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 $
71 #include "opt_inet6.h"
72 #include "opt_ipsec.h"
73 #include "opt_tcpdebug.h"
74 #include "opt_tcp_input.h"
76 #include <sys/param.h>
77 #include <sys/systm.h>
78 #include <sys/kernel.h>
79 #include <sys/sysctl.h>
80 #include <sys/malloc.h>
82 #include <sys/proc.h> /* for proc0 declaration */
83 #include <sys/protosw.h>
84 #include <sys/socket.h>
85 #include <sys/socketvar.h>
86 #include <sys/syslog.h>
87 #include <sys/in_cksum.h>
89 #include <sys/socketvar2.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 static int tcp_do_eifel_detect = 1;
171 SYSCTL_INT(_net_inet_tcp, OID_AUTO, eifel, CTLFLAG_RW,
172 &tcp_do_eifel_detect, 0, "Eifel detection algorithm (RFC 3522)");
174 static int tcp_do_abc = 1;
175 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc, CTLFLAG_RW,
177 "TCP Appropriate Byte Counting (RFC 3465)");
180 * Define as tunable for easy testing with SACK on and off.
181 * Warning: do not change setting in the middle of an existing active TCP flow,
182 * else strange things might happen to that flow.
185 SYSCTL_INT(_net_inet_tcp, OID_AUTO, sack, CTLFLAG_RW,
186 &tcp_do_sack, 0, "Enable SACK Algorithms");
188 int tcp_do_smartsack = 1;
189 SYSCTL_INT(_net_inet_tcp, OID_AUTO, smartsack, CTLFLAG_RW,
190 &tcp_do_smartsack, 0, "Enable Smart SACK Algorithms");
192 int tcp_do_rescuesack = 1;
193 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rescuesack, CTLFLAG_RW,
194 &tcp_do_rescuesack, 0, "Rescue retransmission for SACK");
196 int tcp_aggressive_rescuesack = 0;
197 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rescuesack_agg, CTLFLAG_RW,
198 &tcp_aggressive_rescuesack, 0, "Aggressive rescue retransmission for SACK");
200 int tcp_do_rfc3517bis = 0;
201 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3517bis, CTLFLAG_RW,
202 &tcp_do_rfc3517bis, 0, "Enable RFC3517 update");
204 int tcp_rfc3517bis_rxt = 0;
205 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3517bis_rxt, CTLFLAG_RW,
206 &tcp_rfc3517bis_rxt, 0, "Enable RFC3517 retransmit update");
208 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0,
209 "TCP Segment Reassembly Queue");
211 int tcp_reass_maxseg = 0;
212 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RD,
213 &tcp_reass_maxseg, 0,
214 "Global maximum number of TCP Segments in Reassembly Queue");
216 int tcp_reass_qsize = 0;
217 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD,
219 "Global number of TCP Segments currently in Reassembly Queue");
221 static int tcp_reass_overflows = 0;
222 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD,
223 &tcp_reass_overflows, 0,
224 "Global number of TCP Segment Reassembly Queue Overflows");
226 int tcp_do_autorcvbuf = 1;
227 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_RW,
228 &tcp_do_autorcvbuf, 0, "Enable automatic receive buffer sizing");
230 int tcp_autorcvbuf_inc = 16*1024;
231 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_RW,
232 &tcp_autorcvbuf_inc, 0,
233 "Incrementor step size of automatic receive buffer");
235 int tcp_autorcvbuf_max = 2*1024*1024;
236 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_RW,
237 &tcp_autorcvbuf_max, 0, "Max size of automatic receive buffer");
239 int tcp_sosend_agglim = 2;
240 SYSCTL_INT(_net_inet_tcp, OID_AUTO, sosend_agglim, CTLFLAG_RW,
241 &tcp_sosend_agglim, 0, "TCP sosend mbuf aggregation limit");
243 int tcp_sosend_async = 1;
244 SYSCTL_INT(_net_inet_tcp, OID_AUTO, sosend_async, CTLFLAG_RW,
245 &tcp_sosend_async, 0, "TCP asynchronized pru_send");
247 static int tcp_ignore_redun_dsack = 1;
248 SYSCTL_INT(_net_inet_tcp, OID_AUTO, ignore_redun_dsack, CTLFLAG_RW,
249 &tcp_ignore_redun_dsack, 0, "Ignore redundant DSACK");
251 static void tcp_dooptions(struct tcpopt *, u_char *, int, boolean_t,
253 static void tcp_pulloutofband(struct socket *,
254 struct tcphdr *, struct mbuf *, int);
255 static int tcp_reass(struct tcpcb *, struct tcphdr *, int *,
257 static void tcp_xmit_timer(struct tcpcb *, int, tcp_seq);
258 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *, int);
259 static void tcp_sack_rexmt(struct tcpcb *, struct tcphdr *);
260 static boolean_t tcp_sack_limitedxmit(struct tcpcb *);
261 static int tcp_rmx_msl(const struct tcpcb *);
262 static void tcp_established(struct tcpcb *);
264 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
266 #define ND6_HINT(tp) \
268 if ((tp) && (tp)->t_inpcb && \
269 ((tp)->t_inpcb->inp_vflag & INP_IPV6) && \
270 (tp)->t_inpcb->in6p_route.ro_rt) \
271 nd6_nud_hint((tp)->t_inpcb->in6p_route.ro_rt, NULL, 0); \
278 * Indicate whether this ack should be delayed. We can delay the ack if
279 * - delayed acks are enabled and
280 * - there is no delayed ack timer in progress and
281 * - our last ack wasn't a 0-sized window. We never want to delay
282 * the ack that opens up a 0-sized window.
284 #define DELAY_ACK(tp) \
285 (tcp_delack_enabled && !tcp_callout_pending(tp, tp->tt_delack) && \
286 !(tp->t_flags & TF_RXWIN0SENT))
288 #define acceptable_window_update(tp, th, tiwin) \
289 (SEQ_LT(tp->snd_wl1, th->th_seq) || \
290 (tp->snd_wl1 == th->th_seq && \
291 (SEQ_LT(tp->snd_wl2, th->th_ack) || \
292 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))
294 #define iceildiv(n, d) (((n)+(d)-1) / (d))
295 #define need_early_retransmit(tp, ownd) \
296 (tcp_do_early_retransmit && \
297 (tcp_do_eifel_detect && (tp->t_flags & TF_RCVD_TSTMP)) && \
298 ownd < (4 * tp->t_maxseg) && \
299 tp->t_dupacks + 1 >= iceildiv(ownd, tp->t_maxseg) && \
300 (!TCP_DO_SACK(tp) || ownd <= tp->t_maxseg || \
301 tcp_sack_has_sacked(&tp->scb, ownd - tp->t_maxseg)))
304 tcp_reass(struct tcpcb *tp, struct tcphdr *th, int *tlenp, struct mbuf *m)
307 struct tseg_qent *p = NULL;
308 struct tseg_qent *te;
309 struct socket *so = tp->t_inpcb->inp_socket;
313 * Call with th == NULL after become established to
314 * force pre-ESTABLISHED data up to user socket.
320 * Limit the number of segments in the reassembly queue to prevent
321 * holding on to too many segments (and thus running out of mbufs).
322 * Make sure to let the missing segment through which caused this
323 * queue. Always keep one global queue entry spare to be able to
324 * process the missing segment.
326 if (th->th_seq != tp->rcv_nxt &&
327 tcp_reass_qsize + 1 >= tcp_reass_maxseg) {
328 tcp_reass_overflows++;
329 tcpstat.tcps_rcvmemdrop++;
331 /* no SACK block to report */
332 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
336 /* Allocate a new queue entry. */
337 te = kmalloc(sizeof(struct tseg_qent), M_TSEGQ, M_INTWAIT | M_NULLOK);
339 tcpstat.tcps_rcvmemdrop++;
341 /* no SACK block to report */
342 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
345 atomic_add_int(&tcp_reass_qsize, 1);
348 * Find a segment which begins after this one does.
350 LIST_FOREACH(q, &tp->t_segq, tqe_q) {
351 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
357 * If there is a preceding segment, it may provide some of
358 * our data already. If so, drop the data from the incoming
359 * segment. If it provides all of our data, drop us.
364 /* conversion to int (in i) handles seq wraparound */
365 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
366 if (i > 0) { /* overlaps preceding segment */
367 tp->t_flags |= (TF_DUPSEG | TF_ENCLOSESEG);
368 /* enclosing block starts w/ preceding segment */
369 tp->encloseblk.rblk_start = p->tqe_th->th_seq;
371 /* preceding encloses incoming segment */
372 tp->encloseblk.rblk_end = TCP_SACK_BLKEND(
373 p->tqe_th->th_seq + p->tqe_len,
374 p->tqe_th->th_flags);
375 tcpstat.tcps_rcvduppack++;
376 tcpstat.tcps_rcvdupbyte += *tlenp;
379 atomic_add_int(&tcp_reass_qsize, -1);
381 * Try to present any queued data
382 * at the left window edge to the user.
383 * This is needed after the 3-WHS
386 goto present; /* ??? */
391 /* incoming segment end is enclosing block end */
392 tp->encloseblk.rblk_end = TCP_SACK_BLKEND(
393 th->th_seq + *tlenp, th->th_flags);
394 /* trim end of reported D-SACK block */
395 tp->reportblk.rblk_end = th->th_seq;
398 tcpstat.tcps_rcvoopack++;
399 tcpstat.tcps_rcvoobyte += *tlenp;
402 * While we overlap succeeding segments trim them or,
403 * if they are completely covered, dequeue them.
406 tcp_seq_diff_t i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
407 tcp_seq qend = q->tqe_th->th_seq + q->tqe_len;
408 tcp_seq qend_sack = TCP_SACK_BLKEND(qend, q->tqe_th->th_flags);
409 struct tseg_qent *nq;
413 if (!(tp->t_flags & TF_DUPSEG)) { /* first time through */
414 tp->t_flags |= (TF_DUPSEG | TF_ENCLOSESEG);
415 tp->encloseblk = tp->reportblk;
416 /* report trailing duplicate D-SACK segment */
417 tp->reportblk.rblk_start = q->tqe_th->th_seq;
419 if ((tp->t_flags & TF_ENCLOSESEG) &&
420 SEQ_GT(qend_sack, tp->encloseblk.rblk_end)) {
421 /* extend enclosing block if one exists */
422 tp->encloseblk.rblk_end = qend_sack;
424 if (i < q->tqe_len) {
425 q->tqe_th->th_seq += i;
431 nq = LIST_NEXT(q, tqe_q);
432 LIST_REMOVE(q, tqe_q);
435 atomic_add_int(&tcp_reass_qsize, -1);
439 /* Insert the new segment queue entry into place. */
442 te->tqe_len = *tlenp;
444 /* check if can coalesce with following segment */
445 if (q != NULL && (th->th_seq + *tlenp == q->tqe_th->th_seq)) {
446 tcp_seq tend = te->tqe_th->th_seq + te->tqe_len;
447 tcp_seq tend_sack = TCP_SACK_BLKEND(tend, te->tqe_th->th_flags);
449 te->tqe_len += q->tqe_len;
450 if (q->tqe_th->th_flags & TH_FIN)
451 te->tqe_th->th_flags |= TH_FIN;
452 m_cat(te->tqe_m, q->tqe_m);
453 tp->encloseblk.rblk_end = tend_sack;
455 * When not reporting a duplicate segment, use
456 * the larger enclosing block as the SACK block.
458 if (!(tp->t_flags & TF_DUPSEG))
459 tp->reportblk.rblk_end = tend_sack;
460 LIST_REMOVE(q, tqe_q);
462 atomic_add_int(&tcp_reass_qsize, -1);
466 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
468 /* check if can coalesce with preceding segment */
469 if (p->tqe_th->th_seq + p->tqe_len == th->th_seq) {
470 p->tqe_len += te->tqe_len;
471 m_cat(p->tqe_m, te->tqe_m);
472 tp->encloseblk.rblk_start = p->tqe_th->th_seq;
474 * When not reporting a duplicate segment, use
475 * the larger enclosing block as the SACK block.
477 if (!(tp->t_flags & TF_DUPSEG))
478 tp->reportblk.rblk_start = p->tqe_th->th_seq;
480 atomic_add_int(&tcp_reass_qsize, -1);
482 LIST_INSERT_AFTER(p, te, tqe_q);
488 * Present data to user, advancing rcv_nxt through
489 * completed sequence space.
