2 * Copyright (c) 2002, 2003, 2004 Jeffrey M. Hsu. All rights reserved.
3 * Copyright (c) 2002, 2003, 2004 The DragonFly Project. All rights reserved.
5 * This code is derived from software contributed to The DragonFly Project
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. Neither the name of The DragonFly Project nor the names of its
17 * contributors may be used to endorse or promote products derived
18 * from this software without specific, prior written permission.
20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
23 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
24 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
25 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
26 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
27 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
28 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
29 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
30 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
36 * The Regents of the University of California. All rights reserved.
38 * Redistribution and use in source and binary forms, with or without
39 * modification, are permitted provided that the following conditions
41 * 1. Redistributions of source code must retain the above copyright
42 * notice, this list of conditions and the following disclaimer.
43 * 2. Redistributions in binary form must reproduce the above copyright
44 * notice, this list of conditions and the following disclaimer in the
45 * documentation and/or other materials provided with the distribution.
46 * 3. All advertising materials mentioning features or use of this software
47 * must display the following acknowledgement:
48 * This product includes software developed by the University of
49 * California, Berkeley and its contributors.
50 * 4. Neither the name of the University nor the names of its contributors
51 * may be used to endorse or promote products derived from this software
52 * without specific prior written permission.
54 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
55 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
56 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
57 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
58 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
59 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
60 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
61 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
62 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
63 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
66 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95
67 * $FreeBSD: src/sys/netinet/tcp_input.c,v 1.107.2.38 2003/05/21 04:46:41 cjc Exp $
68 * $DragonFly: src/sys/netinet/tcp_input.c,v 1.68 2008/08/22 09:14:17 sephe Exp $
71 #include "opt_ipfw.h" /* for ipfw_fwd */
73 #include "opt_inet6.h"
74 #include "opt_ipsec.h"
75 #include "opt_tcpdebug.h"
76 #include "opt_tcp_input.h"
78 #include <sys/param.h>
79 #include <sys/systm.h>
80 #include <sys/kernel.h>
81 #include <sys/sysctl.h>
82 #include <sys/malloc.h>
84 #include <sys/proc.h> /* for proc0 declaration */
85 #include <sys/protosw.h>
86 #include <sys/socket.h>
87 #include <sys/socketvar.h>
88 #include <sys/syslog.h>
89 #include <sys/in_cksum.h>
91 #include <sys/socketvar2.h>
93 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
94 #include <machine/stdarg.h>
97 #include <net/route.h>
99 #include <netinet/in.h>
100 #include <netinet/in_systm.h>
101 #include <netinet/ip.h>
102 #include <netinet/ip_icmp.h> /* for ICMP_BANDLIM */
103 #include <netinet/in_var.h>
104 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */
105 #include <netinet/in_pcb.h>
106 #include <netinet/ip_var.h>
107 #include <netinet/ip6.h>
108 #include <netinet/icmp6.h>
109 #include <netinet6/nd6.h>
110 #include <netinet6/ip6_var.h>
111 #include <netinet6/in6_pcb.h>
112 #include <netinet/tcp.h>
113 #include <netinet/tcp_fsm.h>
114 #include <netinet/tcp_seq.h>
115 #include <netinet/tcp_timer.h>
116 #include <netinet/tcp_timer2.h>
117 #include <netinet/tcp_var.h>
118 #include <netinet6/tcp6_var.h>
119 #include <netinet/tcpip.h>
122 #include <netinet/tcp_debug.h>
124 u_char tcp_saveipgen[40]; /* the size must be of max ip header, now IPv6 */
125 struct tcphdr tcp_savetcp;
129 #include <netproto/ipsec/ipsec.h>
130 #include <netproto/ipsec/ipsec6.h>
134 #include <netinet6/ipsec.h>
135 #include <netinet6/ipsec6.h>
136 #include <netproto/key/key.h>
139 MALLOC_DEFINE(M_TSEGQ, "tseg_qent", "TCP segment queue entry");
141 static int log_in_vain = 0;
142 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
143 &log_in_vain, 0, "Log all incoming TCP connections");
145 static int blackhole = 0;
146 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW,
147 &blackhole, 0, "Do not send RST when dropping refused connections");
149 int tcp_delack_enabled = 1;
150 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW,
151 &tcp_delack_enabled, 0,
152 "Delay ACK to try and piggyback it onto a data packet");
154 #ifdef TCP_DROP_SYNFIN
155 static int drop_synfin = 0;
156 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW,
157 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set");
160 static int tcp_do_limitedtransmit = 1;
161 SYSCTL_INT(_net_inet_tcp, OID_AUTO, limitedtransmit, CTLFLAG_RW,
162 &tcp_do_limitedtransmit, 0, "Enable RFC 3042 (Limited Transmit)");
164 static int tcp_do_early_retransmit = 1;
165 SYSCTL_INT(_net_inet_tcp, OID_AUTO, earlyretransmit, CTLFLAG_RW,
166 &tcp_do_early_retransmit, 0, "Early retransmit");
168 int tcp_aggregate_acks = 1;
169 SYSCTL_INT(_net_inet_tcp, OID_AUTO, aggregate_acks, CTLFLAG_RW,
170 &tcp_aggregate_acks, 0, "Aggregate built-up acks into one ack");
172 int tcp_do_rfc3390 = 1;
173 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW,
175 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
177 static int tcp_do_eifel_detect = 1;
178 SYSCTL_INT(_net_inet_tcp, OID_AUTO, eifel, CTLFLAG_RW,
179 &tcp_do_eifel_detect, 0, "Eifel detection algorithm (RFC 3522)");
181 static int tcp_do_abc = 1;
182 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc, CTLFLAG_RW,
184 "TCP Appropriate Byte Counting (RFC 3465)");
187 * Define as tunable for easy testing with SACK on and off.
188 * Warning: do not change setting in the middle of an existing active TCP flow,
189 * else strange things might happen to that flow.
192 SYSCTL_INT(_net_inet_tcp, OID_AUTO, sack, CTLFLAG_RW,
193 &tcp_do_sack, 0, "Enable SACK Algorithms");
195 int tcp_do_smartsack = 1;
196 SYSCTL_INT(_net_inet_tcp, OID_AUTO, smartsack, CTLFLAG_RW,
197 &tcp_do_smartsack, 0, "Enable Smart SACK Algorithms");
199 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0,
200 "TCP Segment Reassembly Queue");
202 int tcp_reass_maxseg = 0;
203 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RD,
204 &tcp_reass_maxseg, 0,
205 "Global maximum number of TCP Segments in Reassembly Queue");
207 int tcp_reass_qsize = 0;
208 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD,
210 "Global number of TCP Segments currently in Reassembly Queue");
212 static int tcp_reass_overflows = 0;
213 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD,
214 &tcp_reass_overflows, 0,
215 "Global number of TCP Segment Reassembly Queue Overflows");
217 int tcp_do_autorcvbuf = 1;
218 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_RW,
219 &tcp_do_autorcvbuf, 0, "Enable automatic receive buffer sizing");
221 int tcp_autorcvbuf_inc = 16*1024;
222 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_RW,
223 &tcp_autorcvbuf_inc, 0,
224 "Incrementor step size of automatic receive buffer");
226 int tcp_autorcvbuf_max = 2*1024*1024;
227 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_RW,
228 &tcp_autorcvbuf_max, 0, "Max size of automatic receive buffer");
230 int tcp_sosnd_agglim = 2;
231 SYSCTL_INT(_net_inet_tcp, OID_AUTO, sosnd_agglim, CTLFLAG_RW,
232 &tcp_sosnd_agglim, 0, "TCP sosend mbuf aggregation limit");
234 int tcp_sosnd_async = 1;
235 SYSCTL_INT(_net_inet_tcp, OID_AUTO, sosnd_async, CTLFLAG_RW,
236 &tcp_sosnd_async, 0, "TCP asynchronized pru_send");
238 static void tcp_dooptions(struct tcpopt *, u_char *, int, boolean_t);
239 static void tcp_pulloutofband(struct socket *,
240 struct tcphdr *, struct mbuf *, int);
241 static int tcp_reass(struct tcpcb *, struct tcphdr *, int *,
243 static void tcp_xmit_timer(struct tcpcb *, int);
244 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *, int);
245 static void tcp_sack_rexmt(struct tcpcb *, struct tcphdr *);
246 static int tcp_rmx_msl(const struct tcpcb *);
248 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
250 #define ND6_HINT(tp) \
252 if ((tp) && (tp)->t_inpcb && \
253 ((tp)->t_inpcb->inp_vflag & INP_IPV6) && \
254 (tp)->t_inpcb->in6p_route.ro_rt) \
255 nd6_nud_hint((tp)->t_inpcb->in6p_route.ro_rt, NULL, 0); \
262 * Indicate whether this ack should be delayed. We can delay the ack if
263 * - delayed acks are enabled and
264 * - there is no delayed ack timer in progress and
265 * - our last ack wasn't a 0-sized window. We never want to delay
266 * the ack that opens up a 0-sized window.
268 #define DELAY_ACK(tp) \
269 (tcp_delack_enabled && !tcp_callout_pending(tp, tp->tt_delack) && \
270 !(tp->t_flags & TF_RXWIN0SENT))
272 #define acceptable_window_update(tp, th, tiwin) \
273 (SEQ_LT(tp->snd_wl1, th->th_seq) || \
274 (tp->snd_wl1 == th->th_seq && \
275 (SEQ_LT(tp->snd_wl2, th->th_ack) || \
276 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))
279 tcp_reass(struct tcpcb *tp, struct tcphdr *th, int *tlenp, struct mbuf *m)
282 struct tseg_qent *p = NULL;
283 struct tseg_qent *te;
284 struct socket *so = tp->t_inpcb->inp_socket;
288 * Call with th == NULL after become established to
289 * force pre-ESTABLISHED data up to user socket.
295 * Limit the number of segments in the reassembly queue to prevent
296 * holding on to too many segments (and thus running out of mbufs).
297 * Make sure to let the missing segment through which caused this
298 * queue. Always keep one global queue entry spare to be able to
299 * process the missing segment.
301 if (th->th_seq != tp->rcv_nxt &&
302 tcp_reass_qsize + 1 >= tcp_reass_maxseg) {
303 tcp_reass_overflows++;
304 tcpstat.tcps_rcvmemdrop++;
306 /* no SACK block to report */
307 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
311 /* Allocate a new queue entry. */
312 MALLOC(te, struct tseg_qent *, sizeof(struct tseg_qent), M_TSEGQ,
313 M_INTWAIT | M_NULLOK);
315 tcpstat.tcps_rcvmemdrop++;
317 /* no SACK block to report */
318 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
321 atomic_add_int(&tcp_reass_qsize, 1);
324 * Find a segment which begins after this one does.
326 LIST_FOREACH(q, &tp->t_segq, tqe_q) {
327 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
333 * If there is a preceding segment, it may provide some of
334 * our data already. If so, drop the data from the incoming
335 * segment. If it provides all of our data, drop us.
340 /* conversion to int (in i) handles seq wraparound */
341 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
342 if (i > 0) { /* overlaps preceding segment */
343 tp->t_flags |= (TF_DUPSEG | TF_ENCLOSESEG);
344 /* enclosing block starts w/ preceding segment */
345 tp->encloseblk.rblk_start = p->tqe_th->th_seq;
347 /* preceding encloses incoming segment */
348 tp->encloseblk.rblk_end = p->tqe_th->th_seq +
350 tcpstat.tcps_rcvduppack++;
351 tcpstat.tcps_rcvdupbyte += *tlenp;
354 atomic_add_int(&tcp_reass_qsize, -1);
356 * Try to present any queued data
357 * at the left window edge to the user.
358 * This is needed after the 3-WHS
361 goto present; /* ??? */
366 /* incoming segment end is enclosing block end */
367 tp->encloseblk.rblk_end = th->th_seq + *tlenp +
368 ((th->th_flags & TH_FIN) != 0);
369 /* trim end of reported D-SACK block */
370 tp->reportblk.rblk_end = th->th_seq;
373 tcpstat.tcps_rcvoopack++;
374 tcpstat.tcps_rcvoobyte += *tlenp;
377 * While we overlap succeeding segments trim them or,
378 * if they are completely covered, dequeue them.
