2 * Copyright (c) 2002, 2003, 2004 Jeffrey M. Hsu. All rights reserved.
3 * Copyright (c) 2002, 2003, 2004 The DragonFly Project. All rights reserved.
5 * This code is derived from software contributed to The DragonFly Project
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 3. Neither the name of The DragonFly Project nor the names of its
17 * contributors may be used to endorse or promote products derived
18 * from this software without specific, prior written permission.
20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
23 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
24 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
25 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
26 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
27 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
28 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
29 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
30 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
36 * The Regents of the University of California. All rights reserved.
38 * Redistribution and use in source and binary forms, with or without
39 * modification, are permitted provided that the following conditions
41 * 1. Redistributions of source code must retain the above copyright
42 * notice, this list of conditions and the following disclaimer.
43 * 2. Redistributions in binary form must reproduce the above copyright
44 * notice, this list of conditions and the following disclaimer in the
45 * documentation and/or other materials provided with the distribution.
46 * 3. All advertising materials mentioning features or use of this software
47 * must display the following acknowledgement:
48 * This product includes software developed by the University of
49 * California, Berkeley and its contributors.
50 * 4. Neither the name of the University nor the names of its contributors
51 * may be used to endorse or promote products derived from this software
52 * without specific prior written permission.
54 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
55 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
56 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
57 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
58 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
59 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
60 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
61 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
62 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
63 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
66 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95
67 * $FreeBSD: src/sys/netinet/tcp_input.c,v 1.107.2.38 2003/05/21 04:46:41 cjc Exp $
68 * $DragonFly: src/sys/netinet/tcp_input.c,v 1.68 2008/08/22 09:14:17 sephe Exp $
71 #include "opt_ipfw.h" /* for ipfw_fwd */
73 #include "opt_inet6.h"
74 #include "opt_ipsec.h"
75 #include "opt_tcpdebug.h"
76 #include "opt_tcp_input.h"
78 #include <sys/param.h>
79 #include <sys/systm.h>
80 #include <sys/kernel.h>
81 #include <sys/sysctl.h>
82 #include <sys/malloc.h>
84 #include <sys/proc.h> /* for proc0 declaration */
85 #include <sys/protosw.h>
86 #include <sys/socket.h>
87 #include <sys/socketvar.h>
88 #include <sys/syslog.h>
89 #include <sys/in_cksum.h>
91 #include <sys/socketvar2.h>
93 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
94 #include <machine/stdarg.h>
97 #include <net/route.h>
99 #include <netinet/in.h>
100 #include <netinet/in_systm.h>
101 #include <netinet/ip.h>
102 #include <netinet/ip_icmp.h> /* for ICMP_BANDLIM */
103 #include <netinet/in_var.h>
104 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */
105 #include <netinet/in_pcb.h>
106 #include <netinet/ip_var.h>
107 #include <netinet/ip6.h>
108 #include <netinet/icmp6.h>
109 #include <netinet6/nd6.h>
110 #include <netinet6/ip6_var.h>
111 #include <netinet6/in6_pcb.h>
112 #include <netinet/tcp.h>
113 #include <netinet/tcp_fsm.h>
114 #include <netinet/tcp_seq.h>
115 #include <netinet/tcp_timer.h>
116 #include <netinet/tcp_timer2.h>
117 #include <netinet/tcp_var.h>
118 #include <netinet6/tcp6_var.h>
119 #include <netinet/tcpip.h>
122 #include <netinet/tcp_debug.h>
124 u_char tcp_saveipgen[40]; /* the size must be of max ip header, now IPv6 */
125 struct tcphdr tcp_savetcp;
129 #include <netproto/ipsec/ipsec.h>
130 #include <netproto/ipsec/ipsec6.h>
134 #include <netinet6/ipsec.h>
135 #include <netinet6/ipsec6.h>
136 #include <netproto/key/key.h>
139 MALLOC_DEFINE(M_TSEGQ, "tseg_qent", "TCP segment queue entry");
141 static int log_in_vain = 0;
142 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
143 &log_in_vain, 0, "Log all incoming TCP connections");
145 static int blackhole = 0;
146 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW,
147 &blackhole, 0, "Do not send RST when dropping refused connections");
149 int tcp_delack_enabled = 1;
150 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW,
151 &tcp_delack_enabled, 0,
152 "Delay ACK to try and piggyback it onto a data packet");
154 #ifdef TCP_DROP_SYNFIN
155 static int drop_synfin = 0;
156 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW,
157 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set");
160 static int tcp_do_limitedtransmit = 1;
161 SYSCTL_INT(_net_inet_tcp, OID_AUTO, limitedtransmit, CTLFLAG_RW,
162 &tcp_do_limitedtransmit, 0, "Enable RFC 3042 (Limited Transmit)");
164 static int tcp_do_early_retransmit = 1;
165 SYSCTL_INT(_net_inet_tcp, OID_AUTO, earlyretransmit, CTLFLAG_RW,
166 &tcp_do_early_retransmit, 0, "Early retransmit");
168 int tcp_aggregate_acks = 1;
169 SYSCTL_INT(_net_inet_tcp, OID_AUTO, aggregate_acks, CTLFLAG_RW,
170 &tcp_aggregate_acks, 0, "Aggregate built-up acks into one ack");
172 int tcp_do_rfc3390 = 1;
173 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW,
175 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
177 static int tcp_do_eifel_detect = 1;
178 SYSCTL_INT(_net_inet_tcp, OID_AUTO, eifel, CTLFLAG_RW,
179 &tcp_do_eifel_detect, 0, "Eifel detection algorithm (RFC 3522)");
181 static int tcp_do_abc = 1;
182 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc, CTLFLAG_RW,
184 "TCP Appropriate Byte Counting (RFC 3465)");
187 * Define as tunable for easy testing with SACK on and off.
188 * Warning: do not change setting in the middle of an existing active TCP flow,
189 * else strange things might happen to that flow.
192 SYSCTL_INT(_net_inet_tcp, OID_AUTO, sack, CTLFLAG_RW,
193 &tcp_do_sack, 0, "Enable SACK Algorithms");
195 int tcp_do_smartsack = 1;
196 SYSCTL_INT(_net_inet_tcp, OID_AUTO, smartsack, CTLFLAG_RW,
197 &tcp_do_smartsack, 0, "Enable Smart SACK Algorithms");
199 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0,
200 "TCP Segment Reassembly Queue");
202 int tcp_reass_maxseg = 0;
203 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RD,
204 &tcp_reass_maxseg, 0,
205 "Global maximum number of TCP Segments in Reassembly Queue");
207 int tcp_reass_qsize = 0;
208 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD,
210 "Global number of TCP Segments currently in Reassembly Queue");
212 static int tcp_reass_overflows = 0;
213 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD,
214 &tcp_reass_overflows, 0,
215 "Global number of TCP Segment Reassembly Queue Overflows");
217 int tcp_do_autorcvbuf = 1;
218 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_RW,
219 &tcp_do_autorcvbuf, 0, "Enable automatic receive buffer sizing");
221 int tcp_autorcvbuf_inc = 16*1024;
222 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_RW,
223 &tcp_autorcvbuf_inc, 0,
224 "Incrementor step size of automatic receive buffer");
226 int tcp_autorcvbuf_max = 2*1024*1024;
227 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_RW,
228 &tcp_autorcvbuf_max, 0, "Max size of automatic receive buffer");
231 static void tcp_dooptions(struct tcpopt *, u_char *, int, boolean_t);
232 static void tcp_pulloutofband(struct socket *,
233 struct tcphdr *, struct mbuf *, int);
234 static int tcp_reass(struct tcpcb *, struct tcphdr *, int *,
236 static void tcp_xmit_timer(struct tcpcb *, int);
237 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *, int);
238 static void tcp_sack_rexmt(struct tcpcb *, struct tcphdr *);
240 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
242 #define ND6_HINT(tp) \
244 if ((tp) && (tp)->t_inpcb && \
245 ((tp)->t_inpcb->inp_vflag & INP_IPV6) && \
246 (tp)->t_inpcb->in6p_route.ro_rt) \
247 nd6_nud_hint((tp)->t_inpcb->in6p_route.ro_rt, NULL, 0); \
254 * Indicate whether this ack should be delayed. We can delay the ack if
255 * - delayed acks are enabled and
256 * - there is no delayed ack timer in progress and
257 * - our last ack wasn't a 0-sized window. We never want to delay
258 * the ack that opens up a 0-sized window.
260 #define DELAY_ACK(tp) \
261 (tcp_delack_enabled && !tcp_callout_pending(tp, tp->tt_delack) && \
262 !(tp->t_flags & TF_RXWIN0SENT))
264 #define acceptable_window_update(tp, th, tiwin) \
265 (SEQ_LT(tp->snd_wl1, th->th_seq) || \
266 (tp->snd_wl1 == th->th_seq && \
267 (SEQ_LT(tp->snd_wl2, th->th_ack) || \
268 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))
271 tcp_reass(struct tcpcb *tp, struct tcphdr *th, int *tlenp, struct mbuf *m)
274 struct tseg_qent *p = NULL;
275 struct tseg_qent *te;
276 struct socket *so = tp->t_inpcb->inp_socket;
280 * Call with th == NULL after become established to
281 * force pre-ESTABLISHED data up to user socket.
287 * Limit the number of segments in the reassembly queue to prevent
288 * holding on to too many segments (and thus running out of mbufs).
289 * Make sure to let the missing segment through which caused this
290 * queue. Always keep one global queue entry spare to be able to
291 * process the missing segment.
293 if (th->th_seq != tp->rcv_nxt &&
294 tcp_reass_qsize + 1 >= tcp_reass_maxseg) {
295 tcp_reass_overflows++;
296 tcpstat.tcps_rcvmemdrop++;
298 /* no SACK block to report */
299 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
303 /* Allocate a new queue entry. */
304 MALLOC(te, struct tseg_qent *, sizeof(struct tseg_qent), M_TSEGQ,
305 M_INTWAIT | M_NULLOK);
307 tcpstat.tcps_rcvmemdrop++;
309 /* no SACK block to report */
310 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
313 atomic_add_int(&tcp_reass_qsize, 1);
316 * Find a segment which begins after this one does.
318 LIST_FOREACH(q, &tp->t_segq, tqe_q) {
319 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
325 * If there is a preceding segment, it may provide some of
326 * our data already. If so, drop the data from the incoming
327 * segment. If it provides all of our data, drop us.
332 /* conversion to int (in i) handles seq wraparound */
333 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
334 if (i > 0) { /* overlaps preceding segment */
335 tp->t_flags |= (TF_DUPSEG | TF_ENCLOSESEG);
336 /* enclosing block starts w/ preceding segment */
337 tp->encloseblk.rblk_start = p->tqe_th->th_seq;
339 /* preceding encloses incoming segment */
340 tp->encloseblk.rblk_end = p->tqe_th->th_seq +
342 tcpstat.tcps_rcvduppack++;
343 tcpstat.tcps_rcvdupbyte += *tlenp;
346 atomic_add_int(&tcp_reass_qsize, -1);
348 * Try to present any queued data
349 * at the left window edge to the user.
350 * This is needed after the 3-WHS
353 goto present; /* ??? */
358 /* incoming segment end is enclosing block end */
359 tp->encloseblk.rblk_end = th->th_seq + *tlenp +
360 ((th->th_flags & TH_FIN) != 0);
361 /* trim end of reported D-SACK block */
362 tp->reportblk.rblk_end = th->th_seq;
365 tcpstat.tcps_rcvoopack++;
366 tcpstat.tcps_rcvoobyte += *tlenp;
369 * While we overlap succeeding segments trim them or,
370 * if they are completely covered, dequeue them.