491 if (!TCPS_HAVEESTABLISHED(tp->t_state))
493 q = LIST_FIRST(&tp->t_segq);
494 if (q == NULL || q->tqe_th->th_seq != tp->rcv_nxt)
496 tp->rcv_nxt += q->tqe_len;
497 if (!(tp->t_flags & TF_DUPSEG)) {
498 /* no SACK block to report since ACK advanced */
499 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
501 /* no enclosing block to report since ACK advanced */
502 tp->t_flags &= ~TF_ENCLOSESEG;
503 flags = q->tqe_th->th_flags & TH_FIN;
504 LIST_REMOVE(q, tqe_q);
505 KASSERT(LIST_EMPTY(&tp->t_segq) ||
506 LIST_FIRST(&tp->t_segq)->tqe_th->th_seq != tp->rcv_nxt,
507 ("segment not coalesced"));
508 if (so->so_state & SS_CANTRCVMORE) {
511 lwkt_gettoken(&so->so_rcv.ssb_token);
512 ssb_appendstream(&so->so_rcv, q->tqe_m);
513 lwkt_reltoken(&so->so_rcv.ssb_token);
516 atomic_add_int(&tcp_reass_qsize, -1);
523 * TCP input routine, follows pages 65-76 of the
524 * protocol specification dated September, 1981 very closely.
528 tcp6_input(struct mbuf **mp, int *offp, int proto)
530 struct mbuf *m = *mp;
531 struct in6_ifaddr *ia6;
533 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
536 * draft-itojun-ipv6-tcp-to-anycast
537 * better place to put this in?
539 ia6 = ip6_getdstifaddr(m);
540 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
543 ip6 = mtod(m, struct ip6_hdr *);
544 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
545 offsetof(struct ip6_hdr, ip6_dst));
546 return (IPPROTO_DONE);
549 tcp_input(mp, offp, proto);
550 return (IPPROTO_DONE);
555 tcp_input(struct mbuf **mp, int *offp, int proto)
559 struct ip *ip = NULL;
561 struct inpcb *inp = NULL;
567 struct tcpcb *tp = NULL;
569 struct socket *so = NULL;
571 boolean_t ourfinisacked, needoutput = FALSE;
574 struct tcpopt to; /* options in this segment */
575 struct sockaddr_in *next_hop = NULL;
576 int rstreason; /* For badport_bandlim accounting purposes */
578 struct ip6_hdr *ip6 = NULL;
583 const boolean_t isipv6 = FALSE;
593 tcpstat.tcps_rcvtotal++;
595 if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
598 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
599 KKASSERT(mtag != NULL);
600 next_hop = m_tag_data(mtag);
604 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? TRUE : FALSE;
608 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */
609 ip6 = mtod(m, struct ip6_hdr *);
610 tlen = (sizeof *ip6) + ntohs(ip6->ip6_plen) - off0;
611 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) {
612 tcpstat.tcps_rcvbadsum++;
615 th = (struct tcphdr *)((caddr_t)ip6 + off0);
618 * Be proactive about unspecified IPv6 address in source.
619 * As we use all-zero to indicate unbounded/unconnected pcb,
620 * unspecified IPv6 address can be used to confuse us.
622 * Note that packets with unspecified IPv6 destination is
623 * already dropped in ip6_input.
625 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
631 * Get IP and TCP header together in first mbuf.
632 * Note: IP leaves IP header in first mbuf.
634 if (off0 > sizeof(struct ip)) {
636 off0 = sizeof(struct ip);
638 /* already checked and pulled up in ip_demux() */
639 KASSERT(m->m_len >= sizeof(struct tcpiphdr),
640 ("TCP header not in one mbuf: m->m_len %d", m->m_len));
641 ip = mtod(m, struct ip *);
642 ipov = (struct ipovly *)ip;
643 th = (struct tcphdr *)((caddr_t)ip + off0);
646 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
647 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
648 th->th_sum = m->m_pkthdr.csum_data;
650 th->th_sum = in_pseudo(ip->ip_src.s_addr,
652 htonl(m->m_pkthdr.csum_data +
655 th->th_sum ^= 0xffff;
658 * Checksum extended TCP header and data.
660 len = sizeof(struct ip) + tlen;
661 bzero(ipov->ih_x1, sizeof ipov->ih_x1);
662 ipov->ih_len = (u_short)tlen;
663 ipov->ih_len = htons(ipov->ih_len);
664 th->th_sum = in_cksum(m, len);
667 tcpstat.tcps_rcvbadsum++;
671 /* Re-initialization for later version check */
672 ip->ip_v = IPVERSION;
677 * Check that TCP offset makes sense,
678 * pull out TCP options and adjust length. XXX
680 off = th->th_off << 2;
681 /* already checked and pulled up in ip_demux() */
682 KASSERT(off >= sizeof(struct tcphdr) && off <= tlen,
683 ("bad TCP data offset %d (tlen %d)", off, tlen));
684 tlen -= off; /* tlen is used instead of ti->ti_len */
685 if (off > sizeof(struct tcphdr)) {
687 IP6_EXTHDR_CHECK(m, off0, off, IPPROTO_DONE);
688 ip6 = mtod(m, struct ip6_hdr *);
689 th = (struct tcphdr *)((caddr_t)ip6 + off0);
691 /* already pulled up in ip_demux() */
692 KASSERT(m->m_len >= sizeof(struct ip) + off,
693 ("TCP header and options not in one mbuf: "
694 "m_len %d, off %d", m->m_len, off));
696 optlen = off - sizeof(struct tcphdr);
697 optp = (u_char *)(th + 1);
699 thflags = th->th_flags;
701 #ifdef TCP_DROP_SYNFIN
703 * If the drop_synfin option is enabled, drop all packets with
704 * both the SYN and FIN bits set. This prevents e.g. nmap from
705 * identifying the TCP/IP stack.
707 * This is a violation of the TCP specification.
709 if (drop_synfin && (thflags & (TH_SYN | TH_FIN)) == (TH_SYN | TH_FIN))
714 * Convert TCP protocol specific fields to host format.
716 th->th_seq = ntohl(th->th_seq);
717 th->th_ack = ntohl(th->th_ack);
718 th->th_win = ntohs(th->th_win);
719 th->th_urp = ntohs(th->th_urp);
722 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options,
723 * until after ip6_savecontrol() is called and before other functions
724 * which don't want those proto headers.
725 * Because ip6_savecontrol() is going to parse the mbuf to
726 * search for data to be passed up to user-land, it wants mbuf
727 * parameters to be unchanged.
728 * XXX: the call of ip6_savecontrol() has been obsoleted based on
729 * latest version of the advanced API (20020110).
731 drop_hdrlen = off0 + off;
734 * Locate pcb for segment.
737 /* IPFIREWALL_FORWARD section */
738 if (next_hop != NULL && !isipv6) { /* IPv6 support is not there yet */
740 * Transparently forwarded. Pretend to be the destination.
741 * already got one like this?
743 cpu = mycpu->gd_cpuid;
744 inp = in_pcblookup_hash(&tcbinfo[cpu],
745 ip->ip_src, th->th_sport,
746 ip->ip_dst, th->th_dport,
747 0, m->m_pkthdr.rcvif);
750 * It's new. Try to find the ambushing socket.
754 * The rest of the ipfw code stores the port in
756 * (The IP address is still in network order.)
758 in_port_t dport = next_hop->sin_port ?
759 htons(next_hop->sin_port) :
762 cpu = tcp_addrcpu(ip->ip_src.s_addr, th->th_sport,
763 next_hop->sin_addr.s_addr, dport);
764 inp = in_pcblookup_hash(&tcbinfo[cpu],
765 ip->ip_src, th->th_sport,
766 next_hop->sin_addr, dport,
767 1, m->m_pkthdr.rcvif);
771 inp = in6_pcblookup_hash(&tcbinfo[0],
772 &ip6->ip6_src, th->th_sport,
773 &ip6->ip6_dst, th->th_dport,
774 1, m->m_pkthdr.rcvif);
776 cpu = mycpu->gd_cpuid;
777 inp = in_pcblookup_hash(&tcbinfo[cpu],
778 ip->ip_src, th->th_sport,
779 ip->ip_dst, th->th_dport,
780 1, m->m_pkthdr.rcvif);
785 * If the state is CLOSED (i.e., TCB does not exist) then
786 * all data in the incoming segment is discarded.
787 * If the TCB exists but is in CLOSED state, it is embryonic,
788 * but should either do a listen or a connect soon.
793 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2];
795 char dbuf[sizeof "aaa.bbb.ccc.ddd"];
796 char sbuf[sizeof "aaa.bbb.ccc.ddd"];
800 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst));
803 strcat(sbuf, ip6_sprintf(&ip6->ip6_src));
806 strcpy(dbuf, inet_ntoa(ip->ip_dst));
807 strcpy(sbuf, inet_ntoa(ip->ip_src));
809 switch (log_in_vain) {
811 if (!(thflags & TH_SYN))
815 "Connection attempt to TCP %s:%d "
816 "from %s:%d flags:0x%02x\n",
817 dbuf, ntohs(th->th_dport), sbuf,
818 ntohs(th->th_sport), thflags);
827 if (thflags & TH_SYN)
836 rstreason = BANDLIM_RST_CLOSEDPORT;
842 if (ipsec6_in_reject_so(m, inp->inp_socket)) {
843 ipsec6stat.in_polvio++;
847 if (ipsec4_in_reject_so(m, inp->inp_socket)) {
848 ipsecstat.in_polvio++;
855 if (ipsec6_in_reject(m, inp))
858 if (ipsec4_in_reject(m, inp))
862 /* Check the minimum TTL for socket. */
864 if ((isipv6 ? ip6->ip6_hlim : ip->ip_ttl) < inp->inp_ip_minttl)
870 rstreason = BANDLIM_RST_CLOSEDPORT;
873 if (tp->t_state <= TCPS_CLOSED)
876 so = inp->inp_socket;
879 if (so->so_options & SO_DEBUG) {
880 ostate = tp->t_state;
882 bcopy(ip6, tcp_saveipgen, sizeof(*ip6));
884 bcopy(ip, tcp_saveipgen, sizeof(*ip));
889 bzero(&to, sizeof to);
891 if (so->so_options & SO_ACCEPTCONN) {
892 struct in_conninfo inc;
895 inc.inc_isipv6 = (isipv6 == TRUE);
898 inc.inc6_faddr = ip6->ip6_src;
899 inc.inc6_laddr = ip6->ip6_dst;
900 inc.inc6_route.ro_rt = NULL; /* XXX */
902 inc.inc_faddr = ip->ip_src;
903 inc.inc_laddr = ip->ip_dst;
904 inc.inc_route.ro_rt = NULL; /* XXX */
906 inc.inc_fport = th->th_sport;
907 inc.inc_lport = th->th_dport;
910 * If the state is LISTEN then ignore segment if it contains
911 * a RST. If the segment contains an ACK then it is bad and
912 * send a RST. If it does not contain a SYN then it is not
913 * interesting; drop it.
915 * If the state is SYN_RECEIVED (syncache) and seg contains
916 * an ACK, but not for our SYN/ACK, send a RST. If the seg
917 * contains a RST, check the sequence number to see if it
918 * is a valid reset segment.
920 if ((thflags & (TH_RST | TH_ACK | TH_SYN)) != TH_SYN) {
921 if ((thflags & (TH_RST | TH_ACK | TH_SYN)) == TH_ACK) {
922 if (!syncache_expand(&inc, th, &so, m)) {
924 * No syncache entry, or ACK was not
925 * for our SYN/ACK. Send a RST.
927 tcpstat.tcps_badsyn++;
928 rstreason = BANDLIM_RST_OPENPORT;
933 * Could not complete 3-way handshake,
934 * connection is being closed down, and
935 * syncache will free mbuf.
938 return(IPPROTO_DONE);
941 * We must be in the correct protocol thread
942 * for this connection.
944 KKASSERT(so->so_port == &curthread->td_msgport);
947 * Socket is created in state SYN_RECEIVED.
948 * Continue processing segment.
953 * This is what would have happened in
954 * tcp_output() when the SYN,ACK was sent.
956 tp->snd_up = tp->snd_una;
957 tp->snd_max = tp->snd_nxt = tp->iss + 1;
958 tp->last_ack_sent = tp->rcv_nxt;
962 if (thflags & TH_RST) {
963 syncache_chkrst(&inc, th);
966 if (thflags & TH_ACK) {
967 syncache_badack(&inc);
968 tcpstat.tcps_badsyn++;
969 rstreason = BANDLIM_RST_OPENPORT;
976 * Segment's flags are (SYN) or (SYN | FIN).
980 * If deprecated address is forbidden,
981 * we do not accept SYN to deprecated interface
982 * address to prevent any new inbound connection from
983 * getting established.
984 * When we do not accept SYN, we send a TCP RST,
985 * with deprecated source address (instead of dropping
986 * it). We compromise it as it is much better for peer
987 * to send a RST, and RST will be the final packet
990 * If we do not forbid deprecated addresses, we accept
991 * the SYN packet. RFC2462 does not suggest dropping
993 * If we decipher RFC2462 5.5.4, it says like this:
994 * 1. use of deprecated addr with existing
995 * communication is okay - "SHOULD continue to be
997 * 2. use of it with new communication:
998 * (2a) "SHOULD NOT be used if alternate address
999 * with sufficient scope is available"
1000 * (2b) nothing mentioned otherwise.