381 tcp_seq_diff_t i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
382 tcp_seq qend = q->tqe_th->th_seq + q->tqe_len;
383 struct tseg_qent *nq;
387 if (!(tp->t_flags & TF_DUPSEG)) { /* first time through */
388 tp->t_flags |= (TF_DUPSEG | TF_ENCLOSESEG);
389 tp->encloseblk = tp->reportblk;
390 /* report trailing duplicate D-SACK segment */
391 tp->reportblk.rblk_start = q->tqe_th->th_seq;
393 if ((tp->t_flags & TF_ENCLOSESEG) &&
394 SEQ_GT(qend, tp->encloseblk.rblk_end)) {
395 /* extend enclosing block if one exists */
396 tp->encloseblk.rblk_end = qend;
398 if (i < q->tqe_len) {
399 q->tqe_th->th_seq += i;
405 nq = LIST_NEXT(q, tqe_q);
406 LIST_REMOVE(q, tqe_q);
409 atomic_add_int(&tcp_reass_qsize, -1);
413 /* Insert the new segment queue entry into place. */
416 te->tqe_len = *tlenp;
418 /* check if can coalesce with following segment */
419 if (q != NULL && (th->th_seq + *tlenp == q->tqe_th->th_seq)) {
420 tcp_seq tend = te->tqe_th->th_seq + te->tqe_len;
422 te->tqe_len += q->tqe_len;
423 if (q->tqe_th->th_flags & TH_FIN)
424 te->tqe_th->th_flags |= TH_FIN;
425 m_cat(te->tqe_m, q->tqe_m);
426 tp->encloseblk.rblk_end = tend;
428 * When not reporting a duplicate segment, use
429 * the larger enclosing block as the SACK block.
431 if (!(tp->t_flags & TF_DUPSEG))
432 tp->reportblk.rblk_end = tend;
433 LIST_REMOVE(q, tqe_q);
435 atomic_add_int(&tcp_reass_qsize, -1);
439 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
441 /* check if can coalesce with preceding segment */
442 if (p->tqe_th->th_seq + p->tqe_len == th->th_seq) {
443 p->tqe_len += te->tqe_len;
444 m_cat(p->tqe_m, te->tqe_m);
445 tp->encloseblk.rblk_start = p->tqe_th->th_seq;
447 * When not reporting a duplicate segment, use
448 * the larger enclosing block as the SACK block.
450 if (!(tp->t_flags & TF_DUPSEG))
451 tp->reportblk.rblk_start = p->tqe_th->th_seq;
453 atomic_add_int(&tcp_reass_qsize, -1);
455 LIST_INSERT_AFTER(p, te, tqe_q);
461 * Present data to user, advancing rcv_nxt through
462 * completed sequence space.
464 if (!TCPS_HAVEESTABLISHED(tp->t_state))
466 q = LIST_FIRST(&tp->t_segq);
467 if (q == NULL || q->tqe_th->th_seq != tp->rcv_nxt)
469 tp->rcv_nxt += q->tqe_len;
470 if (!(tp->t_flags & TF_DUPSEG)) {
471 /* no SACK block to report since ACK advanced */
472 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
474 /* no enclosing block to report since ACK advanced */
475 tp->t_flags &= ~TF_ENCLOSESEG;
476 flags = q->tqe_th->th_flags & TH_FIN;
477 LIST_REMOVE(q, tqe_q);
478 KASSERT(LIST_EMPTY(&tp->t_segq) ||
479 LIST_FIRST(&tp->t_segq)->tqe_th->th_seq != tp->rcv_nxt,
480 ("segment not coalesced"));
481 if (so->so_state & SS_CANTRCVMORE) {
484 lwkt_gettoken(&so->so_rcv.ssb_token);
485 ssb_appendstream(&so->so_rcv, q->tqe_m);
486 lwkt_reltoken(&so->so_rcv.ssb_token);
489 atomic_add_int(&tcp_reass_qsize, -1);
496 * TCP input routine, follows pages 65-76 of the
497 * protocol specification dated September, 1981 very closely.
501 tcp6_input(struct mbuf **mp, int *offp, int proto)
503 struct mbuf *m = *mp;
504 struct in6_ifaddr *ia6;
506 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
509 * draft-itojun-ipv6-tcp-to-anycast
510 * better place to put this in?
512 ia6 = ip6_getdstifaddr(m);
513 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
516 ip6 = mtod(m, struct ip6_hdr *);
517 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
518 offsetof(struct ip6_hdr, ip6_dst));
519 return (IPPROTO_DONE);
522 tcp_input(mp, offp, proto);
523 return (IPPROTO_DONE);
528 tcp_input(struct mbuf **mp, int *offp, int proto)
532 struct ip *ip = NULL;
534 struct inpcb *inp = NULL;
540 struct tcpcb *tp = NULL;
542 struct socket *so = 0;
544 boolean_t ourfinisacked, needoutput = FALSE;
547 struct tcpopt to; /* options in this segment */
548 struct sockaddr_in *next_hop = NULL;
549 int rstreason; /* For badport_bandlim accounting purposes */
551 struct ip6_hdr *ip6 = NULL;
556 const boolean_t isipv6 = FALSE;
566 tcpstat.tcps_rcvtotal++;
568 if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
571 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
572 KKASSERT(mtag != NULL);
573 next_hop = m_tag_data(mtag);
577 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? TRUE : FALSE;
581 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */
582 ip6 = mtod(m, struct ip6_hdr *);
583 tlen = (sizeof *ip6) + ntohs(ip6->ip6_plen) - off0;
584 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) {
585 tcpstat.tcps_rcvbadsum++;
588 th = (struct tcphdr *)((caddr_t)ip6 + off0);
591 * Be proactive about unspecified IPv6 address in source.
592 * As we use all-zero to indicate unbounded/unconnected pcb,
593 * unspecified IPv6 address can be used to confuse us.
595 * Note that packets with unspecified IPv6 destination is
596 * already dropped in ip6_input.
598 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
604 * Get IP and TCP header together in first mbuf.
605 * Note: IP leaves IP header in first mbuf.
607 if (off0 > sizeof(struct ip)) {
609 off0 = sizeof(struct ip);
611 /* already checked and pulled up in ip_demux() */
612 KASSERT(m->m_len >= sizeof(struct tcpiphdr),
613 ("TCP header not in one mbuf: m->m_len %d", m->m_len));
614 ip = mtod(m, struct ip *);
615 ipov = (struct ipovly *)ip;
616 th = (struct tcphdr *)((caddr_t)ip + off0);
619 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
620 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
621 th->th_sum = m->m_pkthdr.csum_data;
623 th->th_sum = in_pseudo(ip->ip_src.s_addr,
625 htonl(m->m_pkthdr.csum_data +
628 th->th_sum ^= 0xffff;
631 * Checksum extended TCP header and data.
633 len = sizeof(struct ip) + tlen;
634 bzero(ipov->ih_x1, sizeof ipov->ih_x1);
635 ipov->ih_len = (u_short)tlen;
636 ipov->ih_len = htons(ipov->ih_len);
637 th->th_sum = in_cksum(m, len);
640 tcpstat.tcps_rcvbadsum++;
644 /* Re-initialization for later version check */
645 ip->ip_v = IPVERSION;
650 * Check that TCP offset makes sense,
651 * pull out TCP options and adjust length. XXX
653 off = th->th_off << 2;
654 /* already checked and pulled up in ip_demux() */
655 KASSERT(off >= sizeof(struct tcphdr) && off <= tlen,
656 ("bad TCP data offset %d (tlen %d)", off, tlen));
657 tlen -= off; /* tlen is used instead of ti->ti_len */
658 if (off > sizeof(struct tcphdr)) {
660 IP6_EXTHDR_CHECK(m, off0, off, IPPROTO_DONE);
661 ip6 = mtod(m, struct ip6_hdr *);
662 th = (struct tcphdr *)((caddr_t)ip6 + off0);
664 /* already pulled up in ip_demux() */
665 KASSERT(m->m_len >= sizeof(struct ip) + off,
666 ("TCP header and options not in one mbuf: "
667 "m_len %d, off %d", m->m_len, off));
669 optlen = off - sizeof(struct tcphdr);
670 optp = (u_char *)(th + 1);
672 thflags = th->th_flags;
674 #ifdef TCP_DROP_SYNFIN
676 * If the drop_synfin option is enabled, drop all packets with
677 * both the SYN and FIN bits set. This prevents e.g. nmap from
678 * identifying the TCP/IP stack.
680 * This is a violation of the TCP specification.
682 if (drop_synfin && (thflags & (TH_SYN | TH_FIN)) == (TH_SYN | TH_FIN))
687 * Convert TCP protocol specific fields to host format.
689 th->th_seq = ntohl(th->th_seq);
690 th->th_ack = ntohl(th->th_ack);
691 th->th_win = ntohs(th->th_win);
692 th->th_urp = ntohs(th->th_urp);
695 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options,
696 * until after ip6_savecontrol() is called and before other functions
697 * which don't want those proto headers.
698 * Because ip6_savecontrol() is going to parse the mbuf to
699 * search for data to be passed up to user-land, it wants mbuf
700 * parameters to be unchanged.
701 * XXX: the call of ip6_savecontrol() has been obsoleted based on
702 * latest version of the advanced API (20020110).
704 drop_hdrlen = off0 + off;
707 * Locate pcb for segment.
710 /* IPFIREWALL_FORWARD section */
711 if (next_hop != NULL && !isipv6) { /* IPv6 support is not there yet */
713 * Transparently forwarded. Pretend to be the destination.
714 * already got one like this?
716 cpu = mycpu->gd_cpuid;
717 inp = in_pcblookup_hash(&tcbinfo[cpu],
718 ip->ip_src, th->th_sport,
719 ip->ip_dst, th->th_dport,
720 0, m->m_pkthdr.rcvif);
723 * It's new. Try to find the ambushing socket.
727 * The rest of the ipfw code stores the port in
729 * (The IP address is still in network order.)
731 in_port_t dport = next_hop->sin_port ?
732 htons(next_hop->sin_port) :
735 cpu = tcp_addrcpu(ip->ip_src.s_addr, th->th_sport,
736 next_hop->sin_addr.s_addr, dport);
737 inp = in_pcblookup_hash(&tcbinfo[cpu],
738 ip->ip_src, th->th_sport,
739 next_hop->sin_addr, dport,
740 1, m->m_pkthdr.rcvif);
744 inp = in6_pcblookup_hash(&tcbinfo[0],
745 &ip6->ip6_src, th->th_sport,
746 &ip6->ip6_dst, th->th_dport,
747 1, m->m_pkthdr.rcvif);
749 cpu = mycpu->gd_cpuid;
750 inp = in_pcblookup_hash(&tcbinfo[cpu],
751 ip->ip_src, th->th_sport,
752 ip->ip_dst, th->th_dport,
753 1, m->m_pkthdr.rcvif);
758 * If the state is CLOSED (i.e., TCB does not exist) then
759 * all data in the incoming segment is discarded.
760 * If the TCB exists but is in CLOSED state, it is embryonic,
761 * but should either do a listen or a connect soon.
766 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2];
768 char dbuf[sizeof "aaa.bbb.ccc.ddd"];
769 char sbuf[sizeof "aaa.bbb.ccc.ddd"];
773 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst));
776 strcat(sbuf, ip6_sprintf(&ip6->ip6_src));
779 strcpy(dbuf, inet_ntoa(ip->ip_dst));
780 strcpy(sbuf, inet_ntoa(ip->ip_src));
782 switch (log_in_vain) {
784 if (!(thflags & TH_SYN))
788 "Connection attempt to TCP %s:%d "
789 "from %s:%d flags:0x%02x\n",
790 dbuf, ntohs(th->th_dport), sbuf,
791 ntohs(th->th_sport), thflags);
800 if (thflags & TH_SYN)
809 rstreason = BANDLIM_RST_CLOSEDPORT;
815 if (ipsec6_in_reject_so(m, inp->inp_socket)) {
816 ipsec6stat.in_polvio++;
820 if (ipsec4_in_reject_so(m, inp->inp_socket)) {
821 ipsecstat.in_polvio++;
828 if (ipsec6_in_reject(m, inp))
831 if (ipsec4_in_reject(m, inp))
835 /* Check the minimum TTL for socket. */
837 if ((isipv6 ? ip6->ip6_hlim : ip->ip_ttl) < inp->inp_ip_minttl)
843 rstreason = BANDLIM_RST_CLOSEDPORT;
846 if (tp->t_state <= TCPS_CLOSED)
849 /* Unscale the window into a 32-bit value. */
850 if (!(thflags & TH_SYN))
851 tiwin = th->th_win << tp->snd_scale;
855 so = inp->inp_socket;
858 if (so->so_options & SO_DEBUG) {
859 ostate = tp->t_state;
861 bcopy(ip6, tcp_saveipgen, sizeof(*ip6));
863 bcopy(ip, tcp_saveipgen, sizeof(*ip));
868 bzero(&to, sizeof to);
870 if (so->so_options & SO_ACCEPTCONN) {
871 struct in_conninfo inc;
874 inc.inc_isipv6 = (isipv6 == TRUE);
877 inc.inc6_faddr = ip6->ip6_src;
878 inc.inc6_laddr = ip6->ip6_dst;
879 inc.inc6_route.ro_rt = NULL; /* XXX */
881 inc.inc_faddr = ip->ip_src;
882 inc.inc_laddr = ip->ip_dst;
883 inc.inc_route.ro_rt = NULL; /* XXX */
885 inc.inc_fport = th->th_sport;
886 inc.inc_lport = th->th_dport;
889 * If the state is LISTEN then ignore segment if it contains
890 * a RST. If the segment contains an ACK then it is bad and
891 * send a RST. If it does not contain a SYN then it is not
892 * interesting; drop it.