373 tcp_seq_diff_t i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
374 tcp_seq qend = q->tqe_th->th_seq + q->tqe_len;
375 struct tseg_qent *nq;
379 if (!(tp->t_flags & TF_DUPSEG)) { /* first time through */
380 tp->t_flags |= (TF_DUPSEG | TF_ENCLOSESEG);
381 tp->encloseblk = tp->reportblk;
382 /* report trailing duplicate D-SACK segment */
383 tp->reportblk.rblk_start = q->tqe_th->th_seq;
385 if ((tp->t_flags & TF_ENCLOSESEG) &&
386 SEQ_GT(qend, tp->encloseblk.rblk_end)) {
387 /* extend enclosing block if one exists */
388 tp->encloseblk.rblk_end = qend;
390 if (i < q->tqe_len) {
391 q->tqe_th->th_seq += i;
397 nq = LIST_NEXT(q, tqe_q);
398 LIST_REMOVE(q, tqe_q);
401 atomic_add_int(&tcp_reass_qsize, -1);
405 /* Insert the new segment queue entry into place. */
408 te->tqe_len = *tlenp;
410 /* check if can coalesce with following segment */
411 if (q != NULL && (th->th_seq + *tlenp == q->tqe_th->th_seq)) {
412 tcp_seq tend = te->tqe_th->th_seq + te->tqe_len;
414 te->tqe_len += q->tqe_len;
415 if (q->tqe_th->th_flags & TH_FIN)
416 te->tqe_th->th_flags |= TH_FIN;
417 m_cat(te->tqe_m, q->tqe_m);
418 tp->encloseblk.rblk_end = tend;
420 * When not reporting a duplicate segment, use
421 * the larger enclosing block as the SACK block.
423 if (!(tp->t_flags & TF_DUPSEG))
424 tp->reportblk.rblk_end = tend;
425 LIST_REMOVE(q, tqe_q);
427 atomic_add_int(&tcp_reass_qsize, -1);
431 LIST_INSERT_HEAD(&tp->t_segq, te, tqe_q);
433 /* check if can coalesce with preceding segment */
434 if (p->tqe_th->th_seq + p->tqe_len == th->th_seq) {
435 p->tqe_len += te->tqe_len;
436 m_cat(p->tqe_m, te->tqe_m);
437 tp->encloseblk.rblk_start = p->tqe_th->th_seq;
439 * When not reporting a duplicate segment, use
440 * the larger enclosing block as the SACK block.
442 if (!(tp->t_flags & TF_DUPSEG))
443 tp->reportblk.rblk_start = p->tqe_th->th_seq;
445 atomic_add_int(&tcp_reass_qsize, -1);
447 LIST_INSERT_AFTER(p, te, tqe_q);
453 * Present data to user, advancing rcv_nxt through
454 * completed sequence space.
456 if (!TCPS_HAVEESTABLISHED(tp->t_state))
458 q = LIST_FIRST(&tp->t_segq);
459 if (q == NULL || q->tqe_th->th_seq != tp->rcv_nxt)
461 tp->rcv_nxt += q->tqe_len;
462 if (!(tp->t_flags & TF_DUPSEG)) {
463 /* no SACK block to report since ACK advanced */
464 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
466 /* no enclosing block to report since ACK advanced */
467 tp->t_flags &= ~TF_ENCLOSESEG;
468 flags = q->tqe_th->th_flags & TH_FIN;
469 LIST_REMOVE(q, tqe_q);
470 KASSERT(LIST_EMPTY(&tp->t_segq) ||
471 LIST_FIRST(&tp->t_segq)->tqe_th->th_seq != tp->rcv_nxt,
472 ("segment not coalesced"));
473 if (so->so_state & SS_CANTRCVMORE) {
476 lwkt_gettoken(&so->so_rcv.ssb_token);
477 ssb_appendstream(&so->so_rcv, q->tqe_m);
478 lwkt_reltoken(&so->so_rcv.ssb_token);
481 atomic_add_int(&tcp_reass_qsize, -1);
488 * TCP input routine, follows pages 65-76 of the
489 * protocol specification dated September, 1981 very closely.
493 tcp6_input(struct mbuf **mp, int *offp, int proto)
495 struct mbuf *m = *mp;
496 struct in6_ifaddr *ia6;
498 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
501 * draft-itojun-ipv6-tcp-to-anycast
502 * better place to put this in?
504 ia6 = ip6_getdstifaddr(m);
505 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
508 ip6 = mtod(m, struct ip6_hdr *);
509 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
510 offsetof(struct ip6_hdr, ip6_dst));
511 return (IPPROTO_DONE);
514 tcp_input(mp, offp, proto);
515 return (IPPROTO_DONE);
520 tcp_input(struct mbuf **mp, int *offp, int proto)
524 struct ip *ip = NULL;
526 struct inpcb *inp = NULL;
532 struct tcpcb *tp = NULL;
534 struct socket *so = 0;
536 boolean_t ourfinisacked, needoutput = FALSE;
539 struct tcpopt to; /* options in this segment */
540 struct sockaddr_in *next_hop = NULL;
541 int rstreason; /* For badport_bandlim accounting purposes */
543 struct ip6_hdr *ip6 = NULL;
548 const boolean_t isipv6 = FALSE;
558 tcpstat.tcps_rcvtotal++;
560 if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
563 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
564 KKASSERT(mtag != NULL);
565 next_hop = m_tag_data(mtag);
569 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? TRUE : FALSE;
573 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */
574 ip6 = mtod(m, struct ip6_hdr *);
575 tlen = (sizeof *ip6) + ntohs(ip6->ip6_plen) - off0;
576 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) {
577 tcpstat.tcps_rcvbadsum++;
580 th = (struct tcphdr *)((caddr_t)ip6 + off0);
583 * Be proactive about unspecified IPv6 address in source.
584 * As we use all-zero to indicate unbounded/unconnected pcb,
585 * unspecified IPv6 address can be used to confuse us.
587 * Note that packets with unspecified IPv6 destination is
588 * already dropped in ip6_input.
590 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
596 * Get IP and TCP header together in first mbuf.
597 * Note: IP leaves IP header in first mbuf.
599 if (off0 > sizeof(struct ip)) {
601 off0 = sizeof(struct ip);
603 /* already checked and pulled up in ip_demux() */
604 KASSERT(m->m_len >= sizeof(struct tcpiphdr),
605 ("TCP header not in one mbuf: m->m_len %d", m->m_len));
606 ip = mtod(m, struct ip *);
607 ipov = (struct ipovly *)ip;
608 th = (struct tcphdr *)((caddr_t)ip + off0);
611 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
612 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
613 th->th_sum = m->m_pkthdr.csum_data;
615 th->th_sum = in_pseudo(ip->ip_src.s_addr,
617 htonl(m->m_pkthdr.csum_data +
620 th->th_sum ^= 0xffff;
623 * Checksum extended TCP header and data.
625 len = sizeof(struct ip) + tlen;
626 bzero(ipov->ih_x1, sizeof ipov->ih_x1);
627 ipov->ih_len = (u_short)tlen;
628 ipov->ih_len = htons(ipov->ih_len);
629 th->th_sum = in_cksum(m, len);
632 tcpstat.tcps_rcvbadsum++;
636 /* Re-initialization for later version check */
637 ip->ip_v = IPVERSION;
642 * Check that TCP offset makes sense,
643 * pull out TCP options and adjust length. XXX
645 off = th->th_off << 2;
646 /* already checked and pulled up in ip_demux() */
647 KASSERT(off >= sizeof(struct tcphdr) && off <= tlen,
648 ("bad TCP data offset %d (tlen %d)", off, tlen));
649 tlen -= off; /* tlen is used instead of ti->ti_len */
650 if (off > sizeof(struct tcphdr)) {
652 IP6_EXTHDR_CHECK(m, off0, off, IPPROTO_DONE);
653 ip6 = mtod(m, struct ip6_hdr *);
654 th = (struct tcphdr *)((caddr_t)ip6 + off0);
656 /* already pulled up in ip_demux() */
657 KASSERT(m->m_len >= sizeof(struct ip) + off,
658 ("TCP header and options not in one mbuf: "
659 "m_len %d, off %d", m->m_len, off));
661 optlen = off - sizeof(struct tcphdr);
662 optp = (u_char *)(th + 1);
664 thflags = th->th_flags;
666 #ifdef TCP_DROP_SYNFIN
668 * If the drop_synfin option is enabled, drop all packets with
669 * both the SYN and FIN bits set. This prevents e.g. nmap from
670 * identifying the TCP/IP stack.
672 * This is a violation of the TCP specification.
674 if (drop_synfin && (thflags & (TH_SYN | TH_FIN)) == (TH_SYN | TH_FIN))
679 * Convert TCP protocol specific fields to host format.
681 th->th_seq = ntohl(th->th_seq);
682 th->th_ack = ntohl(th->th_ack);
683 th->th_win = ntohs(th->th_win);
684 th->th_urp = ntohs(th->th_urp);
687 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options,
688 * until after ip6_savecontrol() is called and before other functions
689 * which don't want those proto headers.
690 * Because ip6_savecontrol() is going to parse the mbuf to
691 * search for data to be passed up to user-land, it wants mbuf
692 * parameters to be unchanged.
693 * XXX: the call of ip6_savecontrol() has been obsoleted based on
694 * latest version of the advanced API (20020110).
696 drop_hdrlen = off0 + off;
699 * Locate pcb for segment.
702 /* IPFIREWALL_FORWARD section */
703 if (next_hop != NULL && !isipv6) { /* IPv6 support is not there yet */
705 * Transparently forwarded. Pretend to be the destination.
706 * already got one like this?
708 cpu = mycpu->gd_cpuid;
709 inp = in_pcblookup_hash(&tcbinfo[cpu],
710 ip->ip_src, th->th_sport,
711 ip->ip_dst, th->th_dport,
712 0, m->m_pkthdr.rcvif);
715 * It's new. Try to find the ambushing socket.
719 * The rest of the ipfw code stores the port in
721 * (The IP address is still in network order.)
723 in_port_t dport = next_hop->sin_port ?
724 htons(next_hop->sin_port) :
727 cpu = tcp_addrcpu(ip->ip_src.s_addr, th->th_sport,
728 next_hop->sin_addr.s_addr, dport);
729 inp = in_pcblookup_hash(&tcbinfo[cpu],
730 ip->ip_src, th->th_sport,
731 next_hop->sin_addr, dport,
732 1, m->m_pkthdr.rcvif);
736 inp = in6_pcblookup_hash(&tcbinfo[0],
737 &ip6->ip6_src, th->th_sport,
738 &ip6->ip6_dst, th->th_dport,
739 1, m->m_pkthdr.rcvif);
741 cpu = mycpu->gd_cpuid;
742 inp = in_pcblookup_hash(&tcbinfo[cpu],
743 ip->ip_src, th->th_sport,
744 ip->ip_dst, th->th_dport,
745 1, m->m_pkthdr.rcvif);
750 * If the state is CLOSED (i.e., TCB does not exist) then
751 * all data in the incoming segment is discarded.
752 * If the TCB exists but is in CLOSED state, it is embryonic,
753 * but should either do a listen or a connect soon.
758 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2];
760 char dbuf[sizeof "aaa.bbb.ccc.ddd"];
761 char sbuf[sizeof "aaa.bbb.ccc.ddd"];
765 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst));
768 strcat(sbuf, ip6_sprintf(&ip6->ip6_src));
771 strcpy(dbuf, inet_ntoa(ip->ip_dst));
772 strcpy(sbuf, inet_ntoa(ip->ip_src));
774 switch (log_in_vain) {
776 if (!(thflags & TH_SYN))
780 "Connection attempt to TCP %s:%d "
781 "from %s:%d flags:0x%02x\n",
782 dbuf, ntohs(th->th_dport), sbuf,
783 ntohs(th->th_sport), thflags);
792 if (thflags & TH_SYN)
801 rstreason = BANDLIM_RST_CLOSEDPORT;
807 if (ipsec6_in_reject_so(m, inp->inp_socket)) {
808 ipsec6stat.in_polvio++;
812 if (ipsec4_in_reject_so(m, inp->inp_socket)) {
813 ipsecstat.in_polvio++;
820 if (ipsec6_in_reject(m, inp))
823 if (ipsec4_in_reject(m, inp))
827 /* Check the minimum TTL for socket. */
829 if ((isipv6 ? ip6->ip6_hlim : ip->ip_ttl) < inp->inp_ip_minttl)
835 rstreason = BANDLIM_RST_CLOSEDPORT;
838 if (tp->t_state <= TCPS_CLOSED)
841 /* Unscale the window into a 32-bit value. */
842 if (!(thflags & TH_SYN))
843 tiwin = th->th_win << tp->snd_scale;
847 so = inp->inp_socket;
850 if (so->so_options & SO_DEBUG) {
851 ostate = tp->t_state;
853 bcopy(ip6, tcp_saveipgen, sizeof(*ip6));
855 bcopy(ip, tcp_saveipgen, sizeof(*ip));
860 bzero(&to, sizeof to);
862 if (so->so_options & SO_ACCEPTCONN) {
863 struct in_conninfo inc;
866 inc.inc_isipv6 = (isipv6 == TRUE);
869 inc.inc6_faddr = ip6->ip6_src;
870 inc.inc6_laddr = ip6->ip6_dst;
871 inc.inc6_route.ro_rt = NULL; /* XXX */
873 inc.inc_faddr = ip->ip_src;
874 inc.inc_laddr = ip->ip_dst;
875 inc.inc_route.ro_rt = NULL; /* XXX */
877 inc.inc_fport = th->th_sport;
878 inc.inc_lport = th->th_dport;
881 * If the state is LISTEN then ignore segment if it contains
882 * a RST. If the segment contains an ACK then it is bad and
883 * send a RST. If it does not contain a SYN then it is not
884 * interesting; drop it.