1001 * Here we fall into (2b) case as we have no choice in
1002 * our source address selection - we must obey the peer.
1004 * The wording in RFC2462 is confusing, and there are
1005 * multiple description text for deprecated address
1006 * handling - worse, they are not exactly the same.
1007 * I believe 5.5.4 is the best one, so we follow 5.5.4.
1009 if (isipv6 && !ip6_use_deprecated) {
1010 struct in6_ifaddr *ia6;
1012 if ((ia6 = ip6_getdstifaddr(m)) &&
1013 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1015 rstreason = BANDLIM_RST_OPENPORT;
1021 * If it is from this socket, drop it, it must be forged.
1022 * Don't bother responding if the destination was a broadcast.
1024 if (th->th_dport == th->th_sport) {
1026 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
1030 if (ip->ip_dst.s_addr == ip->ip_src.s_addr)
1035 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
1037 * Note that it is quite possible to receive unicast
1038 * link-layer packets with a broadcast IP address. Use
1039 * in_broadcast() to find them.
1041 if (m->m_flags & (M_BCAST | M_MCAST))
1044 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1045 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
1048 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1049 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1050 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1051 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
1055 * SYN appears to be valid; create compressed TCP state
1056 * for syncache, or perform t/tcp connection.
1058 if (so->so_qlen <= so->so_qlimit) {
1059 tcp_dooptions(&to, optp, optlen, TRUE, th->th_ack);
1060 if (!syncache_add(&inc, &to, th, so, m))
1064 * Entry added to syncache, mbuf used to
1065 * send SYN,ACK packet.
1067 return(IPPROTO_DONE);
1074 * Should not happen - syncache should pick up these connections.
1076 * Once we are past handling listen sockets we must be in the
1077 * correct protocol processing thread.
1079 KASSERT(tp->t_state != TCPS_LISTEN, ("tcp_input: TCPS_LISTEN state"));
1080 KKASSERT(so->so_port == &curthread->td_msgport);
1082 /* Unscale the window into a 32-bit value. */
1083 if (!(thflags & TH_SYN))
1084 tiwin = th->th_win << tp->snd_scale;
1089 * This is the second part of the MSS DoS prevention code (after
1090 * minmss on the sending side) and it deals with too many too small
1091 * tcp packets in a too short timeframe (1 second).
1093 * XXX Removed. This code was crap. It does not scale to network
1094 * speed, and default values break NFS. Gone.
1099 * Segment received on connection.
1101 * Reset idle time and keep-alive timer. Don't waste time if less
1102 * then a second has elapsed.
1104 if ((int)(ticks - tp->t_rcvtime) > hz)
1105 tcp_timer_keep_activity(tp, thflags);
1109 * XXX this is tradtitional behavior, may need to be cleaned up.
1111 tcp_dooptions(&to, optp, optlen, (thflags & TH_SYN) != 0, th->th_ack);
1112 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1113 if ((to.to_flags & TOF_SCALE) && (tp->t_flags & TF_REQ_SCALE)) {
1114 tp->t_flags |= TF_RCVD_SCALE;
1115 tp->snd_scale = to.to_requested_s_scale;
1119 * Initial send window; will be updated upon next ACK
1121 tp->snd_wnd = th->th_win;
1123 if (to.to_flags & TOF_TS) {
1124 tp->t_flags |= TF_RCVD_TSTMP;
1125 tp->ts_recent = to.to_tsval;
1126 tp->ts_recent_age = ticks;
1128 if (!(to.to_flags & TOF_MSS))
1130 tcp_mss(tp, to.to_mss);
1132 * Only set the TF_SACK_PERMITTED per-connection flag
1133 * if we got a SACK_PERMITTED option from the other side
1134 * and the global tcp_do_sack variable is true.
1136 if (tcp_do_sack && (to.to_flags & TOF_SACK_PERMITTED))
1137 tp->t_flags |= TF_SACK_PERMITTED;
1141 * Header prediction: check for the two common cases
1142 * of a uni-directional data xfer. If the packet has
1143 * no control flags, is in-sequence, the window didn't
1144 * change and we're not retransmitting, it's a
1145 * candidate. If the length is zero and the ack moved
1146 * forward, we're the sender side of the xfer. Just
1147 * free the data acked & wake any higher level process
1148 * that was blocked waiting for space. If the length
1149 * is non-zero and the ack didn't move, we're the
1150 * receiver side. If we're getting packets in-order
1151 * (the reassembly queue is empty), add the data to
1152 * the socket buffer and note that we need a delayed ack.
1153 * Make sure that the hidden state-flags are also off.
1154 * Since we check for TCPS_ESTABLISHED above, it can only
1157 if (tp->t_state == TCPS_ESTABLISHED &&
1158 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1159 !(tp->t_flags & (TF_NEEDSYN | TF_NEEDFIN)) &&
1160 (!(to.to_flags & TOF_TS) ||
1161 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) &&
1162 th->th_seq == tp->rcv_nxt &&
1163 tp->snd_nxt == tp->snd_max) {
1166 * If last ACK falls within this segment's sequence numbers,
1167 * record the timestamp.
1168 * NOTE that the test is modified according to the latest
1169 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1171 if ((to.to_flags & TOF_TS) &&
1172 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1173 tp->ts_recent_age = ticks;
1174 tp->ts_recent = to.to_tsval;
1178 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1179 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1180 tp->snd_cwnd >= tp->snd_wnd &&
1181 !IN_FASTRECOVERY(tp)) {
1183 * This is a pure ack for outstanding data.
1185 ++tcpstat.tcps_predack;
1187 * "bad retransmit" recovery
1189 * If Eifel detection applies, then
1190 * it is deterministic, so use it
1191 * unconditionally over the old heuristic.
1192 * Otherwise, fall back to the old heuristic.
1194 if (tcp_do_eifel_detect &&
1195 (to.to_flags & TOF_TS) && to.to_tsecr &&
1196 (tp->t_flags & TF_FIRSTACCACK)) {
1197 /* Eifel detection applicable. */
1198 if (to.to_tsecr < tp->t_rexmtTS) {
1199 tcp_revert_congestion_state(tp);
1200 ++tcpstat.tcps_eifeldetected;
1201 if (tp->t_rxtshift != 1 ||
1202 ticks >= tp->t_badrxtwin)
1203 ++tcpstat.tcps_rttcantdetect;
1205 } else if (tp->t_rxtshift == 1 &&
1206 ticks < tp->t_badrxtwin) {
1207 tcp_revert_congestion_state(tp);
1208 ++tcpstat.tcps_rttdetected;
1210 tp->t_flags &= ~(TF_FIRSTACCACK |
1211 TF_FASTREXMT | TF_EARLYREXMT);
1213 * Recalculate the retransmit timer / rtt.
1215 * Some machines (certain windows boxes)
1216 * send broken timestamp replies during the
1217 * SYN+ACK phase, ignore timestamps of 0.
1219 if ((to.to_flags & TOF_TS) && to.to_tsecr) {
1221 ticks - to.to_tsecr + 1,
1223 } else if (tp->t_rtttime &&
1224 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1226 ticks - tp->t_rtttime,
1229 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1230 acked = th->th_ack - tp->snd_una;
1231 tcpstat.tcps_rcvackpack++;
1232 tcpstat.tcps_rcvackbyte += acked;
1233 sbdrop(&so->so_snd.sb, acked);
1234 tp->snd_recover = th->th_ack - 1;
1235 tp->snd_una = th->th_ack;
1238 * Update window information.
1240 if (tiwin != tp->snd_wnd &&
1241 acceptable_window_update(tp, th, tiwin)) {
1242 /* keep track of pure window updates */
1243 if (tp->snd_wl2 == th->th_ack &&
1244 tiwin > tp->snd_wnd)
1245 tcpstat.tcps_rcvwinupd++;
1246 tp->snd_wnd = tiwin;
1247 tp->snd_wl1 = th->th_seq;
1248 tp->snd_wl2 = th->th_ack;
1249 if (tp->snd_wnd > tp->max_sndwnd)
1250 tp->max_sndwnd = tp->snd_wnd;
1253 ND6_HINT(tp); /* some progress has been done */
1255 * If all outstanding data are acked, stop
1256 * retransmit timer, otherwise restart timer
1257 * using current (possibly backed-off) value.
1258 * If process is waiting for space,
1259 * wakeup/selwakeup/signal. If data
1260 * are ready to send, let tcp_output
1261 * decide between more output or persist.
1263 if (tp->snd_una == tp->snd_max) {
1264 tcp_callout_stop(tp, tp->tt_rexmt);
1265 } else if (!tcp_callout_active(tp,
1267 tcp_callout_reset(tp, tp->tt_rexmt,
1268 tp->t_rxtcur, tcp_timer_rexmt);
1271 if (so->so_snd.ssb_cc > 0)
1273 return(IPPROTO_DONE);
1275 } else if (tiwin == tp->snd_wnd &&
1276 th->th_ack == tp->snd_una &&
1277 LIST_EMPTY(&tp->t_segq) &&
1278 tlen <= ssb_space(&so->so_rcv)) {
1279 u_long newsize = 0; /* automatic sockbuf scaling */
1281 * This is a pure, in-sequence data packet
1282 * with nothing on the reassembly queue and
1283 * we have enough buffer space to take it.
1285 ++tcpstat.tcps_preddat;
1286 tp->rcv_nxt += tlen;
1287 tcpstat.tcps_rcvpack++;
1288 tcpstat.tcps_rcvbyte += tlen;
1289 ND6_HINT(tp); /* some progress has been done */
1291 * Automatic sizing of receive socket buffer. Often the send
1292 * buffer size is not optimally adjusted to the actual network
1293 * conditions at hand (delay bandwidth product). Setting the
1294 * buffer size too small limits throughput on links with high
1295 * bandwidth and high delay (eg. trans-continental/oceanic links).
1297 * On the receive side the socket buffer memory is only rarely
1298 * used to any significant extent. This allows us to be much
1299 * more aggressive in scaling the receive socket buffer. For
1300 * the case that the buffer space is actually used to a large
1301 * extent and we run out of kernel memory we can simply drop
1302 * the new segments; TCP on the sender will just retransmit it
1303 * later. Setting the buffer size too big may only consume too
1304 * much kernel memory if the application doesn't read() from
1305 * the socket or packet loss or reordering makes use of the
1308 * The criteria to step up the receive buffer one notch are:
1309 * 1. the number of bytes received during the time it takes
1310 * one timestamp to be reflected back to us (the RTT);
1311 * 2. received bytes per RTT is within seven eighth of the
1312 * current socket buffer size;
1313 * 3. receive buffer size has not hit maximal automatic size;
1315 * This algorithm does one step per RTT at most and only if
1316 * we receive a bulk stream w/o packet losses or reorderings.
1317 * Shrinking the buffer during idle times is not necessary as
1318 * it doesn't consume any memory when idle.
1320 * TODO: Only step up if the application is actually serving
1321 * the buffer to better manage the socket buffer resources.
1323 if (tcp_do_autorcvbuf &&
1325 (so->so_rcv.ssb_flags & SSB_AUTOSIZE)) {
1326 if (to.to_tsecr > tp->rfbuf_ts &&
1327 to.to_tsecr - tp->rfbuf_ts < hz) {
1329 (so->so_rcv.ssb_hiwat / 8 * 7) &&
1330 so->so_rcv.ssb_hiwat <
1331 tcp_autorcvbuf_max) {
1333 ulmin(so->so_rcv.ssb_hiwat +
1335 tcp_autorcvbuf_max);
1337 /* Start over with next RTT. */
1341 tp->rfbuf_cnt += tlen; /* add up */
1344 * Add data to socket buffer.
1346 if (so->so_state & SS_CANTRCVMORE) {
1350 * Set new socket buffer size, give up when
1353 * Adjusting the size can mess up ACK
1354 * sequencing when pure window updates are
1355 * being avoided (which is the default),
1358 lwkt_gettoken(&so->so_rcv.ssb_token);
1360 tp->t_flags |= TF_RXRESIZED;
1361 if (!ssb_reserve(&so->so_rcv, newsize,
1363 atomic_clear_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE);
1366 (TCP_MAXWIN << tp->rcv_scale)) {
1367 atomic_clear_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE);
1370 m_adj(m, drop_hdrlen); /* delayed header drop */
1371 ssb_appendstream(&so->so_rcv, m);
1372 lwkt_reltoken(&so->so_rcv.ssb_token);
1376 * This code is responsible for most of the ACKs
1377 * the TCP stack sends back after receiving a data
1378 * packet. Note that the DELAY_ACK check fails if
1379 * the delack timer is already running, which results
1380 * in an ack being sent every other packet (which is
1383 * We then further aggregate acks by not actually
1384 * sending one until the protocol thread has completed
1385 * processing the current backlog of packets. This
1386 * does not delay the ack any further, but allows us
1387 * to take advantage of the packet aggregation that
1388 * high speed NICs do (usually blocks of 8-10 packets)
1389 * to send a single ack rather then four or five acks,
1390 * greatly reducing the ack rate, the return channel
1391 * bandwidth, and the protocol overhead on both ends.