894 * If the state is SYN_RECEIVED (syncache) and seg contains
895 * an ACK, but not for our SYN/ACK, send a RST. If the seg
896 * contains a RST, check the sequence number to see if it
897 * is a valid reset segment.
899 if ((thflags & (TH_RST | TH_ACK | TH_SYN)) != TH_SYN) {
900 if ((thflags & (TH_RST | TH_ACK | TH_SYN)) == TH_ACK) {
901 if (!syncache_expand(&inc, th, &so, m)) {
903 * No syncache entry, or ACK was not
904 * for our SYN/ACK. Send a RST.
906 tcpstat.tcps_badsyn++;
907 rstreason = BANDLIM_RST_OPENPORT;
912 * Could not complete 3-way handshake,
913 * connection is being closed down, and
914 * syncache will free mbuf.
917 return(IPPROTO_DONE);
920 * We must be in the correct protocol thread
921 * for this connection.
923 KKASSERT(so->so_port == &curthread->td_msgport);
926 * Socket is created in state SYN_RECEIVED.
927 * Continue processing segment.
932 * This is what would have happened in
933 * tcp_output() when the SYN,ACK was sent.
935 tp->snd_up = tp->snd_una;
936 tp->snd_max = tp->snd_nxt = tp->iss + 1;
937 tp->last_ack_sent = tp->rcv_nxt;
939 * XXX possible bug - it doesn't appear that tp->snd_wnd is unscaled
940 * until the _second_ ACK is received:
941 * rcv SYN (set wscale opts) --> send SYN/ACK, set snd_wnd = window.
942 * rcv ACK, calculate tiwin --> process SYN_RECEIVED, determine wscale,
943 * move to ESTAB, set snd_wnd to tiwin.
945 tp->snd_wnd = tiwin; /* unscaled */
948 if (thflags & TH_RST) {
949 syncache_chkrst(&inc, th);
952 if (thflags & TH_ACK) {
953 syncache_badack(&inc);
954 tcpstat.tcps_badsyn++;
955 rstreason = BANDLIM_RST_OPENPORT;
962 * Segment's flags are (SYN) or (SYN | FIN).
966 * If deprecated address is forbidden,
967 * we do not accept SYN to deprecated interface
968 * address to prevent any new inbound connection from
969 * getting established.
970 * When we do not accept SYN, we send a TCP RST,
971 * with deprecated source address (instead of dropping
972 * it). We compromise it as it is much better for peer
973 * to send a RST, and RST will be the final packet
976 * If we do not forbid deprecated addresses, we accept
977 * the SYN packet. RFC2462 does not suggest dropping
979 * If we decipher RFC2462 5.5.4, it says like this:
980 * 1. use of deprecated addr with existing
981 * communication is okay - "SHOULD continue to be
983 * 2. use of it with new communication:
984 * (2a) "SHOULD NOT be used if alternate address
985 * with sufficient scope is available"
986 * (2b) nothing mentioned otherwise.
987 * Here we fall into (2b) case as we have no choice in
988 * our source address selection - we must obey the peer.
990 * The wording in RFC2462 is confusing, and there are
991 * multiple description text for deprecated address
992 * handling - worse, they are not exactly the same.
993 * I believe 5.5.4 is the best one, so we follow 5.5.4.
995 if (isipv6 && !ip6_use_deprecated) {
996 struct in6_ifaddr *ia6;
998 if ((ia6 = ip6_getdstifaddr(m)) &&
999 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1001 rstreason = BANDLIM_RST_OPENPORT;
1007 * If it is from this socket, drop it, it must be forged.
1008 * Don't bother responding if the destination was a broadcast.
1010 if (th->th_dport == th->th_sport) {
1012 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
1016 if (ip->ip_dst.s_addr == ip->ip_src.s_addr)
1021 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
1023 * Note that it is quite possible to receive unicast
1024 * link-layer packets with a broadcast IP address. Use
1025 * in_broadcast() to find them.
1027 if (m->m_flags & (M_BCAST | M_MCAST))
1030 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1031 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
1034 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1035 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1036 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1037 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
1041 * SYN appears to be valid; create compressed TCP state
1042 * for syncache, or perform t/tcp connection.
1044 if (so->so_qlen <= so->so_qlimit) {
1045 tcp_dooptions(&to, optp, optlen, TRUE);
1046 if (!syncache_add(&inc, &to, th, &so, m))
1050 * Entry added to syncache, mbuf used to
1051 * send SYN,ACK packet.
1054 return(IPPROTO_DONE);
1057 * We must be in the correct protocol thread for
1060 KKASSERT(so->so_port == &curthread->td_msgport);
1063 tp = intotcpcb(inp);
1064 tp->snd_wnd = tiwin;
1065 tp->t_starttime = ticks;
1066 tp->t_state = TCPS_ESTABLISHED;
1069 * If there is a FIN, or if there is data and the
1070 * connection is local, then delay SYN,ACK(SYN) in
1071 * the hope of piggy-backing it on a response
1072 * segment. Otherwise must send ACK now in case
1073 * the other side is slow starting.
1075 if (DELAY_ACK(tp) &&
1076 ((thflags & TH_FIN) ||
1078 ((isipv6 && in6_localaddr(&inp->in6p_faddr)) ||
1079 (!isipv6 && in_localaddr(inp->inp_faddr)))))) {
1080 tcp_callout_reset(tp, tp->tt_delack,
1081 tcp_delacktime, tcp_timer_delack);
1082 tp->t_flags |= TF_NEEDSYN;
1084 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
1087 tcpstat.tcps_connects++;
1096 * Should not happen - syncache should pick up these connections.
1098 * Once we are past handling listen sockets we must be in the
1099 * correct protocol processing thread.
1101 KASSERT(tp->t_state != TCPS_LISTEN, ("tcp_input: TCPS_LISTEN state"));
1102 KKASSERT(so->so_port == &curthread->td_msgport);
1105 * This is the second part of the MSS DoS prevention code (after
1106 * minmss on the sending side) and it deals with too many too small
1107 * tcp packets in a too short timeframe (1 second).
1109 * XXX Removed. This code was crap. It does not scale to network
1110 * speed, and default values break NFS. Gone.
1115 * Segment received on connection.
1117 * Reset idle time and keep-alive timer. Don't waste time if less
1118 * then a second has elapsed.
1120 if ((int)(ticks - tp->t_rcvtime) > hz)
1121 tcp_timer_keep_activity(tp, thflags);
1125 * XXX this is tradtitional behavior, may need to be cleaned up.
1127 tcp_dooptions(&to, optp, optlen, (thflags & TH_SYN) != 0);
1128 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1129 if (to.to_flags & TOF_SCALE) {
1130 tp->t_flags |= TF_RCVD_SCALE;
1131 tp->requested_s_scale = to.to_requested_s_scale;
1133 if (to.to_flags & TOF_TS) {
1134 tp->t_flags |= TF_RCVD_TSTMP;
1135 tp->ts_recent = to.to_tsval;
1136 tp->ts_recent_age = ticks;
1138 if (to.to_flags & TOF_MSS)
1139 tcp_mss(tp, to.to_mss);
1141 * Only set the TF_SACK_PERMITTED per-connection flag
1142 * if we got a SACK_PERMITTED option from the other side
1143 * and the global tcp_do_sack variable is true.
1145 if (tcp_do_sack && (to.to_flags & TOF_SACK_PERMITTED))
1146 tp->t_flags |= TF_SACK_PERMITTED;
1150 * Header prediction: check for the two common cases
1151 * of a uni-directional data xfer. If the packet has
1152 * no control flags, is in-sequence, the window didn't
1153 * change and we're not retransmitting, it's a
1154 * candidate. If the length is zero and the ack moved
1155 * forward, we're the sender side of the xfer. Just
1156 * free the data acked & wake any higher level process
1157 * that was blocked waiting for space. If the length
1158 * is non-zero and the ack didn't move, we're the
1159 * receiver side. If we're getting packets in-order
1160 * (the reassembly queue is empty), add the data to
1161 * the socket buffer and note that we need a delayed ack.
1162 * Make sure that the hidden state-flags are also off.
1163 * Since we check for TCPS_ESTABLISHED above, it can only
1166 if (tp->t_state == TCPS_ESTABLISHED &&
1167 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1168 !(tp->t_flags & (TF_NEEDSYN | TF_NEEDFIN)) &&
1169 (!(to.to_flags & TOF_TS) ||
1170 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) &&
1171 th->th_seq == tp->rcv_nxt &&
1172 tp->snd_nxt == tp->snd_max) {
1175 * If last ACK falls within this segment's sequence numbers,
1176 * record the timestamp.
1177 * NOTE that the test is modified according to the latest
1178 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1180 if ((to.to_flags & TOF_TS) &&
1181 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1182 tp->ts_recent_age = ticks;
1183 tp->ts_recent = to.to_tsval;
1187 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1188 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1189 tp->snd_cwnd >= tp->snd_wnd &&
1190 !IN_FASTRECOVERY(tp)) {
1192 * This is a pure ack for outstanding data.
1194 ++tcpstat.tcps_predack;
1196 * "bad retransmit" recovery
1198 * If Eifel detection applies, then
1199 * it is deterministic, so use it
1200 * unconditionally over the old heuristic.
1201 * Otherwise, fall back to the old heuristic.
1203 if (tcp_do_eifel_detect &&
1204 (to.to_flags & TOF_TS) && to.to_tsecr &&
1205 (tp->t_flags & TF_FIRSTACCACK)) {
1206 /* Eifel detection applicable. */
1207 if (to.to_tsecr < tp->t_rexmtTS) {
1208 tcp_revert_congestion_state(tp);
1209 ++tcpstat.tcps_eifeldetected;
1211 } else if (tp->t_rxtshift == 1 &&
1212 ticks < tp->t_badrxtwin) {
1213 tcp_revert_congestion_state(tp);
1214 ++tcpstat.tcps_rttdetected;
1216 tp->t_flags &= ~(TF_FIRSTACCACK |
1217 TF_FASTREXMT | TF_EARLYREXMT);
1219 * Recalculate the retransmit timer / rtt.
1221 * Some machines (certain windows boxes)
1222 * send broken timestamp replies during the
1223 * SYN+ACK phase, ignore timestamps of 0.
1225 if ((to.to_flags & TOF_TS) && to.to_tsecr) {
1227 ticks - to.to_tsecr + 1);
1228 } else if (tp->t_rtttime &&
1229 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1231 ticks - tp->t_rtttime);
1233 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1234 acked = th->th_ack - tp->snd_una;
1235 tcpstat.tcps_rcvackpack++;
1236 tcpstat.tcps_rcvackbyte += acked;
1237 sbdrop(&so->so_snd.sb, acked);
1238 tp->snd_recover = th->th_ack - 1;
1239 tp->snd_una = th->th_ack;
1242 * Update window information.