886 * If the state is SYN_RECEIVED (syncache) and seg contains
887 * an ACK, but not for our SYN/ACK, send a RST. If the seg
888 * contains a RST, check the sequence number to see if it
889 * is a valid reset segment.
891 if ((thflags & (TH_RST | TH_ACK | TH_SYN)) != TH_SYN) {
892 if ((thflags & (TH_RST | TH_ACK | TH_SYN)) == TH_ACK) {
893 if (!syncache_expand(&inc, th, &so, m)) {
895 * No syncache entry, or ACK was not
896 * for our SYN/ACK. Send a RST.
898 tcpstat.tcps_badsyn++;
899 rstreason = BANDLIM_RST_OPENPORT;
904 * Could not complete 3-way handshake,
905 * connection is being closed down, and
906 * syncache will free mbuf.
909 return(IPPROTO_DONE);
912 * We must be in the correct protocol thread
913 * for this connection.
915 KKASSERT(so->so_port == &curthread->td_msgport);
918 * Socket is created in state SYN_RECEIVED.
919 * Continue processing segment.
924 * This is what would have happened in
925 * tcp_output() when the SYN,ACK was sent.
927 tp->snd_up = tp->snd_una;
928 tp->snd_max = tp->snd_nxt = tp->iss + 1;
929 tp->last_ack_sent = tp->rcv_nxt;
931 * XXX possible bug - it doesn't appear that tp->snd_wnd is unscaled
932 * until the _second_ ACK is received:
933 * rcv SYN (set wscale opts) --> send SYN/ACK, set snd_wnd = window.
934 * rcv ACK, calculate tiwin --> process SYN_RECEIVED, determine wscale,
935 * move to ESTAB, set snd_wnd to tiwin.
937 tp->snd_wnd = tiwin; /* unscaled */
940 if (thflags & TH_RST) {
941 syncache_chkrst(&inc, th);
944 if (thflags & TH_ACK) {
945 syncache_badack(&inc);
946 tcpstat.tcps_badsyn++;
947 rstreason = BANDLIM_RST_OPENPORT;
954 * Segment's flags are (SYN) or (SYN | FIN).
958 * If deprecated address is forbidden,
959 * we do not accept SYN to deprecated interface
960 * address to prevent any new inbound connection from
961 * getting established.
962 * When we do not accept SYN, we send a TCP RST,
963 * with deprecated source address (instead of dropping
964 * it). We compromise it as it is much better for peer
965 * to send a RST, and RST will be the final packet
968 * If we do not forbid deprecated addresses, we accept
969 * the SYN packet. RFC2462 does not suggest dropping
971 * If we decipher RFC2462 5.5.4, it says like this:
972 * 1. use of deprecated addr with existing
973 * communication is okay - "SHOULD continue to be
975 * 2. use of it with new communication:
976 * (2a) "SHOULD NOT be used if alternate address
977 * with sufficient scope is available"
978 * (2b) nothing mentioned otherwise.
979 * Here we fall into (2b) case as we have no choice in
980 * our source address selection - we must obey the peer.
982 * The wording in RFC2462 is confusing, and there are
983 * multiple description text for deprecated address
984 * handling - worse, they are not exactly the same.
985 * I believe 5.5.4 is the best one, so we follow 5.5.4.
987 if (isipv6 && !ip6_use_deprecated) {
988 struct in6_ifaddr *ia6;
990 if ((ia6 = ip6_getdstifaddr(m)) &&
991 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
993 rstreason = BANDLIM_RST_OPENPORT;
999 * If it is from this socket, drop it, it must be forged.
1000 * Don't bother responding if the destination was a broadcast.
1002 if (th->th_dport == th->th_sport) {
1004 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
1008 if (ip->ip_dst.s_addr == ip->ip_src.s_addr)
1013 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
1015 * Note that it is quite possible to receive unicast
1016 * link-layer packets with a broadcast IP address. Use
1017 * in_broadcast() to find them.
1019 if (m->m_flags & (M_BCAST | M_MCAST))
1022 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1023 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
1026 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1027 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1028 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1029 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
1033 * SYN appears to be valid; create compressed TCP state
1034 * for syncache, or perform t/tcp connection.
1036 if (so->so_qlen <= so->so_qlimit) {
1037 tcp_dooptions(&to, optp, optlen, TRUE);
1038 if (!syncache_add(&inc, &to, th, &so, m))
1042 * Entry added to syncache, mbuf used to
1043 * send SYN,ACK packet.
1046 return(IPPROTO_DONE);
1049 * We must be in the correct protocol thread for
1052 KKASSERT(so->so_port == &curthread->td_msgport);
1055 tp = intotcpcb(inp);
1056 tp->snd_wnd = tiwin;
1057 tp->t_starttime = ticks;
1058 tp->t_state = TCPS_ESTABLISHED;
1061 * If there is a FIN, or if there is data and the
1062 * connection is local, then delay SYN,ACK(SYN) in
1063 * the hope of piggy-backing it on a response
1064 * segment. Otherwise must send ACK now in case
1065 * the other side is slow starting.
1067 if (DELAY_ACK(tp) &&
1068 ((thflags & TH_FIN) ||
1070 ((isipv6 && in6_localaddr(&inp->in6p_faddr)) ||
1071 (!isipv6 && in_localaddr(inp->inp_faddr)))))) {
1072 tcp_callout_reset(tp, tp->tt_delack,
1073 tcp_delacktime, tcp_timer_delack);
1074 tp->t_flags |= TF_NEEDSYN;
1076 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
1079 tcpstat.tcps_connects++;
1088 * Should not happen - syncache should pick up these connections.
1090 * Once we are past handling listen sockets we must be in the
1091 * correct protocol processing thread.
1093 KASSERT(tp->t_state != TCPS_LISTEN, ("tcp_input: TCPS_LISTEN state"));
1094 KKASSERT(so->so_port == &curthread->td_msgport);
1097 * This is the second part of the MSS DoS prevention code (after
1098 * minmss on the sending side) and it deals with too many too small
1099 * tcp packets in a too short timeframe (1 second).
1101 * XXX Removed. This code was crap. It does not scale to network
1102 * speed, and default values break NFS. Gone.
1107 * Segment received on connection.
1109 * Reset idle time and keep-alive timer. Don't waste time if less
1110 * then a second has elapsed. Only update t_rcvtime for non-SYN
1113 * Handle the case where one side thinks the connection is established
1114 * but the other side has, say, rebooted without cleaning out the
1115 * connection. The SYNs could be construed as an attack and wind
1116 * up ignored, but in case it isn't an attack we can validate the
1117 * connection by forcing a keepalive.
1119 if (TCPS_HAVEESTABLISHED(tp->t_state) && (ticks - tp->t_rcvtime) > hz) {
1120 if ((thflags & (TH_SYN | TH_ACK)) == TH_SYN) {
1121 tp->t_flags |= TF_KEEPALIVE;
1122 tcp_callout_reset(tp, tp->tt_keep, hz / 2,
1125 tp->t_rcvtime = ticks;
1126 tp->t_flags &= ~TF_KEEPALIVE;
1127 tcp_callout_reset(tp, tp->tt_keep, tcp_keepidle,
1134 * XXX this is tradtitional behavior, may need to be cleaned up.
1136 tcp_dooptions(&to, optp, optlen, (thflags & TH_SYN) != 0);
1137 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1138 if (to.to_flags & TOF_SCALE) {
1139 tp->t_flags |= TF_RCVD_SCALE;
1140 tp->requested_s_scale = to.to_requested_s_scale;
1142 if (to.to_flags & TOF_TS) {
1143 tp->t_flags |= TF_RCVD_TSTMP;
1144 tp->ts_recent = to.to_tsval;
1145 tp->ts_recent_age = ticks;
1147 if (to.to_flags & TOF_MSS)
1148 tcp_mss(tp, to.to_mss);
1150 * Only set the TF_SACK_PERMITTED per-connection flag
1151 * if we got a SACK_PERMITTED option from the other side
1152 * and the global tcp_do_sack variable is true.
1154 if (tcp_do_sack && (to.to_flags & TOF_SACK_PERMITTED))
1155 tp->t_flags |= TF_SACK_PERMITTED;
1159 * Header prediction: check for the two common cases
1160 * of a uni-directional data xfer. If the packet has
1161 * no control flags, is in-sequence, the window didn't
1162 * change and we're not retransmitting, it's a
1163 * candidate. If the length is zero and the ack moved
1164 * forward, we're the sender side of the xfer. Just
1165 * free the data acked & wake any higher level process
1166 * that was blocked waiting for space. If the length
1167 * is non-zero and the ack didn't move, we're the
1168 * receiver side. If we're getting packets in-order
1169 * (the reassembly queue is empty), add the data to
1170 * the socket buffer and note that we need a delayed ack.
1171 * Make sure that the hidden state-flags are also off.
1172 * Since we check for TCPS_ESTABLISHED above, it can only
1175 if (tp->t_state == TCPS_ESTABLISHED &&
1176 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1177 !(tp->t_flags & (TF_NEEDSYN | TF_NEEDFIN)) &&
1178 (!(to.to_flags & TOF_TS) ||
1179 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) &&
1180 th->th_seq == tp->rcv_nxt &&
1181 tp->snd_nxt == tp->snd_max) {
1184 * If last ACK falls within this segment's sequence numbers,
1185 * record the timestamp.
1186 * NOTE that the test is modified according to the latest
1187 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1189 if ((to.to_flags & TOF_TS) &&
1190 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1191 tp->ts_recent_age = ticks;
1192 tp->ts_recent = to.to_tsval;
1196 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1197 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1198 tp->snd_cwnd >= tp->snd_wnd &&
1199 !IN_FASTRECOVERY(tp)) {
1201 * This is a pure ack for outstanding data.
1203 ++tcpstat.tcps_predack;
1205 * "bad retransmit" recovery
1207 * If Eifel detection applies, then
1208 * it is deterministic, so use it
1209 * unconditionally over the old heuristic.
1210 * Otherwise, fall back to the old heuristic.
1212 if (tcp_do_eifel_detect &&
1213 (to.to_flags & TOF_TS) && to.to_tsecr &&
1214 (tp->t_flags & TF_FIRSTACCACK)) {
1215 /* Eifel detection applicable. */
1216 if (to.to_tsecr < tp->t_rexmtTS) {
1217 tcp_revert_congestion_state(tp);
1218 ++tcpstat.tcps_eifeldetected;
1220 } else if (tp->t_rxtshift == 1 &&
1221 ticks < tp->t_badrxtwin) {
1222 tcp_revert_congestion_state(tp);
1223 ++tcpstat.tcps_rttdetected;
1225 tp->t_flags &= ~(TF_FIRSTACCACK |
1226 TF_FASTREXMT | TF_EARLYREXMT);
1228 * Recalculate the retransmit timer / rtt.
1230 * Some machines (certain windows boxes)
1231 * send broken timestamp replies during the
1232 * SYN+ACK phase, ignore timestamps of 0.
