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 < ((tp->t_rxtthresh + 1) * 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 */
368 (TSACK_F_DUPSEG | TSACK_F_ENCLOSESEG);
369 /* enclosing block starts w/ preceding segment */
370 tp->encloseblk.rblk_start = p->tqe_th->th_seq;
372 /* preceding encloses incoming segment */
373 tp->encloseblk.rblk_end = TCP_SACK_BLKEND(
374 p->tqe_th->th_seq + p->tqe_len,
375 p->tqe_th->th_flags);
376 tcpstat.tcps_rcvduppack++;
377 tcpstat.tcps_rcvdupbyte += *tlenp;
380 atomic_add_int(&tcp_reass_qsize, -1);
382 * Try to present any queued data
383 * at the left window edge to the user.
384 * This is needed after the 3-WHS
387 goto present; /* ??? */
392 /* incoming segment end is enclosing block end */
393 tp->encloseblk.rblk_end = TCP_SACK_BLKEND(
394 th->th_seq + *tlenp, th->th_flags);
395 /* trim end of reported D-SACK block */
396 tp->reportblk.rblk_end = th->th_seq;
399 tcpstat.tcps_rcvoopack++;
400 tcpstat.tcps_rcvoobyte += *tlenp;
403 * While we overlap succeeding segments trim them or,
404 * if they are completely covered, dequeue them.
407 tcp_seq_diff_t i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
408 tcp_seq qend = q->tqe_th->th_seq + q->tqe_len;
409 tcp_seq qend_sack = TCP_SACK_BLKEND(qend, q->tqe_th->th_flags);
410 struct tseg_qent *nq;
414 if (!(tp->sack_flags & TSACK_F_DUPSEG)) {
415 /* first time through */
416 tp->sack_flags |= (TSACK_F_DUPSEG | TSACK_F_ENCLOSESEG);
417 tp->encloseblk = tp->reportblk;
418 /* report trailing duplicate D-SACK segment */
419 tp->reportblk.rblk_start = q->tqe_th->th_seq;
421 if ((tp->sack_flags & TSACK_F_ENCLOSESEG) &&
422 SEQ_GT(qend_sack, tp->encloseblk.rblk_end)) {
423 /* extend enclosing block if one exists */
424 tp->encloseblk.rblk_end = qend_sack;
426 if (i < q->tqe_len) {
427 q->tqe_th->th_seq += i;
433 nq = LIST_NEXT(q, tqe_q);
434 LIST_REMOVE(q, tqe_q);
437 atomic_add_int(&tcp_reass_qsize, -1);
441 /* Insert the new segment queue entry into place. */
444 te->tqe_len = *tlenp;
446 /* check if can coalesce with following segment */
447 if (q != NULL && (th->th_seq + *tlenp == q->tqe_th->th_seq)) {
448 tcp_seq tend = te->tqe_th->th_seq + te->tqe_len;
449 tcp_seq tend_sack = TCP_SACK_BLKEND(tend, te->tqe_th->th_flags);
451 te->tqe_len += q->tqe_len;
452 if (q->tqe_th->th_flags & TH_FIN)
453 te->tqe_th->th_flags |= TH_FIN;
454 m_cat(te->tqe_m, q->tqe_m);
455 tp->encloseblk.rblk_end = tend_sack;
457 * When not reporting a duplicate segment, use
458 * the larger enclosing block as the SACK block.
460 if (!(tp->sack_flags & TSACK_F_DUPSEG))
461 tp->reportblk.rblk_end = tend_sack;
462 LIST_REMOVE(q, tqe_q);
464 atomic_add_int(&tcp_reass_qsize, -1);
468 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
470 /* check if can coalesce with preceding segment */
471 if (p->tqe_th->th_seq + p->tqe_len == th->th_seq) {
472 p->tqe_len += te->tqe_len;
473 m_cat(p->tqe_m, te->tqe_m);
474 tp->encloseblk.rblk_start = p->tqe_th->th_seq;
476 * When not reporting a duplicate segment, use
477 * the larger enclosing block as the SACK block.
479 if (!(tp->sack_flags & TSACK_F_DUPSEG))
480 tp->reportblk.rblk_start = p->tqe_th->th_seq;
482 atomic_add_int(&tcp_reass_qsize, -1);
484 LIST_INSERT_AFTER(p, te, tqe_q);
490 * Present data to user, advancing rcv_nxt through
491 * completed sequence space.
493 if (!TCPS_HAVEESTABLISHED(tp->t_state))
495 q = LIST_FIRST(&tp->t_segq);
496 if (q == NULL || q->tqe_th->th_seq != tp->rcv_nxt)
498 tp->rcv_nxt += q->tqe_len;
499 if (!(tp->sack_flags & TSACK_F_DUPSEG)) {
500 /* no SACK block to report since ACK advanced */
501 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
503 /* no enclosing block to report since ACK advanced */
504 tp->sack_flags &= ~TSACK_F_ENCLOSESEG;
505 flags = q->tqe_th->th_flags & TH_FIN;
506 LIST_REMOVE(q, tqe_q);
507 KASSERT(LIST_EMPTY(&tp->t_segq) ||
508 LIST_FIRST(&tp->t_segq)->tqe_th->th_seq != tp->rcv_nxt,
509 ("segment not coalesced"));
510 if (so->so_state & SS_CANTRCVMORE) {
513 lwkt_gettoken(&so->so_rcv.ssb_token);
514 ssb_appendstream(&so->so_rcv, q->tqe_m);
515 lwkt_reltoken(&so->so_rcv.ssb_token);
518 atomic_add_int(&tcp_reass_qsize, -1);
525 * TCP input routine, follows pages 65-76 of the
526 * protocol specification dated September, 1981 very closely.
530 tcp6_input(struct mbuf **mp, int *offp, int proto)
532 struct mbuf *m = *mp;
533 struct in6_ifaddr *ia6;
535 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
538 * draft-itojun-ipv6-tcp-to-anycast
539 * better place to put this in?
541 ia6 = ip6_getdstifaddr(m);
542 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
545 ip6 = mtod(m, struct ip6_hdr *);
546 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
547 offsetof(struct ip6_hdr, ip6_dst));
548 return (IPPROTO_DONE);
551 tcp_input(mp, offp, proto);
552 return (IPPROTO_DONE);
557 tcp_input(struct mbuf **mp, int *offp, int proto)
561 struct ip *ip = NULL;
563 struct inpcb *inp = NULL;
569 struct tcpcb *tp = NULL;
571 struct socket *so = NULL;
573 boolean_t ourfinisacked, needoutput = FALSE;
576 struct tcpopt to; /* options in this segment */
577 struct sockaddr_in *next_hop = NULL;
578 int rstreason; /* For badport_bandlim accounting purposes */
580 struct ip6_hdr *ip6 = NULL;
585 const boolean_t isipv6 = FALSE;
595 tcpstat.tcps_rcvtotal++;
597 if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
600 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
601 KKASSERT(mtag != NULL);
602 next_hop = m_tag_data(mtag);
606 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? TRUE : FALSE;
610 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */
611 ip6 = mtod(m, struct ip6_hdr *);
612 tlen = (sizeof *ip6) + ntohs(ip6->ip6_plen) - off0;
613 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) {
614 tcpstat.tcps_rcvbadsum++;
617 th = (struct tcphdr *)((caddr_t)ip6 + off0);
620 * Be proactive about unspecified IPv6 address in source.
621 * As we use all-zero to indicate unbounded/unconnected pcb,
622 * unspecified IPv6 address can be used to confuse us.
624 * Note that packets with unspecified IPv6 destination is
625 * already dropped in ip6_input.
627 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
633 * Get IP and TCP header together in first mbuf.
634 * Note: IP leaves IP header in first mbuf.
636 if (off0 > sizeof(struct ip)) {
638 off0 = sizeof(struct ip);
640 /* already checked and pulled up in ip_demux() */
641 KASSERT(m->m_len >= sizeof(struct tcpiphdr),
642 ("TCP header not in one mbuf: m->m_len %d", m->m_len));
643 ip = mtod(m, struct ip *);
644 ipov = (struct ipovly *)ip;
645 th = (struct tcphdr *)((caddr_t)ip + off0);
648 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
649 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
650 th->th_sum = m->m_pkthdr.csum_data;
652 th->th_sum = in_pseudo(ip->ip_src.s_addr,
654 htonl(m->m_pkthdr.csum_data +
657 th->th_sum ^= 0xffff;
660 * Checksum extended TCP header and data.
662 len = sizeof(struct ip) + tlen;
663 bzero(ipov->ih_x1, sizeof ipov->ih_x1);
664 ipov->ih_len = (u_short)tlen;
665 ipov->ih_len = htons(ipov->ih_len);
666 th->th_sum = in_cksum(m, len);
669 tcpstat.tcps_rcvbadsum++;
673 /* Re-initialization for later version check */
674 ip->ip_v = IPVERSION;
679 * Check that TCP offset makes sense,
680 * pull out TCP options and adjust length. XXX
682 off = th->th_off << 2;
683 /* already checked and pulled up in ip_demux() */
684 KASSERT(off >= sizeof(struct tcphdr) && off <= tlen,
685 ("bad TCP data offset %d (tlen %d)", off, tlen));
686 tlen -= off; /* tlen is used instead of ti->ti_len */
687 if (off > sizeof(struct tcphdr)) {
689 IP6_EXTHDR_CHECK(m, off0, off, IPPROTO_DONE);
690 ip6 = mtod(m, struct ip6_hdr *);
691 th = (struct tcphdr *)((caddr_t)ip6 + off0);
693 /* already pulled up in ip_demux() */
694 KASSERT(m->m_len >= sizeof(struct ip) + off,
695 ("TCP header and options not in one mbuf: "
696 "m_len %d, off %d", m->m_len, off));
698 optlen = off - sizeof(struct tcphdr);
699 optp = (u_char *)(th + 1);
701 thflags = th->th_flags;
703 #ifdef TCP_DROP_SYNFIN
705 * If the drop_synfin option is enabled, drop all packets with
706 * both the SYN and FIN bits set. This prevents e.g. nmap from
707 * identifying the TCP/IP stack.
709 * This is a violation of the TCP specification.
711 if (drop_synfin && (thflags & (TH_SYN | TH_FIN)) == (TH_SYN | TH_FIN))
716 * Convert TCP protocol specific fields to host format.
718 th->th_seq = ntohl(th->th_seq);
719 th->th_ack = ntohl(th->th_ack);
720 th->th_win = ntohs(th->th_win);
721 th->th_urp = ntohs(th->th_urp);
724 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options,
725 * until after ip6_savecontrol() is called and before other functions
726 * which don't want those proto headers.
727 * Because ip6_savecontrol() is going to parse the mbuf to
728 * search for data to be passed up to user-land, it wants mbuf
729 * parameters to be unchanged.
730 * XXX: the call of ip6_savecontrol() has been obsoleted based on
731 * latest version of the advanced API (20020110).
733 drop_hdrlen = off0 + off;
736 * Locate pcb for segment.
739 /* IPFIREWALL_FORWARD section */
740 if (next_hop != NULL && !isipv6) { /* IPv6 support is not there yet */
742 * Transparently forwarded. Pretend to be the destination.
743 * already got one like this?
745 cpu = mycpu->gd_cpuid;
746 inp = in_pcblookup_hash(&tcbinfo[cpu],
747 ip->ip_src, th->th_sport,
748 ip->ip_dst, th->th_dport,
749 0, m->m_pkthdr.rcvif);
752 * It's new. Try to find the ambushing socket.
756 * The rest of the ipfw code stores the port in
758 * (The IP address is still in network order.)
760 in_port_t dport = next_hop->sin_port ?