1244 if (tiwin != tp->snd_wnd &&
1245 acceptable_window_update(tp, th, tiwin)) {
1246 /* keep track of pure window updates */
1247 if (tp->snd_wl2 == th->th_ack &&
1248 tiwin > tp->snd_wnd)
1249 tcpstat.tcps_rcvwinupd++;
1250 tp->snd_wnd = tiwin;
1251 tp->snd_wl1 = th->th_seq;
1252 tp->snd_wl2 = th->th_ack;
1253 if (tp->snd_wnd > tp->max_sndwnd)
1254 tp->max_sndwnd = tp->snd_wnd;
1257 ND6_HINT(tp); /* some progress has been done */
1259 * If all outstanding data are acked, stop
1260 * retransmit timer, otherwise restart timer
1261 * using current (possibly backed-off) value.
1262 * If process is waiting for space,
1263 * wakeup/selwakeup/signal. If data
1264 * are ready to send, let tcp_output
1265 * decide between more output or persist.
1267 if (tp->snd_una == tp->snd_max) {
1268 tcp_callout_stop(tp, tp->tt_rexmt);
1269 } else if (!tcp_callout_active(tp,
1271 tcp_callout_reset(tp, tp->tt_rexmt,
1272 tp->t_rxtcur, tcp_timer_rexmt);
1275 if (so->so_snd.ssb_cc > 0)
1277 return(IPPROTO_DONE);
1279 } else if (tiwin == tp->snd_wnd &&
1280 th->th_ack == tp->snd_una &&
1281 LIST_EMPTY(&tp->t_segq) &&
1282 tlen <= ssb_space(&so->so_rcv)) {
1283 u_long newsize = 0; /* automatic sockbuf scaling */
1285 * This is a pure, in-sequence data packet
1286 * with nothing on the reassembly queue and
1287 * we have enough buffer space to take it.
1289 ++tcpstat.tcps_preddat;
1290 tp->rcv_nxt += tlen;
1291 tcpstat.tcps_rcvpack++;
1292 tcpstat.tcps_rcvbyte += tlen;
1293 ND6_HINT(tp); /* some progress has been done */
1295 * Automatic sizing of receive socket buffer. Often the send
1296 * buffer size is not optimally adjusted to the actual network
1297 * conditions at hand (delay bandwidth product). Setting the
1298 * buffer size too small limits throughput on links with high
1299 * bandwidth and high delay (eg. trans-continental/oceanic links).
1301 * On the receive side the socket buffer memory is only rarely
1302 * used to any significant extent. This allows us to be much
1303 * more aggressive in scaling the receive socket buffer. For
1304 * the case that the buffer space is actually used to a large
1305 * extent and we run out of kernel memory we can simply drop
1306 * the new segments; TCP on the sender will just retransmit it
1307 * later. Setting the buffer size too big may only consume too
1308 * much kernel memory if the application doesn't read() from
1309 * the socket or packet loss or reordering makes use of the
1312 * The criteria to step up the receive buffer one notch are:
1313 * 1. the number of bytes received during the time it takes
1314 * one timestamp to be reflected back to us (the RTT);
1315 * 2. received bytes per RTT is within seven eighth of the
1316 * current socket buffer size;
1317 * 3. receive buffer size has not hit maximal automatic size;
1319 * This algorithm does one step per RTT at most and only if
1320 * we receive a bulk stream w/o packet losses or reorderings.
1321 * Shrinking the buffer during idle times is not necessary as
1322 * it doesn't consume any memory when idle.
1324 * TODO: Only step up if the application is actually serving
1325 * the buffer to better manage the socket buffer resources.
1327 if (tcp_do_autorcvbuf &&
1329 (so->so_rcv.ssb_flags & SSB_AUTOSIZE)) {
1330 if (to.to_tsecr > tp->rfbuf_ts &&
1331 to.to_tsecr - tp->rfbuf_ts < hz) {
1333 (so->so_rcv.ssb_hiwat / 8 * 7) &&
1334 so->so_rcv.ssb_hiwat <
1335 tcp_autorcvbuf_max) {
1337 ulmin(so->so_rcv.ssb_hiwat +
1339 tcp_autorcvbuf_max);
1341 /* Start over with next RTT. */
1345 tp->rfbuf_cnt += tlen; /* add up */
1348 * Add data to socket buffer.
1350 if (so->so_state & SS_CANTRCVMORE) {
1354 * Set new socket buffer size, give up when
1357 * Adjusting the size can mess up ACK
1358 * sequencing when pure window updates are
1359 * being avoided (which is the default),
1362 lwkt_gettoken(&so->so_rcv.ssb_token);
1364 tp->t_flags |= TF_RXRESIZED;
1365 if (!ssb_reserve(&so->so_rcv, newsize,
1367 atomic_clear_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE);
1370 (TCP_MAXWIN << tp->rcv_scale)) {
1371 atomic_clear_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE);
1374 m_adj(m, drop_hdrlen); /* delayed header drop */
1375 ssb_appendstream(&so->so_rcv, m);
1376 lwkt_reltoken(&so->so_rcv.ssb_token);
1380 * This code is responsible for most of the ACKs
1381 * the TCP stack sends back after receiving a data
1382 * packet. Note that the DELAY_ACK check fails if
1383 * the delack timer is already running, which results
1384 * in an ack being sent every other packet (which is
1387 * We then further aggregate acks by not actually
1388 * sending one until the protocol thread has completed
1389 * processing the current backlog of packets. This
1390 * does not delay the ack any further, but allows us
1391 * to take advantage of the packet aggregation that
1392 * high speed NICs do (usually blocks of 8-10 packets)
1393 * to send a single ack rather then four or five acks,
1394 * greatly reducing the ack rate, the return channel
1395 * bandwidth, and the protocol overhead on both ends.
1397 * Since this also has the effect of slowing down
1398 * the exponential slow-start ramp-up, systems with
1399 * very large bandwidth-delay products might want
1400 * to turn the feature off.
1402 if (DELAY_ACK(tp)) {
1403 tcp_callout_reset(tp, tp->tt_delack,
1404 tcp_delacktime, tcp_timer_delack);
1405 } else if (tcp_aggregate_acks) {
1406 tp->t_flags |= TF_ACKNOW;
1407 if (!(tp->t_flags & TF_ONOUTPUTQ)) {
1408 tp->t_flags |= TF_ONOUTPUTQ;
1409 tp->tt_cpu = mycpu->gd_cpuid;
1411 &tcpcbackq[tp->tt_cpu],
1415 tp->t_flags |= TF_ACKNOW;
1418 return(IPPROTO_DONE);
1423 * Calculate amount of space in receive window,
1424 * and then do TCP input processing.
1425 * Receive window is amount of space in rcv queue,
1426 * but not less than advertised window.
1428 recvwin = ssb_space(&so->so_rcv);
1431 tp->rcv_wnd = imax(recvwin, (int)(tp->rcv_adv - tp->rcv_nxt));
1433 /* Reset receive buffer auto scaling when not in bulk receive mode. */
1437 switch (tp->t_state) {
1439 * If the state is SYN_RECEIVED:
1440 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1442 case TCPS_SYN_RECEIVED:
1443 if ((thflags & TH_ACK) &&
1444 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1445 SEQ_GT(th->th_ack, tp->snd_max))) {
1446 rstreason = BANDLIM_RST_OPENPORT;
1452 * If the state is SYN_SENT:
1453 * if seg contains an ACK, but not for our SYN, drop the input.
1454 * if seg contains a RST, then drop the connection.
1455 * if seg does not contain SYN, then drop it.
1456 * Otherwise this is an acceptable SYN segment
1457 * initialize tp->rcv_nxt and tp->irs
1458 * if seg contains ack then advance tp->snd_una
1459 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1460 * arrange for segment to be acked (eventually)
1461 * continue processing rest of data/controls, beginning with URG
1464 if ((thflags & TH_ACK) &&
1465 (SEQ_LEQ(th->th_ack, tp->iss) ||
1466 SEQ_GT(th->th_ack, tp->snd_max))) {
1467 rstreason = BANDLIM_UNLIMITED;
1470 if (thflags & TH_RST) {
1471 if (thflags & TH_ACK)
1472 tp = tcp_drop(tp, ECONNREFUSED);
1475 if (!(thflags & TH_SYN))
1477 tp->snd_wnd = th->th_win; /* initial send window */
1479 tp->irs = th->th_seq;
1481 if (thflags & TH_ACK) {
1482 /* Our SYN was acked. */
1483 tcpstat.tcps_connects++;
1485 /* Do window scaling on this connection? */
1486 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
1487 (TF_RCVD_SCALE | TF_REQ_SCALE)) {
1488 tp->snd_scale = tp->requested_s_scale;
1489 tp->rcv_scale = tp->request_r_scale;
1491 tp->rcv_adv += tp->rcv_wnd;
1492 tp->snd_una++; /* SYN is acked */
1493 tcp_callout_stop(tp, tp->tt_rexmt);
1495 * If there's data, delay ACK; if there's also a FIN
1496 * ACKNOW will be turned on later.
1498 if (DELAY_ACK(tp) && tlen != 0) {
1499 tcp_callout_reset(tp, tp->tt_delack,
1500 tcp_delacktime, tcp_timer_delack);
1502 tp->t_flags |= TF_ACKNOW;
1505 * Received <SYN,ACK> in SYN_SENT[*] state.
1507 * SYN_SENT --> ESTABLISHED
1508 * SYN_SENT* --> FIN_WAIT_1
1510 tp->t_starttime = ticks;
1511 if (tp->t_flags & TF_NEEDFIN) {
1512 tp->t_state = TCPS_FIN_WAIT_1;
1513 tp->t_flags &= ~TF_NEEDFIN;
1516 tp->t_state = TCPS_ESTABLISHED;
1517 tcp_callout_reset(tp, tp->tt_keep,
1518 tcp_getkeepidle(tp),
1523 * Received initial SYN in SYN-SENT[*] state =>
1524 * simultaneous open.
1525 * Do 3-way handshake:
1526 * SYN-SENT -> SYN-RECEIVED
1527 * SYN-SENT* -> SYN-RECEIVED*
1529 tp->t_flags |= TF_ACKNOW;
1530 tcp_callout_stop(tp, tp->tt_rexmt);
1531 tp->t_state = TCPS_SYN_RECEIVED;
1536 * Advance th->th_seq to correspond to first data byte.
1537 * If data, trim to stay within window,
1538 * dropping FIN if necessary.
1541 if (tlen > tp->rcv_wnd) {
1542 todrop = tlen - tp->rcv_wnd;
1546 tcpstat.tcps_rcvpackafterwin++;
1547 tcpstat.tcps_rcvbyteafterwin += todrop;
1549 tp->snd_wl1 = th->th_seq - 1;
1550 tp->rcv_up = th->th_seq;
1552 * Client side of transaction: already sent SYN and data.
1553 * If the remote host used T/TCP to validate the SYN,
1554 * our data will be ACK'd; if so, enter normal data segment
1555 * processing in the middle of step 5, ack processing.
1556 * Otherwise, goto step 6.
1558 if (thflags & TH_ACK)
1564 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
1565 * do normal processing (we no longer bother with T/TCP).
1569 case TCPS_TIME_WAIT:
1570 break; /* continue normal processing */
1574 * States other than LISTEN or SYN_SENT.
1575 * First check the RST flag and sequence number since reset segments
1576 * are exempt from the timestamp and connection count tests. This
1577 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
1578 * below which allowed reset segments in half the sequence space
1579 * to fall though and be processed (which gives forged reset
1580 * segments with a random sequence number a 50 percent chance of
1581 * killing a connection).
1582 * Then check timestamp, if present.
1583 * Then check the connection count, if present.
1584 * Then check that at least some bytes of segment are within
1585 * receive window. If segment begins before rcv_nxt,
1586 * drop leading data (and SYN); if nothing left, just ack.
1589 * If the RST bit is set, check the sequence number to see
1590 * if this is a valid reset segment.
1592 * In all states except SYN-SENT, all reset (RST) segments
1593 * are validated by checking their SEQ-fields. A reset is
1594 * valid if its sequence number is in the window.
1595 * Note: this does not take into account delayed ACKs, so
1596 * we should test against last_ack_sent instead of rcv_nxt.
1597 * The sequence number in the reset segment is normally an
1598 * echo of our outgoing acknowledgement numbers, but some hosts
1599 * send a reset with the sequence number at the rightmost edge
1600 * of our receive window, and we have to handle this case.
1601 * If we have multiple segments in flight, the intial reset
1602 * segment sequence numbers will be to the left of last_ack_sent,
1603 * but they will eventually catch up.