1234 if ((to.to_flags & TOF_TS) && to.to_tsecr) {
1236 ticks - to.to_tsecr + 1);
1237 } else if (tp->t_rtttime &&
1238 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1240 ticks - tp->t_rtttime);
1242 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1243 acked = th->th_ack - tp->snd_una;
1244 tcpstat.tcps_rcvackpack++;
1245 tcpstat.tcps_rcvackbyte += acked;
1246 sbdrop(&so->so_snd.sb, acked);
1247 tp->snd_recover = th->th_ack - 1;
1248 tp->snd_una = th->th_ack;
1251 * Update window information.
1253 if (tiwin != tp->snd_wnd &&
1254 acceptable_window_update(tp, th, tiwin)) {
1255 /* keep track of pure window updates */
1256 if (tp->snd_wl2 == th->th_ack &&
1257 tiwin > tp->snd_wnd)
1258 tcpstat.tcps_rcvwinupd++;
1259 tp->snd_wnd = tiwin;
1260 tp->snd_wl1 = th->th_seq;
1261 tp->snd_wl2 = th->th_ack;
1262 if (tp->snd_wnd > tp->max_sndwnd)
1263 tp->max_sndwnd = tp->snd_wnd;
1266 ND6_HINT(tp); /* some progress has been done */
1268 * If all outstanding data are acked, stop
1269 * retransmit timer, otherwise restart timer
1270 * using current (possibly backed-off) value.
1271 * If process is waiting for space,
1272 * wakeup/selwakeup/signal. If data
1273 * are ready to send, let tcp_output
1274 * decide between more output or persist.
1276 if (tp->snd_una == tp->snd_max) {
1277 tcp_callout_stop(tp, tp->tt_rexmt);
1278 } else if (!tcp_callout_active(tp,
1280 tcp_callout_reset(tp, tp->tt_rexmt,
1281 tp->t_rxtcur, tcp_timer_rexmt);
1284 if (so->so_snd.ssb_cc > 0)
1286 return(IPPROTO_DONE);
1288 } else if (tiwin == tp->snd_wnd &&
1289 th->th_ack == tp->snd_una &&
1290 LIST_EMPTY(&tp->t_segq) &&
1291 tlen <= ssb_space(&so->so_rcv)) {
1292 u_long newsize = 0; /* automatic sockbuf scaling */
1294 * This is a pure, in-sequence data packet
1295 * with nothing on the reassembly queue and
1296 * we have enough buffer space to take it.
1298 ++tcpstat.tcps_preddat;
1299 tp->rcv_nxt += tlen;
1300 tcpstat.tcps_rcvpack++;
1301 tcpstat.tcps_rcvbyte += tlen;
1302 ND6_HINT(tp); /* some progress has been done */
1304 * Automatic sizing of receive socket buffer. Often the send
1305 * buffer size is not optimally adjusted to the actual network
1306 * conditions at hand (delay bandwidth product). Setting the
1307 * buffer size too small limits throughput on links with high
1308 * bandwidth and high delay (eg. trans-continental/oceanic links).
1310 * On the receive side the socket buffer memory is only rarely
1311 * used to any significant extent. This allows us to be much
1312 * more aggressive in scaling the receive socket buffer. For
1313 * the case that the buffer space is actually used to a large
1314 * extent and we run out of kernel memory we can simply drop
1315 * the new segments; TCP on the sender will just retransmit it
1316 * later. Setting the buffer size too big may only consume too
1317 * much kernel memory if the application doesn't read() from
1318 * the socket or packet loss or reordering makes use of the
1321 * The criteria to step up the receive buffer one notch are:
1322 * 1. the number of bytes received during the time it takes
1323 * one timestamp to be reflected back to us (the RTT);
1324 * 2. received bytes per RTT is within seven eighth of the
1325 * current socket buffer size;
1326 * 3. receive buffer size has not hit maximal automatic size;
1328 * This algorithm does one step per RTT at most and only if
1329 * we receive a bulk stream w/o packet losses or reorderings.
1330 * Shrinking the buffer during idle times is not necessary as
1331 * it doesn't consume any memory when idle.
1333 * TODO: Only step up if the application is actually serving
1334 * the buffer to better manage the socket buffer resources.
1336 if (tcp_do_autorcvbuf &&
1338 (so->so_rcv.ssb_flags & SSB_AUTOSIZE)) {
1339 if (to.to_tsecr > tp->rfbuf_ts &&
1340 to.to_tsecr - tp->rfbuf_ts < hz) {
1342 (so->so_rcv.ssb_hiwat / 8 * 7) &&
1343 so->so_rcv.ssb_hiwat <
1344 tcp_autorcvbuf_max) {
1346 ulmin(so->so_rcv.ssb_hiwat +
1348 tcp_autorcvbuf_max);
1350 /* Start over with next RTT. */
1354 tp->rfbuf_cnt += tlen; /* add up */
1357 * Add data to socket buffer.
1359 if (so->so_state & SS_CANTRCVMORE) {
1363 * Set new socket buffer size, give up when
1366 * Adjusting the size can mess up ACK
1367 * sequencing when pure window updates are
1368 * being avoided (which is the default),
1371 lwkt_gettoken(&so->so_rcv.ssb_token);
1373 tp->t_flags |= TF_RXRESIZED;
1374 if (!ssb_reserve(&so->so_rcv, newsize,
1376 atomic_clear_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE);
1379 (TCP_MAXWIN << tp->rcv_scale)) {
1380 atomic_clear_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE);
1383 m_adj(m, drop_hdrlen); /* delayed header drop */
1384 ssb_appendstream(&so->so_rcv, m);
1385 lwkt_reltoken(&so->so_rcv.ssb_token);
1389 * This code is responsible for most of the ACKs
1390 * the TCP stack sends back after receiving a data
1391 * packet. Note that the DELAY_ACK check fails if
1392 * the delack timer is already running, which results
1393 * in an ack being sent every other packet (which is
1396 * We then further aggregate acks by not actually
1397 * sending one until the protocol thread has completed
1398 * processing the current backlog of packets. This
1399 * does not delay the ack any further, but allows us
1400 * to take advantage of the packet aggregation that
1401 * high speed NICs do (usually blocks of 8-10 packets)
1402 * to send a single ack rather then four or five acks,
1403 * greatly reducing the ack rate, the return channel
1404 * bandwidth, and the protocol overhead on both ends.
1406 * Since this also has the effect of slowing down
1407 * the exponential slow-start ramp-up, systems with
1408 * very large bandwidth-delay products might want
1409 * to turn the feature off.
1411 if (DELAY_ACK(tp)) {
1412 tcp_callout_reset(tp, tp->tt_delack,
1413 tcp_delacktime, tcp_timer_delack);
1414 } else if (tcp_aggregate_acks) {
1415 tp->t_flags |= TF_ACKNOW;
1416 if (!(tp->t_flags & TF_ONOUTPUTQ)) {
1417 tp->t_flags |= TF_ONOUTPUTQ;
1418 tp->tt_cpu = mycpu->gd_cpuid;
1420 &tcpcbackq[tp->tt_cpu],
1424 tp->t_flags |= TF_ACKNOW;
1427 return(IPPROTO_DONE);
1432 * Calculate amount of space in receive window,
1433 * and then do TCP input processing.
1434 * Receive window is amount of space in rcv queue,
1435 * but not less than advertised window.
1437 recvwin = ssb_space(&so->so_rcv);
1440 tp->rcv_wnd = imax(recvwin, (int)(tp->rcv_adv - tp->rcv_nxt));
1442 /* Reset receive buffer auto scaling when not in bulk receive mode. */
1446 switch (tp->t_state) {
1448 * If the state is SYN_RECEIVED:
1449 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1451 case TCPS_SYN_RECEIVED:
1452 if ((thflags & TH_ACK) &&
1453 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1454 SEQ_GT(th->th_ack, tp->snd_max))) {
1455 rstreason = BANDLIM_RST_OPENPORT;
1461 * If the state is SYN_SENT:
1462 * if seg contains an ACK, but not for our SYN, drop the input.
1463 * if seg contains a RST, then drop the connection.
1464 * if seg does not contain SYN, then drop it.
1465 * Otherwise this is an acceptable SYN segment
1466 * initialize tp->rcv_nxt and tp->irs
1467 * if seg contains ack then advance tp->snd_una
1468 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1469 * arrange for segment to be acked (eventually)
1470 * continue processing rest of data/controls, beginning with URG
1473 if ((thflags & TH_ACK) &&
1474 (SEQ_LEQ(th->th_ack, tp->iss) ||
1475 SEQ_GT(th->th_ack, tp->snd_max))) {
1476 rstreason = BANDLIM_UNLIMITED;
1479 if (thflags & TH_RST) {
1480 if (thflags & TH_ACK)
1481 tp = tcp_drop(tp, ECONNREFUSED);
1484 if (!(thflags & TH_SYN))
1486 tp->snd_wnd = th->th_win; /* initial send window */
1488 tp->irs = th->th_seq;
1490 if (thflags & TH_ACK) {
1491 /* Our SYN was acked. */
1492 tcpstat.tcps_connects++;
1494 /* Do window scaling on this connection? */
1495 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
1496 (TF_RCVD_SCALE | TF_REQ_SCALE)) {
1497 tp->snd_scale = tp->requested_s_scale;
1498 tp->rcv_scale = tp->request_r_scale;
1500 tp->rcv_adv += tp->rcv_wnd;
1501 tp->snd_una++; /* SYN is acked */
1502 tcp_callout_stop(tp, tp->tt_rexmt);
1504 * If there's data, delay ACK; if there's also a FIN
1505 * ACKNOW will be turned on later.
1507 if (DELAY_ACK(tp) && tlen != 0) {
1508 tcp_callout_reset(tp, tp->tt_delack,
1509 tcp_delacktime, tcp_timer_delack);
1511 tp->t_flags |= TF_ACKNOW;
1514 * Received <SYN,ACK> in SYN_SENT[*] state.
1516 * SYN_SENT --> ESTABLISHED
1517 * SYN_SENT* --> FIN_WAIT_1
1519 tp->t_starttime = ticks;
1520 if (tp->t_flags & TF_NEEDFIN) {
1521 tp->t_state = TCPS_FIN_WAIT_1;
1522 tp->t_flags &= ~TF_NEEDFIN;
1525 tp->t_state = TCPS_ESTABLISHED;
1526 tcp_callout_reset(tp, tp->tt_keep, tcp_keepidle,
1531 * Received initial SYN in SYN-SENT[*] state =>
1532 * simultaneous open.
1533 * Do 3-way handshake:
1534 * SYN-SENT -> SYN-RECEIVED
1535 * SYN-SENT* -> SYN-RECEIVED*
1537 tp->t_flags |= TF_ACKNOW;
1538 tcp_callout_stop(tp, tp->tt_rexmt);
1539 tp->t_state = TCPS_SYN_RECEIVED;
1544 * Advance th->th_seq to correspond to first data byte.
1545 * If data, trim to stay within window,
1546 * dropping FIN if necessary.
1549 if (tlen > tp->rcv_wnd) {
1550 todrop = tlen - tp->rcv_wnd;
1554 tcpstat.tcps_rcvpackafterwin++;
1555 tcpstat.tcps_rcvbyteafterwin += todrop;
1557 tp->snd_wl1 = th->th_seq - 1;
1558 tp->rcv_up = th->th_seq;
1560 * Client side of transaction: already sent SYN and data.
1561 * If the remote host used T/TCP to validate the SYN,
1562 * our data will be ACK'd; if so, enter normal data segment
1563 * processing in the middle of step 5, ack processing.
1564 * Otherwise, goto step 6.
1566 if (thflags & TH_ACK)
1572 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
1573 * do normal processing (we no longer bother with T/TCP).
1577 case TCPS_TIME_WAIT:
1578 break; /* continue normal processing */
1582 * States other than LISTEN or SYN_SENT.
1583 * First check the RST flag and sequence number since reset segments
1584 * are exempt from the timestamp and connection count tests. This
1585 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
1586 * below which allowed reset segments in half the sequence space
1587 * to fall though and be processed (which gives forged reset
1588 * segments with a random sequence number a 50 percent chance of
1589 * killing a connection).
1590 * Then check timestamp, if present.
1591 * Then check the connection count, if present.