761 htons(next_hop->sin_port) :
764 cpu = tcp_addrcpu(ip->ip_src.s_addr, th->th_sport,
765 next_hop->sin_addr.s_addr, dport);
766 inp = in_pcblookup_hash(&tcbinfo[cpu],
767 ip->ip_src, th->th_sport,
768 next_hop->sin_addr, dport,
769 1, m->m_pkthdr.rcvif);
773 inp = in6_pcblookup_hash(&tcbinfo[0],
774 &ip6->ip6_src, th->th_sport,
775 &ip6->ip6_dst, th->th_dport,
776 1, m->m_pkthdr.rcvif);
778 cpu = mycpu->gd_cpuid;
779 inp = in_pcblookup_hash(&tcbinfo[cpu],
780 ip->ip_src, th->th_sport,
781 ip->ip_dst, th->th_dport,
782 1, m->m_pkthdr.rcvif);
787 * If the state is CLOSED (i.e., TCB does not exist) then
788 * all data in the incoming segment is discarded.
789 * If the TCB exists but is in CLOSED state, it is embryonic,
790 * but should either do a listen or a connect soon.
795 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2];
797 char dbuf[sizeof "aaa.bbb.ccc.ddd"];
798 char sbuf[sizeof "aaa.bbb.ccc.ddd"];
802 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst));
805 strcat(sbuf, ip6_sprintf(&ip6->ip6_src));
808 strcpy(dbuf, inet_ntoa(ip->ip_dst));
809 strcpy(sbuf, inet_ntoa(ip->ip_src));
811 switch (log_in_vain) {
813 if (!(thflags & TH_SYN))
817 "Connection attempt to TCP %s:%d "
818 "from %s:%d flags:0x%02x\n",
819 dbuf, ntohs(th->th_dport), sbuf,
820 ntohs(th->th_sport), thflags);
829 if (thflags & TH_SYN)
838 rstreason = BANDLIM_RST_CLOSEDPORT;
844 if (ipsec6_in_reject_so(m, inp->inp_socket)) {
845 ipsec6stat.in_polvio++;
849 if (ipsec4_in_reject_so(m, inp->inp_socket)) {
850 ipsecstat.in_polvio++;
857 if (ipsec6_in_reject(m, inp))
860 if (ipsec4_in_reject(m, inp))
864 /* Check the minimum TTL for socket. */
866 if ((isipv6 ? ip6->ip6_hlim : ip->ip_ttl) < inp->inp_ip_minttl)
872 rstreason = BANDLIM_RST_CLOSEDPORT;
875 if (tp->t_state <= TCPS_CLOSED)
878 so = inp->inp_socket;
881 if (so->so_options & SO_DEBUG) {
882 ostate = tp->t_state;
884 bcopy(ip6, tcp_saveipgen, sizeof(*ip6));
886 bcopy(ip, tcp_saveipgen, sizeof(*ip));
891 bzero(&to, sizeof to);
893 if (so->so_options & SO_ACCEPTCONN) {
894 struct in_conninfo inc;
897 inc.inc_isipv6 = (isipv6 == TRUE);
900 inc.inc6_faddr = ip6->ip6_src;
901 inc.inc6_laddr = ip6->ip6_dst;
902 inc.inc6_route.ro_rt = NULL; /* XXX */
904 inc.inc_faddr = ip->ip_src;
905 inc.inc_laddr = ip->ip_dst;
906 inc.inc_route.ro_rt = NULL; /* XXX */
908 inc.inc_fport = th->th_sport;
909 inc.inc_lport = th->th_dport;
912 * If the state is LISTEN then ignore segment if it contains
913 * a RST. If the segment contains an ACK then it is bad and
914 * send a RST. If it does not contain a SYN then it is not
915 * interesting; drop it.
917 * If the state is SYN_RECEIVED (syncache) and seg contains
918 * an ACK, but not for our SYN/ACK, send a RST. If the seg
919 * contains a RST, check the sequence number to see if it
920 * is a valid reset segment.
922 if ((thflags & (TH_RST | TH_ACK | TH_SYN)) != TH_SYN) {
923 if ((thflags & (TH_RST | TH_ACK | TH_SYN)) == TH_ACK) {
924 if (!syncache_expand(&inc, th, &so, m)) {
926 * No syncache entry, or ACK was not
927 * for our SYN/ACK. Send a RST.
929 tcpstat.tcps_badsyn++;
930 rstreason = BANDLIM_RST_OPENPORT;
935 * Could not complete 3-way handshake,
936 * connection is being closed down, and
937 * syncache will free mbuf.
940 return(IPPROTO_DONE);
943 * We must be in the correct protocol thread
944 * for this connection.
946 KKASSERT(so->so_port == &curthread->td_msgport);
949 * Socket is created in state SYN_RECEIVED.
950 * Continue processing segment.
955 * This is what would have happened in
956 * tcp_output() when the SYN,ACK was sent.
958 tp->snd_up = tp->snd_una;
959 tp->snd_max = tp->snd_nxt = tp->iss + 1;
960 tp->last_ack_sent = tp->rcv_nxt;
964 if (thflags & TH_RST) {
965 syncache_chkrst(&inc, th);
968 if (thflags & TH_ACK) {
969 syncache_badack(&inc);
970 tcpstat.tcps_badsyn++;
971 rstreason = BANDLIM_RST_OPENPORT;
978 * Segment's flags are (SYN) or (SYN | FIN).
982 * If deprecated address is forbidden,
983 * we do not accept SYN to deprecated interface
984 * address to prevent any new inbound connection from
985 * getting established.
986 * When we do not accept SYN, we send a TCP RST,
987 * with deprecated source address (instead of dropping
988 * it). We compromise it as it is much better for peer
989 * to send a RST, and RST will be the final packet
992 * If we do not forbid deprecated addresses, we accept
993 * the SYN packet. RFC2462 does not suggest dropping
995 * If we decipher RFC2462 5.5.4, it says like this:
996 * 1. use of deprecated addr with existing
997 * communication is okay - "SHOULD continue to be
999 * 2. use of it with new communication:
1000 * (2a) "SHOULD NOT be used if alternate address
1001 * with sufficient scope is available"
1002 * (2b) nothing mentioned otherwise.
1003 * Here we fall into (2b) case as we have no choice in
1004 * our source address selection - we must obey the peer.
1006 * The wording in RFC2462 is confusing, and there are
1007 * multiple description text for deprecated address
1008 * handling - worse, they are not exactly the same.
1009 * I believe 5.5.4 is the best one, so we follow 5.5.4.
1011 if (isipv6 && !ip6_use_deprecated) {
1012 struct in6_ifaddr *ia6;
1014 if ((ia6 = ip6_getdstifaddr(m)) &&
1015 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1017 rstreason = BANDLIM_RST_OPENPORT;
1023 * If it is from this socket, drop it, it must be forged.
1024 * Don't bother responding if the destination was a broadcast.
1026 if (th->th_dport == th->th_sport) {
1028 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
1032 if (ip->ip_dst.s_addr == ip->ip_src.s_addr)
1037 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
1039 * Note that it is quite possible to receive unicast
1040 * link-layer packets with a broadcast IP address. Use
1041 * in_broadcast() to find them.
1043 if (m->m_flags & (M_BCAST | M_MCAST))
1046 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1047 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
1050 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1051 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1052 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1053 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
1057 * SYN appears to be valid; create compressed TCP state
1058 * for syncache, or perform t/tcp connection.
1060 if (so->so_qlen <= so->so_qlimit) {
1061 tcp_dooptions(&to, optp, optlen, TRUE, th->th_ack);
1062 if (!syncache_add(&inc, &to, th, so, m))
1066 * Entry added to syncache, mbuf used to
1067 * send SYN,ACK packet.
1069 return(IPPROTO_DONE);
1076 * Should not happen - syncache should pick up these connections.
1078 * Once we are past handling listen sockets we must be in the
1079 * correct protocol processing thread.
1081 KASSERT(tp->t_state != TCPS_LISTEN, ("tcp_input: TCPS_LISTEN state"));
1082 KKASSERT(so->so_port == &curthread->td_msgport);
1084 /* Unscale the window into a 32-bit value. */
1085 if (!(thflags & TH_SYN))
1086 tiwin = th->th_win << tp->snd_scale;
1091 * This is the second part of the MSS DoS prevention code (after
1092 * minmss on the sending side) and it deals with too many too small
1093 * tcp packets in a too short timeframe (1 second).
1095 * XXX Removed. This code was crap. It does not scale to network
1096 * speed, and default values break NFS. Gone.
1101 * Segment received on connection.
1103 * Reset idle time and keep-alive timer. Don't waste time if less
1104 * then a second has elapsed.
1106 if ((int)(ticks - tp->t_rcvtime) > hz)
1107 tcp_timer_keep_activity(tp, thflags);
1111 * XXX this is tradtitional behavior, may need to be cleaned up.
1113 tcp_dooptions(&to, optp, optlen, (thflags & TH_SYN) != 0, th->th_ack);
1114 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1115 if ((to.to_flags & TOF_SCALE) && (tp->t_flags & TF_REQ_SCALE)) {
1116 tp->t_flags |= TF_RCVD_SCALE;
1117 tp->snd_scale = to.to_requested_s_scale;
1121 * Initial send window; will be updated upon next ACK
1123 tp->snd_wnd = th->th_win;
1125 if (to.to_flags & TOF_TS) {
1126 tp->t_flags |= TF_RCVD_TSTMP;
1127 tp->ts_recent = to.to_tsval;
1128 tp->ts_recent_age = ticks;
1130 if (!(to.to_flags & TOF_MSS))
1132 tcp_mss(tp, to.to_mss);
1134 * Only set the TF_SACK_PERMITTED per-connection flag
1135 * if we got a SACK_PERMITTED option from the other side
1136 * and the global tcp_do_sack variable is true.
1138 if (tcp_do_sack && (to.to_flags & TOF_SACK_PERMITTED))
1139 tp->t_flags |= TF_SACK_PERMITTED;
1143 * Header prediction: check for the two common cases
1144 * of a uni-directional data xfer. If the packet has
1145 * no control flags, is in-sequence, the window didn't
1146 * change and we're not retransmitting, it's a
1147 * candidate. If the length is zero and the ack moved
1148 * forward, we're the sender side of the xfer. Just
1149 * free the data acked & wake any higher level process
1150 * that was blocked waiting for space. If the length
1151 * is non-zero and the ack didn't move, we're the
1152 * receiver side. If we're getting packets in-order
1153 * (the reassembly queue is empty), add the data to
1154 * the socket buffer and note that we need a delayed ack.
1155 * Make sure that the hidden state-flags are also off.
1156 * Since we check for TCPS_ESTABLISHED above, it can only
1159 if (tp->t_state == TCPS_ESTABLISHED &&
1160 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1161 !(tp->t_flags & (TF_NEEDSYN | TF_NEEDFIN)) &&
1162 (!(to.to_flags & TOF_TS) ||
1163 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) &&
1164 th->th_seq == tp->rcv_nxt &&
1165 tp->snd_nxt == tp->snd_max) {
1168 * If last ACK falls within this segment's sequence numbers,
1169 * record the timestamp.
1170 * NOTE that the test is modified according to the latest
1171 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1173 if ((to.to_flags & TOF_TS) &&
1174 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1175 tp->ts_recent_age = ticks;
1176 tp->ts_recent = to.to_tsval;
1180 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1181 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1182 tp->snd_cwnd >= tp->snd_wnd &&
1183 !IN_FASTRECOVERY(tp)) {
1185 * This is a pure ack for outstanding data.
1187 ++tcpstat.tcps_predack;
1189 * "bad retransmit" recovery
1191 * If Eifel detection applies, then
1192 * it is deterministic, so use it
1193 * unconditionally over the old heuristic.
1194 * Otherwise, fall back to the old heuristic.
