2 * Copyright (c) 2002, 2003, 2004 Jeffrey M. Hsu. All rights reserved.
3 * Copyright (c) 2002, 2003, 2004 The DragonFly Project. All rights reserved.
5 * This code is derived from software contributed to The DragonFly Project
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. Neither the name of The DragonFly Project nor the names of its
17 * contributors may be used to endorse or promote products derived
18 * from this software without specific, prior written permission.
20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
23 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
24 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
25 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
26 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
27 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
28 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
29 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
30 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
36 * The Regents of the University of California. All rights reserved.
38 * Redistribution and use in source and binary forms, with or without
39 * modification, are permitted provided that the following conditions
41 * 1. Redistributions of source code must retain the above copyright
42 * notice, this list of conditions and the following disclaimer.
43 * 2. Redistributions in binary form must reproduce the above copyright
44 * notice, this list of conditions and the following disclaimer in the
45 * documentation and/or other materials provided with the distribution.
46 * 3. All advertising materials mentioning features or use of this software
47 * must display the following acknowledgement:
48 * This product includes software developed by the University of
49 * California, Berkeley and its contributors.
50 * 4. Neither the name of the University nor the names of its contributors
51 * may be used to endorse or promote products derived from this software
52 * without specific prior written permission.
54 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
55 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
56 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
57 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
58 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
59 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
60 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
61 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
62 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
63 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
66 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95
67 * $FreeBSD: src/sys/netinet/tcp_input.c,v 1.107.2.38 2003/05/21 04:46:41 cjc Exp $
68 * $DragonFly: src/sys/netinet/tcp_input.c,v 1.68 2008/08/22 09:14:17 sephe Exp $
71 #include "opt_ipfw.h" /* for ipfw_fwd */
73 #include "opt_inet6.h"
74 #include "opt_ipsec.h"
75 #include "opt_tcpdebug.h"
76 #include "opt_tcp_input.h"
78 #include <sys/param.h>
79 #include <sys/systm.h>
80 #include <sys/kernel.h>
81 #include <sys/sysctl.h>
82 #include <sys/malloc.h>
84 #include <sys/proc.h> /* for proc0 declaration */
85 #include <sys/protosw.h>
86 #include <sys/socket.h>
87 #include <sys/socketvar.h>
88 #include <sys/syslog.h>
89 #include <sys/in_cksum.h>
91 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
92 #include <machine/stdarg.h>
95 #include <net/route.h>
97 #include <netinet/in.h>
98 #include <netinet/in_systm.h>
99 #include <netinet/ip.h>
100 #include <netinet/ip_icmp.h> /* for ICMP_BANDLIM */
101 #include <netinet/in_var.h>
102 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */
103 #include <netinet/in_pcb.h>
104 #include <netinet/ip_var.h>
105 #include <netinet/ip6.h>
106 #include <netinet/icmp6.h>
107 #include <netinet6/nd6.h>
108 #include <netinet6/ip6_var.h>
109 #include <netinet6/in6_pcb.h>
110 #include <netinet/tcp.h>
111 #include <netinet/tcp_fsm.h>
112 #include <netinet/tcp_seq.h>
113 #include <netinet/tcp_timer.h>
114 #include <netinet/tcp_timer2.h>
115 #include <netinet/tcp_var.h>
116 #include <netinet6/tcp6_var.h>
117 #include <netinet/tcpip.h>
120 #include <netinet/tcp_debug.h>
122 u_char tcp_saveipgen[40]; /* the size must be of max ip header, now IPv6 */
123 struct tcphdr tcp_savetcp;
127 #include <netproto/ipsec/ipsec.h>
128 #include <netproto/ipsec/ipsec6.h>
132 #include <netinet6/ipsec.h>
133 #include <netinet6/ipsec6.h>
134 #include <netproto/key/key.h>
137 MALLOC_DEFINE(M_TSEGQ, "tseg_qent", "TCP segment queue entry");
139 static int log_in_vain = 0;
140 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
141 &log_in_vain, 0, "Log all incoming TCP connections");
143 static int blackhole = 0;
144 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW,
145 &blackhole, 0, "Do not send RST when dropping refused connections");
147 int tcp_delack_enabled = 1;
148 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW,
149 &tcp_delack_enabled, 0,
150 "Delay ACK to try and piggyback it onto a data packet");
152 #ifdef TCP_DROP_SYNFIN
153 static int drop_synfin = 0;
154 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW,
155 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set");
158 static int tcp_do_limitedtransmit = 1;
159 SYSCTL_INT(_net_inet_tcp, OID_AUTO, limitedtransmit, CTLFLAG_RW,
160 &tcp_do_limitedtransmit, 0, "Enable RFC 3042 (Limited Transmit)");
162 static int tcp_do_early_retransmit = 1;
163 SYSCTL_INT(_net_inet_tcp, OID_AUTO, earlyretransmit, CTLFLAG_RW,
164 &tcp_do_early_retransmit, 0, "Early retransmit");
166 int tcp_aggregate_acks = 1;
167 SYSCTL_INT(_net_inet_tcp, OID_AUTO, aggregate_acks, CTLFLAG_RW,
168 &tcp_aggregate_acks, 0, "Aggregate built-up acks into one ack");
170 int tcp_do_rfc3390 = 1;
171 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW,
173 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
175 static int tcp_do_eifel_detect = 1;
176 SYSCTL_INT(_net_inet_tcp, OID_AUTO, eifel, CTLFLAG_RW,
177 &tcp_do_eifel_detect, 0, "Eifel detection algorithm (RFC 3522)");
179 static int tcp_do_abc = 1;
180 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc, CTLFLAG_RW,
182 "TCP Appropriate Byte Counting (RFC 3465)");
185 * Define as tunable for easy testing with SACK on and off.
186 * Warning: do not change setting in the middle of an existing active TCP flow,
187 * else strange things might happen to that flow.
190 SYSCTL_INT(_net_inet_tcp, OID_AUTO, sack, CTLFLAG_RW,
191 &tcp_do_sack, 0, "Enable SACK Algorithms");
193 int tcp_do_smartsack = 1;
194 SYSCTL_INT(_net_inet_tcp, OID_AUTO, smartsack, CTLFLAG_RW,
195 &tcp_do_smartsack, 0, "Enable Smart SACK Algorithms");
197 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0,
198 "TCP Segment Reassembly Queue");
200 int tcp_reass_maxseg = 0;
201 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RD,
202 &tcp_reass_maxseg, 0,
203 "Global maximum number of TCP Segments in Reassembly Queue");
205 int tcp_reass_qsize = 0;
206 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD,
208 "Global number of TCP Segments currently in Reassembly Queue");
210 static int tcp_reass_overflows = 0;
211 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD,
212 &tcp_reass_overflows, 0,
213 "Global number of TCP Segment Reassembly Queue Overflows");
215 int tcp_do_autorcvbuf = 1;
216 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_RW,
217 &tcp_do_autorcvbuf, 0, "Enable automatic receive buffer sizing");
219 int tcp_autorcvbuf_inc = 16*1024;
220 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_RW,
221 &tcp_autorcvbuf_inc, 0,
222 "Incrementor step size of automatic receive buffer");
224 int tcp_autorcvbuf_max = 2*1024*1024;
225 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_RW,
226 &tcp_autorcvbuf_max, 0, "Max size of automatic receive buffer");
229 static void tcp_dooptions(struct tcpopt *, u_char *, int, boolean_t);
230 static void tcp_pulloutofband(struct socket *,
231 struct tcphdr *, struct mbuf *, int);
232 static int tcp_reass(struct tcpcb *, struct tcphdr *, int *,
234 static void tcp_xmit_timer(struct tcpcb *, int);
235 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *, int);
236 static void tcp_sack_rexmt(struct tcpcb *, struct tcphdr *);
238 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
240 #define ND6_HINT(tp) \
242 if ((tp) && (tp)->t_inpcb && \
243 ((tp)->t_inpcb->inp_vflag & INP_IPV6) && \
244 (tp)->t_inpcb->in6p_route.ro_rt) \
245 nd6_nud_hint((tp)->t_inpcb->in6p_route.ro_rt, NULL, 0); \
252 * Indicate whether this ack should be delayed. We can delay the ack if
253 * - delayed acks are enabled and
254 * - there is no delayed ack timer in progress and
255 * - our last ack wasn't a 0-sized window. We never want to delay
256 * the ack that opens up a 0-sized window.
258 #define DELAY_ACK(tp) \
259 (tcp_delack_enabled && !tcp_callout_pending(tp, tp->tt_delack) && \
260 !(tp->t_flags & TF_RXWIN0SENT))
262 #define acceptable_window_update(tp, th, tiwin) \
263 (SEQ_LT(tp->snd_wl1, th->th_seq) || \
264 (tp->snd_wl1 == th->th_seq && \
265 (SEQ_LT(tp->snd_wl2, th->th_ack) || \
266 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))
269 tcp_reass(struct tcpcb *tp, struct tcphdr *th, int *tlenp, struct mbuf *m)
272 struct tseg_qent *p = NULL;
273 struct tseg_qent *te;
274 struct socket *so = tp->t_inpcb->inp_socket;
278 * Call with th == NULL after become established to
279 * force pre-ESTABLISHED data up to user socket.
285 * Limit the number of segments in the reassembly queue to prevent
286 * holding on to too many segments (and thus running out of mbufs).
287 * Make sure to let the missing segment through which caused this
288 * queue. Always keep one global queue entry spare to be able to
289 * process the missing segment.
291 if (th->th_seq != tp->rcv_nxt &&
292 tcp_reass_qsize + 1 >= tcp_reass_maxseg) {
293 tcp_reass_overflows++;
294 tcpstat.tcps_rcvmemdrop++;
296 /* no SACK block to report */
297 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
301 /* Allocate a new queue entry. */
302 MALLOC(te, struct tseg_qent *, sizeof(struct tseg_qent), M_TSEGQ,
303 M_INTWAIT | M_NULLOK);
305 tcpstat.tcps_rcvmemdrop++;
307 /* no SACK block to report */
308 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
314 * Find a segment which begins after this one does.
316 LIST_FOREACH(q, &tp->t_segq, tqe_q) {
317 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
323 * If there is a preceding segment, it may provide some of
324 * our data already. If so, drop the data from the incoming
325 * segment. If it provides all of our data, drop us.
330 /* conversion to int (in i) handles seq wraparound */
331 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
332 if (i > 0) { /* overlaps preceding segment */
333 tp->t_flags |= (TF_DUPSEG | TF_ENCLOSESEG);
334 /* enclosing block starts w/ preceding segment */
335 tp->encloseblk.rblk_start = p->tqe_th->th_seq;
337 /* preceding encloses incoming segment */
338 tp->encloseblk.rblk_end = p->tqe_th->th_seq +
340 tcpstat.tcps_rcvduppack++;
341 tcpstat.tcps_rcvdupbyte += *tlenp;
346 * Try to present any queued data
347 * at the left window edge to the user.
348 * This is needed after the 3-WHS
351 goto present; /* ??? */
356 /* incoming segment end is enclosing block end */
357 tp->encloseblk.rblk_end = th->th_seq + *tlenp +
358 ((th->th_flags & TH_FIN) != 0);
359 /* trim end of reported D-SACK block */
360 tp->reportblk.rblk_end = th->th_seq;
363 tcpstat.tcps_rcvoopack++;
364 tcpstat.tcps_rcvoobyte += *tlenp;
367 * While we overlap succeeding segments trim them or,
368 * if they are completely covered, dequeue them.