1393 * Since this also has the effect of slowing down
1394 * the exponential slow-start ramp-up, systems with
1395 * very large bandwidth-delay products might want
1396 * to turn the feature off.
1398 if (DELAY_ACK(tp)) {
1399 tcp_callout_reset(tp, tp->tt_delack,
1400 tcp_delacktime, tcp_timer_delack);
1401 } else if (tcp_aggregate_acks) {
1402 tp->t_flags |= TF_ACKNOW;
1403 if (!(tp->t_flags & TF_ONOUTPUTQ)) {
1404 tp->t_flags |= TF_ONOUTPUTQ;
1405 tp->tt_cpu = mycpu->gd_cpuid;
1407 &tcpcbackq[tp->tt_cpu],
1411 tp->t_flags |= TF_ACKNOW;
1414 return(IPPROTO_DONE);
1419 * Calculate amount of space in receive window,
1420 * and then do TCP input processing.
1421 * Receive window is amount of space in rcv queue,
1422 * but not less than advertised window.
1424 recvwin = ssb_space(&so->so_rcv);
1427 tp->rcv_wnd = imax(recvwin, (int)(tp->rcv_adv - tp->rcv_nxt));
1429 /* Reset receive buffer auto scaling when not in bulk receive mode. */
1433 switch (tp->t_state) {
1435 * If the state is SYN_RECEIVED:
1436 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1438 case TCPS_SYN_RECEIVED:
1439 if ((thflags & TH_ACK) &&
1440 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1441 SEQ_GT(th->th_ack, tp->snd_max))) {
1442 rstreason = BANDLIM_RST_OPENPORT;
1448 * If the state is SYN_SENT:
1449 * if seg contains an ACK, but not for our SYN, drop the input.
1450 * if seg contains a RST, then drop the connection.
1451 * if seg does not contain SYN, then drop it.
1452 * Otherwise this is an acceptable SYN segment
1453 * initialize tp->rcv_nxt and tp->irs
1454 * if seg contains ack then advance tp->snd_una
1455 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1456 * arrange for segment to be acked (eventually)
1457 * continue processing rest of data/controls, beginning with URG
1460 if ((thflags & TH_ACK) &&
1461 (SEQ_LEQ(th->th_ack, tp->iss) ||
1462 SEQ_GT(th->th_ack, tp->snd_max))) {
1463 rstreason = BANDLIM_UNLIMITED;
1466 if (thflags & TH_RST) {
1467 if (thflags & TH_ACK)
1468 tp = tcp_drop(tp, ECONNREFUSED);
1471 if (!(thflags & TH_SYN))
1474 tp->irs = th->th_seq;
1476 if (thflags & TH_ACK) {
1477 /* Our SYN was acked. */
1478 tcpstat.tcps_connects++;
1480 /* Do window scaling on this connection? */
1481 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
1482 (TF_RCVD_SCALE | TF_REQ_SCALE))
1483 tp->rcv_scale = tp->request_r_scale;
1484 tp->rcv_adv += tp->rcv_wnd;
1485 tp->snd_una++; /* SYN is acked */
1486 tcp_callout_stop(tp, tp->tt_rexmt);
1488 * If there's data, delay ACK; if there's also a FIN
1489 * ACKNOW will be turned on later.
1491 if (DELAY_ACK(tp) && tlen != 0) {
1492 tcp_callout_reset(tp, tp->tt_delack,
1493 tcp_delacktime, tcp_timer_delack);
1495 tp->t_flags |= TF_ACKNOW;
1498 * Received <SYN,ACK> in SYN_SENT[*] state.
1500 * SYN_SENT --> ESTABLISHED
1501 * SYN_SENT* --> FIN_WAIT_1
1503 tp->t_starttime = ticks;
1504 if (tp->t_flags & TF_NEEDFIN) {
1505 tp->t_state = TCPS_FIN_WAIT_1;
1506 tp->t_flags &= ~TF_NEEDFIN;
1509 tcp_established(tp);
1513 * Received initial SYN in SYN-SENT[*] state =>
1514 * simultaneous open.
1515 * Do 3-way handshake:
1516 * SYN-SENT -> SYN-RECEIVED
1517 * SYN-SENT* -> SYN-RECEIVED*
1519 tp->t_flags |= TF_ACKNOW;
1520 tcp_callout_stop(tp, tp->tt_rexmt);
1521 tp->t_state = TCPS_SYN_RECEIVED;
1525 * Advance th->th_seq to correspond to first data byte.
1526 * If data, trim to stay within window,
1527 * dropping FIN if necessary.
1530 if (tlen > tp->rcv_wnd) {
1531 todrop = tlen - tp->rcv_wnd;
1535 tcpstat.tcps_rcvpackafterwin++;
1536 tcpstat.tcps_rcvbyteafterwin += todrop;
1538 tp->snd_wl1 = th->th_seq - 1;
1539 tp->rcv_up = th->th_seq;
1541 * Client side of transaction: already sent SYN and data.
1542 * If the remote host used T/TCP to validate the SYN,
1543 * our data will be ACK'd; if so, enter normal data segment
1544 * processing in the middle of step 5, ack processing.
1545 * Otherwise, goto step 6.
1547 if (thflags & TH_ACK)
1553 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
1554 * do normal processing (we no longer bother with T/TCP).
1558 case TCPS_TIME_WAIT:
1559 break; /* continue normal processing */
1563 * States other than LISTEN or SYN_SENT.
1564 * First check the RST flag and sequence number since reset segments
1565 * are exempt from the timestamp and connection count tests. This
1566 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
1567 * below which allowed reset segments in half the sequence space
1568 * to fall though and be processed (which gives forged reset
1569 * segments with a random sequence number a 50 percent chance of
1570 * killing a connection).
1571 * Then check timestamp, if present.
1572 * Then check the connection count, if present.
1573 * Then check that at least some bytes of segment are within
1574 * receive window. If segment begins before rcv_nxt,
1575 * drop leading data (and SYN); if nothing left, just ack.
1578 * If the RST bit is set, check the sequence number to see
1579 * if this is a valid reset segment.
1581 * In all states except SYN-SENT, all reset (RST) segments
1582 * are validated by checking their SEQ-fields. A reset is
1583 * valid if its sequence number is in the window.
1584 * Note: this does not take into account delayed ACKs, so
1585 * we should test against last_ack_sent instead of rcv_nxt.
1586 * The sequence number in the reset segment is normally an
1587 * echo of our outgoing acknowledgement numbers, but some hosts
1588 * send a reset with the sequence number at the rightmost edge
1589 * of our receive window, and we have to handle this case.
1590 * If we have multiple segments in flight, the intial reset
1591 * segment sequence numbers will be to the left of last_ack_sent,
1592 * but they will eventually catch up.
1593 * In any case, it never made sense to trim reset segments to
1594 * fit the receive window since RFC 1122 says:
1595 * 4.2.2.12 RST Segment: RFC-793 Section 3.4
1597 * A TCP SHOULD allow a received RST segment to include data.
1600 * It has been suggested that a RST segment could contain
1601 * ASCII text that encoded and explained the cause of the
1602 * RST. No standard has yet been established for such
1605 * If the reset segment passes the sequence number test examine
1607 * SYN_RECEIVED STATE:
1608 * If passive open, return to LISTEN state.
1609 * If active open, inform user that connection was refused.
1610 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
1611 * Inform user that connection was reset, and close tcb.
1612 * CLOSING, LAST_ACK STATES:
1615 * Drop the segment - see Stevens, vol. 2, p. 964 and
1618 if (thflags & TH_RST) {
1619 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
1620 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
1621 switch (tp->t_state) {
1623 case TCPS_SYN_RECEIVED:
1624 so->so_error = ECONNREFUSED;
1627 case TCPS_ESTABLISHED:
1628 case TCPS_FIN_WAIT_1:
1629 case TCPS_FIN_WAIT_2:
1630 case TCPS_CLOSE_WAIT:
1631 so->so_error = ECONNRESET;
1633 tp->t_state = TCPS_CLOSED;
1634 tcpstat.tcps_drops++;
1643 case TCPS_TIME_WAIT:
1651 * RFC 1323 PAWS: If we have a timestamp reply on this segment
1652 * and it's less than ts_recent, drop it.
1654 if ((to.to_flags & TOF_TS) && tp->ts_recent != 0 &&
1655 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
1657 /* Check to see if ts_recent is over 24 days old. */
1658 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1660 * Invalidate ts_recent. If this segment updates
1661 * ts_recent, the age will be reset later and ts_recent
1662 * will get a valid value. If it does not, setting
1663 * ts_recent to zero will at least satisfy the
1664 * requirement that zero be placed in the timestamp
1665 * echo reply when ts_recent isn't valid. The
1666 * age isn't reset until we get a valid ts_recent
1667 * because we don't want out-of-order segments to be
1668 * dropped when ts_recent is old.
1672 tcpstat.tcps_rcvduppack++;
1673 tcpstat.tcps_rcvdupbyte += tlen;
1674 tcpstat.tcps_pawsdrop++;
1682 * In the SYN-RECEIVED state, validate that the packet belongs to
1683 * this connection before trimming the data to fit the receive
1684 * window. Check the sequence number versus IRS since we know
1685 * the sequence numbers haven't wrapped. This is a partial fix
1686 * for the "LAND" DoS attack.
1688 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
1689 rstreason = BANDLIM_RST_OPENPORT;
1693 todrop = tp->rcv_nxt - th->th_seq;
1695 if (TCP_DO_SACK(tp)) {
1696 /* Report duplicate segment at head of packet. */
1697 tp->reportblk.rblk_start = th->th_seq;
1698 tp->reportblk.rblk_end = TCP_SACK_BLKEND(
1699 th->th_seq + tlen, thflags);
1700 if (SEQ_GT(tp->reportblk.rblk_end, tp->rcv_nxt))
1701 tp->reportblk.rblk_end = tp->rcv_nxt;
1702 tp->t_flags |= (TF_DUPSEG | TF_SACKLEFT | TF_ACKNOW);
1704 if (thflags & TH_SYN) {
1714 * Following if statement from Stevens, vol. 2, p. 960.
1716 if (todrop > tlen ||
1717 (todrop == tlen && !(thflags & TH_FIN))) {
1719 * Any valid FIN must be to the left of the window.
1720 * At this point the FIN must be a duplicate or out
1721 * of sequence; drop it.
1726 * Send an ACK to resynchronize and drop any data.
1727 * But keep on processing for RST or ACK.
1729 tp->t_flags |= TF_ACKNOW;
1731 tcpstat.tcps_rcvduppack++;
1732 tcpstat.tcps_rcvdupbyte += todrop;
1734 tcpstat.tcps_rcvpartduppack++;
1735 tcpstat.tcps_rcvpartdupbyte += todrop;
1737 drop_hdrlen += todrop; /* drop from the top afterwards */
1738 th->th_seq += todrop;
1740 if (th->th_urp > todrop)
1741 th->th_urp -= todrop;
1749 * If new data are received on a connection after the
1750 * user processes are gone, then RST the other end.
1752 if ((so->so_state & SS_NOFDREF) &&
1753 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1755 tcpstat.tcps_rcvafterclose++;
1756 rstreason = BANDLIM_UNLIMITED;
1761 * If segment ends after window, drop trailing data
1762 * (and PUSH and FIN); if nothing left, just ACK.
1764 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
1766 tcpstat.tcps_rcvpackafterwin++;
1767 if (todrop >= tlen) {
1768 tcpstat.tcps_rcvbyteafterwin += tlen;
1770 * If a new connection request is received
1771 * while in TIME_WAIT, drop the old connection
1772 * and start over if the sequence numbers
1773 * are above the previous ones.
1775 if (thflags & TH_SYN &&
1776 tp->t_state == TCPS_TIME_WAIT &&
1777 SEQ_GT(th->th_seq, tp->rcv_nxt)) {
1782 * If window is closed can only take segments at
1783 * window edge, and have to drop data and PUSH from
1784 * incoming segments. Continue processing, but
1785 * remember to ack. Otherwise, drop segment
1788 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1789 tp->t_flags |= TF_ACKNOW;
1790 tcpstat.tcps_rcvwinprobe++;
1794 tcpstat.tcps_rcvbyteafterwin += todrop;
1797 thflags &= ~(TH_PUSH | TH_FIN);
1801 * If last ACK falls within this segment's sequence numbers,
1802 * record its timestamp.