1196 if (tcp_do_eifel_detect &&
1197 (to.to_flags & TOF_TS) && to.to_tsecr &&
1198 (tp->t_flags & TF_FIRSTACCACK)) {
1199 /* Eifel detection applicable. */
1200 if (to.to_tsecr < tp->t_rexmtTS) {
1201 tcp_revert_congestion_state(tp);
1202 ++tcpstat.tcps_eifeldetected;
1203 if (tp->t_rxtshift != 1 ||
1204 ticks >= tp->t_badrxtwin)
1205 ++tcpstat.tcps_rttcantdetect;
1207 } else if (tp->t_rxtshift == 1 &&
1208 ticks < tp->t_badrxtwin) {
1209 tcp_revert_congestion_state(tp);
1210 ++tcpstat.tcps_rttdetected;
1212 tp->t_flags &= ~(TF_FIRSTACCACK |
1213 TF_FASTREXMT | TF_EARLYREXMT);
1215 * Recalculate the retransmit timer / rtt.
1217 * Some machines (certain windows boxes)
1218 * send broken timestamp replies during the
1219 * SYN+ACK phase, ignore timestamps of 0.
1221 if ((to.to_flags & TOF_TS) && to.to_tsecr) {
1223 ticks - to.to_tsecr + 1,
1225 } else if (tp->t_rtttime &&
1226 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1228 ticks - tp->t_rtttime,
1231 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1232 acked = th->th_ack - tp->snd_una;
1233 tcpstat.tcps_rcvackpack++;
1234 tcpstat.tcps_rcvackbyte += acked;
1235 sbdrop(&so->so_snd.sb, acked);
1236 tp->snd_recover = th->th_ack - 1;
1237 tp->snd_una = th->th_ack;
1240 * Update window information.
1242 if (tiwin != tp->snd_wnd &&
1243 acceptable_window_update(tp, th, tiwin)) {
1244 /* keep track of pure window updates */
1245 if (tp->snd_wl2 == th->th_ack &&
1246 tiwin > tp->snd_wnd)
1247 tcpstat.tcps_rcvwinupd++;
1248 tp->snd_wnd = tiwin;
1249 tp->snd_wl1 = th->th_seq;
1250 tp->snd_wl2 = th->th_ack;
1251 if (tp->snd_wnd > tp->max_sndwnd)
1252 tp->max_sndwnd = tp->snd_wnd;
1255 ND6_HINT(tp); /* some progress has been done */
1257 * If all outstanding data are acked, stop
1258 * retransmit timer, otherwise restart timer
1259 * using current (possibly backed-off) value.
1260 * If process is waiting for space,
1261 * wakeup/selwakeup/signal. If data
1262 * are ready to send, let tcp_output
1263 * decide between more output or persist.
1265 if (tp->snd_una == tp->snd_max) {
1266 tcp_callout_stop(tp, tp->tt_rexmt);
1267 } else if (!tcp_callout_active(tp,
1269 tcp_callout_reset(tp, tp->tt_rexmt,
1270 tp->t_rxtcur, tcp_timer_rexmt);
1273 if (so->so_snd.ssb_cc > 0)
1275 return(IPPROTO_DONE);
1277 } else if (tiwin == tp->snd_wnd &&
1278 th->th_ack == tp->snd_una &&
1279 LIST_EMPTY(&tp->t_segq) &&
1280 tlen <= ssb_space(&so->so_rcv)) {
1281 u_long newsize = 0; /* automatic sockbuf scaling */
1283 * This is a pure, in-sequence data packet
1284 * with nothing on the reassembly queue and
1285 * we have enough buffer space to take it.
1287 ++tcpstat.tcps_preddat;
1288 tp->rcv_nxt += tlen;
1289 tcpstat.tcps_rcvpack++;
1290 tcpstat.tcps_rcvbyte += tlen;
1291 ND6_HINT(tp); /* some progress has been done */
1293 * Automatic sizing of receive socket buffer. Often the send
1294 * buffer size is not optimally adjusted to the actual network
1295 * conditions at hand (delay bandwidth product). Setting the
1296 * buffer size too small limits throughput on links with high
1297 * bandwidth and high delay (eg. trans-continental/oceanic links).
1299 * On the receive side the socket buffer memory is only rarely
1300 * used to any significant extent. This allows us to be much
1301 * more aggressive in scaling the receive socket buffer. For
1302 * the case that the buffer space is actually used to a large
1303 * extent and we run out of kernel memory we can simply drop
1304 * the new segments; TCP on the sender will just retransmit it
1305 * later. Setting the buffer size too big may only consume too
1306 * much kernel memory if the application doesn't read() from
1307 * the socket or packet loss or reordering makes use of the
1310 * The criteria to step up the receive buffer one notch are:
1311 * 1. the number of bytes received during the time it takes
1312 * one timestamp to be reflected back to us (the RTT);
1313 * 2. received bytes per RTT is within seven eighth of the
1314 * current socket buffer size;
1315 * 3. receive buffer size has not hit maximal automatic size;
1317 * This algorithm does one step per RTT at most and only if
1318 * we receive a bulk stream w/o packet losses or reorderings.
1319 * Shrinking the buffer during idle times is not necessary as
1320 * it doesn't consume any memory when idle.
1322 * TODO: Only step up if the application is actually serving
1323 * the buffer to better manage the socket buffer resources.
1325 if (tcp_do_autorcvbuf &&
1327 (so->so_rcv.ssb_flags & SSB_AUTOSIZE)) {
1328 if (to.to_tsecr > tp->rfbuf_ts &&
1329 to.to_tsecr - tp->rfbuf_ts < hz) {
1331 (so->so_rcv.ssb_hiwat / 8 * 7) &&
1332 so->so_rcv.ssb_hiwat <
1333 tcp_autorcvbuf_max) {
1335 ulmin(so->so_rcv.ssb_hiwat +
1337 tcp_autorcvbuf_max);
1339 /* Start over with next RTT. */
1343 tp->rfbuf_cnt += tlen; /* add up */
1346 * Add data to socket buffer.
1348 if (so->so_state & SS_CANTRCVMORE) {
1352 * Set new socket buffer size, give up when
1355 * Adjusting the size can mess up ACK
1356 * sequencing when pure window updates are
1357 * being avoided (which is the default),
1360 lwkt_gettoken(&so->so_rcv.ssb_token);
1362 tp->t_flags |= TF_RXRESIZED;
1363 if (!ssb_reserve(&so->so_rcv, newsize,
1365 atomic_clear_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE);
1368 (TCP_MAXWIN << tp->rcv_scale)) {
1369 atomic_clear_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE);
1372 m_adj(m, drop_hdrlen); /* delayed header drop */
1373 ssb_appendstream(&so->so_rcv, m);
1374 lwkt_reltoken(&so->so_rcv.ssb_token);
1378 * This code is responsible for most of the ACKs
1379 * the TCP stack sends back after receiving a data
1380 * packet. Note that the DELAY_ACK check fails if
1381 * the delack timer is already running, which results
1382 * in an ack being sent every other packet (which is
1385 * We then further aggregate acks by not actually
1386 * sending one until the protocol thread has completed
1387 * processing the current backlog of packets. This
1388 * does not delay the ack any further, but allows us
1389 * to take advantage of the packet aggregation that
1390 * high speed NICs do (usually blocks of 8-10 packets)
1391 * to send a single ack rather then four or five acks,
1392 * greatly reducing the ack rate, the return channel
1393 * bandwidth, and the protocol overhead on both ends.
1395 * Since this also has the effect of slowing down
1396 * the exponential slow-start ramp-up, systems with
1397 * very large bandwidth-delay products might want
1398 * to turn the feature off.
1400 if (DELAY_ACK(tp)) {
1401 tcp_callout_reset(tp, tp->tt_delack,
1402 tcp_delacktime, tcp_timer_delack);
1403 } else if (tcp_aggregate_acks) {
1404 tp->t_flags |= TF_ACKNOW;
1405 if (!(tp->t_flags & TF_ONOUTPUTQ)) {
1406 tp->t_flags |= TF_ONOUTPUTQ;
1407 tp->tt_cpu = mycpu->gd_cpuid;
1409 &tcpcbackq[tp->tt_cpu],
1413 tp->t_flags |= TF_ACKNOW;
1416 return(IPPROTO_DONE);
1421 * Calculate amount of space in receive window,
1422 * and then do TCP input processing.
1423 * Receive window is amount of space in rcv queue,
1424 * but not less than advertised window.
1426 recvwin = ssb_space(&so->so_rcv);
1429 tp->rcv_wnd = imax(recvwin, (int)(tp->rcv_adv - tp->rcv_nxt));
1431 /* Reset receive buffer auto scaling when not in bulk receive mode. */
1435 switch (tp->t_state) {
1437 * If the state is SYN_RECEIVED:
1438 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1440 case TCPS_SYN_RECEIVED:
1441 if ((thflags & TH_ACK) &&
1442 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1443 SEQ_GT(th->th_ack, tp->snd_max))) {
1444 rstreason = BANDLIM_RST_OPENPORT;
1450 * If the state is SYN_SENT:
1451 * if seg contains an ACK, but not for our SYN, drop the input.
1452 * if seg contains a RST, then drop the connection.
1453 * if seg does not contain SYN, then drop it.
1454 * Otherwise this is an acceptable SYN segment
1455 * initialize tp->rcv_nxt and tp->irs
1456 * if seg contains ack then advance tp->snd_una
1457 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1458 * arrange for segment to be acked (eventually)
1459 * continue processing rest of data/controls, beginning with URG
1462 if ((thflags & TH_ACK) &&
1463 (SEQ_LEQ(th->th_ack, tp->iss) ||
1464 SEQ_GT(th->th_ack, tp->snd_max))) {
1465 rstreason = BANDLIM_UNLIMITED;
1468 if (thflags & TH_RST) {
1469 if (thflags & TH_ACK)
1470 tp = tcp_drop(tp, ECONNREFUSED);
1473 if (!(thflags & TH_SYN))
1476 tp->irs = th->th_seq;
1478 if (thflags & TH_ACK) {
1479 /* Our SYN was acked. */
1480 tcpstat.tcps_connects++;
1482 /* Do window scaling on this connection? */
1483 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
1484 (TF_RCVD_SCALE | TF_REQ_SCALE))
1485 tp->rcv_scale = tp->request_r_scale;
1486 tp->rcv_adv += tp->rcv_wnd;
1487 tp->snd_una++; /* SYN is acked */
1488 tcp_callout_stop(tp, tp->tt_rexmt);
1490 * If there's data, delay ACK; if there's also a FIN
1491 * ACKNOW will be turned on later.
1493 if (DELAY_ACK(tp) && tlen != 0) {
1494 tcp_callout_reset(tp, tp->tt_delack,
1495 tcp_delacktime, tcp_timer_delack);
1497 tp->t_flags |= TF_ACKNOW;
1500 * Received <SYN,ACK> in SYN_SENT[*] state.
1502 * SYN_SENT --> ESTABLISHED
1503 * SYN_SENT* --> FIN_WAIT_1
1505 tp->t_starttime = ticks;
1506 if (tp->t_flags & TF_NEEDFIN) {
1507 tp->t_state = TCPS_FIN_WAIT_1;
1508 tp->t_flags &= ~TF_NEEDFIN;
1511 tcp_established(tp);
1515 * Received initial SYN in SYN-SENT[*] state =>
1516 * simultaneous open.
1517 * Do 3-way handshake:
1518 * SYN-SENT -> SYN-RECEIVED
1519 * SYN-SENT* -> SYN-RECEIVED*
1521 tp->t_flags |= TF_ACKNOW;
1522 tcp_callout_stop(tp, tp->tt_rexmt);
1523 tp->t_state = TCPS_SYN_RECEIVED;
1527 * Advance th->th_seq to correspond to first data byte.
1528 * If data, trim to stay within window,
1529 * dropping FIN if necessary.