371 tcp_seq_diff_t i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
372 tcp_seq qend = q->tqe_th->th_seq + q->tqe_len;
373 struct tseg_qent *nq;
377 if (!(tp->t_flags & TF_DUPSEG)) { /* first time through */
378 tp->t_flags |= (TF_DUPSEG | TF_ENCLOSESEG);
379 tp->encloseblk = tp->reportblk;
380 /* report trailing duplicate D-SACK segment */
381 tp->reportblk.rblk_start = q->tqe_th->th_seq;
383 if ((tp->t_flags & TF_ENCLOSESEG) &&
384 SEQ_GT(qend, tp->encloseblk.rblk_end)) {
385 /* extend enclosing block if one exists */
386 tp->encloseblk.rblk_end = qend;
388 if (i < q->tqe_len) {
389 q->tqe_th->th_seq += i;
395 nq = LIST_NEXT(q, tqe_q);
396 LIST_REMOVE(q, tqe_q);
403 /* Insert the new segment queue entry into place. */
406 te->tqe_len = *tlenp;
408 /* check if can coalesce with following segment */
409 if (q != NULL && (th->th_seq + *tlenp == q->tqe_th->th_seq)) {
410 tcp_seq tend = te->tqe_th->th_seq + te->tqe_len;
412 te->tqe_len += q->tqe_len;
413 if (q->tqe_th->th_flags & TH_FIN)
414 te->tqe_th->th_flags |= TH_FIN;
415 m_cat(te->tqe_m, q->tqe_m);
416 tp->encloseblk.rblk_end = tend;
418 * When not reporting a duplicate segment, use
419 * the larger enclosing block as the SACK block.
421 if (!(tp->t_flags & TF_DUPSEG))
422 tp->reportblk.rblk_end = tend;
423 LIST_REMOVE(q, tqe_q);
429 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
431 /* check if can coalesce with preceding segment */
432 if (p->tqe_th->th_seq + p->tqe_len == th->th_seq) {
433 p->tqe_len += te->tqe_len;
434 m_cat(p->tqe_m, te->tqe_m);
435 tp->encloseblk.rblk_start = p->tqe_th->th_seq;
437 * When not reporting a duplicate segment, use
438 * the larger enclosing block as the SACK block.
440 if (!(tp->t_flags & TF_DUPSEG))
441 tp->reportblk.rblk_start = p->tqe_th->th_seq;
445 LIST_INSERT_AFTER(p, te, tqe_q);
450 * Present data to user, advancing rcv_nxt through
451 * completed sequence space.
453 if (!TCPS_HAVEESTABLISHED(tp->t_state))
455 q = LIST_FIRST(&tp->t_segq);
456 if (q == NULL || q->tqe_th->th_seq != tp->rcv_nxt)
458 tp->rcv_nxt += q->tqe_len;
459 if (!(tp->t_flags & TF_DUPSEG)) {
460 /* no SACK block to report since ACK advanced */
461 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
463 /* no enclosing block to report since ACK advanced */
464 tp->t_flags &= ~TF_ENCLOSESEG;
465 flags = q->tqe_th->th_flags & TH_FIN;
466 LIST_REMOVE(q, tqe_q);
467 KASSERT(LIST_EMPTY(&tp->t_segq) ||
468 LIST_FIRST(&tp->t_segq)->tqe_th->th_seq != tp->rcv_nxt,
469 ("segment not coalesced"));
470 if (so->so_state & SS_CANTRCVMORE)
473 ssb_appendstream(&so->so_rcv, q->tqe_m);
482 * TCP input routine, follows pages 65-76 of the
483 * protocol specification dated September, 1981 very closely.
487 tcp6_input(struct mbuf **mp, int *offp, int proto)
489 struct mbuf *m = *mp;
490 struct in6_ifaddr *ia6;
492 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
495 * draft-itojun-ipv6-tcp-to-anycast
496 * better place to put this in?
498 ia6 = ip6_getdstifaddr(m);
499 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
502 ip6 = mtod(m, struct ip6_hdr *);
503 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
504 offsetof(struct ip6_hdr, ip6_dst));
505 return (IPPROTO_DONE);
508 tcp_input(m, *offp, proto);
509 return (IPPROTO_DONE);
514 tcp_input(struct mbuf *m, ...)
519 struct ip *ip = NULL;
521 struct inpcb *inp = NULL;
527 struct tcpcb *tp = NULL;
529 struct socket *so = 0;
531 boolean_t ourfinisacked, needoutput = FALSE;
534 struct tcpopt to; /* options in this segment */
535 struct sockaddr_in *next_hop = NULL;
536 int rstreason; /* For badport_bandlim accounting purposes */
538 struct ip6_hdr *ip6 = NULL;
542 const boolean_t isipv6 = FALSE;
549 off0 = __va_arg(ap, int);
550 proto = __va_arg(ap, int);
553 tcpstat.tcps_rcvtotal++;
555 if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
558 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
559 KKASSERT(mtag != NULL);
560 next_hop = m_tag_data(mtag);
564 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? TRUE : FALSE;
568 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */
569 ip6 = mtod(m, struct ip6_hdr *);
570 tlen = (sizeof *ip6) + ntohs(ip6->ip6_plen) - off0;
571 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) {
572 tcpstat.tcps_rcvbadsum++;
575 th = (struct tcphdr *)((caddr_t)ip6 + off0);
578 * Be proactive about unspecified IPv6 address in source.
579 * As we use all-zero to indicate unbounded/unconnected pcb,
580 * unspecified IPv6 address can be used to confuse us.
582 * Note that packets with unspecified IPv6 destination is
583 * already dropped in ip6_input.
585 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
591 * Get IP and TCP header together in first mbuf.
592 * Note: IP leaves IP header in first mbuf.
594 if (off0 > sizeof(struct ip)) {
596 off0 = sizeof(struct ip);
598 /* already checked and pulled up in ip_demux() */
599 KASSERT(m->m_len >= sizeof(struct tcpiphdr),
600 ("TCP header not in one mbuf: m->m_len %d", m->m_len));
601 ip = mtod(m, struct ip *);
602 ipov = (struct ipovly *)ip;
603 th = (struct tcphdr *)((caddr_t)ip + off0);
606 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
607 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
608 th->th_sum = m->m_pkthdr.csum_data;
610 th->th_sum = in_pseudo(ip->ip_src.s_addr,
612 htonl(m->m_pkthdr.csum_data +
615 th->th_sum ^= 0xffff;
618 * Checksum extended TCP header and data.
620 len = sizeof(struct ip) + tlen;
621 bzero(ipov->ih_x1, sizeof ipov->ih_x1);
622 ipov->ih_len = (u_short)tlen;
623 ipov->ih_len = htons(ipov->ih_len);
624 th->th_sum = in_cksum(m, len);
627 tcpstat.tcps_rcvbadsum++;
631 /* Re-initialization for later version check */
632 ip->ip_v = IPVERSION;
637 * Check that TCP offset makes sense,
638 * pull out TCP options and adjust length. XXX
640 off = th->th_off << 2;
641 /* already checked and pulled up in ip_demux() */
642 KASSERT(off >= sizeof(struct tcphdr) && off <= tlen,
643 ("bad TCP data offset %d (tlen %d)", off, tlen));
644 tlen -= off; /* tlen is used instead of ti->ti_len */
645 if (off > sizeof(struct tcphdr)) {
647 IP6_EXTHDR_CHECK(m, off0, off, );
648 ip6 = mtod(m, struct ip6_hdr *);
649 th = (struct tcphdr *)((caddr_t)ip6 + off0);
651 /* already pulled up in ip_demux() */
652 KASSERT(m->m_len >= sizeof(struct ip) + off,
653 ("TCP header and options not in one mbuf: "
654 "m_len %d, off %d", m->m_len, off));
656 optlen = off - sizeof(struct tcphdr);
657 optp = (u_char *)(th + 1);
659 thflags = th->th_flags;
661 #ifdef TCP_DROP_SYNFIN
663 * If the drop_synfin option is enabled, drop all packets with
664 * both the SYN and FIN bits set. This prevents e.g. nmap from
665 * identifying the TCP/IP stack.
667 * This is a violation of the TCP specification.
669 if (drop_synfin && (thflags & (TH_SYN | TH_FIN)) == (TH_SYN | TH_FIN))
674 * Convert TCP protocol specific fields to host format.
676 th->th_seq = ntohl(th->th_seq);
677 th->th_ack = ntohl(th->th_ack);
678 th->th_win = ntohs(th->th_win);
679 th->th_urp = ntohs(th->th_urp);
682 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options,
683 * until after ip6_savecontrol() is called and before other functions
684 * which don't want those proto headers.
685 * Because ip6_savecontrol() is going to parse the mbuf to
686 * search for data to be passed up to user-land, it wants mbuf
687 * parameters to be unchanged.
688 * XXX: the call of ip6_savecontrol() has been obsoleted based on
689 * latest version of the advanced API (20020110).
691 drop_hdrlen = off0 + off;
694 * Locate pcb for segment.
697 /* IPFIREWALL_FORWARD section */
698 if (next_hop != NULL && !isipv6) { /* IPv6 support is not there yet */
700 * Transparently forwarded. Pretend to be the destination.
701 * already got one like this?
703 cpu = mycpu->gd_cpuid;
704 inp = in_pcblookup_hash(&tcbinfo[cpu],
705 ip->ip_src, th->th_sport,
706 ip->ip_dst, th->th_dport,
707 0, m->m_pkthdr.rcvif);
710 * It's new. Try to find the ambushing socket.
714 * The rest of the ipfw code stores the port in
716 * (The IP address is still in network order.)
718 in_port_t dport = next_hop->sin_port ?
719 htons(next_hop->sin_port) :
722 cpu = tcp_addrcpu(ip->ip_src.s_addr, th->th_sport,
723 next_hop->sin_addr.s_addr, dport);
724 inp = in_pcblookup_hash(&tcbinfo[cpu],
725 ip->ip_src, th->th_sport,
726 next_hop->sin_addr, dport,
727 1, m->m_pkthdr.rcvif);
731 inp = in6_pcblookup_hash(&tcbinfo[0],
732 &ip6->ip6_src, th->th_sport,
733 &ip6->ip6_dst, th->th_dport,
734 1, m->m_pkthdr.rcvif);
736 cpu = mycpu->gd_cpuid;
737 inp = in_pcblookup_hash(&tcbinfo[cpu],
738 ip->ip_src, th->th_sport,
739 ip->ip_dst, th->th_dport,
740 1, m->m_pkthdr.rcvif);
745 * If the state is CLOSED (i.e., TCB does not exist) then
746 * all data in the incoming segment is discarded.
747 * If the TCB exists but is in CLOSED state, it is embryonic,
748 * but should either do a listen or a connect soon.
753 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2];
755 char dbuf[sizeof "aaa.bbb.ccc.ddd"];
756 char sbuf[sizeof "aaa.bbb.ccc.ddd"];
760 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst));
763 strcat(sbuf, ip6_sprintf(&ip6->ip6_src));
766 strcpy(dbuf, inet_ntoa(ip->ip_dst));
767 strcpy(sbuf, inet_ntoa(ip->ip_src));
769 switch (log_in_vain) {
771 if (!(thflags & TH_SYN))
775 "Connection attempt to TCP %s:%d "
776 "from %s:%d flags:0x%02x\n",
777 dbuf, ntohs(th->th_dport), sbuf,
778 ntohs(th->th_sport), thflags);
787 if (thflags & TH_SYN)
796 rstreason = BANDLIM_RST_CLOSEDPORT;
802 if (ipsec6_in_reject_so(m, inp->inp_socket)) {
803 ipsec6stat.in_polvio++;
807 if (ipsec4_in_reject_so(m, inp->inp_socket)) {
808 ipsecstat.in_polvio++;
815 if (ipsec6_in_reject(m, inp))
818 if (ipsec4_in_reject(m, inp))
822 /* Check the minimum TTL for socket. */
824 if ((isipv6 ? ip6->ip6_hlim : ip->ip_ttl) < inp->inp_ip_minttl)
830 rstreason = BANDLIM_RST_CLOSEDPORT;
833 if (tp->t_state <= TCPS_CLOSED)
836 /* Unscale the window into a 32-bit value. */
837 if (!(thflags & TH_SYN))
838 tiwin = th->th_win << tp->snd_scale;
842 so = inp->inp_socket;
845 if (so->so_options & SO_DEBUG) {
846 ostate = tp->t_state;
848 bcopy(ip6, tcp_saveipgen, sizeof(*ip6));
850 bcopy(ip, tcp_saveipgen, sizeof(*ip));
855 bzero(&to, sizeof to);
857 if (so->so_options & SO_ACCEPTCONN) {
858 struct in_conninfo inc;
861 inc.inc_isipv6 = (isipv6 == TRUE);
864 inc.inc6_faddr = ip6->ip6_src;
865 inc.inc6_laddr = ip6->ip6_dst;
866 inc.inc6_route.ro_rt = NULL; /* XXX */
868 inc.inc_faddr = ip->ip_src;
869 inc.inc_laddr = ip->ip_dst;
870 inc.inc_route.ro_rt = NULL; /* XXX */
872 inc.inc_fport = th->th_sport;
873 inc.inc_lport = th->th_dport;
876 * If the state is LISTEN then ignore segment if it contains
877 * a RST. If the segment contains an ACK then it is bad and
878 * send a RST. If it does not contain a SYN then it is not
879 * interesting; drop it.
