1 /*
2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
3 * The Regents of the University of California. All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 3. All advertising materials mentioning features or use of this software
14 * must display the following acknowledgement:
15 * This product includes software developed by the University of
16 * California, Berkeley and its contributors.
17 * 4. Neither the name of the University nor the names of its contributors
18 * may be used to endorse or promote products derived from this software
19 * without specific prior written permission.
20 *
21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * SUCH DAMAGE.
32 *
33 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95
34 * $FreeBSD$
35 */
36
37 #include "opt_ipfw.h" /* for ipfw_fwd */
38 #include "opt_tcpdebug.h"
39 #include "opt_tcp_input.h"
40
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/kernel.h>
44 #include <sys/sysctl.h>
45 #include <sys/malloc.h>
46 #include <sys/mbuf.h>
47 #include <sys/proc.h> /* for proc0 declaration */
48 #include <sys/protosw.h>
49 #include <sys/socket.h>
50 #include <sys/socketvar.h>
51 #include <sys/syslog.h>
52
53 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
54
55 #include <net/if.h>
56 #include <net/route.h>
57
58 #include <netinet/in.h>
59 #include <netinet/in_systm.h>
60 #include <netinet/ip.h>
61 #include <netinet/ip_icmp.h> /* for ICMP_BANDLIM */
62 #include <netinet/in_pcb.h>
63 #include <netinet/ip_var.h>
64 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */
65 #include <netinet/tcp.h>
66 #include <netinet/tcp_fsm.h>
67 #include <netinet/tcp_seq.h>
68 #include <netinet/tcp_timer.h>
69 #include <netinet/tcp_var.h>
70 #include <netinet/tcpip.h>
71 #ifdef TCPDEBUG
72 #include <netinet/tcp_debug.h>
73 static struct tcpiphdr tcp_saveti;
74 #endif
75
76 static int tcprexmtthresh = 3;
77 tcp_seq tcp_iss;
78 tcp_cc tcp_ccgen;
79
80 struct tcpstat tcpstat;
81 SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats,
82 CTLFLAG_RD, &tcpstat , tcpstat, "");
83
84 static int log_in_vain = 0;
85 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
86 &log_in_vain, 0, "");
87
88 int tcp_delack_enabled = 1;
89 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW,
90 &tcp_delack_enabled, 0, "");
91
92 #ifdef TCP_DROP_SYNFIN
93 static int drop_synfin = 0;
94 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW,
95 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set");
96 #endif
97
98 #ifdef TCP_RESTRICT_RST
99 static int restrict_rst = 0;
100 SYSCTL_INT(_net_inet_tcp, OID_AUTO, restrict_rst, CTLFLAG_RW,
101 &restrict_rst, 0, "Restrict RST emission");
102 #endif
103
104 u_long tcp_now;
105 struct inpcbhead tcb;
106 struct inpcbinfo tcbinfo;
107
108 static void tcp_dooptions __P((struct tcpcb *,
109 u_char *, int, struct tcpiphdr *, struct tcpopt *));
110 static void tcp_pulloutofband __P((struct socket *,
111 struct tcpiphdr *, struct mbuf *));
112 static int tcp_reass __P((struct tcpcb *, struct tcpiphdr *, struct mbuf *));
113 static void tcp_xmit_timer __P((struct tcpcb *, int));
114
115
116 /*
117 * Insert segment ti into reassembly queue of tcp with
118 * control block tp. Return TH_FIN if reassembly now includes
119 * a segment with FIN. The macro form does the common case inline
120 * (segment is the next to be received on an established connection,
121 * and the queue is empty), avoiding linkage into and removal
122 * from the queue and repetition of various conversions.
123 * Set DELACK for segments received in order, but ack immediately
124 * when segments are out of order (so fast retransmit can work).
125 */
126 #define TCP_REASS(tp, ti, m, so, flags) { \
127 if ((ti)->ti_seq == (tp)->rcv_nxt && \
128 (tp)->t_segq == NULL && \
129 (tp)->t_state == TCPS_ESTABLISHED) { \
130 if (tcp_delack_enabled) \
131 tp->t_flags |= TF_DELACK; \
132 else \
133 tp->t_flags |= TF_ACKNOW; \
134 (tp)->rcv_nxt += (ti)->ti_len; \
135 flags = (ti)->ti_flags & TH_FIN; \
136 tcpstat.tcps_rcvpack++;\
137 tcpstat.tcps_rcvbyte += (ti)->ti_len;\
138 sbappend(&(so)->so_rcv, (m)); \
139 sorwakeup(so); \
140 } else { \
141 (flags) = tcp_reass((tp), (ti), (m)); \
142 tp->t_flags |= TF_ACKNOW; \
143 } \
144 }
145
146 static int
147 tcp_reass(tp, ti, m)
148 register struct tcpcb *tp;
149 register struct tcpiphdr *ti;
150 struct mbuf *m;
151 {
152 struct mbuf *q;
153 struct mbuf *p;
154 struct mbuf *nq;
155 struct socket *so = tp->t_inpcb->inp_socket;
156 int flags;
157
158 #define GETTCP(m) ((struct tcpiphdr *)m->m_pkthdr.header)
159
160 /*
161 * Call with ti==0 after become established to
162 * force pre-ESTABLISHED data up to user socket.