1592 * Then check that at least some bytes of segment are within
1593 * receive window. If segment begins before rcv_nxt,
1594 * drop leading data (and SYN); if nothing left, just ack.
1597 * If the RST bit is set, check the sequence number to see
1598 * if this is a valid reset segment.
1600 * In all states except SYN-SENT, all reset (RST) segments
1601 * are validated by checking their SEQ-fields. A reset is
1602 * valid if its sequence number is in the window.
1603 * Note: this does not take into account delayed ACKs, so
1604 * we should test against last_ack_sent instead of rcv_nxt.
1605 * The sequence number in the reset segment is normally an
1606 * echo of our outgoing acknowledgement numbers, but some hosts
1607 * send a reset with the sequence number at the rightmost edge
1608 * of our receive window, and we have to handle this case.
1609 * If we have multiple segments in flight, the intial reset
1610 * segment sequence numbers will be to the left of last_ack_sent,
1611 * but they will eventually catch up.
1612 * In any case, it never made sense to trim reset segments to
1613 * fit the receive window since RFC 1122 says:
1614 * 4.2.2.12 RST Segment: RFC-793 Section 3.4
1616 * A TCP SHOULD allow a received RST segment to include data.
1619 * It has been suggested that a RST segment could contain
1620 * ASCII text that encoded and explained the cause of the
1621 * RST. No standard has yet been established for such
1624 * If the reset segment passes the sequence number test examine
1626 * SYN_RECEIVED STATE:
1627 * If passive open, return to LISTEN state.
1628 * If active open, inform user that connection was refused.
1629 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
1630 * Inform user that connection was reset, and close tcb.
1631 * CLOSING, LAST_ACK STATES:
1634 * Drop the segment - see Stevens, vol. 2, p. 964 and
1637 if (thflags & TH_RST) {
1638 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
1639 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
1640 switch (tp->t_state) {
1642 case TCPS_SYN_RECEIVED:
1643 so->so_error = ECONNREFUSED;
1646 case TCPS_ESTABLISHED:
1647 case TCPS_FIN_WAIT_1:
1648 case TCPS_FIN_WAIT_2:
1649 case TCPS_CLOSE_WAIT:
1650 so->so_error = ECONNRESET;
1652 tp->t_state = TCPS_CLOSED;
1653 tcpstat.tcps_drops++;
1662 case TCPS_TIME_WAIT:
1670 * RFC 1323 PAWS: If we have a timestamp reply on this segment
1671 * and it's less than ts_recent, drop it.
1673 if ((to.to_flags & TOF_TS) && tp->ts_recent != 0 &&
1674 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
1676 /* Check to see if ts_recent is over 24 days old. */
1677 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1679 * Invalidate ts_recent. If this segment updates
1680 * ts_recent, the age will be reset later and ts_recent
1681 * will get a valid value. If it does not, setting
1682 * ts_recent to zero will at least satisfy the
1683 * requirement that zero be placed in the timestamp
1684 * echo reply when ts_recent isn't valid. The
1685 * age isn't reset until we get a valid ts_recent
1686 * because we don't want out-of-order segments to be
1687 * dropped when ts_recent is old.
1691 tcpstat.tcps_rcvduppack++;
1692 tcpstat.tcps_rcvdupbyte += tlen;
1693 tcpstat.tcps_pawsdrop++;
1701 * In the SYN-RECEIVED state, validate that the packet belongs to
1702 * this connection before trimming the data to fit the receive
1703 * window. Check the sequence number versus IRS since we know
1704 * the sequence numbers haven't wrapped. This is a partial fix
1705 * for the "LAND" DoS attack.
1707 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
1708 rstreason = BANDLIM_RST_OPENPORT;
1712 todrop = tp->rcv_nxt - th->th_seq;
1714 if (TCP_DO_SACK(tp)) {
1715 /* Report duplicate segment at head of packet. */
1716 tp->reportblk.rblk_start = th->th_seq;
1717 tp->reportblk.rblk_end = th->th_seq + tlen;
1718 if (thflags & TH_FIN)
1719 ++tp->reportblk.rblk_end;
1720 if (SEQ_GT(tp->reportblk.rblk_end, tp->rcv_nxt))
1721 tp->reportblk.rblk_end = tp->rcv_nxt;
1722 tp->t_flags |= (TF_DUPSEG | TF_SACKLEFT | TF_ACKNOW);
1724 if (thflags & TH_SYN) {
1734 * Following if statement from Stevens, vol. 2, p. 960.
1736 if (todrop > tlen ||
1737 (todrop == tlen && !(thflags & TH_FIN))) {
1739 * Any valid FIN must be to the left of the window.
1740 * At this point the FIN must be a duplicate or out
1741 * of sequence; drop it.
1746 * Send an ACK to resynchronize and drop any data.
1747 * But keep on processing for RST or ACK.
1749 tp->t_flags |= TF_ACKNOW;
1751 tcpstat.tcps_rcvduppack++;
1752 tcpstat.tcps_rcvdupbyte += todrop;
1754 tcpstat.tcps_rcvpartduppack++;
1755 tcpstat.tcps_rcvpartdupbyte += todrop;
1757 drop_hdrlen += todrop; /* drop from the top afterwards */
1758 th->th_seq += todrop;
1760 if (th->th_urp > todrop)
1761 th->th_urp -= todrop;
1769 * If new data are received on a connection after the
1770 * user processes are gone, then RST the other end.
1772 if ((so->so_state & SS_NOFDREF) &&
1773 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1775 tcpstat.tcps_rcvafterclose++;
1776 rstreason = BANDLIM_UNLIMITED;
1781 * If segment ends after window, drop trailing data
1782 * (and PUSH and FIN); if nothing left, just ACK.
1784 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
1786 tcpstat.tcps_rcvpackafterwin++;
1787 if (todrop >= tlen) {
1788 tcpstat.tcps_rcvbyteafterwin += tlen;
1790 * If a new connection request is received
1791 * while in TIME_WAIT, drop the old connection
1792 * and start over if the sequence numbers
1793 * are above the previous ones.
1795 if (thflags & TH_SYN &&
1796 tp->t_state == TCPS_TIME_WAIT &&
1797 SEQ_GT(th->th_seq, tp->rcv_nxt)) {
1802 * If window is closed can only take segments at
1803 * window edge, and have to drop data and PUSH from
1804 * incoming segments. Continue processing, but
1805 * remember to ack. Otherwise, drop segment
1808 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1809 tp->t_flags |= TF_ACKNOW;
1810 tcpstat.tcps_rcvwinprobe++;
1814 tcpstat.tcps_rcvbyteafterwin += todrop;
1817 thflags &= ~(TH_PUSH | TH_FIN);
1821 * If last ACK falls within this segment's sequence numbers,
1822 * record its timestamp.
1824 * 1) That the test incorporates suggestions from the latest
1825 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1826 * 2) That updating only on newer timestamps interferes with
1827 * our earlier PAWS tests, so this check should be solely
1828 * predicated on the sequence space of this segment.
1829 * 3) That we modify the segment boundary check to be
1830 * Last.ACK.Sent <= SEG.SEQ + SEG.LEN
1831 * instead of RFC1323's
1832 * Last.ACK.Sent < SEG.SEQ + SEG.LEN,
1833 * This modified check allows us to overcome RFC1323's
1834 * limitations as described in Stevens TCP/IP Illustrated
1835 * Vol. 2 p.869. In such cases, we can still calculate the
1836 * RTT correctly when RCV.NXT == Last.ACK.Sent.
1838 if ((to.to_flags & TOF_TS) && SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
1839 SEQ_LEQ(tp->last_ack_sent, (th->th_seq + tlen
1840 + ((thflags & TH_SYN) != 0)
1841 + ((thflags & TH_FIN) != 0)))) {
1842 tp->ts_recent_age = ticks;
1843 tp->ts_recent = to.to_tsval;
1847 * If a SYN is in the window, then this is an
1848 * error and we send an RST and drop the connection.
1850 if (thflags & TH_SYN) {
1851 tp = tcp_drop(tp, ECONNRESET);
1852 rstreason = BANDLIM_UNLIMITED;
1857 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
1858 * flag is on (half-synchronized state), then queue data for
1859 * later processing; else drop segment and return.
1861 if (!(thflags & TH_ACK)) {
1862 if (tp->t_state == TCPS_SYN_RECEIVED ||
1863 (tp->t_flags & TF_NEEDSYN))
1872 switch (tp->t_state) {
1874 * In SYN_RECEIVED state, the ACK acknowledges our SYN, so enter
1875 * ESTABLISHED state and continue processing.
1876 * The ACK was checked above.
1878 case TCPS_SYN_RECEIVED:
1880 tcpstat.tcps_connects++;
1882 /* Do window scaling? */
1883 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
1884 (TF_RCVD_SCALE | TF_REQ_SCALE)) {
1885 tp->snd_scale = tp->requested_s_scale;
1886 tp->rcv_scale = tp->request_r_scale;
1890 * SYN-RECEIVED -> ESTABLISHED
1891 * SYN-RECEIVED* -> FIN-WAIT-1
1893 tp->t_starttime = ticks;
1894 if (tp->t_flags & TF_NEEDFIN) {
1895 tp->t_state = TCPS_FIN_WAIT_1;
1896 tp->t_flags &= ~TF_NEEDFIN;
1898 tp->t_state = TCPS_ESTABLISHED;
1899 tcp_callout_reset(tp, tp->tt_keep, tcp_keepidle,
1903 * If segment contains data or ACK, will call tcp_reass()
1904 * later; if not, do so now to pass queued data to user.
1906 if (tlen == 0 && !(thflags & TH_FIN))
1907 tcp_reass(tp, NULL, NULL, NULL);
1911 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1912 * ACKs. If the ack is in the range
1913 * tp->snd_una < th->th_ack <= tp->snd_max
1914 * then advance tp->snd_una to th->th_ack and drop
1915 * data from the retransmission queue. If this ACK reflects
1916 * more up to date window information we update our window information.
1918 case TCPS_ESTABLISHED:
1919 case TCPS_FIN_WAIT_1:
1920 case TCPS_FIN_WAIT_2:
1921 case TCPS_CLOSE_WAIT:
1924 case TCPS_TIME_WAIT:
1926 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
1927 if (TCP_DO_SACK(tp))
1928 tcp_sack_update_scoreboard(tp, &to);
1929 if (tlen != 0 || tiwin != tp->snd_wnd) {
1933 tcpstat.tcps_rcvdupack++;
1934 if (!tcp_callout_active(tp, tp->tt_rexmt) ||
1935 th->th_ack != tp->snd_una) {
1940 * We have outstanding data (other than
1941 * a window probe), this is a completely
1942 * duplicate ack (ie, window info didn't
1943 * change), the ack is the biggest we've
1944 * seen and we've seen exactly our rexmt
1945 * threshhold of them, so assume a packet
1946 * has been dropped and retransmit it.
1947 * Kludge snd_nxt & the congestion
1948 * window so we send only this one
1951 if (IN_FASTRECOVERY(tp)) {
1952 if (TCP_DO_SACK(tp)) {
1953 /* No artifical cwnd inflation. */
1954 tcp_sack_rexmt(tp, th);
1957 * Dup acks mean that packets
1958 * have left the network
1959 * (they're now cached at the
1960 * receiver) so bump cwnd by
1961 * the amount in the receiver
1962 * to keep a constant cwnd
1963 * packets in the network.