881 * If the state is SYN_RECEIVED (syncache) and seg contains
882 * an ACK, but not for our SYN/ACK, send a RST. If the seg
883 * contains a RST, check the sequence number to see if it
884 * is a valid reset segment.
886 if ((thflags & (TH_RST | TH_ACK | TH_SYN)) != TH_SYN) {
887 if ((thflags & (TH_RST | TH_ACK | TH_SYN)) == TH_ACK) {
888 if (!syncache_expand(&inc, th, &so, m)) {
890 * No syncache entry, or ACK was not
891 * for our SYN/ACK. Send a RST.
893 tcpstat.tcps_badsyn++;
894 rstreason = BANDLIM_RST_OPENPORT;
899 * Could not complete 3-way handshake,
900 * connection is being closed down, and
901 * syncache will free mbuf.
905 * Socket is created in state SYN_RECEIVED.
906 * Continue processing segment.
911 * This is what would have happened in
912 * tcp_output() when the SYN,ACK was sent.
914 tp->snd_up = tp->snd_una;
915 tp->snd_max = tp->snd_nxt = tp->iss + 1;
916 tp->last_ack_sent = tp->rcv_nxt;
918 * XXX possible bug - it doesn't appear that tp->snd_wnd is unscaled
919 * until the _second_ ACK is received:
920 * rcv SYN (set wscale opts) --> send SYN/ACK, set snd_wnd = window.
921 * rcv ACK, calculate tiwin --> process SYN_RECEIVED, determine wscale,
922 * move to ESTAB, set snd_wnd to tiwin.
924 tp->snd_wnd = tiwin; /* unscaled */
927 if (thflags & TH_RST) {
928 syncache_chkrst(&inc, th);
931 if (thflags & TH_ACK) {
932 syncache_badack(&inc);
933 tcpstat.tcps_badsyn++;
934 rstreason = BANDLIM_RST_OPENPORT;
941 * Segment's flags are (SYN) or (SYN | FIN).
945 * If deprecated address is forbidden,
946 * we do not accept SYN to deprecated interface
947 * address to prevent any new inbound connection from
948 * getting established.
949 * When we do not accept SYN, we send a TCP RST,
950 * with deprecated source address (instead of dropping
951 * it). We compromise it as it is much better for peer
952 * to send a RST, and RST will be the final packet
955 * If we do not forbid deprecated addresses, we accept
956 * the SYN packet. RFC2462 does not suggest dropping
958 * If we decipher RFC2462 5.5.4, it says like this:
959 * 1. use of deprecated addr with existing
960 * communication is okay - "SHOULD continue to be
962 * 2. use of it with new communication:
963 * (2a) "SHOULD NOT be used if alternate address
964 * with sufficient scope is available"
965 * (2b) nothing mentioned otherwise.
966 * Here we fall into (2b) case as we have no choice in
967 * our source address selection - we must obey the peer.
969 * The wording in RFC2462 is confusing, and there are
970 * multiple description text for deprecated address
971 * handling - worse, they are not exactly the same.
972 * I believe 5.5.4 is the best one, so we follow 5.5.4.
974 if (isipv6 && !ip6_use_deprecated) {
975 struct in6_ifaddr *ia6;
977 if ((ia6 = ip6_getdstifaddr(m)) &&
978 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
980 rstreason = BANDLIM_RST_OPENPORT;
986 * If it is from this socket, drop it, it must be forged.
987 * Don't bother responding if the destination was a broadcast.
989 if (th->th_dport == th->th_sport) {
991 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
995 if (ip->ip_dst.s_addr == ip->ip_src.s_addr)
1000 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
1002 * Note that it is quite possible to receive unicast
1003 * link-layer packets with a broadcast IP address. Use
1004 * in_broadcast() to find them.
1006 if (m->m_flags & (M_BCAST | M_MCAST))
1009 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1010 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
1013 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1014 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1015 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1016 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
1020 * SYN appears to be valid; create compressed TCP state
1021 * for syncache, or perform t/tcp connection.
1023 if (so->so_qlen <= so->so_qlimit) {
1024 tcp_dooptions(&to, optp, optlen, TRUE);
1025 if (!syncache_add(&inc, &to, th, &so, m))
1029 * Entry added to syncache, mbuf used to
1030 * send SYN,ACK packet.
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 tcp_callout_reset(tp, tp->tt_delack,
1052 tcp_delacktime, tcp_timer_delack);
1053 tp->t_flags |= TF_NEEDSYN;
1055 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
1058 tcpstat.tcps_connects++;
1066 /* should not happen - syncache should pick up these connections */
1067 KASSERT(tp->t_state != TCPS_LISTEN, ("tcp_input: TCPS_LISTEN state"));
1070 * This is the second part of the MSS DoS prevention code (after
1071 * minmss on the sending side) and it deals with too many too small
1072 * tcp packets in a too short timeframe (1 second).
1074 * XXX Removed. This code was crap. It does not scale to network
1075 * speed, and default values break NFS. Gone.
1080 * Segment received on connection.
1082 * Reset idle time and keep-alive timer. Don't waste time if less
1083 * then a second has elapsed. Only update t_rcvtime for non-SYN
1086 * Handle the case where one side thinks the connection is established
1087 * but the other side has, say, rebooted without cleaning out the
1088 * connection. The SYNs could be construed as an attack and wind
1089 * up ignored, but in case it isn't an attack we can validate the
1090 * connection by forcing a keepalive.
1092 if (TCPS_HAVEESTABLISHED(tp->t_state) && (ticks - tp->t_rcvtime) > hz) {
1093 if ((thflags & (TH_SYN | TH_ACK)) == TH_SYN) {
1094 tp->t_flags |= TF_KEEPALIVE;
1095 tcp_callout_reset(tp, tp->tt_keep, hz / 2,
1098 tp->t_rcvtime = ticks;
1099 tp->t_flags &= ~TF_KEEPALIVE;
1100 tcp_callout_reset(tp, tp->tt_keep, tcp_keepidle,
1107 * XXX this is tradtitional behavior, may need to be cleaned up.
1109 tcp_dooptions(&to, optp, optlen, (thflags & TH_SYN) != 0);
1110 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1111 if (to.to_flags & TOF_SCALE) {
1112 tp->t_flags |= TF_RCVD_SCALE;
1113 tp->requested_s_scale = to.to_requested_s_scale;
1115 if (to.to_flags & TOF_TS) {
1116 tp->t_flags |= TF_RCVD_TSTMP;
1117 tp->ts_recent = to.to_tsval;
1118 tp->ts_recent_age = ticks;
1120 if (to.to_flags & TOF_MSS)
1121 tcp_mss(tp, to.to_mss);
1123 * Only set the TF_SACK_PERMITTED per-connection flag
1124 * if we got a SACK_PERMITTED option from the other side
1125 * and the global tcp_do_sack variable is true.
1127 if (tcp_do_sack && (to.to_flags & TOF_SACK_PERMITTED))
1128 tp->t_flags |= TF_SACK_PERMITTED;
1132 * Header prediction: check for the two common cases
1133 * of a uni-directional data xfer. If the packet has
1134 * no control flags, is in-sequence, the window didn't
1135 * change and we're not retransmitting, it's a
1136 * candidate. If the length is zero and the ack moved
1137 * forward, we're the sender side of the xfer. Just
1138 * free the data acked & wake any higher level process
1139 * that was blocked waiting for space. If the length
1140 * is non-zero and the ack didn't move, we're the
1141 * receiver side. If we're getting packets in-order
1142 * (the reassembly queue is empty), add the data to
1143 * the socket buffer and note that we need a delayed ack.
1144 * Make sure that the hidden state-flags are also off.
1145 * Since we check for TCPS_ESTABLISHED above, it can only
1148 if (tp->t_state == TCPS_ESTABLISHED &&
1149 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1150 !(tp->t_flags & (TF_NEEDSYN | TF_NEEDFIN)) &&
1151 (!(to.to_flags & TOF_TS) ||
1152 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) &&
1153 th->th_seq == tp->rcv_nxt &&
1154 tp->snd_nxt == tp->snd_max) {
1157 * If last ACK falls within this segment's sequence numbers,
1158 * record the timestamp.
1159 * NOTE that the test is modified according to the latest
1160 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1162 if ((to.to_flags & TOF_TS) &&
1163 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1164 tp->ts_recent_age = ticks;
1165 tp->ts_recent = to.to_tsval;
1169 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1170 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1171 tp->snd_cwnd >= tp->snd_wnd &&
1172 !IN_FASTRECOVERY(tp)) {
1174 * This is a pure ack for outstanding data.
1176 ++tcpstat.tcps_predack;
1178 * "bad retransmit" recovery
1180 * If Eifel detection applies, then
1181 * it is deterministic, so use it
1182 * unconditionally over the old heuristic.
1183 * Otherwise, fall back to the old heuristic.
1185 if (tcp_do_eifel_detect &&
1186 (to.to_flags & TOF_TS) && to.to_tsecr &&
1187 (tp->t_flags & TF_FIRSTACCACK)) {
1188 /* Eifel detection applicable. */
1189 if (to.to_tsecr < tp->t_rexmtTS) {
1190 tcp_revert_congestion_state(tp);
1191 ++tcpstat.tcps_eifeldetected;
1193 } else if (tp->t_rxtshift == 1 &&
1194 ticks < tp->t_badrxtwin) {
1195 tcp_revert_congestion_state(tp);
1196 ++tcpstat.tcps_rttdetected;
1198 tp->t_flags &= ~(TF_FIRSTACCACK |
1199 TF_FASTREXMT | TF_EARLYREXMT);
1201 * Recalculate the retransmit timer / rtt.
1203 * Some machines (certain windows boxes)
1204 * send broken timestamp replies during the
1205 * SYN+ACK phase, ignore timestamps of 0.
1207 if ((to.to_flags & TOF_TS) && to.to_tsecr) {
1209 ticks - to.to_tsecr + 1);
1210 } else if (tp->t_rtttime &&
1211 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1213 ticks - tp->t_rtttime);
1215 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1216 acked = th->th_ack - tp->snd_una;
1217 tcpstat.tcps_rcvackpack++;
1218 tcpstat.tcps_rcvackbyte += acked;
1219 sbdrop(&so->so_snd.sb, acked);
1220 tp->snd_recover = th->th_ack - 1;
1221 tp->snd_una = th->th_ack;
1224 * Update window information.