1804 * 1) That the test incorporates suggestions from the latest
1805 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1806 * 2) That updating only on newer timestamps interferes with
1807 * our earlier PAWS tests, so this check should be solely
1808 * predicated on the sequence space of this segment.
1809 * 3) That we modify the segment boundary check to be
1810 * Last.ACK.Sent <= SEG.SEQ + SEG.LEN
1811 * instead of RFC1323's
1812 * Last.ACK.Sent < SEG.SEQ + SEG.LEN,
1813 * This modified check allows us to overcome RFC1323's
1814 * limitations as described in Stevens TCP/IP Illustrated
1815 * Vol. 2 p.869. In such cases, we can still calculate the
1816 * RTT correctly when RCV.NXT == Last.ACK.Sent.
1818 if ((to.to_flags & TOF_TS) && SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
1819 SEQ_LEQ(tp->last_ack_sent, (th->th_seq + tlen
1820 + ((thflags & TH_SYN) != 0)
1821 + ((thflags & TH_FIN) != 0)))) {
1822 tp->ts_recent_age = ticks;
1823 tp->ts_recent = to.to_tsval;
1827 * If a SYN is in the window, then this is an
1828 * error and we send an RST and drop the connection.
1830 if (thflags & TH_SYN) {
1831 tp = tcp_drop(tp, ECONNRESET);
1832 rstreason = BANDLIM_UNLIMITED;
1837 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
1838 * flag is on (half-synchronized state), then queue data for
1839 * later processing; else drop segment and return.
1841 if (!(thflags & TH_ACK)) {
1842 if (tp->t_state == TCPS_SYN_RECEIVED ||
1843 (tp->t_flags & TF_NEEDSYN))
1852 switch (tp->t_state) {
1854 * In SYN_RECEIVED state, the ACK acknowledges our SYN, so enter
1855 * ESTABLISHED state and continue processing.
1856 * The ACK was checked above.
1858 case TCPS_SYN_RECEIVED:
1860 tcpstat.tcps_connects++;
1862 /* Do window scaling? */
1863 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
1864 (TF_RCVD_SCALE | TF_REQ_SCALE))
1865 tp->rcv_scale = tp->request_r_scale;
1868 * SYN-RECEIVED -> ESTABLISHED
1869 * SYN-RECEIVED* -> FIN-WAIT-1
1871 tp->t_starttime = ticks;
1872 if (tp->t_flags & TF_NEEDFIN) {
1873 tp->t_state = TCPS_FIN_WAIT_1;
1874 tp->t_flags &= ~TF_NEEDFIN;
1876 tcp_established(tp);
1879 * If segment contains data or ACK, will call tcp_reass()
1880 * later; if not, do so now to pass queued data to user.
1882 if (tlen == 0 && !(thflags & TH_FIN))
1883 tcp_reass(tp, NULL, NULL, NULL);
1887 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1888 * ACKs. If the ack is in the range
1889 * tp->snd_una < th->th_ack <= tp->snd_max
1890 * then advance tp->snd_una to th->th_ack and drop
1891 * data from the retransmission queue. If this ACK reflects
1892 * more up to date window information we update our window information.
1894 case TCPS_ESTABLISHED:
1895 case TCPS_FIN_WAIT_1:
1896 case TCPS_FIN_WAIT_2:
1897 case TCPS_CLOSE_WAIT:
1900 case TCPS_TIME_WAIT:
1902 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
1903 if (TCP_DO_SACK(tp))
1904 tcp_sack_update_scoreboard(tp, &to);
1905 if (!tcp_callout_active(tp, tp->tt_rexmt) ||
1906 th->th_ack != tp->snd_una) {
1907 tcpstat.tcps_rcvdupack++;
1911 if (tlen != 0 || tiwin != tp->snd_wnd) {
1912 if (!tcp_do_rfc3517bis ||
1915 (TOF_SACK | TOF_SACK_REDUNDANT))
1921 * Update window information.
1923 if (tiwin != tp->snd_wnd &&
1924 acceptable_window_update(tp, th, tiwin)) {
1925 /* keep track of pure window updates */
1927 tp->snd_wl2 == th->th_ack &&
1928 tiwin > tp->snd_wnd)
1929 tcpstat.tcps_rcvwinupd++;
1930 tp->snd_wnd = tiwin;
1931 tp->snd_wl1 = th->th_seq;
1932 tp->snd_wl2 = th->th_ack;
1933 if (tp->snd_wnd > tp->max_sndwnd)
1934 tp->max_sndwnd = tp->snd_wnd;
1937 tcpstat.tcps_rcvdupack++;
1940 * We have outstanding data (other than
1941 * a window probe), this is a completely
1942 * duplicate ack (ie, window info didn't
1943 * change), the ack is the biggest we've
1944 * seen and we've seen exactly our rexmt
1945 * threshhold of them, so assume a packet
1946 * has been dropped and retransmit it.
1947 * Kludge snd_nxt & the congestion
1948 * window so we send only this one
1951 if (IN_FASTRECOVERY(tp)) {
1952 if (TCP_DO_SACK(tp)) {
1953 /* No artifical cwnd inflation. */
1954 tcp_sack_rexmt(tp, th);
1957 * Dup acks mean that packets
1958 * have left the network
1959 * (they're now cached at the
1960 * receiver) so bump cwnd by
1961 * the amount in the receiver
1962 * to keep a constant cwnd
1963 * packets in the network.
1965 tp->snd_cwnd += tp->t_maxseg;
1968 } else if (SEQ_LT(th->th_ack, tp->snd_recover)) {
1971 } else if (tcp_ignore_redun_dsack && TCP_DO_SACK(tp) &&
1972 (to.to_flags & (TOF_DSACK | TOF_SACK_REDUNDANT)) ==
1973 (TOF_DSACK | TOF_SACK_REDUNDANT)) {
1975 * If the ACK carries DSACK and other
1976 * SACK blocks carry information that
1977 * we have already known, don't count
1978 * this ACK as duplicate ACK. This
1979 * prevents spurious early retransmit
1980 * and fast retransmit. This also
1981 * meets the requirement of RFC3042
1982 * that new segments should not be sent
1983 * if the SACK blocks do not contain
1984 * new information (XXX we actually
1985 * loosen the requirment that only DSACK
1988 * This kind of ACKs are usually sent
1989 * after spurious retransmit.
1991 /* Do nothing; don't change t_dupacks */
1992 } else if (++tp->t_dupacks == tp->t_rxtthresh) {
1993 tcp_seq old_snd_nxt;
1997 if (tcp_do_eifel_detect &&
1998 (tp->t_flags & TF_RCVD_TSTMP)) {
1999 tcp_save_congestion_state(tp);
2000 tp->t_flags |= TF_FASTREXMT;
2003 * We know we're losing at the current
2004 * window size, so do congestion avoidance:
2005 * set ssthresh to half the current window
2006 * and pull our congestion window back to the
2009 win = min(tp->snd_wnd, tp->snd_cwnd) / 2 /
2013 tp->snd_ssthresh = win * tp->t_maxseg;
2014 ENTER_FASTRECOVERY(tp);
2015 tp->snd_recover = tp->snd_max;
2016 tcp_callout_stop(tp, tp->tt_rexmt);
2018 old_snd_nxt = tp->snd_nxt;
2019 tp->snd_nxt = th->th_ack;
2020 tp->snd_cwnd = tp->t_maxseg;
2022 ++tcpstat.tcps_sndfastrexmit;
2023 tp->snd_cwnd = tp->snd_ssthresh;
2024 tp->rexmt_high = tp->snd_nxt;
2025 tp->t_flags &= ~TF_SACKRESCUED;
2026 if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
2027 tp->snd_nxt = old_snd_nxt;
2028 KASSERT(tp->snd_limited <= 2,
2029 ("tp->snd_limited too big"));
2030 if (TCP_DO_SACK(tp))
2031 tcp_sack_rexmt(tp, th);
2033 tp->snd_cwnd += tp->t_maxseg *
2034 (tp->t_dupacks - tp->snd_limited);
2035 } else if (tcp_do_rfc3517bis && TCP_DO_SACK(tp)) {
2036 if (tcp_rfc3517bis_rxt &&
2037 tcp_sack_islost(&tp->scb, tp->snd_una))
2038 goto fastretransmit;
2039 if (tcp_do_limitedtransmit) {
2040 /* outstanding data */
2042 tp->snd_max - tp->snd_una;
2044 if (!tcp_sack_limitedxmit(tp) &&
2045 need_early_retransmit(tp, ownd)) {
2046 ++tcpstat.tcps_sndearlyrexmit;
2047 tp->t_flags |= TF_EARLYREXMT;
2048 goto fastretransmit;
2051 } else if (tcp_do_limitedtransmit) {
2052 u_long oldcwnd = tp->snd_cwnd;
2053 tcp_seq oldsndmax = tp->snd_max;
2054 tcp_seq oldsndnxt = tp->snd_nxt;
2055 /* outstanding data */
2056 uint32_t ownd = tp->snd_max - tp->snd_una;
2059 KASSERT(tp->t_dupacks == 1 ||
2061 ("dupacks not 1 or 2"));
2062 if (tp->t_dupacks == 1)
2063 tp->snd_limited = 0;
2064 tp->snd_nxt = tp->snd_max;
2065 tp->snd_cwnd = ownd +
2066 (tp->t_dupacks - tp->snd_limited) *
2070 if (SEQ_LT(oldsndnxt, oldsndmax)) {
2071 KASSERT(SEQ_GEQ(oldsndnxt, tp->snd_una),
2072 ("snd_una moved in other threads"));
2073 tp->snd_nxt = oldsndnxt;
2075 tp->snd_cwnd = oldcwnd;
2076 sent = tp->snd_max - oldsndmax;
2077 if (sent > tp->t_maxseg) {
2078 KASSERT((tp->t_dupacks == 2 &&
2079 tp->snd_limited == 0) ||
2080 (sent == tp->t_maxseg + 1 &&
2081 tp->t_flags & TF_SENTFIN),
2083 KASSERT(sent <= tp->t_maxseg * 2,
2084 ("sent too many segments"));
2085 tp->snd_limited = 2;
2086 tcpstat.tcps_sndlimited += 2;
2087 } else if (sent > 0) {
2089 ++tcpstat.tcps_sndlimited;
2090 } else if (need_early_retransmit(tp, ownd)) {
2091 ++tcpstat.tcps_sndearlyrexmit;
2092 tp->t_flags |= TF_EARLYREXMT;
2093 goto fastretransmit;
2102 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una"));
2104 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2106 * Detected optimistic ACK attack.
2107 * Force slow-start to de-synchronize attack.
2109 tp->snd_cwnd = tp->t_maxseg;
2112 tcpstat.tcps_rcvacktoomuch++;
2116 * If we reach this point, ACK is not a duplicate,
2117 * i.e., it ACKs something we sent.
2119 if (tp->t_flags & TF_NEEDSYN) {
2121 * T/TCP: Connection was half-synchronized, and our
2122 * SYN has been ACK'd (so connection is now fully
2123 * synchronized). Go to non-starred state,
2124 * increment snd_una for ACK of SYN, and check if
2125 * we can do window scaling.
2127 tp->t_flags &= ~TF_NEEDSYN;
2129 /* Do window scaling? */
2130 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
2131 (TF_RCVD_SCALE | TF_REQ_SCALE))
2132 tp->rcv_scale = tp->request_r_scale;
2136 acked = th->th_ack - tp->snd_una;
2137 tcpstat.tcps_rcvackpack++;
2138 tcpstat.tcps_rcvackbyte += acked;
2140 if (tcp_do_eifel_detect && acked > 0 &&
2141 (to.to_flags & TOF_TS) && (to.to_tsecr != 0) &&
2142 (tp->t_flags & TF_FIRSTACCACK)) {
2143 /* Eifel detection applicable. */
2144 if (to.to_tsecr < tp->t_rexmtTS) {
2145 ++tcpstat.tcps_eifeldetected;
2146 tcp_revert_congestion_state(tp);
2147 if (tp->t_rxtshift != 1 ||
2148 ticks >= tp->t_badrxtwin)
2149 ++tcpstat.tcps_rttcantdetect;
2151 } else if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) {
2153 * If we just performed our first retransmit,
2154 * and the ACK arrives within our recovery window,
2155 * then it was a mistake to do the retransmit
2156 * in the first place. Recover our original cwnd
2157 * and ssthresh, and proceed to transmit where we
2160 tcp_revert_congestion_state(tp);
2161 ++tcpstat.tcps_rttdetected;
2165 * If we have a timestamp reply, update smoothed
2166 * round trip time. If no timestamp is present but
2167 * transmit timer is running and timed sequence
2168 * number was acked, update smoothed round trip time.