1604 * In any case, it never made sense to trim reset segments to
1605 * fit the receive window since RFC 1122 says:
1606 * 4.2.2.12 RST Segment: RFC-793 Section 3.4
1608 * A TCP SHOULD allow a received RST segment to include data.
1611 * It has been suggested that a RST segment could contain
1612 * ASCII text that encoded and explained the cause of the
1613 * RST. No standard has yet been established for such
1616 * If the reset segment passes the sequence number test examine
1618 * SYN_RECEIVED STATE:
1619 * If passive open, return to LISTEN state.
1620 * If active open, inform user that connection was refused.
1621 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
1622 * Inform user that connection was reset, and close tcb.
1623 * CLOSING, LAST_ACK STATES:
1626 * Drop the segment - see Stevens, vol. 2, p. 964 and
1629 if (thflags & TH_RST) {
1630 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
1631 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
1632 switch (tp->t_state) {
1634 case TCPS_SYN_RECEIVED:
1635 so->so_error = ECONNREFUSED;
1638 case TCPS_ESTABLISHED:
1639 case TCPS_FIN_WAIT_1:
1640 case TCPS_FIN_WAIT_2:
1641 case TCPS_CLOSE_WAIT:
1642 so->so_error = ECONNRESET;
1644 tp->t_state = TCPS_CLOSED;
1645 tcpstat.tcps_drops++;
1654 case TCPS_TIME_WAIT:
1662 * RFC 1323 PAWS: If we have a timestamp reply on this segment
1663 * and it's less than ts_recent, drop it.
1665 if ((to.to_flags & TOF_TS) && tp->ts_recent != 0 &&
1666 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
1668 /* Check to see if ts_recent is over 24 days old. */
1669 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1671 * Invalidate ts_recent. If this segment updates
1672 * ts_recent, the age will be reset later and ts_recent
1673 * will get a valid value. If it does not, setting
1674 * ts_recent to zero will at least satisfy the
1675 * requirement that zero be placed in the timestamp
1676 * echo reply when ts_recent isn't valid. The
1677 * age isn't reset until we get a valid ts_recent
1678 * because we don't want out-of-order segments to be
1679 * dropped when ts_recent is old.
1683 tcpstat.tcps_rcvduppack++;
1684 tcpstat.tcps_rcvdupbyte += tlen;
1685 tcpstat.tcps_pawsdrop++;
1693 * In the SYN-RECEIVED state, validate that the packet belongs to
1694 * this connection before trimming the data to fit the receive
1695 * window. Check the sequence number versus IRS since we know
1696 * the sequence numbers haven't wrapped. This is a partial fix
1697 * for the "LAND" DoS attack.
1699 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
1700 rstreason = BANDLIM_RST_OPENPORT;
1704 todrop = tp->rcv_nxt - th->th_seq;
1706 if (TCP_DO_SACK(tp)) {
1707 /* Report duplicate segment at head of packet. */
1708 tp->reportblk.rblk_start = th->th_seq;
1709 tp->reportblk.rblk_end = th->th_seq + tlen;
1710 if (thflags & TH_FIN)
1711 ++tp->reportblk.rblk_end;
1712 if (SEQ_GT(tp->reportblk.rblk_end, tp->rcv_nxt))
1713 tp->reportblk.rblk_end = tp->rcv_nxt;
1714 tp->t_flags |= (TF_DUPSEG | TF_SACKLEFT | TF_ACKNOW);
1716 if (thflags & TH_SYN) {
1726 * Following if statement from Stevens, vol. 2, p. 960.
1728 if (todrop > tlen ||
1729 (todrop == tlen && !(thflags & TH_FIN))) {
1731 * Any valid FIN must be to the left of the window.
1732 * At this point the FIN must be a duplicate or out
1733 * of sequence; drop it.
1738 * Send an ACK to resynchronize and drop any data.
1739 * But keep on processing for RST or ACK.
1741 tp->t_flags |= TF_ACKNOW;
1743 tcpstat.tcps_rcvduppack++;
1744 tcpstat.tcps_rcvdupbyte += todrop;
1746 tcpstat.tcps_rcvpartduppack++;
1747 tcpstat.tcps_rcvpartdupbyte += todrop;
1749 drop_hdrlen += todrop; /* drop from the top afterwards */
1750 th->th_seq += todrop;
1752 if (th->th_urp > todrop)
1753 th->th_urp -= todrop;
1761 * If new data are received on a connection after the
1762 * user processes are gone, then RST the other end.
1764 if ((so->so_state & SS_NOFDREF) &&
1765 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1767 tcpstat.tcps_rcvafterclose++;
1768 rstreason = BANDLIM_UNLIMITED;
1773 * If segment ends after window, drop trailing data
1774 * (and PUSH and FIN); if nothing left, just ACK.
1776 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
1778 tcpstat.tcps_rcvpackafterwin++;
1779 if (todrop >= tlen) {
1780 tcpstat.tcps_rcvbyteafterwin += tlen;
1782 * If a new connection request is received
1783 * while in TIME_WAIT, drop the old connection
1784 * and start over if the sequence numbers
1785 * are above the previous ones.
1787 if (thflags & TH_SYN &&
1788 tp->t_state == TCPS_TIME_WAIT &&
1789 SEQ_GT(th->th_seq, tp->rcv_nxt)) {
1794 * If window is closed can only take segments at
1795 * window edge, and have to drop data and PUSH from
1796 * incoming segments. Continue processing, but
1797 * remember to ack. Otherwise, drop segment
1800 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1801 tp->t_flags |= TF_ACKNOW;
1802 tcpstat.tcps_rcvwinprobe++;
1806 tcpstat.tcps_rcvbyteafterwin += todrop;
1809 thflags &= ~(TH_PUSH | TH_FIN);
1813 * If last ACK falls within this segment's sequence numbers,
1814 * record its timestamp.
1816 * 1) That the test incorporates suggestions from the latest
1817 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1818 * 2) That updating only on newer timestamps interferes with
1819 * our earlier PAWS tests, so this check should be solely
1820 * predicated on the sequence space of this segment.
1821 * 3) That we modify the segment boundary check to be
1822 * Last.ACK.Sent <= SEG.SEQ + SEG.LEN
1823 * instead of RFC1323's
1824 * Last.ACK.Sent < SEG.SEQ + SEG.LEN,
1825 * This modified check allows us to overcome RFC1323's
1826 * limitations as described in Stevens TCP/IP Illustrated
1827 * Vol. 2 p.869. In such cases, we can still calculate the
1828 * RTT correctly when RCV.NXT == Last.ACK.Sent.
1830 if ((to.to_flags & TOF_TS) && SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
1831 SEQ_LEQ(tp->last_ack_sent, (th->th_seq + tlen
1832 + ((thflags & TH_SYN) != 0)
1833 + ((thflags & TH_FIN) != 0)))) {
1834 tp->ts_recent_age = ticks;
1835 tp->ts_recent = to.to_tsval;
1839 * If a SYN is in the window, then this is an
1840 * error and we send an RST and drop the connection.
1842 if (thflags & TH_SYN) {
1843 tp = tcp_drop(tp, ECONNRESET);
1844 rstreason = BANDLIM_UNLIMITED;
1849 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
1850 * flag is on (half-synchronized state), then queue data for
1851 * later processing; else drop segment and return.
1853 if (!(thflags & TH_ACK)) {
1854 if (tp->t_state == TCPS_SYN_RECEIVED ||
1855 (tp->t_flags & TF_NEEDSYN))
1864 switch (tp->t_state) {
1866 * In SYN_RECEIVED state, the ACK acknowledges our SYN, so enter
1867 * ESTABLISHED state and continue processing.
1868 * The ACK was checked above.
1870 case TCPS_SYN_RECEIVED:
1872 tcpstat.tcps_connects++;
1874 /* Do window scaling? */
1875 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
1876 (TF_RCVD_SCALE | TF_REQ_SCALE)) {
1877 tp->snd_scale = tp->requested_s_scale;
1878 tp->rcv_scale = tp->request_r_scale;
1882 * SYN-RECEIVED -> ESTABLISHED
1883 * SYN-RECEIVED* -> FIN-WAIT-1
1885 tp->t_starttime = ticks;
1886 if (tp->t_flags & TF_NEEDFIN) {
1887 tp->t_state = TCPS_FIN_WAIT_1;
1888 tp->t_flags &= ~TF_NEEDFIN;
1890 tp->t_state = TCPS_ESTABLISHED;
1891 tcp_callout_reset(tp, tp->tt_keep,
1892 tcp_getkeepidle(tp),
1896 * If segment contains data or ACK, will call tcp_reass()
1897 * later; if not, do so now to pass queued data to user.
1899 if (tlen == 0 && !(thflags & TH_FIN))
1900 tcp_reass(tp, NULL, NULL, NULL);
1904 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1905 * ACKs. If the ack is in the range
1906 * tp->snd_una < th->th_ack <= tp->snd_max
1907 * then advance tp->snd_una to th->th_ack and drop
1908 * data from the retransmission queue. If this ACK reflects
1909 * more up to date window information we update our window information.
1911 case TCPS_ESTABLISHED:
1912 case TCPS_FIN_WAIT_1:
1913 case TCPS_FIN_WAIT_2:
1914 case TCPS_CLOSE_WAIT:
1917 case TCPS_TIME_WAIT:
1919 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
1920 if (TCP_DO_SACK(tp))
1921 tcp_sack_update_scoreboard(tp, &to);
1922 if (tlen != 0 || tiwin != tp->snd_wnd) {
1926 tcpstat.tcps_rcvdupack++;
1927 if (!tcp_callout_active(tp, tp->tt_rexmt) ||
1928 th->th_ack != tp->snd_una) {
1933 * We have outstanding data (other than
1934 * a window probe), this is a completely
1935 * duplicate ack (ie, window info didn't
1936 * change), the ack is the biggest we've
1937 * seen and we've seen exactly our rexmt
1938 * threshhold of them, so assume a packet
1939 * has been dropped and retransmit it.
1940 * Kludge snd_nxt & the congestion
1941 * window so we send only this one
1944 if (IN_FASTRECOVERY(tp)) {
1945 if (TCP_DO_SACK(tp)) {
1946 /* No artifical cwnd inflation. */
1947 tcp_sack_rexmt(tp, th);
1950 * Dup acks mean that packets
1951 * have left the network
1952 * (they're now cached at the
1953 * receiver) so bump cwnd by
1954 * the amount in the receiver
1955 * to keep a constant cwnd
1956 * packets in the network.