1226 if (tiwin != tp->snd_wnd &&
1227 acceptable_window_update(tp, th, tiwin)) {
1228 /* keep track of pure window updates */
1229 if (tp->snd_wl2 == th->th_ack &&
1230 tiwin > tp->snd_wnd)
1231 tcpstat.tcps_rcvwinupd++;
1232 tp->snd_wnd = tiwin;
1233 tp->snd_wl1 = th->th_seq;
1234 tp->snd_wl2 = th->th_ack;
1235 if (tp->snd_wnd > tp->max_sndwnd)
1236 tp->max_sndwnd = tp->snd_wnd;
1239 ND6_HINT(tp); /* some progress has been done */
1241 * If all outstanding data are acked, stop
1242 * retransmit timer, otherwise restart timer
1243 * using current (possibly backed-off) value.
1244 * If process is waiting for space,
1245 * wakeup/selwakeup/signal. If data
1246 * are ready to send, let tcp_output
1247 * decide between more output or persist.
1249 if (tp->snd_una == tp->snd_max) {
1250 tcp_callout_stop(tp, tp->tt_rexmt);
1251 } else if (!tcp_callout_active(tp,
1253 tcp_callout_reset(tp, tp->tt_rexmt,
1254 tp->t_rxtcur, tcp_timer_rexmt);
1257 if (so->so_snd.ssb_cc > 0)
1261 } else if (tiwin == tp->snd_wnd &&
1262 th->th_ack == tp->snd_una &&
1263 LIST_EMPTY(&tp->t_segq) &&
1264 tlen <= ssb_space(&so->so_rcv)) {
1265 u_long newsize = 0; /* automatic sockbuf scaling */
1267 * This is a pure, in-sequence data packet
1268 * with nothing on the reassembly queue and
1269 * we have enough buffer space to take it.
1271 ++tcpstat.tcps_preddat;
1272 tp->rcv_nxt += tlen;
1273 tcpstat.tcps_rcvpack++;
1274 tcpstat.tcps_rcvbyte += tlen;
1275 ND6_HINT(tp); /* some progress has been done */
1277 * Automatic sizing of receive socket buffer. Often the send
1278 * buffer size is not optimally adjusted to the actual network
1279 * conditions at hand (delay bandwidth product). Setting the
1280 * buffer size too small limits throughput on links with high
1281 * bandwidth and high delay (eg. trans-continental/oceanic links).
1283 * On the receive side the socket buffer memory is only rarely
1284 * used to any significant extent. This allows us to be much
1285 * more aggressive in scaling the receive socket buffer. For
1286 * the case that the buffer space is actually used to a large
1287 * extent and we run out of kernel memory we can simply drop
1288 * the new segments; TCP on the sender will just retransmit it
1289 * later. Setting the buffer size too big may only consume too
1290 * much kernel memory if the application doesn't read() from
1291 * the socket or packet loss or reordering makes use of the
1294 * The criteria to step up the receive buffer one notch are:
1295 * 1. the number of bytes received during the time it takes
1296 * one timestamp to be reflected back to us (the RTT);
1297 * 2. received bytes per RTT is within seven eighth of the
1298 * current socket buffer size;
1299 * 3. receive buffer size has not hit maximal automatic size;
1301 * This algorithm does one step per RTT at most and only if
1302 * we receive a bulk stream w/o packet losses or reorderings.
1303 * Shrinking the buffer during idle times is not necessary as
1304 * it doesn't consume any memory when idle.
1306 * TODO: Only step up if the application is actually serving
1307 * the buffer to better manage the socket buffer resources.
1309 if (tcp_do_autorcvbuf &&
1311 (so->so_rcv.ssb_flags & SSB_AUTOSIZE)) {
1312 if (to.to_tsecr > tp->rfbuf_ts &&
1313 to.to_tsecr - tp->rfbuf_ts < hz) {
1315 (so->so_rcv.ssb_hiwat / 8 * 7) &&
1316 so->so_rcv.ssb_hiwat <
1317 tcp_autorcvbuf_max) {
1319 ulmin(so->so_rcv.ssb_hiwat +
1321 tcp_autorcvbuf_max);
1323 /* Start over with next RTT. */
1327 tp->rfbuf_cnt += tlen; /* add up */
1330 * Add data to socket buffer.
1332 if (so->so_state & SS_CANTRCVMORE) {
1336 * Set new socket buffer size, give up when
1339 * Adjusting the size can mess up ACK
1340 * sequencing when pure window updates are
1341 * being avoided (which is the default),
1345 tp->t_flags |= TF_RXRESIZED;
1346 if (!ssb_reserve(&so->so_rcv, newsize,
1348 so->so_rcv.ssb_flags &= ~SSB_AUTOSIZE;
1351 (TCP_MAXWIN << tp->rcv_scale)) {
1352 so->so_rcv.ssb_flags &= ~SSB_AUTOSIZE;
1355 m_adj(m, drop_hdrlen); /* delayed header drop */
1356 ssb_appendstream(&so->so_rcv, m);
1360 * This code is responsible for most of the ACKs
1361 * the TCP stack sends back after receiving a data
1362 * packet. Note that the DELAY_ACK check fails if
1363 * the delack timer is already running, which results
1364 * in an ack being sent every other packet (which is
1367 * We then further aggregate acks by not actually
1368 * sending one until the protocol thread has completed
1369 * processing the current backlog of packets. This
1370 * does not delay the ack any further, but allows us
1371 * to take advantage of the packet aggregation that
1372 * high speed NICs do (usually blocks of 8-10 packets)
1373 * to send a single ack rather then four or five acks,
1374 * greatly reducing the ack rate, the return channel
1375 * bandwidth, and the protocol overhead on both ends.
1377 * Since this also has the effect of slowing down
1378 * the exponential slow-start ramp-up, systems with
1379 * very large bandwidth-delay products might want
1380 * to turn the feature off.
1382 if (DELAY_ACK(tp)) {
1383 tcp_callout_reset(tp, tp->tt_delack,
1384 tcp_delacktime, tcp_timer_delack);
1385 } else if (tcp_aggregate_acks) {
1386 tp->t_flags |= TF_ACKNOW;
1387 if (!(tp->t_flags & TF_ONOUTPUTQ)) {
1388 tp->t_flags |= TF_ONOUTPUTQ;
1389 tp->tt_cpu = mycpu->gd_cpuid;
1391 &tcpcbackq[tp->tt_cpu],
1395 tp->t_flags |= TF_ACKNOW;
1403 * Calculate amount of space in receive window,
1404 * and then do TCP input processing.
1405 * Receive window is amount of space in rcv queue,
1406 * but not less than advertised window.
1408 recvwin = ssb_space(&so->so_rcv);
1411 tp->rcv_wnd = imax(recvwin, (int)(tp->rcv_adv - tp->rcv_nxt));
1413 /* Reset receive buffer auto scaling when not in bulk receive mode. */
1417 switch (tp->t_state) {
1419 * If the state is SYN_RECEIVED:
1420 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1422 case TCPS_SYN_RECEIVED:
1423 if ((thflags & TH_ACK) &&
1424 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1425 SEQ_GT(th->th_ack, tp->snd_max))) {
1426 rstreason = BANDLIM_RST_OPENPORT;
1432 * If the state is SYN_SENT:
1433 * if seg contains an ACK, but not for our SYN, drop the input.
1434 * if seg contains a RST, then drop the connection.
1435 * if seg does not contain SYN, then drop it.
1436 * Otherwise this is an acceptable SYN segment
1437 * initialize tp->rcv_nxt and tp->irs
1438 * if seg contains ack then advance tp->snd_una
1439 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1440 * arrange for segment to be acked (eventually)
1441 * continue processing rest of data/controls, beginning with URG
1444 if ((thflags & TH_ACK) &&
1445 (SEQ_LEQ(th->th_ack, tp->iss) ||
1446 SEQ_GT(th->th_ack, tp->snd_max))) {
1447 rstreason = BANDLIM_UNLIMITED;
1450 if (thflags & TH_RST) {
1451 if (thflags & TH_ACK)
1452 tp = tcp_drop(tp, ECONNREFUSED);
1455 if (!(thflags & TH_SYN))
1457 tp->snd_wnd = th->th_win; /* initial send window */
1459 tp->irs = th->th_seq;
1461 if (thflags & TH_ACK) {
1462 /* Our SYN was acked. */
1463 tcpstat.tcps_connects++;
1465 /* Do window scaling on this connection? */
1466 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
1467 (TF_RCVD_SCALE | TF_REQ_SCALE)) {
1468 tp->snd_scale = tp->requested_s_scale;
1469 tp->rcv_scale = tp->request_r_scale;
1471 tp->rcv_adv += tp->rcv_wnd;
1472 tp->snd_una++; /* SYN is acked */
1473 tcp_callout_stop(tp, tp->tt_rexmt);
1475 * If there's data, delay ACK; if there's also a FIN
1476 * ACKNOW will be turned on later.
1478 if (DELAY_ACK(tp) && tlen != 0) {
1479 tcp_callout_reset(tp, tp->tt_delack,
1480 tcp_delacktime, tcp_timer_delack);
1482 tp->t_flags |= TF_ACKNOW;
1485 * Received <SYN,ACK> in SYN_SENT[*] state.
1487 * SYN_SENT --> ESTABLISHED
1488 * SYN_SENT* --> FIN_WAIT_1
1490 tp->t_starttime = ticks;
1491 if (tp->t_flags & TF_NEEDFIN) {
1492 tp->t_state = TCPS_FIN_WAIT_1;
1493 tp->t_flags &= ~TF_NEEDFIN;
1496 tp->t_state = TCPS_ESTABLISHED;
1497 tcp_callout_reset(tp, tp->tt_keep, tcp_keepidle,
1502 * Received initial SYN in SYN-SENT[*] state =>
1503 * simultaneous open.
1504 * Do 3-way handshake:
1505 * SYN-SENT -> SYN-RECEIVED
1506 * SYN-SENT* -> SYN-RECEIVED*
1508 tp->t_flags |= TF_ACKNOW;
1509 tcp_callout_stop(tp, tp->tt_rexmt);
1510 tp->t_state = TCPS_SYN_RECEIVED;
1515 * Advance th->th_seq to correspond to first data byte.
1516 * If data, trim to stay within window,
1517 * dropping FIN if necessary.
1520 if (tlen > tp->rcv_wnd) {
1521 todrop = tlen - tp->rcv_wnd;
1525 tcpstat.tcps_rcvpackafterwin++;
1526 tcpstat.tcps_rcvbyteafterwin += todrop;
1528 tp->snd_wl1 = th->th_seq - 1;
1529 tp->rcv_up = th->th_seq;
1531 * Client side of transaction: already sent SYN and data.
1532 * If the remote host used T/TCP to validate the SYN,
1533 * our data will be ACK'd; if so, enter normal data segment
1534 * processing in the middle of step 5, ack processing.
1535 * Otherwise, goto step 6.
1537 if (thflags & TH_ACK)
1543 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
1544 * do normal processing (we no longer bother with T/TCP).
1548 case TCPS_TIME_WAIT:
1549 break; /* continue normal processing */
1553 * States other than LISTEN or SYN_SENT.
1554 * First check the RST flag and sequence number since reset segments
1555 * are exempt from the timestamp and connection count tests. This
1556 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
1557 * below which allowed reset segments in half the sequence space
1558 * to fall though and be processed (which gives forged reset
1559 * segments with a random sequence number a 50 percent chance of
1560 * killing a connection).
1561 * Then check timestamp, if present.
1562 * Then check the connection count, if present.
1563 * Then check that at least some bytes of segment are within
1564 * receive window. If segment begins before rcv_nxt,
1565 * drop leading data (and SYN); if nothing left, just ack.
1568 * If the RST bit is set, check the sequence number to see
1569 * if this is a valid reset segment.
1571 * In all states except SYN-SENT, all reset (RST) segments
1572 * are validated by checking their SEQ-fields. A reset is
1573 * valid if its sequence number is in the window.
1574 * Note: this does not take into account delayed ACKs, so
1575 * we should test against last_ack_sent instead of rcv_nxt.
1576 * The sequence number in the reset segment is normally an
1577 * echo of our outgoing acknowledgement numbers, but some hosts
1578 * send a reset with the sequence number at the rightmost edge
1579 * of our receive window, and we have to handle this case.