1532 if (tlen > tp->rcv_wnd) {
1533 todrop = tlen - tp->rcv_wnd;
1537 tcpstat.tcps_rcvpackafterwin++;
1538 tcpstat.tcps_rcvbyteafterwin += todrop;
1540 tp->snd_wl1 = th->th_seq - 1;
1541 tp->rcv_up = th->th_seq;
1543 * Client side of transaction: already sent SYN and data.
1544 * If the remote host used T/TCP to validate the SYN,
1545 * our data will be ACK'd; if so, enter normal data segment
1546 * processing in the middle of step 5, ack processing.
1547 * Otherwise, goto step 6.
1549 if (thflags & TH_ACK)
1555 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
1556 * do normal processing (we no longer bother with T/TCP).
1560 case TCPS_TIME_WAIT:
1561 break; /* continue normal processing */
1565 * States other than LISTEN or SYN_SENT.
1566 * First check the RST flag and sequence number since reset segments
1567 * are exempt from the timestamp and connection count tests. This
1568 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
1569 * below which allowed reset segments in half the sequence space
1570 * to fall though and be processed (which gives forged reset
1571 * segments with a random sequence number a 50 percent chance of
1572 * killing a connection).
1573 * Then check timestamp, if present.
1574 * Then check the connection count, if present.
1575 * Then check that at least some bytes of segment are within
1576 * receive window. If segment begins before rcv_nxt,
1577 * drop leading data (and SYN); if nothing left, just ack.
1580 * If the RST bit is set, check the sequence number to see
1581 * if this is a valid reset segment.
1583 * In all states except SYN-SENT, all reset (RST) segments
1584 * are validated by checking their SEQ-fields. A reset is
1585 * valid if its sequence number is in the window.
1586 * Note: this does not take into account delayed ACKs, so
1587 * we should test against last_ack_sent instead of rcv_nxt.
1588 * The sequence number in the reset segment is normally an
1589 * echo of our outgoing acknowledgement numbers, but some hosts
1590 * send a reset with the sequence number at the rightmost edge
1591 * of our receive window, and we have to handle this case.
1592 * If we have multiple segments in flight, the intial reset
1593 * segment sequence numbers will be to the left of last_ack_sent,
1594 * but they will eventually catch up.
1595 * In any case, it never made sense to trim reset segments to
1596 * fit the receive window since RFC 1122 says:
1597 * 4.2.2.12 RST Segment: RFC-793 Section 3.4
1599 * A TCP SHOULD allow a received RST segment to include data.
1602 * It has been suggested that a RST segment could contain
1603 * ASCII text that encoded and explained the cause of the
1604 * RST. No standard has yet been established for such
1607 * If the reset segment passes the sequence number test examine
1609 * SYN_RECEIVED STATE:
1610 * If passive open, return to LISTEN state.
1611 * If active open, inform user that connection was refused.
1612 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
1613 * Inform user that connection was reset, and close tcb.
1614 * CLOSING, LAST_ACK STATES:
1617 * Drop the segment - see Stevens, vol. 2, p. 964 and
1620 if (thflags & TH_RST) {
1621 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
1622 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
1623 switch (tp->t_state) {
1625 case TCPS_SYN_RECEIVED:
1626 so->so_error = ECONNREFUSED;
1629 case TCPS_ESTABLISHED:
1630 case TCPS_FIN_WAIT_1:
1631 case TCPS_FIN_WAIT_2:
1632 case TCPS_CLOSE_WAIT:
1633 so->so_error = ECONNRESET;
1635 tp->t_state = TCPS_CLOSED;
1636 tcpstat.tcps_drops++;
1645 case TCPS_TIME_WAIT:
1653 * RFC 1323 PAWS: If we have a timestamp reply on this segment
1654 * and it's less than ts_recent, drop it.
1656 if ((to.to_flags & TOF_TS) && tp->ts_recent != 0 &&
1657 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
1659 /* Check to see if ts_recent is over 24 days old. */
1660 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1662 * Invalidate ts_recent. If this segment updates
1663 * ts_recent, the age will be reset later and ts_recent
1664 * will get a valid value. If it does not, setting
1665 * ts_recent to zero will at least satisfy the
1666 * requirement that zero be placed in the timestamp
1667 * echo reply when ts_recent isn't valid. The
1668 * age isn't reset until we get a valid ts_recent
1669 * because we don't want out-of-order segments to be
1670 * dropped when ts_recent is old.
1674 tcpstat.tcps_rcvduppack++;
1675 tcpstat.tcps_rcvdupbyte += tlen;
1676 tcpstat.tcps_pawsdrop++;
1684 * In the SYN-RECEIVED state, validate that the packet belongs to
1685 * this connection before trimming the data to fit the receive
1686 * window. Check the sequence number versus IRS since we know
1687 * the sequence numbers haven't wrapped. This is a partial fix
1688 * for the "LAND" DoS attack.
1690 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
1691 rstreason = BANDLIM_RST_OPENPORT;
1695 todrop = tp->rcv_nxt - th->th_seq;
1697 if (TCP_DO_SACK(tp)) {
1698 /* Report duplicate segment at head of packet. */
1699 tp->reportblk.rblk_start = th->th_seq;
1700 tp->reportblk.rblk_end = TCP_SACK_BLKEND(
1701 th->th_seq + tlen, thflags);
1702 if (SEQ_GT(tp->reportblk.rblk_end, tp->rcv_nxt))
1703 tp->reportblk.rblk_end = tp->rcv_nxt;
1704 tp->sack_flags |= (TSACK_F_DUPSEG | TSACK_F_SACKLEFT);
1705 tp->t_flags |= TF_ACKNOW;
1707 if (thflags & TH_SYN) {
1717 * Following if statement from Stevens, vol. 2, p. 960.
1719 if (todrop > tlen ||
1720 (todrop == tlen && !(thflags & TH_FIN))) {
1722 * Any valid FIN must be to the left of the window.
1723 * At this point the FIN must be a duplicate or out
1724 * of sequence; drop it.
1729 * Send an ACK to resynchronize and drop any data.
1730 * But keep on processing for RST or ACK.
1732 tp->t_flags |= TF_ACKNOW;
1734 tcpstat.tcps_rcvduppack++;
1735 tcpstat.tcps_rcvdupbyte += todrop;
1737 tcpstat.tcps_rcvpartduppack++;
1738 tcpstat.tcps_rcvpartdupbyte += todrop;
1740 drop_hdrlen += todrop; /* drop from the top afterwards */
1741 th->th_seq += todrop;
1743 if (th->th_urp > todrop)
1744 th->th_urp -= todrop;
1752 * If new data are received on a connection after the
1753 * user processes are gone, then RST the other end.
1755 if ((so->so_state & SS_NOFDREF) &&
1756 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1758 tcpstat.tcps_rcvafterclose++;
1759 rstreason = BANDLIM_UNLIMITED;
1764 * If segment ends after window, drop trailing data
1765 * (and PUSH and FIN); if nothing left, just ACK.
1767 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
1769 tcpstat.tcps_rcvpackafterwin++;
1770 if (todrop >= tlen) {
1771 tcpstat.tcps_rcvbyteafterwin += tlen;
1773 * If a new connection request is received
1774 * while in TIME_WAIT, drop the old connection
1775 * and start over if the sequence numbers
1776 * are above the previous ones.
1778 if (thflags & TH_SYN &&
1779 tp->t_state == TCPS_TIME_WAIT &&
1780 SEQ_GT(th->th_seq, tp->rcv_nxt)) {
1785 * If window is closed can only take segments at
1786 * window edge, and have to drop data and PUSH from
1787 * incoming segments. Continue processing, but
1788 * remember to ack. Otherwise, drop segment
1791 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1792 tp->t_flags |= TF_ACKNOW;
1793 tcpstat.tcps_rcvwinprobe++;
1797 tcpstat.tcps_rcvbyteafterwin += todrop;
1800 thflags &= ~(TH_PUSH | TH_FIN);
1804 * If last ACK falls within this segment's sequence numbers,
1805 * record its timestamp.
1807 * 1) That the test incorporates suggestions from the latest
1808 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1809 * 2) That updating only on newer timestamps interferes with
1810 * our earlier PAWS tests, so this check should be solely
1811 * predicated on the sequence space of this segment.
1812 * 3) That we modify the segment boundary check to be
1813 * Last.ACK.Sent <= SEG.SEQ + SEG.LEN
1814 * instead of RFC1323's
1815 * Last.ACK.Sent < SEG.SEQ + SEG.LEN,
1816 * This modified check allows us to overcome RFC1323's
1817 * limitations as described in Stevens TCP/IP Illustrated
1818 * Vol. 2 p.869. In such cases, we can still calculate the
1819 * RTT correctly when RCV.NXT == Last.ACK.Sent.
1821 if ((to.to_flags & TOF_TS) && SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
1822 SEQ_LEQ(tp->last_ack_sent, (th->th_seq + tlen
1823 + ((thflags & TH_SYN) != 0)
1824 + ((thflags & TH_FIN) != 0)))) {
1825 tp->ts_recent_age = ticks;
1826 tp->ts_recent = to.to_tsval;
1830 * If a SYN is in the window, then this is an
1831 * error and we send an RST and drop the connection.
1833 if (thflags & TH_SYN) {
1834 tp = tcp_drop(tp, ECONNRESET);
1835 rstreason = BANDLIM_UNLIMITED;
1840 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
1841 * flag is on (half-synchronized state), then queue data for
1842 * later processing; else drop segment and return.
1844 if (!(thflags & TH_ACK)) {
1845 if (tp->t_state == TCPS_SYN_RECEIVED ||
1846 (tp->t_flags & TF_NEEDSYN))
1855 switch (tp->t_state) {
1857 * In SYN_RECEIVED state, the ACK acknowledges our SYN, so enter
1858 * ESTABLISHED state and continue processing.
1859 * The ACK was checked above.
1861 case TCPS_SYN_RECEIVED:
1863 tcpstat.tcps_connects++;
1865 /* Do window scaling? */
1866 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
1867 (TF_RCVD_SCALE | TF_REQ_SCALE))
1868 tp->rcv_scale = tp->request_r_scale;
1871 * SYN-RECEIVED -> ESTABLISHED
1872 * SYN-RECEIVED* -> FIN-WAIT-1
1874 tp->t_starttime = ticks;
1875 if (tp->t_flags & TF_NEEDFIN) {
1876 tp->t_state = TCPS_FIN_WAIT_1;
1877 tp->t_flags &= ~TF_NEEDFIN;
1879 tcp_established(tp);
1882 * If segment contains data or ACK, will call tcp_reass()
1883 * later; if not, do so now to pass queued data to user.
1885 if (tlen == 0 && !(thflags & TH_FIN))
1886 tcp_reass(tp, NULL, NULL, NULL);
1890 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1891 * ACKs. If the ack is in the range
1892 * tp->snd_una < th->th_ack <= tp->snd_max
1893 * then advance tp->snd_una to th->th_ack and drop
1894 * data from the retransmission queue. If this ACK reflects
1895 * more up to date window information we update our window information.
1897 case TCPS_ESTABLISHED:
1898 case TCPS_FIN_WAIT_1:
1899 case TCPS_FIN_WAIT_2:
1900 case TCPS_CLOSE_WAIT:
1903 case TCPS_TIME_WAIT:
1905 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
1906 if (TCP_DO_SACK(tp))
1907 tcp_sack_update_scoreboard(tp, &to);
1908 if (!tcp_callout_active(tp, tp->tt_rexmt) ||
1909 th->th_ack != tp->snd_una) {
1910 tcpstat.tcps_rcvdupack++;
1914 if (tlen != 0 || tiwin != tp->snd_wnd) {
1915 if (!tcp_do_rfc3517bis ||
1918 (TOF_SACK | TOF_SACK_REDUNDANT))
1924 * Update window information.