1958 tp->snd_cwnd += tp->t_maxseg;
1961 } else if (SEQ_LT(th->th_ack, tp->snd_recover)) {
1964 } else if (++tp->t_dupacks == tcprexmtthresh) {
1965 tcp_seq old_snd_nxt;
1969 if (tcp_do_eifel_detect &&
1970 (tp->t_flags & TF_RCVD_TSTMP)) {
1971 tcp_save_congestion_state(tp);
1972 tp->t_flags |= TF_FASTREXMT;
1975 * We know we're losing at the current
1976 * window size, so do congestion avoidance:
1977 * set ssthresh to half the current window
1978 * and pull our congestion window back to the
1981 win = min(tp->snd_wnd, tp->snd_cwnd) / 2 /
1985 tp->snd_ssthresh = win * tp->t_maxseg;
1986 ENTER_FASTRECOVERY(tp);
1987 tp->snd_recover = tp->snd_max;
1988 tcp_callout_stop(tp, tp->tt_rexmt);
1990 old_snd_nxt = tp->snd_nxt;
1991 tp->snd_nxt = th->th_ack;
1992 tp->snd_cwnd = tp->t_maxseg;
1994 ++tcpstat.tcps_sndfastrexmit;
1995 tp->snd_cwnd = tp->snd_ssthresh;
1996 tp->rexmt_high = tp->snd_nxt;
1997 if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
1998 tp->snd_nxt = old_snd_nxt;
1999 KASSERT(tp->snd_limited <= 2,
2000 ("tp->snd_limited too big"));
2001 if (TCP_DO_SACK(tp))
2002 tcp_sack_rexmt(tp, th);
2004 tp->snd_cwnd += tp->t_maxseg *
2005 (tp->t_dupacks - tp->snd_limited);
2006 } else if (tcp_do_limitedtransmit) {
2007 u_long oldcwnd = tp->snd_cwnd;
2008 tcp_seq oldsndmax = tp->snd_max;
2009 tcp_seq oldsndnxt = tp->snd_nxt;
2010 /* outstanding data */
2011 uint32_t ownd = tp->snd_max - tp->snd_una;
2014 #define iceildiv(n, d) (((n)+(d)-1) / (d))
2016 KASSERT(tp->t_dupacks == 1 ||
2018 ("dupacks not 1 or 2"));
2019 if (tp->t_dupacks == 1)
2020 tp->snd_limited = 0;
2021 tp->snd_nxt = tp->snd_max;
2022 tp->snd_cwnd = ownd +
2023 (tp->t_dupacks - tp->snd_limited) *
2028 * Other acks may have been processed,
2029 * snd_nxt cannot be reset to a value less
2032 if (SEQ_LT(oldsndnxt, oldsndmax)) {
2033 if (SEQ_GT(oldsndnxt, tp->snd_una))
2034 tp->snd_nxt = oldsndnxt;
2036 tp->snd_nxt = tp->snd_una;
2038 tp->snd_cwnd = oldcwnd;
2039 sent = tp->snd_max - oldsndmax;
2040 if (sent > tp->t_maxseg) {
2041 KASSERT((tp->t_dupacks == 2 &&
2042 tp->snd_limited == 0) ||
2043 (sent == tp->t_maxseg + 1 &&
2044 tp->t_flags & TF_SENTFIN),
2046 KASSERT(sent <= tp->t_maxseg * 2,
2047 ("sent too many segments"));
2048 tp->snd_limited = 2;
2049 tcpstat.tcps_sndlimited += 2;
2050 } else if (sent > 0) {
2052 ++tcpstat.tcps_sndlimited;
2053 } else if (tcp_do_early_retransmit &&
2054 (tcp_do_eifel_detect &&
2055 (tp->t_flags & TF_RCVD_TSTMP)) &&
2056 ownd < 4 * tp->t_maxseg &&
2057 tp->t_dupacks + 1 >=
2058 iceildiv(ownd, tp->t_maxseg) &&
2059 (!TCP_DO_SACK(tp) ||
2060 ownd <= tp->t_maxseg ||
2061 tcp_sack_has_sacked(&tp->scb,
2062 ownd - tp->t_maxseg))) {
2063 ++tcpstat.tcps_sndearlyrexmit;
2064 tp->t_flags |= TF_EARLYREXMT;
2065 goto fastretransmit;
2071 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una"));
2073 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2075 * Detected optimistic ACK attack.
2076 * Force slow-start to de-synchronize attack.
2078 tp->snd_cwnd = tp->t_maxseg;
2081 tcpstat.tcps_rcvacktoomuch++;
2085 * If we reach this point, ACK is not a duplicate,
2086 * i.e., it ACKs something we sent.
2088 if (tp->t_flags & TF_NEEDSYN) {
2090 * T/TCP: Connection was half-synchronized, and our
2091 * SYN has been ACK'd (so connection is now fully
2092 * synchronized). Go to non-starred state,
2093 * increment snd_una for ACK of SYN, and check if
2094 * we can do window scaling.
2096 tp->t_flags &= ~TF_NEEDSYN;
2098 /* Do window scaling? */
2099 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
2100 (TF_RCVD_SCALE | TF_REQ_SCALE)) {
2101 tp->snd_scale = tp->requested_s_scale;
2102 tp->rcv_scale = tp->request_r_scale;
2107 acked = th->th_ack - tp->snd_una;
2108 tcpstat.tcps_rcvackpack++;
2109 tcpstat.tcps_rcvackbyte += acked;
2111 if (tcp_do_eifel_detect && acked > 0 &&
2112 (to.to_flags & TOF_TS) && (to.to_tsecr != 0) &&
2113 (tp->t_flags & TF_FIRSTACCACK)) {
2114 /* Eifel detection applicable. */
2115 if (to.to_tsecr < tp->t_rexmtTS) {
2116 ++tcpstat.tcps_eifeldetected;
2117 tcp_revert_congestion_state(tp);
2118 if (tp->t_rxtshift == 1 &&
2119 ticks >= tp->t_badrxtwin)
2120 ++tcpstat.tcps_rttcantdetect;
2122 } else if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) {
2124 * If we just performed our first retransmit,
2125 * and the ACK arrives within our recovery window,
2126 * then it was a mistake to do the retransmit
2127 * in the first place. Recover our original cwnd
2128 * and ssthresh, and proceed to transmit where we
2131 tcp_revert_congestion_state(tp);
2132 ++tcpstat.tcps_rttdetected;
2136 * If we have a timestamp reply, update smoothed
2137 * round trip time. If no timestamp is present but
2138 * transmit timer is running and timed sequence
2139 * number was acked, update smoothed round trip time.
2140 * Since we now have an rtt measurement, cancel the
2141 * timer backoff (cf., Phil Karn's retransmit alg.).
2142 * Recompute the initial retransmit timer.
2144 * Some machines (certain windows boxes) send broken
2145 * timestamp replies during the SYN+ACK phase, ignore
2148 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0))
2149 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1);
2150 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq))
2151 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2152 tcp_xmit_bandwidth_limit(tp, th->th_ack);
2155 * If no data (only SYN) was ACK'd,
2156 * skip rest of ACK processing.
2161 /* Stop looking for an acceptable ACK since one was received. */
2162 tp->t_flags &= ~(TF_FIRSTACCACK | TF_FASTREXMT | TF_EARLYREXMT);
2164 if (acked > so->so_snd.ssb_cc) {
2165 tp->snd_wnd -= so->so_snd.ssb_cc;
2166 sbdrop(&so->so_snd.sb, (int)so->so_snd.ssb_cc);
2167 ourfinisacked = TRUE;
2169 sbdrop(&so->so_snd.sb, acked);
2170 tp->snd_wnd -= acked;
2171 ourfinisacked = FALSE;
2176 * Update window information.
2177 * Don't look at window if no ACK:
2178 * TAC's send garbage on first SYN.
2180 if (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2181 (tp->snd_wl1 == th->th_seq &&
2182 (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2183 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)))) {
2184 /* keep track of pure window updates */
2185 if (tlen == 0 && tp->snd_wl2 == th->th_ack &&
2186 tiwin > tp->snd_wnd)
2187 tcpstat.tcps_rcvwinupd++;
2188 tp->snd_wnd = tiwin;
2189 tp->snd_wl1 = th->th_seq;
2190 tp->snd_wl2 = th->th_ack;
2191 if (tp->snd_wnd > tp->max_sndwnd)
2192 tp->max_sndwnd = tp->snd_wnd;
2196 tp->snd_una = th->th_ack;
2197 if (TCP_DO_SACK(tp))
2198 tcp_sack_update_scoreboard(tp, &to);
2199 if (IN_FASTRECOVERY(tp)) {
2200 if (SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2201 EXIT_FASTRECOVERY(tp);
2204 * If the congestion window was inflated
2205 * to account for the other side's
2206 * cached packets, retract it.
2208 if (!TCP_DO_SACK(tp))
2209 tp->snd_cwnd = tp->snd_ssthresh;
2212 * Window inflation should have left us
2213 * with approximately snd_ssthresh outstanding
2214 * data. But, in case we would be inclined
2215 * to send a burst, better do it using
2218 if (SEQ_GT(th->th_ack + tp->snd_cwnd,
2219 tp->snd_max + 2 * tp->t_maxseg))
2221 (tp->snd_max - tp->snd_una) +
2226 if (TCP_DO_SACK(tp)) {
2227 tp->snd_max_rexmt = tp->snd_max;
2228 tcp_sack_rexmt(tp, th);
2230 tcp_newreno_partial_ack(tp, th, acked);
2236 * Open the congestion window. When in slow-start,
2237 * open exponentially: maxseg per packet. Otherwise,
2238 * open linearly: maxseg per window.
2240 if (tp->snd_cwnd <= tp->snd_ssthresh) {
2242 (SEQ_LT(tp->snd_nxt, tp->snd_max) ?
2243 tp->t_maxseg : 2 * tp->t_maxseg);
2246 tp->snd_cwnd += tcp_do_abc ?
2247 min(acked, abc_sslimit) : tp->t_maxseg;
2249 /* linear increase */
2250 tp->snd_wacked += tcp_do_abc ? acked :
2252 if (tp->snd_wacked >= tp->snd_cwnd) {
2253 tp->snd_wacked -= tp->snd_cwnd;
2254 tp->snd_cwnd += tp->t_maxseg;
2257 tp->snd_cwnd = min(tp->snd_cwnd,
2258 TCP_MAXWIN << tp->snd_scale);
2259 tp->snd_recover = th->th_ack - 1;
2261 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2262 tp->snd_nxt = tp->snd_una;
2265 * If all outstanding data is acked, stop retransmit
2266 * timer and remember to restart (more output or persist).
2267 * If there is more data to be acked, restart retransmit
2268 * timer, using current (possibly backed-off) value.
2270 if (th->th_ack == tp->snd_max) {
2271 tcp_callout_stop(tp, tp->tt_rexmt);
2273 } else if (!tcp_callout_active(tp, tp->tt_persist)) {
2274 tcp_callout_reset(tp, tp->tt_rexmt, tp->t_rxtcur,
2278 switch (tp->t_state) {
2280 * In FIN_WAIT_1 STATE in addition to the processing
2281 * for the ESTABLISHED state if our FIN is now acknowledged
2282 * then enter FIN_WAIT_2.
2284 case TCPS_FIN_WAIT_1:
2285 if (ourfinisacked) {
2287 * If we can't receive any more
2288 * data, then closing user can proceed.
2289 * Starting the timer is contrary to the
2290 * specification, but if we don't get a FIN
2291 * we'll hang forever.
2293 if (so->so_state & SS_CANTRCVMORE) {
2294 soisdisconnected(so);
2295 tcp_callout_reset(tp, tp->tt_2msl,
2296 tp->t_maxidle, tcp_timer_2msl);
2298 tp->t_state = TCPS_FIN_WAIT_2;
2303 * In CLOSING STATE in addition to the processing for
2304 * the ESTABLISHED state if the ACK acknowledges our FIN
2305 * then enter the TIME-WAIT state, otherwise ignore
2309 if (ourfinisacked) {
2310 tp->t_state = TCPS_TIME_WAIT;
2311 tcp_canceltimers(tp);
2312 tcp_callout_reset(tp, tp->tt_2msl,
2313 2 * tcp_rmx_msl(tp),
2315 soisdisconnected(so);
2320 * In LAST_ACK, we may still be waiting for data to drain
2321 * and/or to be acked, as well as for the ack of our FIN.
2322 * If our FIN is now acknowledged, delete the TCB,
2323 * enter the closed state and return.
2326 if (ourfinisacked) {
2333 * In TIME_WAIT state the only thing that should arrive
2334 * is a retransmission of the remote FIN. Acknowledge
2335 * it and restart the finack timer.
2337 case TCPS_TIME_WAIT:
2338 tcp_callout_reset(tp, tp->tt_2msl, 2 * tcp_rmx_msl(tp),
2346 * Update window information.
2347 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2349 if ((thflags & TH_ACK) &&
2350 acceptable_window_update(tp, th, tiwin)) {
2351 /* keep track of pure window updates */
2352 if (tlen == 0 && tp->snd_wl2 == th->th_ack &&
2353 tiwin > tp->snd_wnd)
2354 tcpstat.tcps_rcvwinupd++;
2355 tp->snd_wnd = tiwin;
2356 tp->snd_wl1 = th->th_seq;
2357 tp->snd_wl2 = th->th_ack;
2358 if (tp->snd_wnd > tp->max_sndwnd)
2359 tp->max_sndwnd = tp->snd_wnd;
2364 * Process segments with URG.
2366 if ((thflags & TH_URG) && th->th_urp &&
2367 !TCPS_HAVERCVDFIN(tp->t_state)) {
2369 * This is a kludge, but if we receive and accept
2370 * random urgent pointers, we'll crash in
2371 * soreceive. It's hard to imagine someone
2372 * actually wanting to send this much urgent data.
2374 if (th->th_urp + so->so_rcv.ssb_cc > sb_max) {
2375 th->th_urp = 0; /* XXX */
2376 thflags &= ~TH_URG; /* XXX */
2377 goto dodata; /* XXX */
2380 * If this segment advances the known urgent pointer,
2381 * then mark the data stream. This should not happen
2382 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2383 * a FIN has been received from the remote side.