1580 * If we have multiple segments in flight, the intial reset
1581 * segment sequence numbers will be to the left of last_ack_sent,
1582 * but they will eventually catch up.
1583 * In any case, it never made sense to trim reset segments to
1584 * fit the receive window since RFC 1122 says:
1585 * 4.2.2.12 RST Segment: RFC-793 Section 3.4
1587 * A TCP SHOULD allow a received RST segment to include data.
1590 * It has been suggested that a RST segment could contain
1591 * ASCII text that encoded and explained the cause of the
1592 * RST. No standard has yet been established for such
1595 * If the reset segment passes the sequence number test examine
1597 * SYN_RECEIVED STATE:
1598 * If passive open, return to LISTEN state.
1599 * If active open, inform user that connection was refused.
1600 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
1601 * Inform user that connection was reset, and close tcb.
1602 * CLOSING, LAST_ACK STATES:
1605 * Drop the segment - see Stevens, vol. 2, p. 964 and
1608 if (thflags & TH_RST) {
1609 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
1610 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
1611 switch (tp->t_state) {
1613 case TCPS_SYN_RECEIVED:
1614 so->so_error = ECONNREFUSED;
1617 case TCPS_ESTABLISHED:
1618 case TCPS_FIN_WAIT_1:
1619 case TCPS_FIN_WAIT_2:
1620 case TCPS_CLOSE_WAIT:
1621 so->so_error = ECONNRESET;
1623 tp->t_state = TCPS_CLOSED;
1624 tcpstat.tcps_drops++;
1633 case TCPS_TIME_WAIT:
1641 * RFC 1323 PAWS: If we have a timestamp reply on this segment
1642 * and it's less than ts_recent, drop it.
1644 if ((to.to_flags & TOF_TS) && tp->ts_recent != 0 &&
1645 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
1647 /* Check to see if ts_recent is over 24 days old. */
1648 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1650 * Invalidate ts_recent. If this segment updates
1651 * ts_recent, the age will be reset later and ts_recent
1652 * will get a valid value. If it does not, setting
1653 * ts_recent to zero will at least satisfy the
1654 * requirement that zero be placed in the timestamp
1655 * echo reply when ts_recent isn't valid. The
1656 * age isn't reset until we get a valid ts_recent
1657 * because we don't want out-of-order segments to be
1658 * dropped when ts_recent is old.
1662 tcpstat.tcps_rcvduppack++;
1663 tcpstat.tcps_rcvdupbyte += tlen;
1664 tcpstat.tcps_pawsdrop++;
1672 * In the SYN-RECEIVED state, validate that the packet belongs to
1673 * this connection before trimming the data to fit the receive
1674 * window. Check the sequence number versus IRS since we know
1675 * the sequence numbers haven't wrapped. This is a partial fix
1676 * for the "LAND" DoS attack.
1678 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
1679 rstreason = BANDLIM_RST_OPENPORT;
1683 todrop = tp->rcv_nxt - th->th_seq;
1685 if (TCP_DO_SACK(tp)) {
1686 /* Report duplicate segment at head of packet. */
1687 tp->reportblk.rblk_start = th->th_seq;
1688 tp->reportblk.rblk_end = th->th_seq + tlen;
1689 if (thflags & TH_FIN)
1690 ++tp->reportblk.rblk_end;
1691 if (SEQ_GT(tp->reportblk.rblk_end, tp->rcv_nxt))
1692 tp->reportblk.rblk_end = tp->rcv_nxt;
1693 tp->t_flags |= (TF_DUPSEG | TF_SACKLEFT | TF_ACKNOW);
1695 if (thflags & TH_SYN) {
1705 * Following if statement from Stevens, vol. 2, p. 960.
1707 if (todrop > tlen ||
1708 (todrop == tlen && !(thflags & TH_FIN))) {
1710 * Any valid FIN must be to the left of the window.
1711 * At this point the FIN must be a duplicate or out
1712 * of sequence; drop it.
1717 * Send an ACK to resynchronize and drop any data.
1718 * But keep on processing for RST or ACK.
1720 tp->t_flags |= TF_ACKNOW;
1722 tcpstat.tcps_rcvduppack++;
1723 tcpstat.tcps_rcvdupbyte += todrop;
1725 tcpstat.tcps_rcvpartduppack++;
1726 tcpstat.tcps_rcvpartdupbyte += todrop;
1728 drop_hdrlen += todrop; /* drop from the top afterwards */
1729 th->th_seq += todrop;
1731 if (th->th_urp > todrop)
1732 th->th_urp -= todrop;
1740 * If new data are received on a connection after the
1741 * user processes are gone, then RST the other end.
1743 if ((so->so_state & SS_NOFDREF) &&
1744 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1746 tcpstat.tcps_rcvafterclose++;
1747 rstreason = BANDLIM_UNLIMITED;
1752 * If segment ends after window, drop trailing data
1753 * (and PUSH and FIN); if nothing left, just ACK.
1755 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
1757 tcpstat.tcps_rcvpackafterwin++;
1758 if (todrop >= tlen) {
1759 tcpstat.tcps_rcvbyteafterwin += tlen;
1761 * If a new connection request is received
1762 * while in TIME_WAIT, drop the old connection
1763 * and start over if the sequence numbers
1764 * are above the previous ones.
1766 if (thflags & TH_SYN &&
1767 tp->t_state == TCPS_TIME_WAIT &&
1768 SEQ_GT(th->th_seq, tp->rcv_nxt)) {
1773 * If window is closed can only take segments at
1774 * window edge, and have to drop data and PUSH from
1775 * incoming segments. Continue processing, but
1776 * remember to ack. Otherwise, drop segment
1779 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1780 tp->t_flags |= TF_ACKNOW;
1781 tcpstat.tcps_rcvwinprobe++;
1785 tcpstat.tcps_rcvbyteafterwin += todrop;
1788 thflags &= ~(TH_PUSH | TH_FIN);
1792 * If last ACK falls within this segment's sequence numbers,
1793 * record its timestamp.
1795 * 1) That the test incorporates suggestions from the latest
1796 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1797 * 2) That updating only on newer timestamps interferes with
1798 * our earlier PAWS tests, so this check should be solely
1799 * predicated on the sequence space of this segment.
1800 * 3) That we modify the segment boundary check to be
1801 * Last.ACK.Sent <= SEG.SEQ + SEG.LEN
1802 * instead of RFC1323's
1803 * Last.ACK.Sent < SEG.SEQ + SEG.LEN,
1804 * This modified check allows us to overcome RFC1323's
1805 * limitations as described in Stevens TCP/IP Illustrated
1806 * Vol. 2 p.869. In such cases, we can still calculate the
1807 * RTT correctly when RCV.NXT == Last.ACK.Sent.
1809 if ((to.to_flags & TOF_TS) && SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
1810 SEQ_LEQ(tp->last_ack_sent, (th->th_seq + tlen
1811 + ((thflags & TH_SYN) != 0)
1812 + ((thflags & TH_FIN) != 0)))) {
1813 tp->ts_recent_age = ticks;
1814 tp->ts_recent = to.to_tsval;
1818 * If a SYN is in the window, then this is an
1819 * error and we send an RST and drop the connection.
1821 if (thflags & TH_SYN) {
1822 tp = tcp_drop(tp, ECONNRESET);
1823 rstreason = BANDLIM_UNLIMITED;
1828 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
1829 * flag is on (half-synchronized state), then queue data for
1830 * later processing; else drop segment and return.
1832 if (!(thflags & TH_ACK)) {
1833 if (tp->t_state == TCPS_SYN_RECEIVED ||
1834 (tp->t_flags & TF_NEEDSYN))
1843 switch (tp->t_state) {
1845 * In SYN_RECEIVED state, the ACK acknowledges our SYN, so enter
1846 * ESTABLISHED state and continue processing.
1847 * The ACK was checked above.
1849 case TCPS_SYN_RECEIVED:
1851 tcpstat.tcps_connects++;
1853 /* Do window scaling? */
1854 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
1855 (TF_RCVD_SCALE | TF_REQ_SCALE)) {
1856 tp->snd_scale = tp->requested_s_scale;
1857 tp->rcv_scale = tp->request_r_scale;
1861 * SYN-RECEIVED -> ESTABLISHED
1862 * SYN-RECEIVED* -> FIN-WAIT-1
1864 tp->t_starttime = ticks;
1865 if (tp->t_flags & TF_NEEDFIN) {
1866 tp->t_state = TCPS_FIN_WAIT_1;
1867 tp->t_flags &= ~TF_NEEDFIN;
1869 tp->t_state = TCPS_ESTABLISHED;
1870 tcp_callout_reset(tp, tp->tt_keep, tcp_keepidle,
1874 * If segment contains data or ACK, will call tcp_reass()
1875 * later; if not, do so now to pass queued data to user.
1877 if (tlen == 0 && !(thflags & TH_FIN))
1878 tcp_reass(tp, NULL, NULL, NULL);
1882 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1883 * ACKs. If the ack is in the range
1884 * tp->snd_una < th->th_ack <= tp->snd_max
1885 * then advance tp->snd_una to th->th_ack and drop
1886 * data from the retransmission queue. If this ACK reflects
1887 * more up to date window information we update our window information.
1889 case TCPS_ESTABLISHED:
1890 case TCPS_FIN_WAIT_1:
1891 case TCPS_FIN_WAIT_2:
1892 case TCPS_CLOSE_WAIT:
1895 case TCPS_TIME_WAIT:
1897 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
1898 if (TCP_DO_SACK(tp))
1899 tcp_sack_update_scoreboard(tp, &to);
1900 if (tlen != 0 || tiwin != tp->snd_wnd) {
1904 tcpstat.tcps_rcvdupack++;
1905 if (!tcp_callout_active(tp, tp->tt_rexmt) ||
1906 th->th_ack != tp->snd_una) {
1911 * We have outstanding data (other than
1912 * a window probe), this is a completely
1913 * duplicate ack (ie, window info didn't
1914 * change), the ack is the biggest we've
1915 * seen and we've seen exactly our rexmt
1916 * threshhold of them, so assume a packet
1917 * has been dropped and retransmit it.
1918 * Kludge snd_nxt & the congestion
1919 * window so we send only this one
1922 if (IN_FASTRECOVERY(tp)) {
1923 if (TCP_DO_SACK(tp)) {
1924 /* No artifical cwnd inflation. */
1925 tcp_sack_rexmt(tp, th);
1928 * Dup acks mean that packets
1929 * have left the network
1930 * (they're now cached at the
1931 * receiver) so bump cwnd by
1932 * the amount in the receiver
1933 * to keep a constant cwnd
1934 * packets in the network.