2169 * Since we now have an rtt measurement, cancel the
2170 * timer backoff (cf., Phil Karn's retransmit alg.).
2171 * Recompute the initial retransmit timer.
2173 * Some machines (certain windows boxes) send broken
2174 * timestamp replies during the SYN+ACK phase, ignore
2177 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0))
2178 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1, th->th_ack);
2179 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq))
2180 tcp_xmit_timer(tp, ticks - tp->t_rtttime, th->th_ack);
2181 tcp_xmit_bandwidth_limit(tp, th->th_ack);
2184 * If no data (only SYN) was ACK'd,
2185 * skip rest of ACK processing.
2190 /* Stop looking for an acceptable ACK since one was received. */
2191 tp->t_flags &= ~(TF_FIRSTACCACK | TF_FASTREXMT | TF_EARLYREXMT);
2193 if (acked > so->so_snd.ssb_cc) {
2194 tp->snd_wnd -= so->so_snd.ssb_cc;
2195 sbdrop(&so->so_snd.sb, (int)so->so_snd.ssb_cc);
2196 ourfinisacked = TRUE;
2198 sbdrop(&so->so_snd.sb, acked);
2199 tp->snd_wnd -= acked;
2200 ourfinisacked = FALSE;
2205 * Update window information.
2207 if (acceptable_window_update(tp, th, tiwin)) {
2208 /* keep track of pure window updates */
2209 if (tlen == 0 && tp->snd_wl2 == th->th_ack &&
2210 tiwin > tp->snd_wnd)
2211 tcpstat.tcps_rcvwinupd++;
2212 tp->snd_wnd = tiwin;
2213 tp->snd_wl1 = th->th_seq;
2214 tp->snd_wl2 = th->th_ack;
2215 if (tp->snd_wnd > tp->max_sndwnd)
2216 tp->max_sndwnd = tp->snd_wnd;
2220 tp->snd_una = th->th_ack;
2221 if (TCP_DO_SACK(tp))
2222 tcp_sack_update_scoreboard(tp, &to);
2223 if (IN_FASTRECOVERY(tp)) {
2224 if (SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2225 EXIT_FASTRECOVERY(tp);
2228 * If the congestion window was inflated
2229 * to account for the other side's
2230 * cached packets, retract it.
2232 if (!TCP_DO_SACK(tp))
2233 tp->snd_cwnd = tp->snd_ssthresh;
2236 * Window inflation should have left us
2237 * with approximately snd_ssthresh outstanding
2238 * data. But, in case we would be inclined
2239 * to send a burst, better do it using
2242 if (SEQ_GT(th->th_ack + tp->snd_cwnd,
2243 tp->snd_max + 2 * tp->t_maxseg))
2245 (tp->snd_max - tp->snd_una) +
2250 if (TCP_DO_SACK(tp)) {
2251 tp->snd_max_rexmt = tp->snd_max;
2252 tcp_sack_rexmt(tp, th);
2254 tcp_newreno_partial_ack(tp, th, acked);
2260 * Open the congestion window. When in slow-start,
2261 * open exponentially: maxseg per packet. Otherwise,
2262 * open linearly: maxseg per window.
2264 if (tp->snd_cwnd <= tp->snd_ssthresh) {
2266 (SEQ_LT(tp->snd_nxt, tp->snd_max) ?
2267 tp->t_maxseg : 2 * tp->t_maxseg);
2270 tp->snd_cwnd += tcp_do_abc ?
2271 min(acked, abc_sslimit) : tp->t_maxseg;
2273 /* linear increase */
2274 tp->snd_wacked += tcp_do_abc ? acked :
2276 if (tp->snd_wacked >= tp->snd_cwnd) {
2277 tp->snd_wacked -= tp->snd_cwnd;
2278 tp->snd_cwnd += tp->t_maxseg;
2281 tp->snd_cwnd = min(tp->snd_cwnd,
2282 TCP_MAXWIN << tp->snd_scale);
2283 tp->snd_recover = th->th_ack - 1;
2285 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2286 tp->snd_nxt = tp->snd_una;
2289 * If all outstanding data is acked, stop retransmit
2290 * timer and remember to restart (more output or persist).
2291 * If there is more data to be acked, restart retransmit
2292 * timer, using current (possibly backed-off) value.
2294 if (th->th_ack == tp->snd_max) {
2295 tcp_callout_stop(tp, tp->tt_rexmt);
2297 } else if (!tcp_callout_active(tp, tp->tt_persist)) {
2298 tcp_callout_reset(tp, tp->tt_rexmt, tp->t_rxtcur,
2302 switch (tp->t_state) {
2304 * In FIN_WAIT_1 STATE in addition to the processing
2305 * for the ESTABLISHED state if our FIN is now acknowledged
2306 * then enter FIN_WAIT_2.
2308 case TCPS_FIN_WAIT_1:
2309 if (ourfinisacked) {
2311 * If we can't receive any more
2312 * data, then closing user can proceed.
2313 * Starting the timer is contrary to the
2314 * specification, but if we don't get a FIN
2315 * we'll hang forever.
2317 if (so->so_state & SS_CANTRCVMORE) {
2318 soisdisconnected(so);
2319 tcp_callout_reset(tp, tp->tt_2msl,
2320 tp->t_maxidle, tcp_timer_2msl);
2322 tp->t_state = TCPS_FIN_WAIT_2;
2327 * In CLOSING STATE in addition to the processing for
2328 * the ESTABLISHED state if the ACK acknowledges our FIN
2329 * then enter the TIME-WAIT state, otherwise ignore
2333 if (ourfinisacked) {
2334 tp->t_state = TCPS_TIME_WAIT;
2335 tcp_canceltimers(tp);
2336 tcp_callout_reset(tp, tp->tt_2msl,
2337 2 * tcp_rmx_msl(tp),
2339 soisdisconnected(so);
2344 * In LAST_ACK, we may still be waiting for data to drain
2345 * and/or to be acked, as well as for the ack of our FIN.
2346 * If our FIN is now acknowledged, delete the TCB,
2347 * enter the closed state and return.
2350 if (ourfinisacked) {
2357 * In TIME_WAIT state the only thing that should arrive
2358 * is a retransmission of the remote FIN. Acknowledge
2359 * it and restart the finack timer.
2361 case TCPS_TIME_WAIT:
2362 tcp_callout_reset(tp, tp->tt_2msl, 2 * tcp_rmx_msl(tp),
2370 * Update window information.
2371 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2373 if ((thflags & TH_ACK) &&
2374 acceptable_window_update(tp, th, tiwin)) {
2375 /* keep track of pure window updates */
2376 if (tlen == 0 && tp->snd_wl2 == th->th_ack &&
2377 tiwin > tp->snd_wnd)
2378 tcpstat.tcps_rcvwinupd++;
2379 tp->snd_wnd = tiwin;
2380 tp->snd_wl1 = th->th_seq;
2381 tp->snd_wl2 = th->th_ack;
2382 if (tp->snd_wnd > tp->max_sndwnd)
2383 tp->max_sndwnd = tp->snd_wnd;
2388 * Process segments with URG.
2390 if ((thflags & TH_URG) && th->th_urp &&
2391 !TCPS_HAVERCVDFIN(tp->t_state)) {
2393 * This is a kludge, but if we receive and accept
2394 * random urgent pointers, we'll crash in
2395 * soreceive. It's hard to imagine someone
2396 * actually wanting to send this much urgent data.
2398 if (th->th_urp + so->so_rcv.ssb_cc > sb_max) {
2399 th->th_urp = 0; /* XXX */
2400 thflags &= ~TH_URG; /* XXX */
2401 goto dodata; /* XXX */
2404 * If this segment advances the known urgent pointer,
2405 * then mark the data stream. This should not happen
2406 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2407 * a FIN has been received from the remote side.
2408 * In these states we ignore the URG.
2410 * According to RFC961 (Assigned Protocols),
2411 * the urgent pointer points to the last octet
2412 * of urgent data. We continue, however,
2413 * to consider it to indicate the first octet
2414 * of data past the urgent section as the original
2415 * spec states (in one of two places).
2417 if (SEQ_GT(th->th_seq + th->th_urp, tp->rcv_up)) {
2418 tp->rcv_up = th->th_seq + th->th_urp;
2419 so->so_oobmark = so->so_rcv.ssb_cc +
2420 (tp->rcv_up - tp->rcv_nxt) - 1;
2421 if (so->so_oobmark == 0)
2422 sosetstate(so, SS_RCVATMARK);
2424 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2427 * Remove out of band data so doesn't get presented to user.
2428 * This can happen independent of advancing the URG pointer,
2429 * but if two URG's are pending at once, some out-of-band
2430 * data may creep in... ick.
2432 if (th->th_urp <= (u_long)tlen &&
2433 !(so->so_options & SO_OOBINLINE)) {
2434 /* hdr drop is delayed */
2435 tcp_pulloutofband(so, th, m, drop_hdrlen);
2439 * If no out of band data is expected,
2440 * pull receive urgent pointer along
2441 * with the receive window.
2443 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2444 tp->rcv_up = tp->rcv_nxt;
2449 * Process the segment text, merging it into the TCP sequencing queue,
2450 * and arranging for acknowledgment of receipt if necessary.
2451 * This process logically involves adjusting tp->rcv_wnd as data
2452 * is presented to the user (this happens in tcp_usrreq.c,
2453 * case PRU_RCVD). If a FIN has already been received on this
2454 * connection then we just ignore the text.
2456 if ((tlen || (thflags & TH_FIN)) && !TCPS_HAVERCVDFIN(tp->t_state)) {
2457 m_adj(m, drop_hdrlen); /* delayed header drop */
2459 * Insert segment which includes th into TCP reassembly queue
2460 * with control block tp. Set thflags to whether reassembly now
2461 * includes a segment with FIN. This handles the common case
2462 * inline (segment is the next to be received on an established
2463 * connection, and the queue is empty), avoiding linkage into
2464 * and removal from the queue and repetition of various
2466 * Set DELACK for segments received in order, but ack
2467 * immediately when segments are out of order (so
2468 * fast retransmit can work).
2470 if (th->th_seq == tp->rcv_nxt &&
2471 LIST_EMPTY(&tp->t_segq) &&
2472 TCPS_HAVEESTABLISHED(tp->t_state)) {
2473 if (DELAY_ACK(tp)) {
2474 tcp_callout_reset(tp, tp->tt_delack,
2475 tcp_delacktime, tcp_timer_delack);
2477 tp->t_flags |= TF_ACKNOW;
2479 tp->rcv_nxt += tlen;
2480 thflags = th->th_flags & TH_FIN;
2481 tcpstat.tcps_rcvpack++;
2482 tcpstat.tcps_rcvbyte += tlen;
2484 if (so->so_state & SS_CANTRCVMORE) {
2487 lwkt_gettoken(&so->so_rcv.ssb_token);
2488 ssb_appendstream(&so->so_rcv, m);
2489 lwkt_reltoken(&so->so_rcv.ssb_token);
2493 if (!(tp->t_flags & TF_DUPSEG)) {
2494 /* Initialize SACK report block. */
2495 tp->reportblk.rblk_start = th->th_seq;
2496 tp->reportblk.rblk_end = TCP_SACK_BLKEND(
2497 th->th_seq + tlen, thflags);
2499 thflags = tcp_reass(tp, th, &tlen, m);
2500 tp->t_flags |= TF_ACKNOW;
2504 * Note the amount of data that peer has sent into
2505 * our window, in order to estimate the sender's
2508 len = so->so_rcv.ssb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2515 * If FIN is received ACK the FIN and let the user know
2516 * that the connection is closing.
2518 if (thflags & TH_FIN) {
2519 if (!TCPS_HAVERCVDFIN(tp->t_state)) {
2522 * If connection is half-synchronized
2523 * (ie NEEDSYN flag on) then delay ACK,
2524 * so it may be piggybacked when SYN is sent.
2525 * Otherwise, since we received a FIN then no
2526 * more input can be expected, send ACK now.
2528 if (DELAY_ACK(tp) && (tp->t_flags & TF_NEEDSYN)) {
2529 tcp_callout_reset(tp, tp->tt_delack,
2530 tcp_delacktime, tcp_timer_delack);
2532 tp->t_flags |= TF_ACKNOW;
2537 switch (tp->t_state) {
2539 * In SYN_RECEIVED and ESTABLISHED STATES
2540 * enter the CLOSE_WAIT state.
2542 case TCPS_SYN_RECEIVED:
2543 tp->t_starttime = ticks;
2545 case TCPS_ESTABLISHED:
2546 tp->t_state = TCPS_CLOSE_WAIT;
2550 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2551 * enter the CLOSING state.
