* default variation is +- 1 rtt */
2972 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
2974 TCPT_RANGESET(tp->t_rxtcur,
2975 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
2976 tp->t_rttmin, TCPTV_REXMTMAX);
2979 * if there's an mtu associated with the route, use it
2980 * else, use the link mtu.
2982 if (rt->rt_rmx.rmx_mtu)
2983 mss = rt->rt_rmx.rmx_mtu - min_protoh;
2986 mss = ND_IFINFO(rt->rt_ifp)->linkmtu - min_protoh;
2987 if (!in6_localaddr(&inp->in6p_faddr))
2988 mss = min(mss, tcp_v6mssdflt);
2990 mss = ifp->if_mtu - min_protoh;
2991 if (!in_localaddr(inp->inp_faddr))
2992 mss = min(mss, tcp_mssdflt);
2995 mss = min(mss, offer);
2997 * maxopd stores the maximum length of data AND options
2998 * in a segment; maxseg is the amount of data in a normal
2999 * segment. We need to store this value (maxopd) apart
3000 * from maxseg, because now every segment carries options
3001 * and thus we normally have somewhat less data in segments.
3006 * In case of T/TCP, origoffer==-1 indicates, that no segments
3007 * were received yet. In this case we just guess, otherwise
3008 * we do the same as before T/TCP.
3010 if ((tp->t_flags & (TF_REQ_TSTMP | TF_NOOPT)) == TF_REQ_TSTMP &&
3012 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
3013 mss -= TCPOLEN_TSTAMP_APPA;
3014 if ((tp->t_flags & (TF_REQ_CC | TF_NOOPT)) == TF_REQ_CC &&
3016 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC))
3017 mss -= TCPOLEN_CC_APPA;
3019 #if (MCLBYTES & (MCLBYTES - 1)) == 0
3021 mss &= ~(MCLBYTES-1);
3024 mss = mss / MCLBYTES * MCLBYTES;
3027 * If there's a pipesize, change the socket buffer
3028 * to that size. Make the socket buffers an integral
3029 * number of mss units; if the mss is larger than
3030 * the socket buffer, decrease the mss.
3033 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0)
3035 bufsize = so->so_snd.sb_hiwat;
3039 bufsize = roundup(bufsize, mss);
3040 if (bufsize > sb_max)
3042 if (bufsize > so->so_snd.sb_hiwat)
3043 sbreserve(&so->so_snd, bufsize, so, NULL);
3048 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0)
3050 bufsize = so->so_rcv.sb_hiwat;
3051 if (bufsize > mss) {
3052 bufsize = roundup(bufsize, mss);
3053 if (bufsize > sb_max)
3055 if (bufsize > so->so_rcv.sb_hiwat)
3056 sbreserve(&so->so_rcv, bufsize, so, NULL);
3060 * Set the slow-start flight size depending on whether this
3061 * is a local network or not.
3064 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380));
3068 if (rt->rt_rmx.rmx_ssthresh) {
3070 * There's some sort of gateway or interface
3071 * buffer limit on the path. Use this to set
3072 * the slow start threshhold, but set the
3073 * threshold to no less than 2*mss.
3075 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
3076 tcpstat.tcps_usedssthresh++;
3081 * Determine the MSS option to send on an outgoing SYN.
3084 tcp_mssopt(struct tcpcb *tp)
3089 ((tp->t_inpcb->inp_vflag & INP_IPV6) ? TRUE : FALSE);
3090 int min_protoh = isipv6 ?
3091 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
3092 sizeof(struct tcpiphdr);
3094 const boolean_t isipv6 = FALSE;
3095 const size_t min_protoh = sizeof(struct tcpiphdr);
3099 rt = tcp_rtlookup6(&tp->t_inpcb->inp_inc);
3101 rt = tcp_rtlookup(&tp->t_inpcb->inp_inc);
3103 return (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
3105 return (rt->rt_ifp->if_mtu - min_protoh);
3109 * When a partial ack arrives, force the retransmission of the
3110 * next unacknowledged segment. Do not exit Fast Recovery.
3112 * Implement the Slow-but-Steady variant of NewReno by restarting the
3113 * the retransmission timer. Turn it off here so it can be restarted
3114 * later in tcp_output().
3117 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th, int acked)
3119 tcp_seq old_snd_nxt = tp->snd_nxt;
3120 u_long ocwnd = tp->snd_cwnd;
3122 callout_stop(tp->tt_rexmt);
3124 tp->snd_nxt = th->th_ack;
3125 /* Set snd_cwnd to one segment beyond acknowledged offset. */
3126 tp->snd_cwnd = tp->t_maxseg;
3127 tp->t_flags |= TF_ACKNOW;
3129 if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
3130 tp->snd_nxt = old_snd_nxt;
3131 /* partial window deflation */
3133 tp->snd_cwnd = ocwnd - acked + tp->t_maxseg;
3135 tp->snd_cwnd = tp->t_maxseg;
3139 * In contrast to the Slow-but-Steady NewReno variant,
3140 * we do not reset the retransmission timer for SACK retransmissions,
3141 * except when retransmitting snd_una.
3144 tcp_sack_rexmt(struct tcpcb *tp, struct tcphdr *th)
3146 uint32_t pipe, seglen;
3149 tcp_seq old_snd_nxt = tp->snd_nxt;
3150 u_long ocwnd = tp->snd_cwnd;
3151 int nseg = 0; /* consecutive new segments */
3152 #define MAXBURST 4 /* limit burst of new packets on partial ack */
3155 pipe = tcp_sack_compute_pipe(tp);
3156 while ((tcp_seq_diff_t)(ocwnd - pipe) >= (tcp_seq_diff_t)tp->t_maxseg &&
3157 (!tcp_do_smartsack || nseg < MAXBURST) &&
3158 tcp_sack_nextseg(tp, &nextrexmt, &seglen, &lostdup)) {
3160 tcp_seq old_snd_max;
3163 if (nextrexmt == tp->snd_max) ++nseg;
3164 tp->snd_nxt = nextrexmt;
3165 tp->snd_cwnd = nextrexmt - tp->snd_una + seglen;
3166 old_snd_max = tp->snd_max;
3167 if (nextrexmt == tp->snd_una)
3168 callout_stop(tp->tt_rexmt);
3169 error = tcp_output(tp);
3172 sent = tp->snd_nxt - nextrexmt;
3177 tcpstat.tcps_sndsackpack++;
3178 tcpstat.tcps_sndsackbyte += sent;
3179 if (SEQ_LT(nextrexmt, old_snd_max) &&
3180 SEQ_LT(tp->rexmt_high, tp->snd_nxt))
3181 tp->rexmt_high = seq_min(tp->snd_nxt, old_snd_max);
3183 if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
3184 tp->snd_nxt = old_snd_nxt;
3185 tp->snd_cwnd = ocwnd;