1936 tp->snd_cwnd += tp->t_maxseg;
1939 } else if (SEQ_LT(th->th_ack, tp->snd_recover)) {
1942 } else if (++tp->t_dupacks == tcprexmtthresh) {
1943 tcp_seq old_snd_nxt;
1947 if (tcp_do_eifel_detect &&
1948 (tp->t_flags & TF_RCVD_TSTMP)) {
1949 tcp_save_congestion_state(tp);
1950 tp->t_flags |= TF_FASTREXMT;
1953 * We know we're losing at the current
1954 * window size, so do congestion avoidance:
1955 * set ssthresh to half the current window
1956 * and pull our congestion window back to the
1959 win = min(tp->snd_wnd, tp->snd_cwnd) / 2 /
1963 tp->snd_ssthresh = win * tp->t_maxseg;
1964 ENTER_FASTRECOVERY(tp);
1965 tp->snd_recover = tp->snd_max;
1966 tcp_callout_stop(tp, tp->tt_rexmt);
1968 old_snd_nxt = tp->snd_nxt;
1969 tp->snd_nxt = th->th_ack;
1970 tp->snd_cwnd = tp->t_maxseg;
1972 ++tcpstat.tcps_sndfastrexmit;
1973 tp->snd_cwnd = tp->snd_ssthresh;
1974 tp->rexmt_high = tp->snd_nxt;
1975 if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
1976 tp->snd_nxt = old_snd_nxt;
1977 KASSERT(tp->snd_limited <= 2,
1978 ("tp->snd_limited too big"));
1979 if (TCP_DO_SACK(tp))
1980 tcp_sack_rexmt(tp, th);
1982 tp->snd_cwnd += tp->t_maxseg *
1983 (tp->t_dupacks - tp->snd_limited);
1984 } else if (tcp_do_limitedtransmit) {
1985 u_long oldcwnd = tp->snd_cwnd;
1986 tcp_seq oldsndmax = tp->snd_max;
1987 tcp_seq oldsndnxt = tp->snd_nxt;
1988 /* outstanding data */
1989 uint32_t ownd = tp->snd_max - tp->snd_una;
1992 #define iceildiv(n, d) (((n)+(d)-1) / (d))
1994 KASSERT(tp->t_dupacks == 1 ||
1996 ("dupacks not 1 or 2"));
1997 if (tp->t_dupacks == 1)
1998 tp->snd_limited = 0;
1999 tp->snd_nxt = tp->snd_max;
2000 tp->snd_cwnd = ownd +
2001 (tp->t_dupacks - tp->snd_limited) *
2006 * Other acks may have been processed,
2007 * snd_nxt cannot be reset to a value less
2010 if (SEQ_LT(oldsndnxt, oldsndmax)) {
2011 if (SEQ_GT(oldsndnxt, tp->snd_una))
2012 tp->snd_nxt = oldsndnxt;
2014 tp->snd_nxt = tp->snd_una;
2016 tp->snd_cwnd = oldcwnd;
2017 sent = tp->snd_max - oldsndmax;
2018 if (sent > tp->t_maxseg) {
2019 KASSERT((tp->t_dupacks == 2 &&
2020 tp->snd_limited == 0) ||
2021 (sent == tp->t_maxseg + 1 &&
2022 tp->t_flags & TF_SENTFIN),
2024 KASSERT(sent <= tp->t_maxseg * 2,
2025 ("sent too many segments"));
2026 tp->snd_limited = 2;
2027 tcpstat.tcps_sndlimited += 2;
2028 } else if (sent > 0) {
2030 ++tcpstat.tcps_sndlimited;
2031 } else if (tcp_do_early_retransmit &&
2032 (tcp_do_eifel_detect &&
2033 (tp->t_flags & TF_RCVD_TSTMP)) &&
2034 ownd < 4 * tp->t_maxseg &&
2035 tp->t_dupacks + 1 >=
2036 iceildiv(ownd, tp->t_maxseg) &&
2037 (!TCP_DO_SACK(tp) ||
2038 ownd <= tp->t_maxseg ||
2039 tcp_sack_has_sacked(&tp->scb,
2040 ownd - tp->t_maxseg))) {
2041 ++tcpstat.tcps_sndearlyrexmit;
2042 tp->t_flags |= TF_EARLYREXMT;
2043 goto fastretransmit;
2049 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una"));
2051 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2053 * Detected optimistic ACK attack.
2054 * Force slow-start to de-synchronize attack.
2056 tp->snd_cwnd = tp->t_maxseg;
2059 tcpstat.tcps_rcvacktoomuch++;
2063 * If we reach this point, ACK is not a duplicate,
2064 * i.e., it ACKs something we sent.
2066 if (tp->t_flags & TF_NEEDSYN) {
2068 * T/TCP: Connection was half-synchronized, and our
2069 * SYN has been ACK'd (so connection is now fully
2070 * synchronized). Go to non-starred state,
2071 * increment snd_una for ACK of SYN, and check if
2072 * we can do window scaling.
2074 tp->t_flags &= ~TF_NEEDSYN;
2076 /* Do window scaling? */
2077 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
2078 (TF_RCVD_SCALE | TF_REQ_SCALE)) {
2079 tp->snd_scale = tp->requested_s_scale;
2080 tp->rcv_scale = tp->request_r_scale;
2085 acked = th->th_ack - tp->snd_una;
2086 tcpstat.tcps_rcvackpack++;
2087 tcpstat.tcps_rcvackbyte += acked;
2089 if (tcp_do_eifel_detect && acked > 0 &&
2090 (to.to_flags & TOF_TS) && (to.to_tsecr != 0) &&
2091 (tp->t_flags & TF_FIRSTACCACK)) {
2092 /* Eifel detection applicable. */
2093 if (to.to_tsecr < tp->t_rexmtTS) {
2094 ++tcpstat.tcps_eifeldetected;
2095 tcp_revert_congestion_state(tp);
2096 if (tp->t_rxtshift == 1 &&
2097 ticks >= tp->t_badrxtwin)
2098 ++tcpstat.tcps_rttcantdetect;
2100 } else if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) {
2102 * If we just performed our first retransmit,
2103 * and the ACK arrives within our recovery window,
2104 * then it was a mistake to do the retransmit
2105 * in the first place. Recover our original cwnd
2106 * and ssthresh, and proceed to transmit where we
2109 tcp_revert_congestion_state(tp);
2110 ++tcpstat.tcps_rttdetected;
2114 * If we have a timestamp reply, update smoothed
2115 * round trip time. If no timestamp is present but
2116 * transmit timer is running and timed sequence
2117 * number was acked, update smoothed round trip time.
2118 * Since we now have an rtt measurement, cancel the
2119 * timer backoff (cf., Phil Karn's retransmit alg.).
2120 * Recompute the initial retransmit timer.
2122 * Some machines (certain windows boxes) send broken
2123 * timestamp replies during the SYN+ACK phase, ignore
2126 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0))
2127 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1);
2128 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq))
2129 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2130 tcp_xmit_bandwidth_limit(tp, th->th_ack);
2133 * If no data (only SYN) was ACK'd,
2134 * skip rest of ACK processing.
2139 /* Stop looking for an acceptable ACK since one was received. */
2140 tp->t_flags &= ~(TF_FIRSTACCACK | TF_FASTREXMT | TF_EARLYREXMT);
2142 if (acked > so->so_snd.ssb_cc) {
2143 tp->snd_wnd -= so->so_snd.ssb_cc;
2144 sbdrop(&so->so_snd.sb, (int)so->so_snd.ssb_cc);
2145 ourfinisacked = TRUE;
2147 sbdrop(&so->so_snd.sb, acked);
2148 tp->snd_wnd -= acked;
2149 ourfinisacked = FALSE;
2154 * Update window information.
2155 * Don't look at window if no ACK:
2156 * TAC's send garbage on first SYN.
2158 if (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2159 (tp->snd_wl1 == th->th_seq &&
2160 (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2161 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)))) {
2162 /* keep track of pure window updates */
2163 if (tlen == 0 && tp->snd_wl2 == th->th_ack &&
2164 tiwin > tp->snd_wnd)
2165 tcpstat.tcps_rcvwinupd++;
2166 tp->snd_wnd = tiwin;
2167 tp->snd_wl1 = th->th_seq;
2168 tp->snd_wl2 = th->th_ack;
2169 if (tp->snd_wnd > tp->max_sndwnd)
2170 tp->max_sndwnd = tp->snd_wnd;
2174 tp->snd_una = th->th_ack;
2175 if (TCP_DO_SACK(tp))
2176 tcp_sack_update_scoreboard(tp, &to);
2177 if (IN_FASTRECOVERY(tp)) {
2178 if (SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2179 EXIT_FASTRECOVERY(tp);
2182 * If the congestion window was inflated
2183 * to account for the other side's
2184 * cached packets, retract it.
2186 if (!TCP_DO_SACK(tp))
2187 tp->snd_cwnd = tp->snd_ssthresh;
2190 * Window inflation should have left us
2191 * with approximately snd_ssthresh outstanding
2192 * data. But, in case we would be inclined
2193 * to send a burst, better do it using
2196 if (SEQ_GT(th->th_ack + tp->snd_cwnd,
2197 tp->snd_max + 2 * tp->t_maxseg))
2199 (tp->snd_max - tp->snd_una) +
2204 if (TCP_DO_SACK(tp)) {
2205 tp->snd_max_rexmt = tp->snd_max;
2206 tcp_sack_rexmt(tp, th);
2208 tcp_newreno_partial_ack(tp, th, acked);
2214 * Open the congestion window. When in slow-start,
2215 * open exponentially: maxseg per packet. Otherwise,
2216 * open linearly: maxseg per window.
2218 if (tp->snd_cwnd <= tp->snd_ssthresh) {
2220 (SEQ_LT(tp->snd_nxt, tp->snd_max) ?
2221 tp->t_maxseg : 2 * tp->t_maxseg);
2224 tp->snd_cwnd += tcp_do_abc ?
2225 min(acked, abc_sslimit) : tp->t_maxseg;
2227 /* linear increase */
2228 tp->snd_wacked += tcp_do_abc ? acked :
2230 if (tp->snd_wacked >= tp->snd_cwnd) {
2231 tp->snd_wacked -= tp->snd_cwnd;
2232 tp->snd_cwnd += tp->t_maxseg;
2235 tp->snd_cwnd = min(tp->snd_cwnd,
2236 TCP_MAXWIN << tp->snd_scale);
2237 tp->snd_recover = th->th_ack - 1;
2239 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2240 tp->snd_nxt = tp->snd_una;
2243 * If all outstanding data is acked, stop retransmit
2244 * timer and remember to restart (more output or persist).
2245 * If there is more data to be acked, restart retransmit
2246 * timer, using current (possibly backed-off) value.
2248 if (th->th_ack == tp->snd_max) {
2249 tcp_callout_stop(tp, tp->tt_rexmt);
2251 } else if (!tcp_callout_active(tp, tp->tt_persist)) {
2252 tcp_callout_reset(tp, tp->tt_rexmt, tp->t_rxtcur,
2256 switch (tp->t_state) {
2258 * In FIN_WAIT_1 STATE in addition to the processing
2259 * for the ESTABLISHED state if our FIN is now acknowledged
2260 * then enter FIN_WAIT_2.
2262 case TCPS_FIN_WAIT_1:
2263 if (ourfinisacked) {
2265 * If we can't receive any more
2266 * data, then closing user can proceed.
2267 * Starting the timer is contrary to the
2268 * specification, but if we don't get a FIN
2269 * we'll hang forever.
2271 if (so->so_state & SS_CANTRCVMORE) {
2272 soisdisconnected(so);
2273 tcp_callout_reset(tp, tp->tt_2msl,
2274 tcp_maxidle, tcp_timer_2msl);
2276 tp->t_state = TCPS_FIN_WAIT_2;
2281 * In CLOSING STATE in addition to the processing for
2282 * the ESTABLISHED state if the ACK acknowledges our FIN
2283 * then enter the TIME-WAIT state, otherwise ignore
2287 if (ourfinisacked) {
2288 tp->t_state = TCPS_TIME_WAIT;
2289 tcp_canceltimers(tp);
2290 tcp_callout_reset(tp, tp->tt_2msl,
2291 2 * tcp_msl, tcp_timer_2msl);
2292 soisdisconnected(so);
2297 * In LAST_ACK, we may still be waiting for data to drain
2298 * and/or to be acked, as well as for the ack of our FIN.
2299 * If our FIN is now acknowledged, delete the TCB,
2300 * enter the closed state and return.
2303 if (ourfinisacked) {
2310 * In TIME_WAIT state the only thing that should arrive
2311 * is a retransmission of the remote FIN. Acknowledge
2312 * it and restart the finack timer.
2314 case TCPS_TIME_WAIT:
2315 tcp_callout_reset(tp, tp->tt_2msl, 2 * tcp_msl,
2323 * Update window information.