2553 case TCPS_FIN_WAIT_1:
2554 tp->t_state = TCPS_CLOSING;
2558 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2559 * starting the time-wait timer, turning off the other
2562 case TCPS_FIN_WAIT_2:
2563 tp->t_state = TCPS_TIME_WAIT;
2564 tcp_canceltimers(tp);
2565 tcp_callout_reset(tp, tp->tt_2msl, 2 * tcp_rmx_msl(tp),
2567 soisdisconnected(so);
2571 * In TIME_WAIT state restart the 2 MSL time_wait timer.
2573 case TCPS_TIME_WAIT:
2574 tcp_callout_reset(tp, tp->tt_2msl, 2 * tcp_rmx_msl(tp),
2581 if (so->so_options & SO_DEBUG)
2582 tcp_trace(TA_INPUT, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2586 * Return any desired output.
2588 if (needoutput || (tp->t_flags & TF_ACKNOW))
2590 tcp_sack_report_cleanup(tp);
2591 return(IPPROTO_DONE);
2595 * Generate an ACK dropping incoming segment if it occupies
2596 * sequence space, where the ACK reflects our state.
2598 * We can now skip the test for the RST flag since all
2599 * paths to this code happen after packets containing
2600 * RST have been dropped.
2602 * In the SYN-RECEIVED state, don't send an ACK unless the
2603 * segment we received passes the SYN-RECEIVED ACK test.
2604 * If it fails send a RST. This breaks the loop in the
2605 * "LAND" DoS attack, and also prevents an ACK storm
2606 * between two listening ports that have been sent forged
2607 * SYN segments, each with the source address of the other.
2609 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
2610 (SEQ_GT(tp->snd_una, th->th_ack) ||
2611 SEQ_GT(th->th_ack, tp->snd_max)) ) {
2612 rstreason = BANDLIM_RST_OPENPORT;
2616 if (so->so_options & SO_DEBUG)
2617 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2620 tp->t_flags |= TF_ACKNOW;
2622 tcp_sack_report_cleanup(tp);
2623 return(IPPROTO_DONE);
2627 * Generate a RST, dropping incoming segment.
2628 * Make ACK acceptable to originator of segment.
2629 * Don't bother to respond if destination was broadcast/multicast.
2631 if ((thflags & TH_RST) || m->m_flags & (M_BCAST | M_MCAST))
2634 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
2635 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
2638 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
2639 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
2640 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
2641 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2644 /* IPv6 anycast check is done at tcp6_input() */
2647 * Perform bandwidth limiting.
2650 if (badport_bandlim(rstreason) < 0)
2655 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2656 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2658 if (thflags & TH_ACK)
2659 /* mtod() below is safe as long as hdr dropping is delayed */
2660 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack,
2663 if (thflags & TH_SYN)
2665 /* mtod() below is safe as long as hdr dropping is delayed */
2666 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq + tlen,
2667 (tcp_seq)0, TH_RST | TH_ACK);
2670 tcp_sack_report_cleanup(tp);
2671 return(IPPROTO_DONE);
2675 * Drop space held by incoming segment and return.
2678 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2679 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2683 tcp_sack_report_cleanup(tp);
2684 return(IPPROTO_DONE);
2688 * Parse TCP options and place in tcpopt.
2691 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, boolean_t is_syn,
2697 for (; cnt > 0; cnt -= optlen, cp += optlen) {
2699 if (opt == TCPOPT_EOL)
2701 if (opt == TCPOPT_NOP)
2707 if (optlen < 2 || optlen > cnt)
2712 if (optlen != TCPOLEN_MAXSEG)
2716 to->to_flags |= TOF_MSS;
2717 bcopy(cp + 2, &to->to_mss, sizeof to->to_mss);
2718 to->to_mss = ntohs(to->to_mss);
2721 if (optlen != TCPOLEN_WINDOW)
2725 to->to_flags |= TOF_SCALE;
2726 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
2728 case TCPOPT_TIMESTAMP:
2729 if (optlen != TCPOLEN_TIMESTAMP)
2731 to->to_flags |= TOF_TS;
2732 bcopy(cp + 2, &to->to_tsval, sizeof to->to_tsval);
2733 to->to_tsval = ntohl(to->to_tsval);
2734 bcopy(cp + 6, &to->to_tsecr, sizeof to->to_tsecr);
2735 to->to_tsecr = ntohl(to->to_tsecr);
2737 * If echoed timestamp is later than the current time,
2738 * fall back to non RFC1323 RTT calculation.
2740 if (to->to_tsecr != 0 && TSTMP_GT(to->to_tsecr, ticks))
2743 case TCPOPT_SACK_PERMITTED:
2744 if (optlen != TCPOLEN_SACK_PERMITTED)
2748 to->to_flags |= TOF_SACK_PERMITTED;
2751 if ((optlen - 2) & 0x07) /* not multiple of 8 */
2753 to->to_nsackblocks = (optlen - 2) / 8;
2754 to->to_sackblocks = (struct raw_sackblock *) (cp + 2);
2755 to->to_flags |= TOF_SACK;
2756 for (i = 0; i < to->to_nsackblocks; i++) {
2757 struct raw_sackblock *r = &to->to_sackblocks[i];
2759 r->rblk_start = ntohl(r->rblk_start);
2760 r->rblk_end = ntohl(r->rblk_end);
2762 if (SEQ_LEQ(r->rblk_end, r->rblk_start)) {
2764 * Invalid SACK block; discard all
2767 tcpstat.tcps_rcvbadsackopt++;
2768 to->to_nsackblocks = 0;
2769 to->to_sackblocks = NULL;
2770 to->to_flags &= ~TOF_SACK;
2774 if ((to->to_flags & TOF_SACK) &&
2775 tcp_sack_ndsack_blocks(to->to_sackblocks,
2776 to->to_nsackblocks, ack))
2777 to->to_flags |= TOF_DSACK;
2779 #ifdef TCP_SIGNATURE
2781 * XXX In order to reply to a host which has set the
2782 * TCP_SIGNATURE option in its initial SYN, we have to
2783 * record the fact that the option was observed here
2784 * for the syncache code to perform the correct response.
2786 case TCPOPT_SIGNATURE:
2787 if (optlen != TCPOLEN_SIGNATURE)
2789 to->to_flags |= (TOF_SIGNATURE | TOF_SIGLEN);
2791 #endif /* TCP_SIGNATURE */
2799 * Pull out of band byte out of a segment so
2800 * it doesn't appear in the user's data queue.
2801 * It is still reflected in the segment length for
2802 * sequencing purposes.
2803 * "off" is the delayed to be dropped hdrlen.
2806 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, int off)
2808 int cnt = off + th->th_urp - 1;
2811 if (m->m_len > cnt) {
2812 char *cp = mtod(m, caddr_t) + cnt;
2813 struct tcpcb *tp = sototcpcb(so);
2816 tp->t_oobflags |= TCPOOB_HAVEDATA;
2817 bcopy(cp + 1, cp, m->m_len - cnt - 1);
2819 if (m->m_flags & M_PKTHDR)
2828 panic("tcp_pulloutofband");
2832 * Collect new round-trip time estimate
2833 * and update averages and current timeout.
2836 tcp_xmit_timer(struct tcpcb *tp, int rtt, tcp_seq ack)
2840 tcpstat.tcps_rttupdated++;
2842 if ((tp->t_flags & TF_REBASERTO) && SEQ_GT(ack, tp->snd_max_prev)) {
2843 #ifdef DEBUG_EIFEL_RESPONSE
2844 kprintf("srtt/rttvar, prev %d/%d, cur %d/%d, ",
2845 tp->t_srtt_prev, tp->t_rttvar_prev,
2846 tp->t_srtt, tp->t_rttvar);
2849 tcpstat.tcps_eifelresponse++;
2851 tp->t_flags &= ~TF_REBASERTO;
2852 tp->t_srtt = max(tp->t_srtt_prev, (rtt << TCP_RTT_SHIFT));
2853 tp->t_rttvar = max(tp->t_rttvar_prev,
2854 (rtt << (TCP_RTTVAR_SHIFT - 1)));
2855 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
2856 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2858 #ifdef DEBUG_EIFEL_RESPONSE
2859 kprintf("new %d/%d ", tp->t_srtt, tp->t_rttvar);
2861 } else if (tp->t_srtt != 0) {
2865 * srtt is stored as fixed point with 5 bits after the
2866 * binary point (i.e., scaled by 8). The following magic
2867 * is equivalent to the smoothing algorithm in rfc793 with
2868 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2869 * point). Adjust rtt to origin 0.
2871 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
2872 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
2874 if ((tp->t_srtt += delta) <= 0)
2878 * We accumulate a smoothed rtt variance (actually, a
2879 * smoothed mean difference), then set the retransmit
2880 * timer to smoothed rtt + 4 times the smoothed variance.
2881 * rttvar is stored as fixed point with 4 bits after the
2882 * binary point (scaled by 16). The following is
2883 * equivalent to rfc793 smoothing with an alpha of .75
2884 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
2885 * rfc793's wired-in beta.
2889 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
2890 if ((tp->t_rttvar += delta) <= 0)
2892 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
2893 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2896 * No rtt measurement yet - use the unsmoothed rtt.
2897 * Set the variance to half the rtt (so our first
2898 * retransmit happens at 3*rtt).
2900 tp->t_srtt = rtt << TCP_RTT_SHIFT;
2901 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
2902 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2907 #ifdef DEBUG_EIFEL_RESPONSE
2909 kprintf("| rxtcur prev %d, old %d, ",
2910 tp->t_rxtcur_prev, tp->t_rxtcur);
2915 * the retransmit should happen at rtt + 4 * rttvar.
2916 * Because of the way we do the smoothing, srtt and rttvar
2917 * will each average +1/2 tick of bias. When we compute
2918 * the retransmit timer, we want 1/2 tick of rounding and
2919 * 1 extra tick because of +-1/2 tick uncertainty in the
2920 * firing of the timer. The bias will give us exactly the
2921 * 1.5 tick we need. But, because the bias is
2922 * statistical, we have to test that we don't drop below
2923 * the minimum feasible timer (which is 2 ticks).
2925 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
2926 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
2929 if (tp->t_rxtcur < tp->t_rxtcur_prev + tcp_eifel_rtoinc) {
2931 * RFC4015 requires that the new RTO is at least
2932 * 2*G (tcp_eifel_rtoinc) greater then the RTO
2933 * (t_rxtcur_prev) when the spurious retransmit
2936 * The above condition could be true, if the SRTT
2937 * and RTTVAR used to calculate t_rxtcur_prev
2938 * resulted in a value less than t_rttmin. So
2939 * simply increasing SRTT by tcp_eifel_rtoinc when
2940 * preparing for the Eifel response in
2941 * tcp_save_congestion_state() could not ensure
2942 * that the new RTO will be tcp_eifel_rtoinc greater
2945 tp->t_rxtcur = tp->t_rxtcur_prev + tcp_eifel_rtoinc;
2947 #ifdef DEBUG_EIFEL_RESPONSE
2948 kprintf("new %d\n", tp->t_rxtcur);
2953 * We received an ack for a packet that wasn't retransmitted;
2954 * it is probably safe to discard any error indications we've
2955 * received recently. This isn't quite right, but close enough
2956 * for now (a route might have failed after we sent a segment,
2957 * and the return path might not be symmetrical).
2959 tp->t_softerror = 0;
2963 * Determine a reasonable value for maxseg size.
2964 * If the route is known, check route for mtu.
2965 * If none, use an mss that can be handled on the outgoing
2966 * interface without forcing IP to fragment; if bigger than
2967 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2968 * to utilize large mbufs. If no route is found, route has no mtu,
2969 * or the destination isn't local, use a default, hopefully conservative
2970 * size (usually 512 or the default IP max size, but no more than the mtu
2971 * of the interface), as we can't discover anything about intervening
2972 * gateways or networks. We also initialize the congestion/slow start
2973 * window to be a single segment if the destination isn't local.
2974 * While looking at the routing entry, we also initialize other path-dependent
2975 * parameters from pre-set or cached values in the routing entry.
2977 * Also take into account the space needed for options that we
2978 * send regularly. Make maxseg shorter by that amount to assure
2979 * that we can send maxseg amount of data even when the options
2980 * are present. Store the upper limit of the length of options plus
2983 * NOTE that this routine is only called when we process an incoming
2984 * segment, for outgoing segments only tcp_mssopt is called.
2987 tcp_mss(struct tcpcb *tp, int offer)
2993 struct inpcb *inp = tp->t_inpcb;
2996 boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) ? TRUE : FALSE);
2997 size_t min_protoh = isipv6 ?
2998 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
2999 sizeof(struct tcpiphdr);
3001 const boolean_t isipv6 = FALSE;
3002 const size_t min_protoh = sizeof(struct tcpiphdr);
3006 rt = tcp_rtlookup6(&inp->inp_inc);
3008 rt = tcp_rtlookup(&inp->inp_inc);
3010 tp->t_maxopd = tp->t_maxseg =
3011 (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
3015 so = inp->inp_socket;
3018 * Offer == 0 means that there was no MSS on the SYN segment,
3019 * in this case we use either the interface mtu or tcp_mssdflt.