1926 if (tiwin != tp->snd_wnd &&
1927 acceptable_window_update(tp, th, tiwin)) {
1928 /* keep track of pure window updates */
1930 tp->snd_wl2 == th->th_ack &&
1931 tiwin > tp->snd_wnd)
1932 tcpstat.tcps_rcvwinupd++;
1933 tp->snd_wnd = tiwin;
1934 tp->snd_wl1 = th->th_seq;
1935 tp->snd_wl2 = th->th_ack;
1936 if (tp->snd_wnd > tp->max_sndwnd)
1937 tp->max_sndwnd = tp->snd_wnd;
1940 tcpstat.tcps_rcvdupack++;
1943 * We have outstanding data (other than
1944 * a window probe), this is a completely
1945 * duplicate ack (ie, window info didn't
1946 * change), the ack is the biggest we've
1947 * seen and we've seen exactly our rexmt
1948 * threshhold of them, so assume a packet
1949 * has been dropped and retransmit it.
1950 * Kludge snd_nxt & the congestion
1951 * window so we send only this one
1954 if (IN_FASTRECOVERY(tp)) {
1955 if (TCP_DO_SACK(tp)) {
1956 /* No artifical cwnd inflation. */
1957 tcp_sack_rexmt(tp, th);
1960 * Dup acks mean that packets
1961 * have left the network
1962 * (they're now cached at the
1963 * receiver) so bump cwnd by
1964 * the amount in the receiver
1965 * to keep a constant cwnd
1966 * packets in the network.
1968 tp->snd_cwnd += tp->t_maxseg;
1971 } else if (SEQ_LT(th->th_ack, tp->snd_recover)) {
1974 } else if (tcp_ignore_redun_dsack && TCP_DO_SACK(tp) &&
1975 (to.to_flags & (TOF_DSACK | TOF_SACK_REDUNDANT)) ==
1976 (TOF_DSACK | TOF_SACK_REDUNDANT)) {
1978 * If the ACK carries DSACK and other
1979 * SACK blocks carry information that
1980 * we have already known, don't count
1981 * this ACK as duplicate ACK. This
1982 * prevents spurious early retransmit
1983 * and fast retransmit. This also
1984 * meets the requirement of RFC3042
1985 * that new segments should not be sent
1986 * if the SACK blocks do not contain
1987 * new information (XXX we actually
1988 * loosen the requirment that only DSACK
1991 * This kind of ACKs are usually sent
1992 * after spurious retransmit.
1994 /* Do nothing; don't change t_dupacks */
1995 } else if (++tp->t_dupacks == tp->t_rxtthresh) {
1996 tcp_seq old_snd_nxt;
2000 if (tcp_do_eifel_detect &&
2001 (tp->t_flags & TF_RCVD_TSTMP)) {
2002 tcp_save_congestion_state(tp);
2003 tp->t_flags |= TF_FASTREXMT;
2006 * We know we're losing at the current
2007 * window size, so do congestion avoidance:
2008 * set ssthresh to half the current window
2009 * and pull our congestion window back to the
2012 win = min(tp->snd_wnd, tp->snd_cwnd) / 2 /
2016 tp->snd_ssthresh = win * tp->t_maxseg;
2017 ENTER_FASTRECOVERY(tp);
2018 tp->snd_recover = tp->snd_max;
2019 tcp_callout_stop(tp, tp->tt_rexmt);
2021 old_snd_nxt = tp->snd_nxt;
2022 tp->snd_nxt = th->th_ack;
2023 tp->snd_cwnd = tp->t_maxseg;
2025 ++tcpstat.tcps_sndfastrexmit;
2026 tp->snd_cwnd = tp->snd_ssthresh;
2027 tp->rexmt_high = tp->snd_nxt;
2028 tp->sack_flags &= ~TSACK_F_SACKRESCUED;
2029 if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
2030 tp->snd_nxt = old_snd_nxt;
2031 KASSERT(tp->snd_limited <= 2,
2032 ("tp->snd_limited too big"));
2033 if (TCP_DO_SACK(tp))
2034 tcp_sack_rexmt(tp, th);
2036 tp->snd_cwnd += tp->t_maxseg *
2037 (tp->t_dupacks - tp->snd_limited);
2038 } else if (tcp_do_rfc3517bis && TCP_DO_SACK(tp)) {
2039 if (tcp_rfc3517bis_rxt &&
2040 tcp_sack_islost(&tp->scb, tp->snd_una))
2041 goto fastretransmit;
2042 if (tcp_do_limitedtransmit) {
2043 /* outstanding data */
2045 tp->snd_max - tp->snd_una;
2047 if (!tcp_sack_limitedxmit(tp) &&
2048 need_early_retransmit(tp, ownd)) {
2049 ++tcpstat.tcps_sndearlyrexmit;
2050 tp->t_flags |= TF_EARLYREXMT;
2051 goto fastretransmit;
2054 } else if (tcp_do_limitedtransmit) {
2055 u_long oldcwnd = tp->snd_cwnd;
2056 tcp_seq oldsndmax = tp->snd_max;
2057 tcp_seq oldsndnxt = tp->snd_nxt;
2058 /* outstanding data */
2059 uint32_t ownd = tp->snd_max - tp->snd_una;
2062 KASSERT(tp->t_dupacks == 1 ||
2064 ("dupacks not 1 or 2"));
2065 if (tp->t_dupacks == 1)
2066 tp->snd_limited = 0;
2067 tp->snd_nxt = tp->snd_max;
2068 tp->snd_cwnd = ownd +
2069 (tp->t_dupacks - tp->snd_limited) *
2073 if (SEQ_LT(oldsndnxt, oldsndmax)) {
2074 KASSERT(SEQ_GEQ(oldsndnxt, tp->snd_una),
2075 ("snd_una moved in other threads"));
2076 tp->snd_nxt = oldsndnxt;
2078 tp->snd_cwnd = oldcwnd;
2079 sent = tp->snd_max - oldsndmax;
2080 if (sent > tp->t_maxseg) {
2081 KASSERT((tp->t_dupacks == 2 &&
2082 tp->snd_limited == 0) ||
2083 (sent == tp->t_maxseg + 1 &&
2084 tp->t_flags & TF_SENTFIN),
2086 KASSERT(sent <= tp->t_maxseg * 2,
2087 ("sent too many segments"));
2088 tp->snd_limited = 2;
2089 tcpstat.tcps_sndlimited += 2;
2090 } else if (sent > 0) {
2092 ++tcpstat.tcps_sndlimited;
2093 } else if (need_early_retransmit(tp, ownd)) {
2094 ++tcpstat.tcps_sndearlyrexmit;
2095 tp->t_flags |= TF_EARLYREXMT;
2096 goto fastretransmit;
2105 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una"));
2107 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2109 * Detected optimistic ACK attack.
2110 * Force slow-start to de-synchronize attack.
2112 tp->snd_cwnd = tp->t_maxseg;
2115 tcpstat.tcps_rcvacktoomuch++;
2119 * If we reach this point, ACK is not a duplicate,
2120 * i.e., it ACKs something we sent.
2122 if (tp->t_flags & TF_NEEDSYN) {
2124 * T/TCP: Connection was half-synchronized, and our
2125 * SYN has been ACK'd (so connection is now fully
2126 * synchronized). Go to non-starred state,
2127 * increment snd_una for ACK of SYN, and check if
2128 * we can do window scaling.
2130 tp->t_flags &= ~TF_NEEDSYN;
2132 /* Do window scaling? */
2133 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
2134 (TF_RCVD_SCALE | TF_REQ_SCALE))
2135 tp->rcv_scale = tp->request_r_scale;
2139 acked = th->th_ack - tp->snd_una;
2140 tcpstat.tcps_rcvackpack++;
2141 tcpstat.tcps_rcvackbyte += acked;
2143 if (tcp_do_eifel_detect && acked > 0 &&
2144 (to.to_flags & TOF_TS) && (to.to_tsecr != 0) &&
2145 (tp->t_flags & TF_FIRSTACCACK)) {
2146 /* Eifel detection applicable. */
2147 if (to.to_tsecr < tp->t_rexmtTS) {
2148 ++tcpstat.tcps_eifeldetected;
2149 tcp_revert_congestion_state(tp);
2150 if (tp->t_rxtshift != 1 ||
2151 ticks >= tp->t_badrxtwin)
2152 ++tcpstat.tcps_rttcantdetect;
2154 } else if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) {
2156 * If we just performed our first retransmit,
2157 * and the ACK arrives within our recovery window,
2158 * then it was a mistake to do the retransmit
2159 * in the first place. Recover our original cwnd
2160 * and ssthresh, and proceed to transmit where we
2163 tcp_revert_congestion_state(tp);
2164 ++tcpstat.tcps_rttdetected;
2168 * If we have a timestamp reply, update smoothed
2169 * round trip time. If no timestamp is present but
2170 * transmit timer is running and timed sequence
2171 * number was acked, update smoothed round trip time.
2172 * Since we now have an rtt measurement, cancel the
2173 * timer backoff (cf., Phil Karn's retransmit alg.).
2174 * Recompute the initial retransmit timer.
2176 * Some machines (certain windows boxes) send broken
2177 * timestamp replies during the SYN+ACK phase, ignore
2180 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0))
2181 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1, th->th_ack);
2182 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq))
2183 tcp_xmit_timer(tp, ticks - tp->t_rtttime, th->th_ack);
2184 tcp_xmit_bandwidth_limit(tp, th->th_ack);
2187 * If no data (only SYN) was ACK'd,
2188 * skip rest of ACK processing.
2193 /* Stop looking for an acceptable ACK since one was received. */
2194 tp->t_flags &= ~(TF_FIRSTACCACK | TF_FASTREXMT | TF_EARLYREXMT);
2196 if (acked > so->so_snd.ssb_cc) {
2197 tp->snd_wnd -= so->so_snd.ssb_cc;
2198 sbdrop(&so->so_snd.sb, (int)so->so_snd.ssb_cc);
2199 ourfinisacked = TRUE;
2201 sbdrop(&so->so_snd.sb, acked);
2202 tp->snd_wnd -= acked;
2203 ourfinisacked = FALSE;
2208 * Update window information.
2210 if (acceptable_window_update(tp, th, tiwin)) {
2211 /* keep track of pure window updates */
2212 if (tlen == 0 && tp->snd_wl2 == th->th_ack &&
2213 tiwin > tp->snd_wnd)
2214 tcpstat.tcps_rcvwinupd++;
2215 tp->snd_wnd = tiwin;
2216 tp->snd_wl1 = th->th_seq;
2217 tp->snd_wl2 = th->th_ack;
2218 if (tp->snd_wnd > tp->max_sndwnd)
2219 tp->max_sndwnd = tp->snd_wnd;
2223 tp->snd_una = th->th_ack;
2224 if (TCP_DO_SACK(tp))
2225 tcp_sack_update_scoreboard(tp, &to);
2226 if (IN_FASTRECOVERY(tp)) {
2227 if (SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2228 EXIT_FASTRECOVERY(tp);
2231 * If the congestion window was inflated
2232 * to account for the other side's
2233 * cached packets, retract it.
2235 if (!TCP_DO_SACK(tp))
2236 tp->snd_cwnd = tp->snd_ssthresh;
2239 * Window inflation should have left us
2240 * with approximately snd_ssthresh outstanding
2241 * data. But, in case we would be inclined
2242 * to send a burst, better do it using
2245 if (SEQ_GT(th->th_ack + tp->snd_cwnd,
2246 tp->snd_max + 2 * tp->t_maxseg))
2248 (tp->snd_max - tp->snd_una) +
2253 if (TCP_DO_SACK(tp)) {
2254 tp->snd_max_rexmt = tp->snd_max;
2255 tcp_sack_rexmt(tp, th);
2257 tcp_newreno_partial_ack(tp, th, acked);
2263 * Open the congestion window. When in slow-start,
2264 * open exponentially: maxseg per packet. Otherwise,
2265 * open linearly: maxseg per window.