2324 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2326 if ((thflags & TH_ACK) &&
2327 acceptable_window_update(tp, th, tiwin)) {
2328 /* keep track of pure window updates */
2329 if (tlen == 0 && tp->snd_wl2 == th->th_ack &&
2330 tiwin > tp->snd_wnd)
2331 tcpstat.tcps_rcvwinupd++;
2332 tp->snd_wnd = tiwin;
2333 tp->snd_wl1 = th->th_seq;
2334 tp->snd_wl2 = th->th_ack;
2335 if (tp->snd_wnd > tp->max_sndwnd)
2336 tp->max_sndwnd = tp->snd_wnd;
2341 * Process segments with URG.
2343 if ((thflags & TH_URG) && th->th_urp &&
2344 !TCPS_HAVERCVDFIN(tp->t_state)) {
2346 * This is a kludge, but if we receive and accept
2347 * random urgent pointers, we'll crash in
2348 * soreceive. It's hard to imagine someone
2349 * actually wanting to send this much urgent data.
2351 if (th->th_urp + so->so_rcv.ssb_cc > sb_max) {
2352 th->th_urp = 0; /* XXX */
2353 thflags &= ~TH_URG; /* XXX */
2354 goto dodata; /* XXX */
2357 * If this segment advances the known urgent pointer,
2358 * then mark the data stream. This should not happen
2359 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2360 * a FIN has been received from the remote side.
2361 * In these states we ignore the URG.
2363 * According to RFC961 (Assigned Protocols),
2364 * the urgent pointer points to the last octet
2365 * of urgent data. We continue, however,
2366 * to consider it to indicate the first octet
2367 * of data past the urgent section as the original
2368 * spec states (in one of two places).
2370 if (SEQ_GT(th->th_seq + th->th_urp, tp->rcv_up)) {
2371 tp->rcv_up = th->th_seq + th->th_urp;
2372 so->so_oobmark = so->so_rcv.ssb_cc +
2373 (tp->rcv_up - tp->rcv_nxt) - 1;
2374 if (so->so_oobmark == 0)
2375 so->so_state |= SS_RCVATMARK;
2377 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2380 * Remove out of band data so doesn't get presented to user.
2381 * This can happen independent of advancing the URG pointer,
2382 * but if two URG's are pending at once, some out-of-band
2383 * data may creep in... ick.
2385 if (th->th_urp <= (u_long)tlen &&
2386 !(so->so_options & SO_OOBINLINE)) {
2387 /* hdr drop is delayed */
2388 tcp_pulloutofband(so, th, m, drop_hdrlen);
2392 * If no out of band data is expected,
2393 * pull receive urgent pointer along
2394 * with the receive window.
2396 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2397 tp->rcv_up = tp->rcv_nxt;
2402 * Process the segment text, merging it into the TCP sequencing queue,
2403 * and arranging for acknowledgment of receipt if necessary.
2404 * This process logically involves adjusting tp->rcv_wnd as data
2405 * is presented to the user (this happens in tcp_usrreq.c,
2406 * case PRU_RCVD). If a FIN has already been received on this
2407 * connection then we just ignore the text.
2409 if ((tlen || (thflags & TH_FIN)) && !TCPS_HAVERCVDFIN(tp->t_state)) {
2410 m_adj(m, drop_hdrlen); /* delayed header drop */
2412 * Insert segment which includes th into TCP reassembly queue
2413 * with control block tp. Set thflags to whether reassembly now
2414 * includes a segment with FIN. This handles the common case
2415 * inline (segment is the next to be received on an established
2416 * connection, and the queue is empty), avoiding linkage into
2417 * and removal from the queue and repetition of various
2419 * Set DELACK for segments received in order, but ack
2420 * immediately when segments are out of order (so
2421 * fast retransmit can work).
2423 if (th->th_seq == tp->rcv_nxt &&
2424 LIST_EMPTY(&tp->t_segq) &&
2425 TCPS_HAVEESTABLISHED(tp->t_state)) {
2426 if (DELAY_ACK(tp)) {
2427 tcp_callout_reset(tp, tp->tt_delack,
2428 tcp_delacktime, tcp_timer_delack);
2430 tp->t_flags |= TF_ACKNOW;
2432 tp->rcv_nxt += tlen;
2433 thflags = th->th_flags & TH_FIN;
2434 tcpstat.tcps_rcvpack++;
2435 tcpstat.tcps_rcvbyte += tlen;
2437 if (so->so_state & SS_CANTRCVMORE)
2440 ssb_appendstream(&so->so_rcv, m);
2443 if (!(tp->t_flags & TF_DUPSEG)) {
2444 /* Initialize SACK report block. */
2445 tp->reportblk.rblk_start = th->th_seq;
2446 tp->reportblk.rblk_end = th->th_seq + tlen +
2447 ((thflags & TH_FIN) != 0);
2449 thflags = tcp_reass(tp, th, &tlen, m);
2450 tp->t_flags |= TF_ACKNOW;
2454 * Note the amount of data that peer has sent into
2455 * our window, in order to estimate the sender's
2458 len = so->so_rcv.ssb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2465 * If FIN is received ACK the FIN and let the user know
2466 * that the connection is closing.
2468 if (thflags & TH_FIN) {
2469 if (!TCPS_HAVERCVDFIN(tp->t_state)) {
2472 * If connection is half-synchronized
2473 * (ie NEEDSYN flag on) then delay ACK,
2474 * so it may be piggybacked when SYN is sent.
2475 * Otherwise, since we received a FIN then no
2476 * more input can be expected, send ACK now.
2478 if (DELAY_ACK(tp) && (tp->t_flags & TF_NEEDSYN)) {
2479 tcp_callout_reset(tp, tp->tt_delack,
2480 tcp_delacktime, tcp_timer_delack);
2482 tp->t_flags |= TF_ACKNOW;
2487 switch (tp->t_state) {
2489 * In SYN_RECEIVED and ESTABLISHED STATES
2490 * enter the CLOSE_WAIT state.
2492 case TCPS_SYN_RECEIVED:
2493 tp->t_starttime = ticks;
2495 case TCPS_ESTABLISHED:
2496 tp->t_state = TCPS_CLOSE_WAIT;
2500 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2501 * enter the CLOSING state.
2503 case TCPS_FIN_WAIT_1:
2504 tp->t_state = TCPS_CLOSING;
2508 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2509 * starting the time-wait timer, turning off the other
2512 case TCPS_FIN_WAIT_2:
2513 tp->t_state = TCPS_TIME_WAIT;
2514 tcp_canceltimers(tp);
2515 tcp_callout_reset(tp, tp->tt_2msl, 2 * tcp_msl,
2517 soisdisconnected(so);
2521 * In TIME_WAIT state restart the 2 MSL time_wait timer.
2523 case TCPS_TIME_WAIT:
2524 tcp_callout_reset(tp, tp->tt_2msl, 2 * tcp_msl,
2531 if (so->so_options & SO_DEBUG)
2532 tcp_trace(TA_INPUT, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2536 * Return any desired output.
2538 if (needoutput || (tp->t_flags & TF_ACKNOW))
2544 * Generate an ACK dropping incoming segment if it occupies
2545 * sequence space, where the ACK reflects our state.
2547 * We can now skip the test for the RST flag since all
2548 * paths to this code happen after packets containing
2549 * RST have been dropped.
2551 * In the SYN-RECEIVED state, don't send an ACK unless the
2552 * segment we received passes the SYN-RECEIVED ACK test.
2553 * If it fails send a RST. This breaks the loop in the
2554 * "LAND" DoS attack, and also prevents an ACK storm
2555 * between two listening ports that have been sent forged
2556 * SYN segments, each with the source address of the other.
2558 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
2559 (SEQ_GT(tp->snd_una, th->th_ack) ||
2560 SEQ_GT(th->th_ack, tp->snd_max)) ) {
2561 rstreason = BANDLIM_RST_OPENPORT;
2565 if (so->so_options & SO_DEBUG)
2566 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2569 tp->t_flags |= TF_ACKNOW;
2575 * Generate a RST, dropping incoming segment.
2576 * Make ACK acceptable to originator of segment.
2577 * Don't bother to respond if destination was broadcast/multicast.
2579 if ((thflags & TH_RST) || m->m_flags & (M_BCAST | M_MCAST))
2582 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
2583 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
2586 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
2587 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
2588 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
2589 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2592 /* IPv6 anycast check is done at tcp6_input() */
2595 * Perform bandwidth limiting.
2598 if (badport_bandlim(rstreason) < 0)
2603 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2604 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2606 if (thflags & TH_ACK)
2607 /* mtod() below is safe as long as hdr dropping is delayed */
2608 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack,
2611 if (thflags & TH_SYN)
2613 /* mtod() below is safe as long as hdr dropping is delayed */
2614 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq + tlen,
2615 (tcp_seq)0, TH_RST | TH_ACK);
2621 * Drop space held by incoming segment and return.
2624 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2625 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2632 * Parse TCP options and place in tcpopt.
2635 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, boolean_t is_syn)
2640 for (; cnt > 0; cnt -= optlen, cp += optlen) {
2642 if (opt == TCPOPT_EOL)
2644 if (opt == TCPOPT_NOP)
2650 if (optlen < 2 || optlen > cnt)
2655 if (optlen != TCPOLEN_MAXSEG)
2659 to->to_flags |= TOF_MSS;
2660 bcopy(cp + 2, &to->to_mss, sizeof to->to_mss);
2661 to->to_mss = ntohs(to->to_mss);
2664 if (optlen != TCPOLEN_WINDOW)
2668 to->to_flags |= TOF_SCALE;
2669 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
2671 case TCPOPT_TIMESTAMP:
2672 if (optlen != TCPOLEN_TIMESTAMP)
2674 to->to_flags |= TOF_TS;
2675 bcopy(cp + 2, &to->to_tsval, sizeof to->to_tsval);
2676 to->to_tsval = ntohl(to->to_tsval);
2677 bcopy(cp + 6, &to->to_tsecr, sizeof to->to_tsecr);
2678 to->to_tsecr = ntohl(to->to_tsecr);
2680 * If echoed timestamp is later than the current time,
2681 * fall back to non RFC1323 RTT calculation.
2683 if (to->to_tsecr != 0 && TSTMP_GT(to->to_tsecr, ticks))
2686 case TCPOPT_SACK_PERMITTED:
2687 if (optlen != TCPOLEN_SACK_PERMITTED)
2691 to->to_flags |= TOF_SACK_PERMITTED;
2694 if ((optlen - 2) & 0x07) /* not multiple of 8 */
2696 to->to_nsackblocks = (optlen - 2) / 8;
2697 to->to_sackblocks = (struct raw_sackblock *) (cp + 2);
2698 to->to_flags |= TOF_SACK;
2699 for (i = 0; i < to->to_nsackblocks; i++) {
2700 struct raw_sackblock *r = &to->to_sackblocks[i];
2702 r->rblk_start = ntohl(r->rblk_start);
2703 r->rblk_end = ntohl(r->rblk_end);
2706 #ifdef TCP_SIGNATURE
2708 * XXX In order to reply to a host which has set the
2709 * TCP_SIGNATURE option in its initial SYN, we have to
2710 * record the fact that the option was observed here
2711 * for the syncache code to perform the correct response.
2713 case TCPOPT_SIGNATURE:
2714 if (optlen != TCPOLEN_SIGNATURE)
2716 to->to_flags |= (TOF_SIGNATURE | TOF_SIGLEN);
2718 #endif /* TCP_SIGNATURE */
2726 * Pull out of band byte out of a segment so
2727 * it doesn't appear in the user's data queue.
2728 * It is still reflected in the segment length for
2729 * sequencing purposes.
2730 * "off" is the delayed to be dropped hdrlen.
2733 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, int off)
2735 int cnt = off + th->th_urp - 1;
2738 if (m->m_len > cnt) {
2739 char *cp = mtod(m, caddr_t) + cnt;
2740 struct tcpcb *tp = sototcpcb(so);
2743 tp->t_oobflags |= TCPOOB_HAVEDATA;
2744 bcopy(cp + 1, cp, m->m_len - cnt - 1);
2746 if (m->m_flags & M_PKTHDR)
2755 panic("tcp_pulloutofband");
2759 * Collect new round-trip time estimate
2760 * and update averages and current timeout.