2384 * In these states we ignore the URG.
2386 * According to RFC961 (Assigned Protocols),
2387 * the urgent pointer points to the last octet
2388 * of urgent data. We continue, however,
2389 * to consider it to indicate the first octet
2390 * of data past the urgent section as the original
2391 * spec states (in one of two places).
2393 if (SEQ_GT(th->th_seq + th->th_urp, tp->rcv_up)) {
2394 tp->rcv_up = th->th_seq + th->th_urp;
2395 so->so_oobmark = so->so_rcv.ssb_cc +
2396 (tp->rcv_up - tp->rcv_nxt) - 1;
2397 if (so->so_oobmark == 0)
2398 sosetstate(so, SS_RCVATMARK);
2400 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2403 * Remove out of band data so doesn't get presented to user.
2404 * This can happen independent of advancing the URG pointer,
2405 * but if two URG's are pending at once, some out-of-band
2406 * data may creep in... ick.
2408 if (th->th_urp <= (u_long)tlen &&
2409 !(so->so_options & SO_OOBINLINE)) {
2410 /* hdr drop is delayed */
2411 tcp_pulloutofband(so, th, m, drop_hdrlen);
2415 * If no out of band data is expected,
2416 * pull receive urgent pointer along
2417 * with the receive window.
2419 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2420 tp->rcv_up = tp->rcv_nxt;
2425 * Process the segment text, merging it into the TCP sequencing queue,
2426 * and arranging for acknowledgment of receipt if necessary.
2427 * This process logically involves adjusting tp->rcv_wnd as data
2428 * is presented to the user (this happens in tcp_usrreq.c,
2429 * case PRU_RCVD). If a FIN has already been received on this
2430 * connection then we just ignore the text.
2432 if ((tlen || (thflags & TH_FIN)) && !TCPS_HAVERCVDFIN(tp->t_state)) {
2433 m_adj(m, drop_hdrlen); /* delayed header drop */
2435 * Insert segment which includes th into TCP reassembly queue
2436 * with control block tp. Set thflags to whether reassembly now
2437 * includes a segment with FIN. This handles the common case
2438 * inline (segment is the next to be received on an established
2439 * connection, and the queue is empty), avoiding linkage into
2440 * and removal from the queue and repetition of various
2442 * Set DELACK for segments received in order, but ack
2443 * immediately when segments are out of order (so
2444 * fast retransmit can work).
2446 if (th->th_seq == tp->rcv_nxt &&
2447 LIST_EMPTY(&tp->t_segq) &&
2448 TCPS_HAVEESTABLISHED(tp->t_state)) {
2449 if (DELAY_ACK(tp)) {
2450 tcp_callout_reset(tp, tp->tt_delack,
2451 tcp_delacktime, tcp_timer_delack);
2453 tp->t_flags |= TF_ACKNOW;
2455 tp->rcv_nxt += tlen;
2456 thflags = th->th_flags & TH_FIN;
2457 tcpstat.tcps_rcvpack++;
2458 tcpstat.tcps_rcvbyte += tlen;
2460 if (so->so_state & SS_CANTRCVMORE) {
2463 lwkt_gettoken(&so->so_rcv.ssb_token);
2464 ssb_appendstream(&so->so_rcv, m);
2465 lwkt_reltoken(&so->so_rcv.ssb_token);
2469 if (!(tp->t_flags & TF_DUPSEG)) {
2470 /* Initialize SACK report block. */
2471 tp->reportblk.rblk_start = th->th_seq;
2472 tp->reportblk.rblk_end = th->th_seq + tlen +
2473 ((thflags & TH_FIN) != 0);
2475 thflags = tcp_reass(tp, th, &tlen, m);
2476 tp->t_flags |= TF_ACKNOW;
2480 * Note the amount of data that peer has sent into
2481 * our window, in order to estimate the sender's
2484 len = so->so_rcv.ssb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2491 * If FIN is received ACK the FIN and let the user know
2492 * that the connection is closing.
2494 if (thflags & TH_FIN) {
2495 if (!TCPS_HAVERCVDFIN(tp->t_state)) {
2498 * If connection is half-synchronized
2499 * (ie NEEDSYN flag on) then delay ACK,
2500 * so it may be piggybacked when SYN is sent.
2501 * Otherwise, since we received a FIN then no
2502 * more input can be expected, send ACK now.
2504 if (DELAY_ACK(tp) && (tp->t_flags & TF_NEEDSYN)) {
2505 tcp_callout_reset(tp, tp->tt_delack,
2506 tcp_delacktime, tcp_timer_delack);
2508 tp->t_flags |= TF_ACKNOW;
2513 switch (tp->t_state) {
2515 * In SYN_RECEIVED and ESTABLISHED STATES
2516 * enter the CLOSE_WAIT state.
2518 case TCPS_SYN_RECEIVED:
2519 tp->t_starttime = ticks;
2521 case TCPS_ESTABLISHED:
2522 tp->t_state = TCPS_CLOSE_WAIT;
2526 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2527 * enter the CLOSING state.
2529 case TCPS_FIN_WAIT_1:
2530 tp->t_state = TCPS_CLOSING;
2534 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2535 * starting the time-wait timer, turning off the other
2538 case TCPS_FIN_WAIT_2:
2539 tp->t_state = TCPS_TIME_WAIT;
2540 tcp_canceltimers(tp);
2541 tcp_callout_reset(tp, tp->tt_2msl, 2 * tcp_rmx_msl(tp),
2543 soisdisconnected(so);
2547 * In TIME_WAIT state restart the 2 MSL time_wait timer.
2549 case TCPS_TIME_WAIT:
2550 tcp_callout_reset(tp, tp->tt_2msl, 2 * tcp_rmx_msl(tp),
2557 if (so->so_options & SO_DEBUG)
2558 tcp_trace(TA_INPUT, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2562 * Return any desired output.
2564 if (needoutput || (tp->t_flags & TF_ACKNOW))
2566 return(IPPROTO_DONE);
2570 * Generate an ACK dropping incoming segment if it occupies
2571 * sequence space, where the ACK reflects our state.
2573 * We can now skip the test for the RST flag since all
2574 * paths to this code happen after packets containing
2575 * RST have been dropped.
2577 * In the SYN-RECEIVED state, don't send an ACK unless the
2578 * segment we received passes the SYN-RECEIVED ACK test.
2579 * If it fails send a RST. This breaks the loop in the
2580 * "LAND" DoS attack, and also prevents an ACK storm
2581 * between two listening ports that have been sent forged
2582 * SYN segments, each with the source address of the other.
2584 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
2585 (SEQ_GT(tp->snd_una, th->th_ack) ||
2586 SEQ_GT(th->th_ack, tp->snd_max)) ) {
2587 rstreason = BANDLIM_RST_OPENPORT;
2591 if (so->so_options & SO_DEBUG)
2592 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2595 tp->t_flags |= TF_ACKNOW;
2597 return(IPPROTO_DONE);
2601 * Generate a RST, dropping incoming segment.
2602 * Make ACK acceptable to originator of segment.
2603 * Don't bother to respond if destination was broadcast/multicast.
2605 if ((thflags & TH_RST) || m->m_flags & (M_BCAST | M_MCAST))
2608 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
2609 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
2612 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
2613 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
2614 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
2615 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2618 /* IPv6 anycast check is done at tcp6_input() */
2621 * Perform bandwidth limiting.
2624 if (badport_bandlim(rstreason) < 0)
2629 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2630 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2632 if (thflags & TH_ACK)
2633 /* mtod() below is safe as long as hdr dropping is delayed */
2634 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack,
2637 if (thflags & TH_SYN)
2639 /* mtod() below is safe as long as hdr dropping is delayed */
2640 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq + tlen,
2641 (tcp_seq)0, TH_RST | TH_ACK);
2643 return(IPPROTO_DONE);
2647 * Drop space held by incoming segment and return.
2650 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2651 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2654 return(IPPROTO_DONE);
2658 * Parse TCP options and place in tcpopt.
2661 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, boolean_t is_syn)
2666 for (; cnt > 0; cnt -= optlen, cp += optlen) {
2668 if (opt == TCPOPT_EOL)
2670 if (opt == TCPOPT_NOP)
2676 if (optlen < 2 || optlen > cnt)
2681 if (optlen != TCPOLEN_MAXSEG)
2685 to->to_flags |= TOF_MSS;
2686 bcopy(cp + 2, &to->to_mss, sizeof to->to_mss);
2687 to->to_mss = ntohs(to->to_mss);
2690 if (optlen != TCPOLEN_WINDOW)
2694 to->to_flags |= TOF_SCALE;
2695 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
2697 case TCPOPT_TIMESTAMP:
2698 if (optlen != TCPOLEN_TIMESTAMP)
2700 to->to_flags |= TOF_TS;
2701 bcopy(cp + 2, &to->to_tsval, sizeof to->to_tsval);
2702 to->to_tsval = ntohl(to->to_tsval);
2703 bcopy(cp + 6, &to->to_tsecr, sizeof to->to_tsecr);
2704 to->to_tsecr = ntohl(to->to_tsecr);
2706 * If echoed timestamp is later than the current time,
2707 * fall back to non RFC1323 RTT calculation.
2709 if (to->to_tsecr != 0 && TSTMP_GT(to->to_tsecr, ticks))
2712 case TCPOPT_SACK_PERMITTED:
2713 if (optlen != TCPOLEN_SACK_PERMITTED)
2717 to->to_flags |= TOF_SACK_PERMITTED;
2720 if ((optlen - 2) & 0x07) /* not multiple of 8 */
2722 to->to_nsackblocks = (optlen - 2) / 8;
2723 to->to_sackblocks = (struct raw_sackblock *) (cp + 2);
2724 to->to_flags |= TOF_SACK;
2725 for (i = 0; i < to->to_nsackblocks; i++) {
2726 struct raw_sackblock *r = &to->to_sackblocks[i];
2728 r->rblk_start = ntohl(r->rblk_start);
2729 r->rblk_end = ntohl(r->rblk_end);
2732 #ifdef TCP_SIGNATURE
2734 * XXX In order to reply to a host which has set the
2735 * TCP_SIGNATURE option in its initial SYN, we have to
2736 * record the fact that the option was observed here
2737 * for the syncache code to perform the correct response.
2739 case TCPOPT_SIGNATURE:
2740 if (optlen != TCPOLEN_SIGNATURE)
2742 to->to_flags |= (TOF_SIGNATURE | TOF_SIGLEN);
2744 #endif /* TCP_SIGNATURE */
2752 * Pull out of band byte out of a segment so
2753 * it doesn't appear in the user's data queue.
2754 * It is still reflected in the segment length for
2755 * sequencing purposes.
2756 * "off" is the delayed to be dropped hdrlen.
2759 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, int off)
2761 int cnt = off + th->th_urp - 1;
2764 if (m->m_len > cnt) {
2765 char *cp = mtod(m, caddr_t) + cnt;
2766 struct tcpcb *tp = sototcpcb(so);
2769 tp->t_oobflags |= TCPOOB_HAVEDATA;
2770 bcopy(cp + 1, cp, m->m_len - cnt - 1);
2772 if (m->m_flags & M_PKTHDR)
2781 panic("tcp_pulloutofband");
2785 * Collect new round-trip time estimate
2786 * and update averages and current timeout.
2789 tcp_xmit_timer(struct tcpcb *tp, int rtt)
2793 tcpstat.tcps_rttupdated++;
2795 if (tp->t_srtt != 0) {
2797 * srtt is stored as fixed point with 5 bits after the
2798 * binary point (i.e., scaled by 8). The following magic
2799 * is equivalent to the smoothing algorithm in rfc793 with
2800 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2801 * point). Adjust rtt to origin 0.
2803 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
2804 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
2806 if ((tp->t_srtt += delta) <= 0)
2810 * We accumulate a smoothed rtt variance (actually, a
2811 * smoothed mean difference), then set the retransmit
2812 * timer to smoothed rtt + 4 times the smoothed variance.
2813 * rttvar is stored as fixed point with 4 bits after the
2814 * binary point (scaled by 16). The following is
2815 * equivalent to rfc793 smoothing with an alpha of .75
2816 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
2817 * rfc793's wired-in beta.
2821 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
2822 if ((tp->t_rttvar += delta) <= 0)
2824 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
2825 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2828 * No rtt measurement yet - use the unsmoothed rtt.