163 */
164 if (ti == 0)
165 goto present;
166
167 m->m_pkthdr.header = ti;
168
169 /*
170 * Find a segment which begins after this one does.
171 */
172 for (q = tp->t_segq, p = NULL; q; p = q, q = q->m_nextpkt)
173 if (SEQ_GT(GETTCP(q)->ti_seq, ti->ti_seq))
174 break;
175
176 /*
177 * If there is a preceding segment, it may provide some of
178 * our data already. If so, drop the data from the incoming
179 * segment. If it provides all of our data, drop us.
180 */
181 if (p != NULL) {
182 register int i;
183 /* conversion to int (in i) handles seq wraparound */
184 i = GETTCP(p)->ti_seq + GETTCP(p)->ti_len - ti->ti_seq;
185 if (i > 0) {
186 if (i >= ti->ti_len) {
187 tcpstat.tcps_rcvduppack++;
188 tcpstat.tcps_rcvdupbyte += ti->ti_len;
189 m_freem(m);
190 /*
191 * Try to present any queued data
192 * at the left window edge to the user.
193 * This is needed after the 3-WHS
194 * completes.
195 */
196 goto present; /* ??? */
197 }
198 m_adj(m, i);
199 ti->ti_len -= i;
200 ti->ti_seq += i;
201 }
202 }
203 tcpstat.tcps_rcvoopack++;
204 tcpstat.tcps_rcvoobyte += ti->ti_len;
205
206 /*
207 * While we overlap succeeding segments trim them or,
208 * if they are completely covered, dequeue them.
209 */
210 while (q) {
211 register int i = (ti->ti_seq + ti->ti_len) - GETTCP(q)->ti_seq;
212 if (i <= 0)
213 break;
214 if (i < GETTCP(q)->ti_len) {
215 GETTCP(q)->ti_seq += i;
216 GETTCP(q)->ti_len -= i;
217 m_adj(q, i);
218 break;
219 }
220
221 nq = q->m_nextpkt;
222 if (p)
223 p->m_nextpkt = nq;
224 else
225 tp->t_segq = nq;
226 m_freem(q);
227 q = nq;
228 }
229
230 if (p == NULL) {
231 m->m_nextpkt = tp->t_segq;
232 tp->t_segq = m;
233 } else {
234 m->m_nextpkt = p->m_nextpkt;
235 p->m_nextpkt = m;
236 }
237
238 present:
239 /*
240 * Present data to user, advancing rcv_nxt through
241 * completed sequence space.
242 */
243 if (!TCPS_HAVEESTABLISHED(tp->t_state))
244 return (0);
245 q = tp->t_segq;
246 if (!q || GETTCP(q)->ti_seq != tp->rcv_nxt)
247 return (0);
248 do {
249 tp->rcv_nxt += GETTCP(q)->ti_len;
250 flags = GETTCP(q)->ti_flags & TH_FIN;
251 nq = q->m_nextpkt;
252 tp->t_segq = nq;
253 q->m_nextpkt = NULL;
254 if (so->so_state & SS_CANTRCVMORE)
255 m_freem(q);
256 else
257 sbappend(&so->so_rcv, q);
258 q = nq;
259 } while (q && GETTCP(q)->ti_seq == tp->rcv_nxt);
260 sorwakeup(so);
261 return (flags);
262
263 #undef GETTCP
264 }
265
266 /*
267 * TCP input routine, follows pages 65-76 of the
268 * protocol specification dated September, 1981 very closely.
269 */
270 void
271 tcp_input(m, iphlen)
272 register struct mbuf *m;
273 int iphlen;
274 {
275 register struct tcpiphdr *ti;
276 register struct inpcb *inp;
277 u_char *optp = NULL;
278 int optlen = 0;
279 int len, tlen, off;
280 register struct tcpcb *tp = 0;
281 register int tiflags;
282 struct socket *so = 0;
283 int todrop, acked, ourfinisacked, needoutput = 0;
284 struct in_addr laddr;
285 int dropsocket = 0;
286 int iss = 0;
287 u_long tiwin;
288 struct tcpopt to; /* options in this segment */
289 struct rmxp_tao *taop; /* pointer to our TAO cache entry */
290 struct rmxp_tao tao_noncached; /* in case there's no cached entry */
291 #ifdef TCPDEBUG
292 short ostate = 0;
293 #endif
294
295 bzero((char *)&to, sizeof(to));
296
297 tcpstat.tcps_rcvtotal++;
298 /*
299 * Get IP and TCP header together in first mbuf.