2763 tcp_xmit_timer(struct tcpcb *tp, int rtt)
2767 tcpstat.tcps_rttupdated++;
2769 if (tp->t_srtt != 0) {
2771 * srtt is stored as fixed point with 5 bits after the
2772 * binary point (i.e., scaled by 8). The following magic
2773 * is equivalent to the smoothing algorithm in rfc793 with
2774 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2775 * point). Adjust rtt to origin 0.
2777 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
2778 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
2780 if ((tp->t_srtt += delta) <= 0)
2784 * We accumulate a smoothed rtt variance (actually, a
2785 * smoothed mean difference), then set the retransmit
2786 * timer to smoothed rtt + 4 times the smoothed variance.
2787 * rttvar is stored as fixed point with 4 bits after the
2788 * binary point (scaled by 16). The following is
2789 * equivalent to rfc793 smoothing with an alpha of .75
2790 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
2791 * rfc793's wired-in beta.
2795 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
2796 if ((tp->t_rttvar += delta) <= 0)
2798 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
2799 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2802 * No rtt measurement yet - use the unsmoothed rtt.
2803 * Set the variance to half the rtt (so our first
2804 * retransmit happens at 3*rtt).
2806 tp->t_srtt = rtt << TCP_RTT_SHIFT;
2807 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
2808 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2814 * the retransmit should happen at rtt + 4 * rttvar.
2815 * Because of the way we do the smoothing, srtt and rttvar
2816 * will each average +1/2 tick of bias. When we compute
2817 * the retransmit timer, we want 1/2 tick of rounding and
2818 * 1 extra tick because of +-1/2 tick uncertainty in the
2819 * firing of the timer. The bias will give us exactly the
2820 * 1.5 tick we need. But, because the bias is
2821 * statistical, we have to test that we don't drop below
2822 * the minimum feasible timer (which is 2 ticks).
2824 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
2825 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
2828 * We received an ack for a packet that wasn't retransmitted;
2829 * it is probably safe to discard any error indications we've
2830 * received recently. This isn't quite right, but close enough
2831 * for now (a route might have failed after we sent a segment,
2832 * and the return path might not be symmetrical).
2834 tp->t_softerror = 0;
2838 * Determine a reasonable value for maxseg size.
2839 * If the route is known, check route for mtu.
2840 * If none, use an mss that can be handled on the outgoing
2841 * interface without forcing IP to fragment; if bigger than
2842 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2843 * to utilize large mbufs. If no route is found, route has no mtu,
2844 * or the destination isn't local, use a default, hopefully conservative
2845 * size (usually 512 or the default IP max size, but no more than the mtu
2846 * of the interface), as we can't discover anything about intervening
2847 * gateways or networks. We also initialize the congestion/slow start
2848 * window to be a single segment if the destination isn't local.
2849 * While looking at the routing entry, we also initialize other path-dependent
2850 * parameters from pre-set or cached values in the routing entry.
2852 * Also take into account the space needed for options that we
2853 * send regularly. Make maxseg shorter by that amount to assure
2854 * that we can send maxseg amount of data even when the options
2855 * are present. Store the upper limit of the length of options plus
2858 * NOTE that this routine is only called when we process an incoming
2859 * segment, for outgoing segments only tcp_mssopt is called.
2862 tcp_mss(struct tcpcb *tp, int offer)
2868 struct inpcb *inp = tp->t_inpcb;
2871 boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) ? TRUE : FALSE);
2872 size_t min_protoh = isipv6 ?
2873 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
2874 sizeof(struct tcpiphdr);
2876 const boolean_t isipv6 = FALSE;
2877 const size_t min_protoh = sizeof(struct tcpiphdr);
2881 rt = tcp_rtlookup6(&inp->inp_inc);
2883 rt = tcp_rtlookup(&inp->inp_inc);
2885 tp->t_maxopd = tp->t_maxseg =
2886 (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
2890 so = inp->inp_socket;
2893 * Offer == 0 means that there was no MSS on the SYN segment,
2894 * in this case we use either the interface mtu or tcp_mssdflt.
2896 * An offer which is too large will be cut down later.
2900 if (in6_localaddr(&inp->in6p_faddr)) {
2901 offer = ND_IFINFO(rt->rt_ifp)->linkmtu -
2904 offer = tcp_v6mssdflt;
2907 if (in_localaddr(inp->inp_faddr))
2908 offer = ifp->if_mtu - min_protoh;
2910 offer = tcp_mssdflt;
2915 * Prevent DoS attack with too small MSS. Round up
2916 * to at least minmss.
2918 * Sanity check: make sure that maxopd will be large
2919 * enough to allow some data on segments even is the
2920 * all the option space is used (40bytes). Otherwise
2921 * funny things may happen in tcp_output.
2923 offer = max(offer, tcp_minmss);
2924 offer = max(offer, 64);
2926 rt->rt_rmx.rmx_mssopt = offer;
2929 * While we're here, check if there's an initial rtt
2930 * or rttvar. Convert from the route-table units
2931 * to scaled multiples of the slow timeout timer.
2933 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
2935 * XXX the lock bit for RTT indicates that the value
2936 * is also a minimum value; this is subject to time.
2938 if (rt->rt_rmx.rmx_locks & RTV_RTT)
2939 tp->t_rttmin = rtt / (RTM_RTTUNIT / hz);
2940 tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE));
2941 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
2942 tcpstat.tcps_usedrtt++;
2943 if (rt->rt_rmx.rmx_rttvar) {
2944 tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
2945 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE));
2946 tcpstat.tcps_usedrttvar++;
2948 /* default variation is +- 1 rtt */
2950 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
2952 TCPT_RANGESET(tp->t_rxtcur,
2953 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
2954 tp->t_rttmin, TCPTV_REXMTMAX);
2958 * if there's an mtu associated with the route, use it
2959 * else, use the link mtu. Take the smaller of mss or offer
2962 if (rt->rt_rmx.rmx_mtu) {
2963 mss = rt->rt_rmx.rmx_mtu - min_protoh;
2966 mss = ND_IFINFO(rt->rt_ifp)->linkmtu - min_protoh;
2968 mss = ifp->if_mtu - min_protoh;
2970 mss = min(mss, offer);
2973 * maxopd stores the maximum length of data AND options
2974 * in a segment; maxseg is the amount of data in a normal
2975 * segment. We need to store this value (maxopd) apart
2976 * from maxseg, because now every segment carries options
2977 * and thus we normally have somewhat less data in segments.
2981 if ((tp->t_flags & (TF_REQ_TSTMP | TF_NOOPT)) == TF_REQ_TSTMP &&
2982 ((tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
2983 mss -= TCPOLEN_TSTAMP_APPA;
2985 #if (MCLBYTES & (MCLBYTES - 1)) == 0
2987 mss &= ~(MCLBYTES-1);
2990 mss = mss / MCLBYTES * MCLBYTES;
2993 * If there's a pipesize, change the socket buffer
2994 * to that size. Make the socket buffers an integral
2995 * number of mss units; if the mss is larger than
2996 * the socket buffer, decrease the mss.
2999 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0)
3001 bufsize = so->so_snd.ssb_hiwat;
3005 bufsize = roundup(bufsize, mss);
3006 if (bufsize > sb_max)
3008 if (bufsize > so->so_snd.ssb_hiwat)
3009 ssb_reserve(&so->so_snd, bufsize, so, NULL);
3014 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0)
3016 bufsize = so->so_rcv.ssb_hiwat;
3017 if (bufsize > mss) {
3018 bufsize = roundup(bufsize, mss);
3019 if (bufsize > sb_max)
3021 if (bufsize > so->so_rcv.ssb_hiwat)
3022 ssb_reserve(&so->so_rcv, bufsize, so, NULL);
3026 * Set the slow-start flight size depending on whether this
3027 * is a local network or not.
3030 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380));
3034 if (rt->rt_rmx.rmx_ssthresh) {
3036 * There's some sort of gateway or interface
3037 * buffer limit on the path. Use this to set
3038 * the slow start threshhold, but set the
3039 * threshold to no less than 2*mss.
3041 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
3042 tcpstat.tcps_usedssthresh++;
3047 * Determine the MSS option to send on an outgoing SYN.
3050 tcp_mssopt(struct tcpcb *tp)
3055 ((tp->t_inpcb->inp_vflag & INP_IPV6) ? TRUE : FALSE);
3056 int min_protoh = isipv6 ?
3057 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
3058 sizeof(struct tcpiphdr);
3060 const boolean_t isipv6 = FALSE;
3061 const size_t min_protoh = sizeof(struct tcpiphdr);
3065 rt = tcp_rtlookup6(&tp->t_inpcb->inp_inc);
3067 rt = tcp_rtlookup(&tp->t_inpcb->inp_inc);
3069 return (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
3071 return (rt->rt_ifp->if_mtu - min_protoh);
3075 * When a partial ack arrives, force the retransmission of the
3076 * next unacknowledged segment. Do not exit Fast Recovery.
3078 * Implement the Slow-but-Steady variant of NewReno by restarting the
3079 * the retransmission timer. Turn it off here so it can be restarted
3080 * later in tcp_output().
3083 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th, int acked)
3085 tcp_seq old_snd_nxt = tp->snd_nxt;
3086 u_long ocwnd = tp->snd_cwnd;
3088 tcp_callout_stop(tp, tp->tt_rexmt);
3090 tp->snd_nxt = th->th_ack;
3091 /* Set snd_cwnd to one segment beyond acknowledged offset. */
3092 tp->snd_cwnd = tp->t_maxseg;
3093 tp->t_flags |= TF_ACKNOW;
3095 if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
3096 tp->snd_nxt = old_snd_nxt;
3097 /* partial window deflation */
3099 tp->snd_cwnd = ocwnd - acked + tp->t_maxseg;
3101 tp->snd_cwnd = tp->t_maxseg;
3105 * In contrast to the Slow-but-Steady NewReno variant,
3106 * we do not reset the retransmission timer for SACK retransmissions,
3107 * except when retransmitting snd_una.
3110 tcp_sack_rexmt(struct tcpcb *tp, struct tcphdr *th)
3112 uint32_t pipe, seglen;
3115 tcp_seq old_snd_nxt = tp->snd_nxt;
3116 u_long ocwnd = tp->snd_cwnd;
3117 int nseg = 0; /* consecutive new segments */
3118 #define MAXBURST 4 /* limit burst of new packets on partial ack */
3121 pipe = tcp_sack_compute_pipe(tp);
3122 while ((tcp_seq_diff_t)(ocwnd - pipe) >= (tcp_seq_diff_t)tp->t_maxseg &&
3123 (!tcp_do_smartsack || nseg < MAXBURST) &&
3124 tcp_sack_nextseg(tp, &nextrexmt, &seglen, &lostdup)) {
3126 tcp_seq old_snd_max;
3129 if (nextrexmt == tp->snd_max)
3131 tp->snd_nxt = nextrexmt;
3132 tp->snd_cwnd = nextrexmt - tp->snd_una + seglen;
3133 old_snd_max = tp->snd_max;
3134 if (nextrexmt == tp->snd_una)
3135 tcp_callout_stop(tp, tp->tt_rexmt);
3136 error = tcp_output(tp);
3139 sent = tp->snd_nxt - nextrexmt;
3144 tcpstat.tcps_sndsackpack++;
3145 tcpstat.tcps_sndsackbyte += sent;
3146 if (SEQ_LT(nextrexmt, old_snd_max) &&
3147 SEQ_LT(tp->rexmt_high, tp->snd_nxt))
3148 tp->rexmt_high = seq_min(tp->snd_nxt, old_snd_max);
3150 if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
3151 tp->snd_nxt = old_snd_nxt;
3152 tp->snd_cwnd = ocwnd;
projects / dragonfly.git / blob