3021 * An offer which is too large will be cut down later.
3025 if (in6_localaddr(&inp->in6p_faddr)) {
3026 offer = ND_IFINFO(rt->rt_ifp)->linkmtu -
3029 offer = tcp_v6mssdflt;
3032 if (in_localaddr(inp->inp_faddr))
3033 offer = ifp->if_mtu - min_protoh;
3035 offer = tcp_mssdflt;
3040 * Prevent DoS attack with too small MSS. Round up
3041 * to at least minmss.
3043 * Sanity check: make sure that maxopd will be large
3044 * enough to allow some data on segments even is the
3045 * all the option space is used (40bytes). Otherwise
3046 * funny things may happen in tcp_output.
3048 offer = max(offer, tcp_minmss);
3049 offer = max(offer, 64);
3051 rt->rt_rmx.rmx_mssopt = offer;
3054 * While we're here, check if there's an initial rtt
3055 * or rttvar. Convert from the route-table units
3056 * to scaled multiples of the slow timeout timer.
3058 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
3060 * XXX the lock bit for RTT indicates that the value
3061 * is also a minimum value; this is subject to time.
3063 if (rt->rt_rmx.rmx_locks & RTV_RTT)
3064 tp->t_rttmin = rtt / (RTM_RTTUNIT / hz);
3065 tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE));
3066 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
3067 tcpstat.tcps_usedrtt++;
3068 if (rt->rt_rmx.rmx_rttvar) {
3069 tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
3070 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE));
3071 tcpstat.tcps_usedrttvar++;
3073 /* default variation is +- 1 rtt */
3075 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
3077 TCPT_RANGESET(tp->t_rxtcur,
3078 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
3079 tp->t_rttmin, TCPTV_REXMTMAX);
3083 * if there's an mtu associated with the route, use it
3084 * else, use the link mtu. Take the smaller of mss or offer
3087 if (rt->rt_rmx.rmx_mtu) {
3088 mss = rt->rt_rmx.rmx_mtu - min_protoh;
3091 mss = ND_IFINFO(rt->rt_ifp)->linkmtu - min_protoh;
3093 mss = ifp->if_mtu - min_protoh;
3095 mss = min(mss, offer);
3098 * maxopd stores the maximum length of data AND options
3099 * in a segment; maxseg is the amount of data in a normal
3100 * segment. We need to store this value (maxopd) apart
3101 * from maxseg, because now every segment carries options
3102 * and thus we normally have somewhat less data in segments.
3106 if ((tp->t_flags & (TF_REQ_TSTMP | TF_NOOPT)) == TF_REQ_TSTMP &&
3107 ((tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
3108 mss -= TCPOLEN_TSTAMP_APPA;
3110 #if (MCLBYTES & (MCLBYTES - 1)) == 0
3112 mss &= ~(MCLBYTES-1);
3115 mss = mss / MCLBYTES * MCLBYTES;
3118 * If there's a pipesize, change the socket buffer
3119 * to that size. Make the socket buffers an integral
3120 * number of mss units; if the mss is larger than
3121 * the socket buffer, decrease the mss.
3124 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0)
3126 bufsize = so->so_snd.ssb_hiwat;
3130 bufsize = roundup(bufsize, mss);
3131 if (bufsize > sb_max)
3133 if (bufsize > so->so_snd.ssb_hiwat)
3134 ssb_reserve(&so->so_snd, bufsize, so, NULL);
3139 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0)
3141 bufsize = so->so_rcv.ssb_hiwat;
3142 if (bufsize > mss) {
3143 bufsize = roundup(bufsize, mss);
3144 if (bufsize > sb_max)
3146 if (bufsize > so->so_rcv.ssb_hiwat) {
3147 lwkt_gettoken(&so->so_rcv.ssb_token);
3148 ssb_reserve(&so->so_rcv, bufsize, so, NULL);
3149 lwkt_reltoken(&so->so_rcv.ssb_token);
3154 * Set the slow-start flight size
3156 * NOTE: t_maxseg must have been configured!
3158 tp->snd_cwnd = tcp_initial_window(tp);
3160 if (rt->rt_rmx.rmx_ssthresh) {
3162 * There's some sort of gateway or interface
3163 * buffer limit on the path. Use this to set
3164 * the slow start threshhold, but set the
3165 * threshold to no less than 2*mss.
3167 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
3168 tcpstat.tcps_usedssthresh++;
3173 * Determine the MSS option to send on an outgoing SYN.
3176 tcp_mssopt(struct tcpcb *tp)
3181 ((tp->t_inpcb->inp_vflag & INP_IPV6) ? TRUE : FALSE);
3182 int min_protoh = isipv6 ?
3183 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
3184 sizeof(struct tcpiphdr);
3186 const boolean_t isipv6 = FALSE;
3187 const size_t min_protoh = sizeof(struct tcpiphdr);
3191 rt = tcp_rtlookup6(&tp->t_inpcb->inp_inc);
3193 rt = tcp_rtlookup(&tp->t_inpcb->inp_inc);
3195 return (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
3197 return (rt->rt_ifp->if_mtu - min_protoh);
3201 * When a partial ack arrives, force the retransmission of the
3202 * next unacknowledged segment. Do not exit Fast Recovery.
3204 * Implement the Slow-but-Steady variant of NewReno by restarting the
3205 * the retransmission timer. Turn it off here so it can be restarted
3206 * later in tcp_output().
3209 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th, int acked)
3211 tcp_seq old_snd_nxt = tp->snd_nxt;
3212 u_long ocwnd = tp->snd_cwnd;
3214 tcp_callout_stop(tp, tp->tt_rexmt);
3216 tp->snd_nxt = th->th_ack;
3217 /* Set snd_cwnd to one segment beyond acknowledged offset. */
3218 tp->snd_cwnd = tp->t_maxseg;
3219 tp->t_flags |= TF_ACKNOW;
3221 if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
3222 tp->snd_nxt = old_snd_nxt;
3223 /* partial window deflation */
3225 tp->snd_cwnd = ocwnd - acked + tp->t_maxseg;
3227 tp->snd_cwnd = tp->t_maxseg;
3231 * In contrast to the Slow-but-Steady NewReno variant,
3232 * we do not reset the retransmission timer for SACK retransmissions,
3233 * except when retransmitting snd_una.
3236 tcp_sack_rexmt(struct tcpcb *tp, struct tcphdr *th)
3238 tcp_seq old_snd_nxt = tp->snd_nxt;
3239 u_long ocwnd = tp->snd_cwnd;
3241 int nseg = 0; /* consecutive new segments */
3242 int nseg_rexmt = 0; /* retransmitted segments */
3243 #define MAXBURST 4 /* limit burst of new packets on partial ack */
3246 pipe = tcp_sack_compute_pipe(tp);
3247 while ((tcp_seq_diff_t)(ocwnd - pipe) >= (tcp_seq_diff_t)tp->t_maxseg &&
3248 (!tcp_do_smartsack || nseg < MAXBURST)) {
3249 tcp_seq old_snd_max, old_rexmt_high, nextrexmt;
3250 uint32_t sent, seglen;
3254 old_rexmt_high = tp->rexmt_high;
3255 if (!tcp_sack_nextseg(tp, &nextrexmt, &seglen, &rescue)) {
3256 tp->rexmt_high = old_rexmt_high;
3261 * If the next tranmission is a rescue retranmission,
3262 * we check whether we have already sent some data
3263 * (either new segments or retransmitted segments)
3264 * into the the network or not. Since the idea of rescue
3265 * retransmission is to sustain ACK clock, as long as
3266 * some segments are in the network, ACK clock will be
3269 if (rescue && (nseg_rexmt > 0 || nseg > 0)) {
3270 tp->rexmt_high = old_rexmt_high;
3274 if (nextrexmt == tp->snd_max)
3278 tp->snd_nxt = nextrexmt;
3279 tp->snd_cwnd = nextrexmt - tp->snd_una + seglen;
3280 old_snd_max = tp->snd_max;
3281 if (nextrexmt == tp->snd_una)
3282 tcp_callout_stop(tp, tp->tt_rexmt);
3283 error = tcp_output(tp);
3285 tp->rexmt_high = old_rexmt_high;
3288 sent = tp->snd_nxt - nextrexmt;
3290 tp->rexmt_high = old_rexmt_high;
3294 tcpstat.tcps_sndsackpack++;
3295 tcpstat.tcps_sndsackbyte += sent;
3298 tcpstat.tcps_sackrescue++;
3299 tp->rexmt_rescue = tp->snd_nxt;
3300 tp->t_flags |= TF_SACKRESCUED;
3303 if (SEQ_LT(nextrexmt, old_snd_max) &&
3304 SEQ_LT(tp->rexmt_high, tp->snd_nxt)) {
3305 tp->rexmt_high = seq_min(tp->snd_nxt, old_snd_max);
3306 if (tcp_aggressive_rescuesack &&
3307 (tp->t_flags & TF_SACKRESCUED) &&
3308 SEQ_LT(tp->rexmt_rescue, tp->rexmt_high)) {
3309 /* Drag RescueRxt along with HighRxt */
3310 tp->rexmt_rescue = tp->rexmt_high;
3314 if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
3315 tp->snd_nxt = old_snd_nxt;
3316 tp->snd_cwnd = ocwnd;
3320 tcp_sack_limitedxmit(struct tcpcb *tp)
3322 tcp_seq oldsndnxt = tp->snd_nxt;
3323 tcp_seq oldsndmax = tp->snd_max;
3324 u_long ocwnd = tp->snd_cwnd;
3326 boolean_t ret = FALSE;
3328 tp->rexmt_high = tp->snd_una - 1;
3329 pipe = tcp_sack_compute_pipe(tp);
3330 while ((tcp_seq_diff_t)(ocwnd - pipe) >= (tcp_seq_diff_t)tp->t_maxseg) {
3335 next = tp->snd_nxt = tp->snd_max;
3336 tp->snd_cwnd = tp->snd_nxt - tp->snd_una + tp->t_maxseg;
3338 error = tcp_output(tp);
3342 sent = tp->snd_nxt - next;
3346 ++tcpstat.tcps_sndlimited;
3350 if (SEQ_LT(oldsndnxt, oldsndmax)) {
3351 KASSERT(SEQ_GEQ(oldsndnxt, tp->snd_una),
3352 ("snd_una moved in other threads"));
3353 tp->snd_nxt = oldsndnxt;
3355 tp->snd_cwnd = ocwnd;
3361 * Reset idle time and keep-alive timer, typically called when a valid
3362 * tcp packet is received but may also be called when FASTKEEP is set
3363 * to prevent the previous long-timeout from calculating to a drop.
3365 * Only update t_rcvtime for non-SYN packets.
3367 * Handle the case where one side thinks the connection is established
3368 * but the other side has, say, rebooted without cleaning out the
3369 * connection. The SYNs could be construed as an attack and wind
3370 * up ignored, but in case it isn't an attack we can validate the
3371 * connection by forcing a keepalive.
3374 tcp_timer_keep_activity(struct tcpcb *tp, int thflags)
3376 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
3377 if ((thflags & (TH_SYN | TH_ACK)) == TH_SYN) {
3378 tp->t_flags |= TF_KEEPALIVE;
3379 tcp_callout_reset(tp, tp->tt_keep, hz / 2,
3382 tp->t_rcvtime = ticks;
3383 tp->t_flags &= ~TF_KEEPALIVE;
3384 tcp_callout_reset(tp, tp->tt_keep,
3392 tcp_rmx_msl(const struct tcpcb *tp)
3395 struct inpcb *inp = tp->t_inpcb;
3398 boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) ? TRUE : FALSE);
3400 const boolean_t isipv6 = FALSE;
3404 rt = tcp_rtlookup6(&inp->inp_inc);
3406 rt = tcp_rtlookup(&inp->inp_inc);
3407 if (rt == NULL || rt->rt_rmx.rmx_msl == 0)
3410 msl = (rt->rt_rmx.rmx_msl * hz) / 1000;
3418 tcp_established(struct tcpcb *tp)
3420 tp->t_state = TCPS_ESTABLISHED;
3421 tcp_callout_reset(tp, tp->tt_keep, tp->t_keepidle, tcp_timer_keep);
3423 if (tp->t_rxtsyn > 0) {
3426 * "If the timer expires awaiting the ACK of a SYN segment
3427 * and the TCP implementation is using an RTO less than 3
3428 * seconds, the RTO MUST be re-initialized to 3 seconds
3429 * when data transmission begins"
3431 if (tp->t_rxtcur < TCPTV_RTOBASE3)
3432 tp->t_rxtcur = TCPTV_RTOBASE3;