1965 tp->snd_cwnd += tp->t_maxseg;
1968 } else if (SEQ_LT(th->th_ack, tp->snd_recover)) {
1971 } else if (++tp->t_dupacks == tcprexmtthresh) {
1972 tcp_seq old_snd_nxt;
1976 if (tcp_do_eifel_detect &&
1977 (tp->t_flags & TF_RCVD_TSTMP)) {
1978 tcp_save_congestion_state(tp);
1979 tp->t_flags |= TF_FASTREXMT;
1982 * We know we're losing at the current
1983 * window size, so do congestion avoidance:
1984 * set ssthresh to half the current window
1985 * and pull our congestion window back to the
1988 win = min(tp->snd_wnd, tp->snd_cwnd) / 2 /
1992 tp->snd_ssthresh = win * tp->t_maxseg;
1993 ENTER_FASTRECOVERY(tp);
1994 tp->snd_recover = tp->snd_max;
1995 tcp_callout_stop(tp, tp->tt_rexmt);
1997 old_snd_nxt = tp->snd_nxt;
1998 tp->snd_nxt = th->th_ack;
1999 tp->snd_cwnd = tp->t_maxseg;
2001 ++tcpstat.tcps_sndfastrexmit;
2002 tp->snd_cwnd = tp->snd_ssthresh;
2003 tp->rexmt_high = tp->snd_nxt;
2004 if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
2005 tp->snd_nxt = old_snd_nxt;
2006 KASSERT(tp->snd_limited <= 2,
2007 ("tp->snd_limited too big"));
2008 if (TCP_DO_SACK(tp))
2009 tcp_sack_rexmt(tp, th);
2011 tp->snd_cwnd += tp->t_maxseg *
2012 (tp->t_dupacks - tp->snd_limited);
2013 } else if (tcp_do_limitedtransmit) {
2014 u_long oldcwnd = tp->snd_cwnd;
2015 tcp_seq oldsndmax = tp->snd_max;
2016 tcp_seq oldsndnxt = tp->snd_nxt;
2017 /* outstanding data */
2018 uint32_t ownd = tp->snd_max - tp->snd_una;
2021 #define iceildiv(n, d) (((n)+(d)-1) / (d))
2023 KASSERT(tp->t_dupacks == 1 ||
2025 ("dupacks not 1 or 2"));
2026 if (tp->t_dupacks == 1)
2027 tp->snd_limited = 0;
2028 tp->snd_nxt = tp->snd_max;
2029 tp->snd_cwnd = ownd +
2030 (tp->t_dupacks - tp->snd_limited) *
2035 * Other acks may have been processed,
2036 * snd_nxt cannot be reset to a value less
2039 if (SEQ_LT(oldsndnxt, oldsndmax)) {
2040 if (SEQ_GT(oldsndnxt, tp->snd_una))
2041 tp->snd_nxt = oldsndnxt;
2043 tp->snd_nxt = tp->snd_una;
2045 tp->snd_cwnd = oldcwnd;
2046 sent = tp->snd_max - oldsndmax;
2047 if (sent > tp->t_maxseg) {
2048 KASSERT((tp->t_dupacks == 2 &&
2049 tp->snd_limited == 0) ||
2050 (sent == tp->t_maxseg + 1 &&
2051 tp->t_flags & TF_SENTFIN),
2053 KASSERT(sent <= tp->t_maxseg * 2,
2054 ("sent too many segments"));
2055 tp->snd_limited = 2;
2056 tcpstat.tcps_sndlimited += 2;
2057 } else if (sent > 0) {
2059 ++tcpstat.tcps_sndlimited;
2060 } else if (tcp_do_early_retransmit &&
2061 (tcp_do_eifel_detect &&
2062 (tp->t_flags & TF_RCVD_TSTMP)) &&
2063 ownd < 4 * tp->t_maxseg &&
2064 tp->t_dupacks + 1 >=
2065 iceildiv(ownd, tp->t_maxseg) &&
2066 (!TCP_DO_SACK(tp) ||
2067 ownd <= tp->t_maxseg ||
2068 tcp_sack_has_sacked(&tp->scb,
2069 ownd - tp->t_maxseg))) {
2070 ++tcpstat.tcps_sndearlyrexmit;
2071 tp->t_flags |= TF_EARLYREXMT;
2072 goto fastretransmit;
2078 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una"));
2080 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2082 * Detected optimistic ACK attack.
2083 * Force slow-start to de-synchronize attack.
2085 tp->snd_cwnd = tp->t_maxseg;
2088 tcpstat.tcps_rcvacktoomuch++;
2092 * If we reach this point, ACK is not a duplicate,
2093 * i.e., it ACKs something we sent.
2095 if (tp->t_flags & TF_NEEDSYN) {
2097 * T/TCP: Connection was half-synchronized, and our
2098 * SYN has been ACK'd (so connection is now fully
2099 * synchronized). Go to non-starred state,
2100 * increment snd_una for ACK of SYN, and check if
2101 * we can do window scaling.
2103 tp->t_flags &= ~TF_NEEDSYN;
2105 /* Do window scaling? */
2106 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
2107 (TF_RCVD_SCALE | TF_REQ_SCALE)) {
2108 tp->snd_scale = tp->requested_s_scale;
2109 tp->rcv_scale = tp->request_r_scale;
2114 acked = th->th_ack - tp->snd_una;
2115 tcpstat.tcps_rcvackpack++;
2116 tcpstat.tcps_rcvackbyte += acked;
2118 if (tcp_do_eifel_detect && acked > 0 &&
2119 (to.to_flags & TOF_TS) && (to.to_tsecr != 0) &&
2120 (tp->t_flags & TF_FIRSTACCACK)) {
2121 /* Eifel detection applicable. */
2122 if (to.to_tsecr < tp->t_rexmtTS) {
2123 ++tcpstat.tcps_eifeldetected;
2124 tcp_revert_congestion_state(tp);
2125 if (tp->t_rxtshift == 1 &&
2126 ticks >= tp->t_badrxtwin)
2127 ++tcpstat.tcps_rttcantdetect;
2129 } else if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) {
2131 * If we just performed our first retransmit,
2132 * and the ACK arrives within our recovery window,
2133 * then it was a mistake to do the retransmit
2134 * in the first place. Recover our original cwnd
2135 * and ssthresh, and proceed to transmit where we
2138 tcp_revert_congestion_state(tp);
2139 ++tcpstat.tcps_rttdetected;
2143 * If we have a timestamp reply, update smoothed
2144 * round trip time. If no timestamp is present but
2145 * transmit timer is running and timed sequence
2146 * number was acked, update smoothed round trip time.
2147 * Since we now have an rtt measurement, cancel the
2148 * timer backoff (cf., Phil Karn's retransmit alg.).
2149 * Recompute the initial retransmit timer.
2151 * Some machines (certain windows boxes) send broken
2152 * timestamp replies during the SYN+ACK phase, ignore
2155 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0))
2156 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1);
2157 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq))
2158 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2159 tcp_xmit_bandwidth_limit(tp, th->th_ack);
2162 * If no data (only SYN) was ACK'd,
2163 * skip rest of ACK processing.
2168 /* Stop looking for an acceptable ACK since one was received. */
2169 tp->t_flags &= ~(TF_FIRSTACCACK | TF_FASTREXMT | TF_EARLYREXMT);
2171 if (acked > so->so_snd.ssb_cc) {
2172 tp->snd_wnd -= so->so_snd.ssb_cc;
2173 sbdrop(&so->so_snd.sb, (int)so->so_snd.ssb_cc);
2174 ourfinisacked = TRUE;
2176 sbdrop(&so->so_snd.sb, acked);
2177 tp->snd_wnd -= acked;
2178 ourfinisacked = FALSE;
2183 * Update window information.
2184 * Don't look at window if no ACK:
2185 * TAC's send garbage on first SYN.
2187 if (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2188 (tp->snd_wl1 == th->th_seq &&
2189 (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2190 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)))) {
2191 /* keep track of pure window updates */
2192 if (tlen == 0 && tp->snd_wl2 == th->th_ack &&
2193 tiwin > tp->snd_wnd)
2194 tcpstat.tcps_rcvwinupd++;
2195 tp->snd_wnd = tiwin;
2196 tp->snd_wl1 = th->th_seq;
2197 tp->snd_wl2 = th->th_ack;
2198 if (tp->snd_wnd > tp->max_sndwnd)
2199 tp->max_sndwnd = tp->snd_wnd;
2203 tp->snd_una = th->th_ack;
2204 if (TCP_DO_SACK(tp))
2205 tcp_sack_update_scoreboard(tp, &to);
2206 if (IN_FASTRECOVERY(tp)) {
2207 if (SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2208 EXIT_FASTRECOVERY(tp);
2211 * If the congestion window was inflated
2212 * to account for the other side's
2213 * cached packets, retract it.
2215 if (!TCP_DO_SACK(tp))
2216 tp->snd_cwnd = tp->snd_ssthresh;
2219 * Window inflation should have left us
2220 * with approximately snd_ssthresh outstanding
2221 * data. But, in case we would be inclined
2222 * to send a burst, better do it using
2225 if (SEQ_GT(th->th_ack + tp->snd_cwnd,
2226 tp->snd_max + 2 * tp->t_maxseg))
2228 (tp->snd_max - tp->snd_una) +
2233 if (TCP_DO_SACK(tp)) {
2234 tp->snd_max_rexmt = tp->snd_max;
2235 tcp_sack_rexmt(tp, th);
2237 tcp_newreno_partial_ack(tp, th, acked);
2243 * Open the congestion window. When in slow-start,
2244 * open exponentially: maxseg per packet. Otherwise,
2245 * open linearly: maxseg per window.
2247 if (tp->snd_cwnd <= tp->snd_ssthresh) {
2249 (SEQ_LT(tp->snd_nxt, tp->snd_max) ?
2250 tp->t_maxseg : 2 * tp->t_maxseg);
2253 tp->snd_cwnd += tcp_do_abc ?
2254 min(acked, abc_sslimit) : tp->t_maxseg;
2256 /* linear increase */
2257 tp->snd_wacked += tcp_do_abc ? acked :
2259 if (tp->snd_wacked >= tp->snd_cwnd) {
2260 tp->snd_wacked -= tp->snd_cwnd;
2261 tp->snd_cwnd += tp->t_maxseg;
2264 tp->snd_cwnd = min(tp->snd_cwnd,
2265 TCP_MAXWIN << tp->snd_scale);
2266 tp->snd_recover = th->th_ack - 1;
2268 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2269 tp->snd_nxt = tp->snd_una;
2272 * If all outstanding data is acked, stop retransmit
2273 * timer and remember to restart (more output or persist).
2274 * If there is more data to be acked, restart retransmit
2275 * timer, using current (possibly backed-off) value.
2277 if (th->th_ack == tp->snd_max) {
2278 tcp_callout_stop(tp, tp->tt_rexmt);
2280 } else if (!tcp_callout_active(tp, tp->tt_persist)) {
2281 tcp_callout_reset(tp, tp->tt_rexmt, tp->t_rxtcur,
2285 switch (tp->t_state) {
2287 * In FIN_WAIT_1 STATE in addition to the processing
2288 * for the ESTABLISHED state if our FIN is now acknowledged
2289 * then enter FIN_WAIT_2.
2291 case TCPS_FIN_WAIT_1:
2292 if (ourfinisacked) {
2294 * If we can't receive any more
2295 * data, then closing user can proceed.
2296 * Starting the timer is contrary to the
2297 * specification, but if we don't get a FIN
2298 * we'll hang forever.
2300 if (so->so_state & SS_CANTRCVMORE) {
2301 soisdisconnected(so);
2302 tcp_callout_reset(tp, tp->tt_2msl,
2303 tcp_maxidle, tcp_timer_2msl);
2305 tp->t_state = TCPS_FIN_WAIT_2;
2310 * In CLOSING STATE in addition to the processing for
2311 * the ESTABLISHED state if the ACK acknowledges our FIN
2312 * then enter the TIME-WAIT state, otherwise ignore
2316 if (ourfinisacked) {
2317 tp->t_state = TCPS_TIME_WAIT;
2318 tcp_canceltimers(tp);
2319 tcp_callout_reset(tp, tp->tt_2msl,
2320 2 * tcp_msl, tcp_timer_2msl);
2321 soisdisconnected(so);
2326 * In LAST_ACK, we may still be waiting for data to drain
2327 * and/or to be acked, as well as for the ack of our FIN.
2328 * If our FIN is now acknowledged, delete the TCB,
2329 * enter the closed state and return.
2332 if (ourfinisacked) {
2339 * In TIME_WAIT state the only thing that should arrive
2340 * is a retransmission of the remote FIN. Acknowledge
2341 * it and restart the finack timer.
2343 case TCPS_TIME_WAIT:
2344 tcp_callout_reset(tp, tp->tt_2msl, 2 * tcp_msl,
2352 * Update window information.