2267 if (tp->snd_cwnd <= tp->snd_ssthresh) {
2269 (SEQ_LT(tp->snd_nxt, tp->snd_max) ?
2270 tp->t_maxseg : 2 * tp->t_maxseg);
2273 tp->snd_cwnd += tcp_do_abc ?
2274 min(acked, abc_sslimit) : tp->t_maxseg;
2276 /* linear increase */
2277 tp->snd_wacked += tcp_do_abc ? acked :
2279 if (tp->snd_wacked >= tp->snd_cwnd) {
2280 tp->snd_wacked -= tp->snd_cwnd;
2281 tp->snd_cwnd += tp->t_maxseg;
2284 tp->snd_cwnd = min(tp->snd_cwnd,
2285 TCP_MAXWIN << tp->snd_scale);
2286 tp->snd_recover = th->th_ack - 1;
2288 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2289 tp->snd_nxt = tp->snd_una;
2292 * If all outstanding data is acked, stop retransmit
2293 * timer and remember to restart (more output or persist).
2294 * If there is more data to be acked, restart retransmit
2295 * timer, using current (possibly backed-off) value.
2297 if (th->th_ack == tp->snd_max) {
2298 tcp_callout_stop(tp, tp->tt_rexmt);
2300 } else if (!tcp_callout_active(tp, tp->tt_persist)) {
2301 tcp_callout_reset(tp, tp->tt_rexmt, tp->t_rxtcur,
2305 switch (tp->t_state) {
2307 * In FIN_WAIT_1 STATE in addition to the processing
2308 * for the ESTABLISHED state if our FIN is now acknowledged
2309 * then enter FIN_WAIT_2.
2311 case TCPS_FIN_WAIT_1:
2312 if (ourfinisacked) {
2314 * If we can't receive any more
2315 * data, then closing user can proceed.
2316 * Starting the timer is contrary to the
2317 * specification, but if we don't get a FIN
2318 * we'll hang forever.
2320 if (so->so_state & SS_CANTRCVMORE) {
2321 soisdisconnected(so);
2322 tcp_callout_reset(tp, tp->tt_2msl,
2323 tp->t_maxidle, tcp_timer_2msl);
2325 tp->t_state = TCPS_FIN_WAIT_2;
2330 * In CLOSING STATE in addition to the processing for
2331 * the ESTABLISHED state if the ACK acknowledges our FIN
2332 * then enter the TIME-WAIT state, otherwise ignore
2336 if (ourfinisacked) {
2337 tp->t_state = TCPS_TIME_WAIT;
2338 tcp_canceltimers(tp);
2339 tcp_callout_reset(tp, tp->tt_2msl,
2340 2 * tcp_rmx_msl(tp),
2342 soisdisconnected(so);
2347 * In LAST_ACK, we may still be waiting for data to drain
2348 * and/or to be acked, as well as for the ack of our FIN.
2349 * If our FIN is now acknowledged, delete the TCB,
2350 * enter the closed state and return.
2353 if (ourfinisacked) {
2360 * In TIME_WAIT state the only thing that should arrive
2361 * is a retransmission of the remote FIN. Acknowledge
2362 * it and restart the finack timer.
2364 case TCPS_TIME_WAIT:
2365 tcp_callout_reset(tp, tp->tt_2msl, 2 * tcp_rmx_msl(tp),
2373 * Update window information.
2374 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2376 if ((thflags & TH_ACK) &&
2377 acceptable_window_update(tp, th, tiwin)) {
2378 /* keep track of pure window updates */
2379 if (tlen == 0 && tp->snd_wl2 == th->th_ack &&
2380 tiwin > tp->snd_wnd)
2381 tcpstat.tcps_rcvwinupd++;
2382 tp->snd_wnd = tiwin;
2383 tp->snd_wl1 = th->th_seq;
2384 tp->snd_wl2 = th->th_ack;
2385 if (tp->snd_wnd > tp->max_sndwnd)
2386 tp->max_sndwnd = tp->snd_wnd;
2391 * Process segments with URG.
2393 if ((thflags & TH_URG) && th->th_urp &&
2394 !TCPS_HAVERCVDFIN(tp->t_state)) {
2396 * This is a kludge, but if we receive and accept
2397 * random urgent pointers, we'll crash in
2398 * soreceive. It's hard to imagine someone
2399 * actually wanting to send this much urgent data.
2401 if (th->th_urp + so->so_rcv.ssb_cc > sb_max) {
2402 th->th_urp = 0; /* XXX */
2403 thflags &= ~TH_URG; /* XXX */
2404 goto dodata; /* XXX */
2407 * If this segment advances the known urgent pointer,
2408 * then mark the data stream. This should not happen
2409 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2410 * a FIN has been received from the remote side.
2411 * In these states we ignore the URG.
2413 * According to RFC961 (Assigned Protocols),
2414 * the urgent pointer points to the last octet
2415 * of urgent data. We continue, however,
2416 * to consider it to indicate the first octet
2417 * of data past the urgent section as the original
2418 * spec states (in one of two places).
2420 if (SEQ_GT(th->th_seq + th->th_urp, tp->rcv_up)) {
2421 tp->rcv_up = th->th_seq + th->th_urp;
2422 so->so_oobmark = so->so_rcv.ssb_cc +
2423 (tp->rcv_up - tp->rcv_nxt) - 1;
2424 if (so->so_oobmark == 0)
2425 sosetstate(so, SS_RCVATMARK);
2427 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2430 * Remove out of band data so doesn't get presented to user.
2431 * This can happen independent of advancing the URG pointer,
2432 * but if two URG's are pending at once, some out-of-band
2433 * data may creep in... ick.
2435 if (th->th_urp <= (u_long)tlen &&
2436 !(so->so_options & SO_OOBINLINE)) {
2437 /* hdr drop is delayed */
2438 tcp_pulloutofband(so, th, m, drop_hdrlen);
2442 * If no out of band data is expected,
2443 * pull receive urgent pointer along
2444 * with the receive window.
2446 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2447 tp->rcv_up = tp->rcv_nxt;
2452 * Process the segment text, merging it into the TCP sequencing queue,
2453 * and arranging for acknowledgment of receipt if necessary.
2454 * This process logically involves adjusting tp->rcv_wnd as data
2455 * is presented to the user (this happens in tcp_usrreq.c,
2456 * case PRU_RCVD). If a FIN has already been received on this
2457 * connection then we just ignore the text.
2459 if ((tlen || (thflags & TH_FIN)) && !TCPS_HAVERCVDFIN(tp->t_state)) {
2460 m_adj(m, drop_hdrlen); /* delayed header drop */
2462 * Insert segment which includes th into TCP reassembly queue
2463 * with control block tp. Set thflags to whether reassembly now
2464 * includes a segment with FIN. This handles the common case
2465 * inline (segment is the next to be received on an established
2466 * connection, and the queue is empty), avoiding linkage into
2467 * and removal from the queue and repetition of various
2469 * Set DELACK for segments received in order, but ack
2470 * immediately when segments are out of order (so
2471 * fast retransmit can work).
2473 if (th->th_seq == tp->rcv_nxt &&
2474 LIST_EMPTY(&tp->t_segq) &&
2475 TCPS_HAVEESTABLISHED(tp->t_state)) {
2476 if (DELAY_ACK(tp)) {
2477 tcp_callout_reset(tp, tp->tt_delack,
2478 tcp_delacktime, tcp_timer_delack);
2480 tp->t_flags |= TF_ACKNOW;
2482 tp->rcv_nxt += tlen;
2483 thflags = th->th_flags & TH_FIN;
2484 tcpstat.tcps_rcvpack++;
2485 tcpstat.tcps_rcvbyte += tlen;
2487 if (so->so_state & SS_CANTRCVMORE) {
2490 lwkt_gettoken(&so->so_rcv.ssb_token);
2491 ssb_appendstream(&so->so_rcv, m);
2492 lwkt_reltoken(&so->so_rcv.ssb_token);
2496 if (!(tp->sack_flags & TSACK_F_DUPSEG)) {
2497 /* Initialize SACK report block. */
2498 tp->reportblk.rblk_start = th->th_seq;
2499 tp->reportblk.rblk_end = TCP_SACK_BLKEND(
2500 th->th_seq + tlen, thflags);
2502 thflags = tcp_reass(tp, th, &tlen, m);
2503 tp->t_flags |= TF_ACKNOW;
2507 * Note the amount of data that peer has sent into
2508 * our window, in order to estimate the sender's
2511 len = so->so_rcv.ssb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2518 * If FIN is received ACK the FIN and let the user know
2519 * that the connection is closing.
2521 if (thflags & TH_FIN) {
2522 if (!TCPS_HAVERCVDFIN(tp->t_state)) {
2525 * If connection is half-synchronized
2526 * (ie NEEDSYN flag on) then delay ACK,
2527 * so it may be piggybacked when SYN is sent.
2528 * Otherwise, since we received a FIN then no
2529 * more input can be expected, send ACK now.
2531 if (DELAY_ACK(tp) && (tp->t_flags & TF_NEEDSYN)) {
2532 tcp_callout_reset(tp, tp->tt_delack,
2533 tcp_delacktime, tcp_timer_delack);
2535 tp->t_flags |= TF_ACKNOW;
2540 switch (tp->t_state) {
2542 * In SYN_RECEIVED and ESTABLISHED STATES
2543 * enter the CLOSE_WAIT state.
2545 case TCPS_SYN_RECEIVED:
2546 tp->t_starttime = ticks;
2548 case TCPS_ESTABLISHED:
2549 tp->t_state = TCPS_CLOSE_WAIT;
2553 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2554 * enter the CLOSING state.
2556 case TCPS_FIN_WAIT_1:
2557 tp->t_state = TCPS_CLOSING;
2561 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2562 * starting the time-wait timer, turning off the other
2565 case TCPS_FIN_WAIT_2:
2566 tp->t_state = TCPS_TIME_WAIT;
2567 tcp_canceltimers(tp);
2568 tcp_callout_reset(tp, tp->tt_2msl, 2 * tcp_rmx_msl(tp),
2570 soisdisconnected(so);
2574 * In TIME_WAIT state restart the 2 MSL time_wait timer.
2576 case TCPS_TIME_WAIT:
2577 tcp_callout_reset(tp, tp->tt_2msl, 2 * tcp_rmx_msl(tp),
2584 if (so->so_options & SO_DEBUG)
2585 tcp_trace(TA_INPUT, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2589 * Return any desired output.
2591 if (needoutput || (tp->t_flags & TF_ACKNOW))
2593 tcp_sack_report_cleanup(tp);
2594 return(IPPROTO_DONE);
2598 * Generate an ACK dropping incoming segment if it occupies
2599 * sequence space, where the ACK reflects our state.
2601 * We can now skip the test for the RST flag since all
2602 * paths to this code happen after packets containing
2603 * RST have been dropped.
2605 * In the SYN-RECEIVED state, don't send an ACK unless the
2606 * segment we received passes the SYN-RECEIVED ACK test.
2607 * If it fails send a RST. This breaks the loop in the
2608 * "LAND" DoS attack, and also prevents an ACK storm
2609 * between two listening ports that have been sent forged
2610 * SYN segments, each with the source address of the other.
2612 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
2613 (SEQ_GT(tp->snd_una, th->th_ack) ||
2614 SEQ_GT(th->th_ack, tp->snd_max)) ) {
2615 rstreason = BANDLIM_RST_OPENPORT;
2619 if (so->so_options & SO_DEBUG)
2620 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2623 tp->t_flags |= TF_ACKNOW;
2625 tcp_sack_report_cleanup(tp);
2626 return(IPPROTO_DONE);
2630 * Generate a RST, dropping incoming segment.
2631 * Make ACK acceptable to originator of segment.
2632 * Don't bother to respond if destination was broadcast/multicast.