2829 * Set the variance to half the rtt (so our first
2830 * retransmit happens at 3*rtt).
2832 tp->t_srtt = rtt << TCP_RTT_SHIFT;
2833 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
2834 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2840 * the retransmit should happen at rtt + 4 * rttvar.
2841 * Because of the way we do the smoothing, srtt and rttvar
2842 * will each average +1/2 tick of bias. When we compute
2843 * the retransmit timer, we want 1/2 tick of rounding and
2844 * 1 extra tick because of +-1/2 tick uncertainty in the
2845 * firing of the timer. The bias will give us exactly the
2846 * 1.5 tick we need. But, because the bias is
2847 * statistical, we have to test that we don't drop below
2848 * the minimum feasible timer (which is 2 ticks).
2850 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
2851 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
2854 * We received an ack for a packet that wasn't retransmitted;
2855 * it is probably safe to discard any error indications we've
2856 * received recently. This isn't quite right, but close enough
2857 * for now (a route might have failed after we sent a segment,
2858 * and the return path might not be symmetrical).
2860 tp->t_softerror = 0;
2864 * Determine a reasonable value for maxseg size.
2865 * If the route is known, check route for mtu.
2866 * If none, use an mss that can be handled on the outgoing
2867 * interface without forcing IP to fragment; if bigger than
2868 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2869 * to utilize large mbufs. If no route is found, route has no mtu,
2870 * or the destination isn't local, use a default, hopefully conservative
2871 * size (usually 512 or the default IP max size, but no more than the mtu
2872 * of the interface), as we can't discover anything about intervening
2873 * gateways or networks. We also initialize the congestion/slow start
2874 * window to be a single segment if the destination isn't local.
2875 * While looking at the routing entry, we also initialize other path-dependent
2876 * parameters from pre-set or cached values in the routing entry.
2878 * Also take into account the space needed for options that we
2879 * send regularly. Make maxseg shorter by that amount to assure
2880 * that we can send maxseg amount of data even when the options
2881 * are present. Store the upper limit of the length of options plus
2884 * NOTE that this routine is only called when we process an incoming
2885 * segment, for outgoing segments only tcp_mssopt is called.
2888 tcp_mss(struct tcpcb *tp, int offer)
2894 struct inpcb *inp = tp->t_inpcb;
2897 boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) ? TRUE : FALSE);
2898 size_t min_protoh = isipv6 ?
2899 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
2900 sizeof(struct tcpiphdr);
2902 const boolean_t isipv6 = FALSE;
2903 const size_t min_protoh = sizeof(struct tcpiphdr);
2907 rt = tcp_rtlookup6(&inp->inp_inc);
2909 rt = tcp_rtlookup(&inp->inp_inc);
2911 tp->t_maxopd = tp->t_maxseg =
2912 (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
2916 so = inp->inp_socket;
2919 * Offer == 0 means that there was no MSS on the SYN segment,
2920 * in this case we use either the interface mtu or tcp_mssdflt.
2922 * An offer which is too large will be cut down later.
2926 if (in6_localaddr(&inp->in6p_faddr)) {
2927 offer = ND_IFINFO(rt->rt_ifp)->linkmtu -
2930 offer = tcp_v6mssdflt;
2933 if (in_localaddr(inp->inp_faddr))
2934 offer = ifp->if_mtu - min_protoh;
2936 offer = tcp_mssdflt;
2941 * Prevent DoS attack with too small MSS. Round up
2942 * to at least minmss.
2944 * Sanity check: make sure that maxopd will be large
2945 * enough to allow some data on segments even is the
2946 * all the option space is used (40bytes). Otherwise
2947 * funny things may happen in tcp_output.
2949 offer = max(offer, tcp_minmss);
2950 offer = max(offer, 64);
2952 rt->rt_rmx.rmx_mssopt = offer;
2955 * While we're here, check if there's an initial rtt
2956 * or rttvar. Convert from the route-table units
2957 * to scaled multiples of the slow timeout timer.
2959 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
2961 * XXX the lock bit for RTT indicates that the value
2962 * is also a minimum value; this is subject to time.
2964 if (rt->rt_rmx.rmx_locks & RTV_RTT)
2965 tp->t_rttmin = rtt / (RTM_RTTUNIT / hz);
2966 tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE));
2967 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
2968 tcpstat.tcps_usedrtt++;
2969 if (rt->rt_rmx.rmx_rttvar) {
2970 tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
2971 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE));
2972 tcpstat.tcps_usedrttvar++;
2974 /* default variation is +- 1 rtt */
2976 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
2978 TCPT_RANGESET(tp->t_rxtcur,
2979 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
2980 tp->t_rttmin, TCPTV_REXMTMAX);
2984 * if there's an mtu associated with the route, use it
2985 * else, use the link mtu. Take the smaller of mss or offer
2988 if (rt->rt_rmx.rmx_mtu) {
2989 mss = rt->rt_rmx.rmx_mtu - min_protoh;
2992 mss = ND_IFINFO(rt->rt_ifp)->linkmtu - min_protoh;
2994 mss = ifp->if_mtu - min_protoh;
2996 mss = min(mss, offer);
2999 * maxopd stores the maximum length of data AND options
3000 * in a segment; maxseg is the amount of data in a normal
3001 * segment. We need to store this value (maxopd) apart
3002 * from maxseg, because now every segment carries options
3003 * and thus we normally have somewhat less data in segments.
3007 if ((tp->t_flags & (TF_REQ_TSTMP | TF_NOOPT)) == TF_REQ_TSTMP &&
3008 ((tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
3009 mss -= TCPOLEN_TSTAMP_APPA;
3011 #if (MCLBYTES & (MCLBYTES - 1)) == 0
3013 mss &= ~(MCLBYTES-1);
3016 mss = mss / MCLBYTES * MCLBYTES;
3019 * If there's a pipesize, change the socket buffer
3020 * to that size. Make the socket buffers an integral
3021 * number of mss units; if the mss is larger than
3022 * the socket buffer, decrease the mss.
3025 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0)
3027 bufsize = so->so_snd.ssb_hiwat;
3031 bufsize = roundup(bufsize, mss);
3032 if (bufsize > sb_max)
3034 if (bufsize > so->so_snd.ssb_hiwat)
3035 ssb_reserve(&so->so_snd, bufsize, so, NULL);
3040 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0)
3042 bufsize = so->so_rcv.ssb_hiwat;
3043 if (bufsize > mss) {
3044 bufsize = roundup(bufsize, mss);
3045 if (bufsize > sb_max)
3047 if (bufsize > so->so_rcv.ssb_hiwat) {
3048 lwkt_gettoken(&so->so_rcv.ssb_token);
3049 ssb_reserve(&so->so_rcv, bufsize, so, NULL);
3050 lwkt_reltoken(&so->so_rcv.ssb_token);
3055 * Set the slow-start flight size depending on whether this
3056 * is a local network or not.
3059 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380));
3063 if (rt->rt_rmx.rmx_ssthresh) {
3065 * There's some sort of gateway or interface
3066 * buffer limit on the path. Use this to set
3067 * the slow start threshhold, but set the
3068 * threshold to no less than 2*mss.
3070 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
3071 tcpstat.tcps_usedssthresh++;
3076 * Determine the MSS option to send on an outgoing SYN.
3079 tcp_mssopt(struct tcpcb *tp)
3084 ((tp->t_inpcb->inp_vflag & INP_IPV6) ? TRUE : FALSE);
3085 int min_protoh = isipv6 ?
3086 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
3087 sizeof(struct tcpiphdr);
3089 const boolean_t isipv6 = FALSE;
3090 const size_t min_protoh = sizeof(struct tcpiphdr);
3094 rt = tcp_rtlookup6(&tp->t_inpcb->inp_inc);
3096 rt = tcp_rtlookup(&tp->t_inpcb->inp_inc);
3098 return (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
3100 return (rt->rt_ifp->if_mtu - min_protoh);
3104 * When a partial ack arrives, force the retransmission of the
3105 * next unacknowledged segment. Do not exit Fast Recovery.
3107 * Implement the Slow-but-Steady variant of NewReno by restarting the
3108 * the retransmission timer. Turn it off here so it can be restarted
3109 * later in tcp_output().
3112 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th, int acked)
3114 tcp_seq old_snd_nxt = tp->snd_nxt;
3115 u_long ocwnd = tp->snd_cwnd;
3117 tcp_callout_stop(tp, tp->tt_rexmt);
3119 tp->snd_nxt = th->th_ack;
3120 /* Set snd_cwnd to one segment beyond acknowledged offset. */
3121 tp->snd_cwnd = tp->t_maxseg;
3122 tp->t_flags |= TF_ACKNOW;
3124 if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
3125 tp->snd_nxt = old_snd_nxt;
3126 /* partial window deflation */
3128 tp->snd_cwnd = ocwnd - acked + tp->t_maxseg;
3130 tp->snd_cwnd = tp->t_maxseg;
3134 * In contrast to the Slow-but-Steady NewReno variant,
3135 * we do not reset the retransmission timer for SACK retransmissions,
3136 * except when retransmitting snd_una.
3139 tcp_sack_rexmt(struct tcpcb *tp, struct tcphdr *th)
3141 uint32_t pipe, seglen;
3144 tcp_seq old_snd_nxt = tp->snd_nxt;
3145 u_long ocwnd = tp->snd_cwnd;
3146 int nseg = 0; /* consecutive new segments */
3147 #define MAXBURST 4 /* limit burst of new packets on partial ack */
3150 pipe = tcp_sack_compute_pipe(tp);
3151 while ((tcp_seq_diff_t)(ocwnd - pipe) >= (tcp_seq_diff_t)tp->t_maxseg &&
3152 (!tcp_do_smartsack || nseg < MAXBURST) &&
3153 tcp_sack_nextseg(tp, &nextrexmt, &seglen, &lostdup)) {
3155 tcp_seq old_snd_max;
3158 if (nextrexmt == tp->snd_max)
3160 tp->snd_nxt = nextrexmt;
3161 tp->snd_cwnd = nextrexmt - tp->snd_una + seglen;
3162 old_snd_max = tp->snd_max;
3163 if (nextrexmt == tp->snd_una)
3164 tcp_callout_stop(tp, tp->tt_rexmt);
3165 error = tcp_output(tp);
3168 sent = tp->snd_nxt - nextrexmt;
3173 tcpstat.tcps_sndsackpack++;
3174 tcpstat.tcps_sndsackbyte += sent;
3175 if (SEQ_LT(nextrexmt, old_snd_max) &&
3176 SEQ_LT(tp->rexmt_high, tp->snd_nxt))
3177 tp->rexmt_high = seq_min(tp->snd_nxt, old_snd_max);
3179 if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
3180 tp->snd_nxt = old_snd_nxt;
3181 tp->snd_cwnd = ocwnd;
3185 * Reset idle time and keep-alive timer, typically called when a valid
3186 * tcp packet is received but may also be called when FASTKEEP is set
3187 * to prevent the previous long-timeout from calculating to a drop.
3189 * Only update t_rcvtime for non-SYN packets.
3191 * Handle the case where one side thinks the connection is established
3192 * but the other side has, say, rebooted without cleaning out the
3193 * connection. The SYNs could be construed as an attack and wind
3194 * up ignored, but in case it isn't an attack we can validate the
3195 * connection by forcing a keepalive.
3198 tcp_timer_keep_activity(struct tcpcb *tp, int thflags)
3200 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
3201 if ((thflags & (TH_SYN | TH_ACK)) == TH_SYN) {
3202 tp->t_flags |= TF_KEEPALIVE;
3203 tcp_callout_reset(tp, tp->tt_keep, hz / 2,
3206 tp->t_rcvtime = ticks;
3207 tp->t_flags &= ~TF_KEEPALIVE;
3208 tcp_callout_reset(tp, tp->tt_keep,
3209 tcp_getkeepidle(tp),
3216 tcp_rmx_msl(const struct tcpcb *tp)
3219 struct inpcb *inp = tp->t_inpcb;
3222 boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) ? TRUE : FALSE);
3224 const boolean_t isipv6 = FALSE;
3228 rt = tcp_rtlookup6(&inp->inp_inc);
3230 rt = tcp_rtlookup(&inp->inp_inc);
3231 if (rt == NULL || rt->rt_rmx.rmx_msl == 0)
3234 msl = (rt->rt_rmx.rmx_msl * hz) / 1000;
tuxillo's projects / dragonfly.git / blob