2353 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2355 if ((thflags & TH_ACK) &&
2356 acceptable_window_update(tp, th, tiwin)) {
2357 /* keep track of pure window updates */
2358 if (tlen == 0 && tp->snd_wl2 == th->th_ack &&
2359 tiwin > tp->snd_wnd)
2360 tcpstat.tcps_rcvwinupd++;
2361 tp->snd_wnd = tiwin;
2362 tp->snd_wl1 = th->th_seq;
2363 tp->snd_wl2 = th->th_ack;
2364 if (tp->snd_wnd > tp->max_sndwnd)
2365 tp->max_sndwnd = tp->snd_wnd;
2370 * Process segments with URG.
2372 if ((thflags & TH_URG) && th->th_urp &&
2373 !TCPS_HAVERCVDFIN(tp->t_state)) {
2375 * This is a kludge, but if we receive and accept
2376 * random urgent pointers, we'll crash in
2377 * soreceive. It's hard to imagine someone
2378 * actually wanting to send this much urgent data.
2380 if (th->th_urp + so->so_rcv.ssb_cc > sb_max) {
2381 th->th_urp = 0; /* XXX */
2382 thflags &= ~TH_URG; /* XXX */
2383 goto dodata; /* XXX */
2386 * If this segment advances the known urgent pointer,
2387 * then mark the data stream. This should not happen
2388 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2389 * a FIN has been received from the remote side.
2390 * In these states we ignore the URG.
2392 * According to RFC961 (Assigned Protocols),
2393 * the urgent pointer points to the last octet
2394 * of urgent data. We continue, however,
2395 * to consider it to indicate the first octet
2396 * of data past the urgent section as the original
2397 * spec states (in one of two places).
2399 if (SEQ_GT(th->th_seq + th->th_urp, tp->rcv_up)) {
2400 tp->rcv_up = th->th_seq + th->th_urp;
2401 so->so_oobmark = so->so_rcv.ssb_cc +
2402 (tp->rcv_up - tp->rcv_nxt) - 1;
2403 if (so->so_oobmark == 0)
2404 sosetstate(so, SS_RCVATMARK);
2406 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2409 * Remove out of band data so doesn't get presented to user.
2410 * This can happen independent of advancing the URG pointer,
2411 * but if two URG's are pending at once, some out-of-band
2412 * data may creep in... ick.
2414 if (th->th_urp <= (u_long)tlen &&
2415 !(so->so_options & SO_OOBINLINE)) {
2416 /* hdr drop is delayed */
2417 tcp_pulloutofband(so, th, m, drop_hdrlen);
2421 * If no out of band data is expected,
2422 * pull receive urgent pointer along
2423 * with the receive window.
2425 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2426 tp->rcv_up = tp->rcv_nxt;
2431 * Process the segment text, merging it into the TCP sequencing queue,
2432 * and arranging for acknowledgment of receipt if necessary.
2433 * This process logically involves adjusting tp->rcv_wnd as data
2434 * is presented to the user (this happens in tcp_usrreq.c,
2435 * case PRU_RCVD). If a FIN has already been received on this
2436 * connection then we just ignore the text.
2438 if ((tlen || (thflags & TH_FIN)) && !TCPS_HAVERCVDFIN(tp->t_state)) {
2439 m_adj(m, drop_hdrlen); /* delayed header drop */
2441 * Insert segment which includes th into TCP reassembly queue
2442 * with control block tp. Set thflags to whether reassembly now
2443 * includes a segment with FIN. This handles the common case
2444 * inline (segment is the next to be received on an established
2445 * connection, and the queue is empty), avoiding linkage into
2446 * and removal from the queue and repetition of various
2448 * Set DELACK for segments received in order, but ack
2449 * immediately when segments are out of order (so
2450 * fast retransmit can work).
2452 if (th->th_seq == tp->rcv_nxt &&
2453 LIST_EMPTY(&tp->t_segq) &&
2454 TCPS_HAVEESTABLISHED(tp->t_state)) {
2455 if (DELAY_ACK(tp)) {
2456 tcp_callout_reset(tp, tp->tt_delack,
2457 tcp_delacktime, tcp_timer_delack);
2459 tp->t_flags |= TF_ACKNOW;
2461 tp->rcv_nxt += tlen;
2462 thflags = th->th_flags & TH_FIN;
2463 tcpstat.tcps_rcvpack++;
2464 tcpstat.tcps_rcvbyte += tlen;
2466 if (so->so_state & SS_CANTRCVMORE) {
2469 lwkt_gettoken(&so->so_rcv.ssb_token);
2470 ssb_appendstream(&so->so_rcv, m);
2471 lwkt_reltoken(&so->so_rcv.ssb_token);
2475 if (!(tp->t_flags & TF_DUPSEG)) {
2476 /* Initialize SACK report block. */
2477 tp->reportblk.rblk_start = th->th_seq;
2478 tp->reportblk.rblk_end = th->th_seq + tlen +
2479 ((thflags & TH_FIN) != 0);
2481 thflags = tcp_reass(tp, th, &tlen, m);
2482 tp->t_flags |= TF_ACKNOW;
2486 * Note the amount of data that peer has sent into
2487 * our window, in order to estimate the sender's
2490 len = so->so_rcv.ssb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2497 * If FIN is received ACK the FIN and let the user know
2498 * that the connection is closing.
2500 if (thflags & TH_FIN) {
2501 if (!TCPS_HAVERCVDFIN(tp->t_state)) {
2504 * If connection is half-synchronized
2505 * (ie NEEDSYN flag on) then delay ACK,
2506 * so it may be piggybacked when SYN is sent.
2507 * Otherwise, since we received a FIN then no
2508 * more input can be expected, send ACK now.
2510 if (DELAY_ACK(tp) && (tp->t_flags & TF_NEEDSYN)) {
2511 tcp_callout_reset(tp, tp->tt_delack,
2512 tcp_delacktime, tcp_timer_delack);
2514 tp->t_flags |= TF_ACKNOW;
2519 switch (tp->t_state) {
2521 * In SYN_RECEIVED and ESTABLISHED STATES
2522 * enter the CLOSE_WAIT state.
2524 case TCPS_SYN_RECEIVED:
2525 tp->t_starttime = ticks;
2527 case TCPS_ESTABLISHED:
2528 tp->t_state = TCPS_CLOSE_WAIT;
2532 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2533 * enter the CLOSING state.
2535 case TCPS_FIN_WAIT_1:
2536 tp->t_state = TCPS_CLOSING;
2540 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2541 * starting the time-wait timer, turning off the other
2544 case TCPS_FIN_WAIT_2:
2545 tp->t_state = TCPS_TIME_WAIT;
2546 tcp_canceltimers(tp);
2547 tcp_callout_reset(tp, tp->tt_2msl, 2 * tcp_msl,
2549 soisdisconnected(so);
2553 * In TIME_WAIT state restart the 2 MSL time_wait timer.
2555 case TCPS_TIME_WAIT:
2556 tcp_callout_reset(tp, tp->tt_2msl, 2 * tcp_msl,
2563 if (so->so_options & SO_DEBUG)
2564 tcp_trace(TA_INPUT, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2568 * Return any desired output.
2570 if (needoutput || (tp->t_flags & TF_ACKNOW))
2572 return(IPPROTO_DONE);
2576 * Generate an ACK dropping incoming segment if it occupies
2577 * sequence space, where the ACK reflects our state.
2579 * We can now skip the test for the RST flag since all
2580 * paths to this code happen after packets containing
2581 * RST have been dropped.
2583 * In the SYN-RECEIVED state, don't send an ACK unless the
2584 * segment we received passes the SYN-RECEIVED ACK test.
2585 * If it fails send a RST. This breaks the loop in the
2586 * "LAND" DoS attack, and also prevents an ACK storm
2587 * between two listening ports that have been sent forged
2588 * SYN segments, each with the source address of the other.
2590 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
2591 (SEQ_GT(tp->snd_una, th->th_ack) ||
2592 SEQ_GT(th->th_ack, tp->snd_max)) ) {
2593 rstreason = BANDLIM_RST_OPENPORT;
2597 if (so->so_options & SO_DEBUG)
2598 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2601 tp->t_flags |= TF_ACKNOW;
2603 return(IPPROTO_DONE);
2607 * Generate a RST, dropping incoming segment.
2608 * Make ACK acceptable to originator of segment.
2609 * Don't bother to respond if destination was broadcast/multicast.
2611 if ((thflags & TH_RST) || m->m_flags & (M_BCAST | M_MCAST))
2614 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
2615 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
2618 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
2619 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
2620 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
2621 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2624 /* IPv6 anycast check is done at tcp6_input() */
2627 * Perform bandwidth limiting.
2630 if (badport_bandlim(rstreason) < 0)
2635 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2636 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2638 if (thflags & TH_ACK)
2639 /* mtod() below is safe as long as hdr dropping is delayed */
2640 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack,
2643 if (thflags & TH_SYN)
2645 /* mtod() below is safe as long as hdr dropping is delayed */
2646 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq + tlen,
2647 (tcp_seq)0, TH_RST | TH_ACK);
2649 return(IPPROTO_DONE);
2653 * Drop space held by incoming segment and return.
2656 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2657 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2660 return(IPPROTO_DONE);
2664 * Parse TCP options and place in tcpopt.
2667 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, boolean_t is_syn)
2672 for (; cnt > 0; cnt -= optlen, cp += optlen) {
2674 if (opt == TCPOPT_EOL)
2676 if (opt == TCPOPT_NOP)
2682 if (optlen < 2 || optlen > cnt)
2687 if (optlen != TCPOLEN_MAXSEG)
2691 to->to_flags |= TOF_MSS;
2692 bcopy(cp + 2, &to->to_mss, sizeof to->to_mss);
2693 to->to_mss = ntohs(to->to_mss);
2696 if (optlen != TCPOLEN_WINDOW)
2700 to->to_flags |= TOF_SCALE;
2701 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
2703 case TCPOPT_TIMESTAMP:
2704 if (optlen != TCPOLEN_TIMESTAMP)
2706 to->to_flags |= TOF_TS;
2707 bcopy(cp + 2, &to->to_tsval, sizeof to->to_tsval);
2708 to->to_tsval = ntohl(to->to_tsval);
2709 bcopy(cp + 6, &to->to_tsecr, sizeof to->to_tsecr);
2710 to->to_tsecr = ntohl(to->to_tsecr);
2712 * If echoed timestamp is later than the current time,
2713 * fall back to non RFC1323 RTT calculation.
2715 if (to->to_tsecr != 0 && TSTMP_GT(to->to_tsecr, ticks))
2718 case TCPOPT_SACK_PERMITTED:
2719 if (optlen != TCPOLEN_SACK_PERMITTED)
2723 to->to_flags |= TOF_SACK_PERMITTED;
2726 if ((optlen - 2) & 0x07) /* not multiple of 8 */
2728 to->to_nsackblocks = (optlen - 2) / 8;
2729 to->to_sackblocks = (struct raw_sackblock *) (cp + 2);
2730 to->to_flags |= TOF_SACK;
2731 for (i = 0; i < to->to_nsackblocks; i++) {
2732 struct raw_sackblock *r = &to->to_sackblocks[i];
2734 r->rblk_start = ntohl(r->rblk_start);
2735 r->rblk_end = ntohl(r->rblk_end);
2738 #ifdef TCP_SIGNATURE
2740 * XXX In order to reply to a host which has set the
2741 * TCP_SIGNATURE option in its initial SYN, we have to
2742 * record the fact that the option was observed here
2743 * for the syncache code to perform the correct response.
2745 case TCPOPT_SIGNATURE:
2746 if (optlen != TCPOLEN_SIGNATURE)
2748 to->to_flags |= (TOF_SIGNATURE | TOF_SIGLEN);
2750 #endif /* TCP_SIGNATURE */
2758 * Pull out of band byte out of a segment so
2759 * it doesn't appear in the user's data queue.
2760 * It is still reflected in the segment length for
2761 * sequencing purposes.
2762 * "off" is the delayed to be dropped hdrlen.
2765 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, int off)
2767 int cnt = off + th->th_urp - 1;
2770 if (m->m_len > cnt) {
2771 char *cp = mtod(m, caddr_t) + cnt;
2772 struct tcpcb *tp = sototcpcb(so);
2775 tp->t_oobflags |= TCPOOB_HAVEDATA;
2776 bcopy(cp + 1, cp, m->m_len - cnt - 1);
2778 if (m->m_flags & M_PKTHDR)
2787 panic("tcp_pulloutofband");
2791 * Collect new round-trip time estimate
2792 * and update averages and current timeout.