2634 if ((thflags & TH_RST) || m->m_flags & (M_BCAST | M_MCAST))
2637 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
2638 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
2641 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
2642 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
2643 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
2644 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2647 /* IPv6 anycast check is done at tcp6_input() */
2650 * Perform bandwidth limiting.
2653 if (badport_bandlim(rstreason) < 0)
2658 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2659 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2661 if (thflags & TH_ACK)
2662 /* mtod() below is safe as long as hdr dropping is delayed */
2663 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack,
2666 if (thflags & TH_SYN)
2668 /* mtod() below is safe as long as hdr dropping is delayed */
2669 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq + tlen,
2670 (tcp_seq)0, TH_RST | TH_ACK);
2673 tcp_sack_report_cleanup(tp);
2674 return(IPPROTO_DONE);
2678 * Drop space held by incoming segment and return.
2681 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2682 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2686 tcp_sack_report_cleanup(tp);
2687 return(IPPROTO_DONE);
2691 * Parse TCP options and place in tcpopt.
2694 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, boolean_t is_syn,
2700 for (; cnt > 0; cnt -= optlen, cp += optlen) {
2702 if (opt == TCPOPT_EOL)
2704 if (opt == TCPOPT_NOP)
2710 if (optlen < 2 || optlen > cnt)
2715 if (optlen != TCPOLEN_MAXSEG)
2719 to->to_flags |= TOF_MSS;
2720 bcopy(cp + 2, &to->to_mss, sizeof to->to_mss);
2721 to->to_mss = ntohs(to->to_mss);
2724 if (optlen != TCPOLEN_WINDOW)
2728 to->to_flags |= TOF_SCALE;
2729 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
2731 case TCPOPT_TIMESTAMP:
2732 if (optlen != TCPOLEN_TIMESTAMP)
2734 to->to_flags |= TOF_TS;
2735 bcopy(cp + 2, &to->to_tsval, sizeof to->to_tsval);
2736 to->to_tsval = ntohl(to->to_tsval);
2737 bcopy(cp + 6, &to->to_tsecr, sizeof to->to_tsecr);
2738 to->to_tsecr = ntohl(to->to_tsecr);
2740 * If echoed timestamp is later than the current time,
2741 * fall back to non RFC1323 RTT calculation.
2743 if (to->to_tsecr != 0 && TSTMP_GT(to->to_tsecr, ticks))
2746 case TCPOPT_SACK_PERMITTED:
2747 if (optlen != TCPOLEN_SACK_PERMITTED)
2751 to->to_flags |= TOF_SACK_PERMITTED;
2754 if ((optlen - 2) & 0x07) /* not multiple of 8 */
2756 to->to_nsackblocks = (optlen - 2) / 8;
2757 to->to_sackblocks = (struct raw_sackblock *) (cp + 2);
2758 to->to_flags |= TOF_SACK;
2759 for (i = 0; i < to->to_nsackblocks; i++) {
2760 struct raw_sackblock *r = &to->to_sackblocks[i];
2762 r->rblk_start = ntohl(r->rblk_start);
2763 r->rblk_end = ntohl(r->rblk_end);
2765 if (SEQ_LEQ(r->rblk_end, r->rblk_start)) {
2767 * Invalid SACK block; discard all
2770 tcpstat.tcps_rcvbadsackopt++;
2771 to->to_nsackblocks = 0;
2772 to->to_sackblocks = NULL;
2773 to->to_flags &= ~TOF_SACK;
2777 if ((to->to_flags & TOF_SACK) &&
2778 tcp_sack_ndsack_blocks(to->to_sackblocks,
2779 to->to_nsackblocks, ack))
2780 to->to_flags |= TOF_DSACK;
2782 #ifdef TCP_SIGNATURE
2784 * XXX In order to reply to a host which has set the
2785 * TCP_SIGNATURE option in its initial SYN, we have to
2786 * record the fact that the option was observed here
2787 * for the syncache code to perform the correct response.
2789 case TCPOPT_SIGNATURE:
2790 if (optlen != TCPOLEN_SIGNATURE)
2792 to->to_flags |= (TOF_SIGNATURE | TOF_SIGLEN);
2794 #endif /* TCP_SIGNATURE */
2802 * Pull out of band byte out of a segment so
2803 * it doesn't appear in the user's data queue.
2804 * It is still reflected in the segment length for
2805 * sequencing purposes.
2806 * "off" is the delayed to be dropped hdrlen.
2809 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, int off)
2811 int cnt = off + th->th_urp - 1;
2814 if (m->m_len > cnt) {
2815 char *cp = mtod(m, caddr_t) + cnt;
2816 struct tcpcb *tp = sototcpcb(so);
2819 tp->t_oobflags |= TCPOOB_HAVEDATA;
2820 bcopy(cp + 1, cp, m->m_len - cnt - 1);
2822 if (m->m_flags & M_PKTHDR)
2831 panic("tcp_pulloutofband");
2835 * Collect new round-trip time estimate
2836 * and update averages and current timeout.
2839 tcp_xmit_timer(struct tcpcb *tp, int rtt, tcp_seq ack)
2843 tcpstat.tcps_rttupdated++;
2845 if ((tp->t_flags & TF_REBASERTO) && SEQ_GT(ack, tp->snd_max_prev)) {
2846 #ifdef DEBUG_EIFEL_RESPONSE
2847 kprintf("srtt/rttvar, prev %d/%d, cur %d/%d, ",
2848 tp->t_srtt_prev, tp->t_rttvar_prev,
2849 tp->t_srtt, tp->t_rttvar);
2852 tcpstat.tcps_eifelresponse++;
2854 tp->t_flags &= ~TF_REBASERTO;
2855 tp->t_srtt = max(tp->t_srtt_prev, (rtt << TCP_RTT_SHIFT));
2856 tp->t_rttvar = max(tp->t_rttvar_prev,
2857 (rtt << (TCP_RTTVAR_SHIFT - 1)));
2858 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
2859 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2861 #ifdef DEBUG_EIFEL_RESPONSE
2862 kprintf("new %d/%d ", tp->t_srtt, tp->t_rttvar);
2864 } else if (tp->t_srtt != 0) {
2868 * srtt is stored as fixed point with 5 bits after the
2869 * binary point (i.e., scaled by 8). The following magic
2870 * is equivalent to the smoothing algorithm in rfc793 with
2871 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2872 * point). Adjust rtt to origin 0.
2874 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
2875 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
2877 if ((tp->t_srtt += delta) <= 0)
2881 * We accumulate a smoothed rtt variance (actually, a
2882 * smoothed mean difference), then set the retransmit
2883 * timer to smoothed rtt + 4 times the smoothed variance.
2884 * rttvar is stored as fixed point with 4 bits after the
2885 * binary point (scaled by 16). The following is
2886 * equivalent to rfc793 smoothing with an alpha of .75
2887 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
2888 * rfc793's wired-in beta.
2892 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
2893 if ((tp->t_rttvar += delta) <= 0)
2895 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
2896 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2899 * No rtt measurement yet - use the unsmoothed rtt.
2900 * Set the variance to half the rtt (so our first
2901 * retransmit happens at 3*rtt).
2903 tp->t_srtt = rtt << TCP_RTT_SHIFT;
2904 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
2905 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2910 #ifdef DEBUG_EIFEL_RESPONSE
2912 kprintf("| rxtcur prev %d, old %d, ",
2913 tp->t_rxtcur_prev, tp->t_rxtcur);
2918 * the retransmit should happen at rtt + 4 * rttvar.
2919 * Because of the way we do the smoothing, srtt and rttvar
2920 * will each average +1/2 tick of bias. When we compute
2921 * the retransmit timer, we want 1/2 tick of rounding and
2922 * 1 extra tick because of +-1/2 tick uncertainty in the
2923 * firing of the timer. The bias will give us exactly the
2924 * 1.5 tick we need. But, because the bias is
2925 * statistical, we have to test that we don't drop below
2926 * the minimum feasible timer (which is 2 ticks).
2928 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
2929 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
2932 if (tp->t_rxtcur < tp->t_rxtcur_prev + tcp_eifel_rtoinc) {
2934 * RFC4015 requires that the new RTO is at least
2935 * 2*G (tcp_eifel_rtoinc) greater then the RTO
2936 * (t_rxtcur_prev) when the spurious retransmit
2939 * The above condition could be true, if the SRTT
2940 * and RTTVAR used to calculate t_rxtcur_prev
2941 * resulted in a value less than t_rttmin. So
2942 * simply increasing SRTT by tcp_eifel_rtoinc when
2943 * preparing for the Eifel response in
2944 * tcp_save_congestion_state() could not ensure
2945 * that the new RTO will be tcp_eifel_rtoinc greater
2948 tp->t_rxtcur = tp->t_rxtcur_prev + tcp_eifel_rtoinc;
2950 #ifdef DEBUG_EIFEL_RESPONSE
2951 kprintf("new %d\n", tp->t_rxtcur);
2956 * We received an ack for a packet that wasn't retransmitted;
2957 * it is probably safe to discard any error indications we've
2958 * received recently. This isn't quite right, but close enough
2959 * for now (a route might have failed after we sent a segment,
2960 * and the return path might not be symmetrical).
2962 tp->t_softerror = 0;
2966 * Determine a reasonable value for maxseg size.
2967 * If the route is known, check route for mtu.
2968 * If none, use an mss that can be handled on the outgoing
2969 * interface without forcing IP to fragment; if bigger than
2970 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2971 * to utilize large mbufs. If no route is found, route has no mtu,
2972 * or the destination isn't local, use a default, hopefully conservative
2973 * size (usually 512 or the default IP max size, but no more than the mtu
2974 * of the interface), as we can't discover anything about intervening
2975 * gateways or networks. We also initialize the congestion/slow start
2976 * window to be a single segment if the destination isn't local.
2977 * While looking at the routing entry, we also initialize other path-dependent
2978 * parameters from pre-set or cached values in the routing entry.
2980 * Also take into account the space needed for options that we
2981 * send regularly. Make maxseg shorter by that amount to assure
2982 * that we can send maxseg amount of data even when the options
2983 * are present. Store the upper limit of the length of options plus
2986 * NOTE that this routine is only called when we process an incoming
2987 * segment, for outgoing segments only tcp_mssopt is called.
2990 tcp_mss(struct tcpcb *tp, int offer)
2996 struct inpcb *inp = tp->t_inpcb;
2999 boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) ? TRUE : FALSE);
3000 size_t min_protoh = isipv6 ?
3001 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
3002 sizeof(struct tcpiphdr);
3004 const boolean_t isipv6 = FALSE;
3005 const size_t min_protoh = sizeof(struct tcpiphdr);
3009 rt = tcp_rtlookup6(&inp->inp_inc);
3011 rt = tcp_rtlookup(&inp->inp_inc);
3013 tp->t_maxopd = tp->t_maxseg =
3014 (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
3018 so = inp->inp_socket;
3021 * Offer == 0 means that there was no MSS on the SYN segment,
3022 * in this case we use either the interface mtu or tcp_mssdflt.
3024 * An offer which is too large will be cut down later.
3028 if (in6_localaddr(&inp->in6p_faddr)) {
3029 offer = ND_IFINFO(rt->rt_ifp)->linkmtu -
3032 offer = tcp_v6mssdflt;
3035 if (in_localaddr(inp->inp_faddr))
3036 offer = ifp->if_mtu - min_protoh;
3038 offer = tcp_mssdflt;
3043 * Prevent DoS attack with too small MSS. Round up
3044 * to at least minmss.
3046 * Sanity check: make sure that maxopd will be large
3047 * enough to allow some data on segments even is the
3048 * all the option space is used (40bytes). Otherwise
3049 * funny things may happen in tcp_output.