2795 tcp_xmit_timer(struct tcpcb *tp, int rtt)
2799 tcpstat.tcps_rttupdated++;
2801 if (tp->t_srtt != 0) {
2803 * srtt is stored as fixed point with 5 bits after the
2804 * binary point (i.e., scaled by 8). The following magic
2805 * is equivalent to the smoothing algorithm in rfc793 with
2806 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2807 * point). Adjust rtt to origin 0.
2809 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
2810 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
2812 if ((tp->t_srtt += delta) <= 0)
2816 * We accumulate a smoothed rtt variance (actually, a
2817 * smoothed mean difference), then set the retransmit
2818 * timer to smoothed rtt + 4 times the smoothed variance.
2819 * rttvar is stored as fixed point with 4 bits after the
2820 * binary point (scaled by 16). The following is
2821 * equivalent to rfc793 smoothing with an alpha of .75
2822 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
2823 * rfc793's wired-in beta.
2827 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
2828 if ((tp->t_rttvar += delta) <= 0)
2830 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
2831 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2834 * No rtt measurement yet - use the unsmoothed rtt.
2835 * Set the variance to half the rtt (so our first
2836 * retransmit happens at 3*rtt).
2838 tp->t_srtt = rtt << TCP_RTT_SHIFT;
2839 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
2840 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2846 * the retransmit should happen at rtt + 4 * rttvar.
2847 * Because of the way we do the smoothing, srtt and rttvar
2848 * will each average +1/2 tick of bias. When we compute
2849 * the retransmit timer, we want 1/2 tick of rounding and
2850 * 1 extra tick because of +-1/2 tick uncertainty in the
2851 * firing of the timer. The bias will give us exactly the
2852 * 1.5 tick we need. But, because the bias is
2853 * statistical, we have to test that we don't drop below
2854 * the minimum feasible timer (which is 2 ticks).
2856 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
2857 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
2860 * We received an ack for a packet that wasn't retransmitted;
2861 * it is probably safe to discard any error indications we've
2862 * received recently. This isn't quite right, but close enough
2863 * for now (a route might have failed after we sent a segment,
2864 * and the return path might not be symmetrical).
2866 tp->t_softerror = 0;
2870 * Determine a reasonable value for maxseg size.
2871 * If the route is known, check route for mtu.
2872 * If none, use an mss that can be handled on the outgoing
2873 * interface without forcing IP to fragment; if bigger than
2874 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2875 * to utilize large mbufs. If no route is found, route has no mtu,
2876 * or the destination isn't local, use a default, hopefully conservative
2877 * size (usually 512 or the default IP max size, but no more than the mtu
2878 * of the interface), as we can't discover anything about intervening
2879 * gateways or networks. We also initialize the congestion/slow start
2880 * window to be a single segment if the destination isn't local.
2881 * While looking at the routing entry, we also initialize other path-dependent
2882 * parameters from pre-set or cached values in the routing entry.
2884 * Also take into account the space needed for options that we
2885 * send regularly. Make maxseg shorter by that amount to assure
2886 * that we can send maxseg amount of data even when the options
2887 * are present. Store the upper limit of the length of options plus
2890 * NOTE that this routine is only called when we process an incoming
2891 * segment, for outgoing segments only tcp_mssopt is called.
2894 tcp_mss(struct tcpcb *tp, int offer)
2900 struct inpcb *inp = tp->t_inpcb;
2903 boolean_t isipv6 = ((inp->inp_vflag & INP_IPV6) ? TRUE : FALSE);
2904 size_t min_protoh = isipv6 ?
2905 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
2906 sizeof(struct tcpiphdr);
2908 const boolean_t isipv6 = FALSE;
2909 const size_t min_protoh = sizeof(struct tcpiphdr);
2913 rt = tcp_rtlookup6(&inp->inp_inc);
2915 rt = tcp_rtlookup(&inp->inp_inc);
2917 tp->t_maxopd = tp->t_maxseg =
2918 (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
2922 so = inp->inp_socket;
2925 * Offer == 0 means that there was no MSS on the SYN segment,
2926 * in this case we use either the interface mtu or tcp_mssdflt.
2928 * An offer which is too large will be cut down later.
2932 if (in6_localaddr(&inp->in6p_faddr)) {
2933 offer = ND_IFINFO(rt->rt_ifp)->linkmtu -
2936 offer = tcp_v6mssdflt;
2939 if (in_localaddr(inp->inp_faddr))
2940 offer = ifp->if_mtu - min_protoh;
2942 offer = tcp_mssdflt;
2947 * Prevent DoS attack with too small MSS. Round up
2948 * to at least minmss.
2950 * Sanity check: make sure that maxopd will be large
2951 * enough to allow some data on segments even is the
2952 * all the option space is used (40bytes). Otherwise
2953 * funny things may happen in tcp_output.
2955 offer = max(offer, tcp_minmss);
2956 offer = max(offer, 64);
2958 rt->rt_rmx.rmx_mssopt = offer;
2961 * While we're here, check if there's an initial rtt
2962 * or rttvar. Convert from the route-table units
2963 * to scaled multiples of the slow timeout timer.
2965 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
2967 * XXX the lock bit for RTT indicates that the value
2968 * is also a minimum value; this is subject to time.
2970 if (rt->rt_rmx.rmx_locks & RTV_RTT)
2971 tp->t_rttmin = rtt / (RTM_RTTUNIT / hz);
2972 tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE));
2973 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
2974 tcpstat.tcps_usedrtt++;
2975 if (rt->rt_rmx.rmx_rttvar) {
2976 tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
2977 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE));
2978 tcpstat.tcps_usedrttvar++;
2980 /* default variation is +- 1 rtt */
2982 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
2984 TCPT_RANGESET(tp->t_rxtcur,
2985 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
2986 tp->t_rttmin, TCPTV_REXMTMAX);
2990 * if there's an mtu associated with the route, use it
2991 * else, use the link mtu. Take the smaller of mss or offer
2994 if (rt->rt_rmx.rmx_mtu) {
2995 mss = rt->rt_rmx.rmx_mtu - min_protoh;
2998 mss = ND_IFINFO(rt->rt_ifp)->linkmtu - min_protoh;
3000 mss = ifp->if_mtu - min_protoh;
3002 mss = min(mss, offer);
3005 * maxopd stores the maximum length of data AND options
3006 * in a segment; maxseg is the amount of data in a normal
3007 * segment. We need to store this value (maxopd) apart
3008 * from maxseg, because now every segment carries options
3009 * and thus we normally have somewhat less data in segments.
3013 if ((tp->t_flags & (TF_REQ_TSTMP | TF_NOOPT)) == TF_REQ_TSTMP &&
3014 ((tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
3015 mss -= TCPOLEN_TSTAMP_APPA;
3017 #if (MCLBYTES & (MCLBYTES - 1)) == 0
3019 mss &= ~(MCLBYTES-1);
3022 mss = mss / MCLBYTES * MCLBYTES;
3025 * If there's a pipesize, change the socket buffer
3026 * to that size. Make the socket buffers an integral
3027 * number of mss units; if the mss is larger than
3028 * the socket buffer, decrease the mss.
3031 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0)
3033 bufsize = so->so_snd.ssb_hiwat;
3037 bufsize = roundup(bufsize, mss);
3038 if (bufsize > sb_max)
3040 if (bufsize > so->so_snd.ssb_hiwat)
3041 ssb_reserve(&so->so_snd, bufsize, so, NULL);
3046 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0)
3048 bufsize = so->so_rcv.ssb_hiwat;
3049 if (bufsize > mss) {
3050 bufsize = roundup(bufsize, mss);
3051 if (bufsize > sb_max)
3053 if (bufsize > so->so_rcv.ssb_hiwat) {
3054 lwkt_gettoken(&so->so_rcv.ssb_token);
3055 ssb_reserve(&so->so_rcv, bufsize, so, NULL);
3056 lwkt_reltoken(&so->so_rcv.ssb_token);
3061 * Set the slow-start flight size depending on whether this
3062 * is a local network or not.
3065 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380));
3069 if (rt->rt_rmx.rmx_ssthresh) {
3071 * There's some sort of gateway or interface
3072 * buffer limit on the path. Use this to set
3073 * the slow start threshhold, but set the
3074 * threshold to no less than 2*mss.
3076 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
3077 tcpstat.tcps_usedssthresh++;
3082 * Determine the MSS option to send on an outgoing SYN.
3085 tcp_mssopt(struct tcpcb *tp)
3090 ((tp->t_inpcb->inp_vflag & INP_IPV6) ? TRUE : FALSE);
3091 int min_protoh = isipv6 ?
3092 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
3093 sizeof(struct tcpiphdr);
3095 const boolean_t isipv6 = FALSE;
3096 const size_t min_protoh = sizeof(struct tcpiphdr);
3100 rt = tcp_rtlookup6(&tp->t_inpcb->inp_inc);
3102 rt = tcp_rtlookup(&tp->t_inpcb->inp_inc);
3104 return (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
3106 return (rt->rt_ifp->if_mtu - min_protoh);
3110 * When a partial ack arrives, force the retransmission of the
3111 * next unacknowledged segment. Do not exit Fast Recovery.
3113 * Implement the Slow-but-Steady variant of NewReno by restarting the
3114 * the retransmission timer. Turn it off here so it can be restarted
3115 * later in tcp_output().
3118 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th, int acked)
3120 tcp_seq old_snd_nxt = tp->snd_nxt;
3121 u_long ocwnd = tp->snd_cwnd;
3123 tcp_callout_stop(tp, tp->tt_rexmt);
3125 tp->snd_nxt = th->th_ack;
3126 /* Set snd_cwnd to one segment beyond acknowledged offset. */
3127 tp->snd_cwnd = tp->t_maxseg;
3128 tp->t_flags |= TF_ACKNOW;
3130 if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
3131 tp->snd_nxt = old_snd_nxt;
3132 /* partial window deflation */
3134 tp->snd_cwnd = ocwnd - acked + tp->t_maxseg;
3136 tp->snd_cwnd = tp->t_maxseg;
3140 * In contrast to the Slow-but-Steady NewReno variant,
3141 * we do not reset the retransmission timer for SACK retransmissions,
3142 * except when retransmitting snd_una.
3145 tcp_sack_rexmt(struct tcpcb *tp, struct tcphdr *th)
3147 uint32_t pipe, seglen;
3150 tcp_seq old_snd_nxt = tp->snd_nxt;
3151 u_long ocwnd = tp->snd_cwnd;
3152 int nseg = 0; /* consecutive new segments */
3153 #define MAXBURST 4 /* limit burst of new packets on partial ack */
3156 pipe = tcp_sack_compute_pipe(tp);
3157 while ((tcp_seq_diff_t)(ocwnd - pipe) >= (tcp_seq_diff_t)tp->t_maxseg &&
3158 (!tcp_do_smartsack || nseg < MAXBURST) &&
3159 tcp_sack_nextseg(tp, &nextrexmt, &seglen, &lostdup)) {
3161 tcp_seq old_snd_max;
3164 if (nextrexmt == tp->snd_max)
3166 tp->snd_nxt = nextrexmt;
3167 tp->snd_cwnd = nextrexmt - tp->snd_una + seglen;
3168 old_snd_max = tp->snd_max;
3169 if (nextrexmt == tp->snd_una)
3170 tcp_callout_stop(tp, tp->tt_rexmt);
3171 error = tcp_output(tp);
3174 sent = tp->snd_nxt - nextrexmt;
3179 tcpstat.tcps_sndsackpack++;
3180 tcpstat.tcps_sndsackbyte += sent;
3181 if (SEQ_LT(nextrexmt, old_snd_max) &&
3182 SEQ_LT(tp->rexmt_high, tp->snd_nxt))
3183 tp->rexmt_high = seq_min(tp->snd_nxt, old_snd_max);
3185 if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
3186 tp->snd_nxt = old_snd_nxt;
3187 tp->snd_cwnd = ocwnd;
projects / dragonfly.git / blob