3051 offer = max(offer, tcp_minmss);
3052 offer = max(offer, 64);
3054 rt->rt_rmx.rmx_mssopt = offer;
3057 * While we're here, check if there's an initial rtt
3058 * or rttvar. Convert from the route-table units
3059 * to scaled multiples of the slow timeout timer.
3061 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
3063 * XXX the lock bit for RTT indicates that the value
3064 * is also a minimum value; this is subject to time.
3066 if (rt->rt_rmx.rmx_locks & RTV_RTT)
3067 tp->t_rttmin = rtt / (RTM_RTTUNIT / hz);
3068 tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE));
3069 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
3070 tcpstat.tcps_usedrtt++;
3071 if (rt->rt_rmx.rmx_rttvar) {
3072 tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
3073 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE));
3074 tcpstat.tcps_usedrttvar++;
3076 /* default variation is +- 1 rtt */
3078 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
3080 TCPT_RANGESET(tp->t_rxtcur,
3081 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
3082 tp->t_rttmin, TCPTV_REXMTMAX);
3086 * if there's an mtu associated with the route, use it
3087 * else, use the link mtu. Take the smaller of mss or offer
3090 if (rt->rt_rmx.rmx_mtu) {
3091 mss = rt->rt_rmx.rmx_mtu - min_protoh;
3094 mss = ND_IFINFO(rt->rt_ifp)->linkmtu - min_protoh;
3096 mss = ifp->if_mtu - min_protoh;
3098 mss = min(mss, offer);
3101 * maxopd stores the maximum length of data AND options
3102 * in a segment; maxseg is the amount of data in a normal
3103 * segment. We need to store this value (maxopd) apart
3104 * from maxseg, because now every segment carries options
3105 * and thus we normally have somewhat less data in segments.
3109 if ((tp->t_flags & (TF_REQ_TSTMP | TF_NOOPT)) == TF_REQ_TSTMP &&
3110 ((tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
3111 mss -= TCPOLEN_TSTAMP_APPA;
3113 #if (MCLBYTES & (MCLBYTES - 1)) == 0
3115 mss &= ~(MCLBYTES-1);
3118 mss = mss / MCLBYTES * MCLBYTES;
3121 * If there's a pipesize, change the socket buffer
3122 * to that size. Make the socket buffers an integral
3123 * number of mss units; if the mss is larger than
3124 * the socket buffer, decrease the mss.
3127 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0)
3129 bufsize = so->so_snd.ssb_hiwat;
3133 bufsize = roundup(bufsize, mss);
3134 if (bufsize > sb_max)
3136 if (bufsize > so->so_snd.ssb_hiwat)
3137 ssb_reserve(&so->so_snd, bufsize, so, NULL);
3142 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0)
3144 bufsize = so->so_rcv.ssb_hiwat;
3145 if (bufsize > mss) {
3146 bufsize = roundup(bufsize, mss);
3147 if (bufsize > sb_max)
3149 if (bufsize > so->so_rcv.ssb_hiwat) {
3150 lwkt_gettoken(&so->so_rcv.ssb_token);
3151 ssb_reserve(&so->so_rcv, bufsize, so, NULL);
3152 lwkt_reltoken(&so->so_rcv.ssb_token);
3157 * Set the slow-start flight size
3159 * NOTE: t_maxseg must have been configured!
3161 tp->snd_cwnd = tcp_initial_window(tp);
3163 if (rt->rt_rmx.rmx_ssthresh) {
3165 * There's some sort of gateway or interface
3166 * buffer limit on the path. Use this to set
3167 * the slow start threshhold, but set the
3168 * threshold to no less than 2*mss.
3170 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
3171 tcpstat.tcps_usedssthresh++;
3176 * Determine the MSS option to send on an outgoing SYN.
3179 tcp_mssopt(struct tcpcb *tp)
3184 ((tp->t_inpcb->inp_vflag & INP_IPV6) ? TRUE : FALSE);
3185 int min_protoh = isipv6 ?
3186 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
3187 sizeof(struct tcpiphdr);
3189 const boolean_t isipv6 = FALSE;
3190 const size_t min_protoh = sizeof(struct tcpiphdr);
3194 rt = tcp_rtlookup6(&tp->t_inpcb->inp_inc);
3196 rt = tcp_rtlookup(&tp->t_inpcb->inp_inc);
3198 return (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
3200 return (rt->rt_ifp->if_mtu - min_protoh);
3204 * When a partial ack arrives, force the retransmission of the
3205 * next unacknowledged segment. Do not exit Fast Recovery.
3207 * Implement the Slow-but-Steady variant of NewReno by restarting the
3208 * the retransmission timer. Turn it off here so it can be restarted
3209 * later in tcp_output().
3212 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th, int acked)
3214 tcp_seq old_snd_nxt = tp->snd_nxt;
3215 u_long ocwnd = tp->snd_cwnd;
3217 tcp_callout_stop(tp, tp->tt_rexmt);
3219 tp->snd_nxt = th->th_ack;
3220 /* Set snd_cwnd to one segment beyond acknowledged offset. */
3221 tp->snd_cwnd = tp->t_maxseg;
3222 tp->t_flags |= TF_ACKNOW;
3224 if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
3225 tp->snd_nxt = old_snd_nxt;
3226 /* partial window deflation */
3228 tp->snd_cwnd = ocwnd - acked + tp->t_maxseg;
3230 tp->snd_cwnd = tp->t_maxseg;
3234 * In contrast to the Slow-but-Steady NewReno variant,
3235 * we do not reset the retransmission timer for SACK retransmissions,
3236 * except when retransmitting snd_una.
3239 tcp_sack_rexmt(struct tcpcb *tp, struct tcphdr *th)
3241 tcp_seq old_snd_nxt = tp->snd_nxt;
3242 u_long ocwnd = tp->snd_cwnd;
3244 int nseg = 0; /* consecutive new segments */
3245 int nseg_rexmt = 0; /* retransmitted segments */
3246 #define MAXBURST 4 /* limit burst of new packets on partial ack */
3249 pipe = tcp_sack_compute_pipe(tp);
3250 while ((tcp_seq_diff_t)(ocwnd - pipe) >= (tcp_seq_diff_t)tp->t_maxseg &&
3251 (!tcp_do_smartsack || nseg < MAXBURST)) {
3252 tcp_seq old_snd_max, old_rexmt_high, nextrexmt;
3253 uint32_t sent, seglen;
3257 old_rexmt_high = tp->rexmt_high;
3258 if (!tcp_sack_nextseg(tp, &nextrexmt, &seglen, &rescue)) {
3259 tp->rexmt_high = old_rexmt_high;
3264 * If the next tranmission is a rescue retranmission,
3265 * we check whether we have already sent some data
3266 * (either new segments or retransmitted segments)
3267 * into the the network or not. Since the idea of rescue
3268 * retransmission is to sustain ACK clock, as long as
3269 * some segments are in the network, ACK clock will be
3272 if (rescue && (nseg_rexmt > 0 || nseg > 0)) {
3273 tp->rexmt_high = old_rexmt_high;
3277 if (nextrexmt == tp->snd_max)
3281 tp->snd_nxt = nextrexmt;
3282 tp->snd_cwnd = nextrexmt - tp->snd_una + seglen;
3283 old_snd_max = tp->snd_max;
3284 if (nextrexmt == tp->snd_una)
3285 tcp_callout_stop(tp, tp->tt_rexmt);
3286 error = tcp_output(tp);
3288 tp->rexmt_high = old_rexmt_high;
3291 sent = tp->snd_nxt - nextrexmt;
3293 tp->rexmt_high = old_rexmt_high;
3297 tcpstat.tcps_sndsackpack++;
3298 tcpstat.tcps_sndsackbyte += sent;
3301 tcpstat.tcps_sackrescue++;
3302 tp->rexmt_rescue = tp->snd_nxt;
3303 tp->sack_flags |= TSACK_F_SACKRESCUED;
3306 if (SEQ_LT(nextrexmt, old_snd_max) &&
3307 SEQ_LT(tp->rexmt_high, tp->snd_nxt)) {
3308 tp->rexmt_high = seq_min(tp->snd_nxt, old_snd_max);
3309 if (tcp_aggressive_rescuesack &&
3310 (tp->sack_flags & TSACK_F_SACKRESCUED) &&
3311 SEQ_LT(tp->rexmt_rescue, tp->rexmt_high)) {
3312 /* Drag RescueRxt along with HighRxt */
3313 tp->rexmt_rescue = tp->rexmt_high;
3317 if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
3318 tp->snd_nxt = old_snd_nxt;
3319 tp->snd_cwnd = ocwnd;
3323 tcp_sack_limitedxmit(struct tcpcb *tp)
3325 tcp_seq oldsndnxt = tp->snd_nxt;
3326 tcp_seq oldsndmax = tp->snd_max;
3327 u_long ocwnd = tp->snd_cwnd;
3329 boolean_t ret = FALSE;
3331 tp->rexmt_high = tp->snd_una - 1;
3332 pipe = tcp_sack_compute_pipe(tp);
3333 while ((tcp_seq_diff_t)(ocwnd - pipe) >= (tcp_seq_diff_t)tp->t_maxseg) {
3338 next = tp->snd_nxt = tp->snd_max;
3339 tp->snd_cwnd = tp->snd_nxt - tp->snd_una + tp->t_maxseg;
3341 error = tcp_output(tp);
3345 sent = tp->snd_nxt - next;
3349 ++tcpstat.tcps_sndlimited;
3353 if (SEQ_LT(oldsndnxt, oldsndmax)) {
3354 KASSERT(SEQ_GEQ(oldsndnxt, tp->snd_una),
3355 ("snd_una moved in other threads"));
3356 tp->snd_nxt = oldsndnxt;
3358 tp->snd_cwnd = ocwnd;
3364 * Reset idle time and keep-alive timer, typically called when a valid
3365 * tcp packet is received but may also be called when FASTKEEP is set
3366 * to prevent the previous long-timeout from calculating to a drop.
3368 * Only update t_rcvtime for non-SYN packets.
3370 * Handle the case where one side thinks the connection is established
3371 * but the other side has, say, rebooted without cleaning out the
3372 * connection. The SYNs could be construed as an attack and wind
3373 * up ignored, but in case it isn't an attack we can validate the
3374 * connection by forcing a keepalive.
3377 tcp_timer_keep_activity(struct tcpcb *tp, int thflags)
3379 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
3380 if ((thflags & (TH_SYN | TH_ACK)) == TH_SYN) {
3381 tp->t_flags |= TF_KEEPALIVE;
3382 tcp_callout_reset(tp, tp->tt_keep, hz / 2,
3385 tp->t_rcvtime = ticks;
3386 tp->t_flags &= ~TF_KEEPALIVE;
3387 tcp_callout_reset(tp, tp->tt_keep,
3395 tcp_rmx_msl(const struct tcpcb *tp)
3398 struct inpcb *inp = tp->t_inpcb;
3401 boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) ? TRUE : FALSE);
3403 const boolean_t isipv6 = FALSE;
3407 rt = tcp_rtlookup6(&inp->inp_inc);
3409 rt = tcp_rtlookup(&inp->inp_inc);
3410 if (rt == NULL || rt->rt_rmx.rmx_msl == 0)
3413 msl = (rt->rt_rmx.rmx_msl * hz) / 1000;
3421 tcp_established(struct tcpcb *tp)
3423 tp->t_state = TCPS_ESTABLISHED;
3424 tcp_callout_reset(tp, tp->tt_keep, tp->t_keepidle, tcp_timer_keep);
3426 if (tp->t_rxtsyn > 0) {
3429 * "If the timer expires awaiting the ACK of a SYN segment
3430 * and the TCP implementation is using an RTO less than 3
3431 * seconds, the RTO MUST be re-initialized to 3 seconds
3432 * when data transmission begins"
3434 if (tp->t_rxtcur < TCPTV_RTOBASE3)
3435 tp->t_rxtcur = TCPTV_RTOBASE3;