300 * Note: IP leaves IP header in first mbuf.
301 */
302 ti = mtod(m, struct tcpiphdr *);
303 if (iphlen > sizeof (struct ip))
304 ip_stripoptions(m, (struct mbuf *)0);
305 if (m->m_len < sizeof (struct tcpiphdr)) {
306 if ((m = m_pullup(m, sizeof (struct tcpiphdr))) == 0) {
307 tcpstat.tcps_rcvshort++;
308 return;
309 }
310 ti = mtod(m, struct tcpiphdr *);
311 }
312
313 /*
314 * Checksum extended TCP header and data.
315 */
316 tlen = ((struct ip *)ti)->ip_len;
317 len = sizeof (struct ip) + tlen;
318 bzero(ti->ti_x1, sizeof(ti->ti_x1));
319 ti->ti_len = (u_short)tlen;
320 HTONS(ti->ti_len);
321 ti->ti_sum = in_cksum(m, len);
322 if (ti->ti_sum) {
323 tcpstat.tcps_rcvbadsum++;
324 goto drop;
325 }
326
327 /*
328 * Check that TCP offset makes sense,
329 * pull out TCP options and adjust length. XXX
330 */
331 off = ti->ti_off << 2;
332 if (off < sizeof (struct tcphdr) || off > tlen) {
333 tcpstat.tcps_rcvbadoff++;
334 goto drop;
335 }
336 tlen -= off;
337 ti->ti_len = tlen;
338 if (off > sizeof (struct tcphdr)) {
339 if (m->m_len < sizeof(struct ip) + off) {
340 if ((m = m_pullup(m, sizeof (struct ip) + off)) == 0) {
341 tcpstat.tcps_rcvshort++;
342 return;
343 }
344 ti = mtod(m, struct tcpiphdr *);
345 }
346 optlen = off - sizeof (struct tcphdr);
347 optp = mtod(m, u_char *) + sizeof (struct tcpiphdr);
348 }
349 tiflags = ti->ti_flags;
350
351 #ifdef TCP_DROP_SYNFIN
352 /*
353 * If the drop_synfin option is enabled, drop all packets with
354 * both the SYN and FIN bits set. This prevents e.g. nmap from
355 * identifying the TCP/IP stack.
356 *
357 * This is incompatible with RFC1644 extensions (T/TCP).
358 */
359 if (drop_synfin && (tiflags & (TH_SYN|TH_FIN)) == (TH_SYN|TH_FIN))
360 goto drop;
361 #endif
362
363 /*
364 * Convert TCP protocol specific fields to host format.
365 */
366 NTOHL(ti->ti_seq);
367 NTOHL(ti->ti_ack);
368 NTOHS(ti->ti_win);
369 NTOHS(ti->ti_urp);
370
371 /*
372 * Drop TCP, IP headers and TCP options.
373 */
374 m->m_data += sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
375 m->m_len -= sizeof(struct tcpiphdr)+off-sizeof(struct tcphdr);
376
377 /*
378 * Locate pcb for segment.
379 */
380 findpcb:
381 #ifdef IPFIREWALL_FORWARD
382 if (ip_fw_fwd_addr != NULL) {
383 /*
384 * Diverted. Pretend to be the destination.
385 * already got one like this?
386 */
387 inp = in_pcblookup_hash(&tcbinfo, ti->ti_src, ti->ti_sport,
388 ti->ti_dst, ti->ti_dport, 0);
389 if (!inp) {
390 /*
391 * No, then it's new. Try find the ambushing socket
392 */
393 if (!ip_fw_fwd_addr->sin_port) {
394 inp = in_pcblookup_hash(&tcbinfo, ti->ti_src,
395 ti->ti_sport, ip_fw_fwd_addr->sin_addr,
396 ti->ti_dport, 1);
397 } else {
398 inp = in_pcblookup_hash(&tcbinfo,
399 ti->ti_src, ti->ti_sport,
400 ip_fw_fwd_addr->sin_addr,
401 ntohs(ip_fw_fwd_addr->sin_port), 1);
402 }
403 }
404 ip_fw_fwd_addr = NULL;
405 } else
406 #endif /* IPFIREWALL_FORWARD */
407
408 inp = in_pcblookup_hash(&tcbinfo, ti->ti_src, ti->ti_sport,
409 ti->ti_dst, ti->ti_dport, 1);
410
411 /*
412 * If the state is CLOSED (i.e., TCB does not exist) then
413 * all data in the incoming segment is discarded.
414 * If the TCB exists but is in CLOSED state, it is embryonic,
415 * but should either do a listen or a connect soon.
416 */
417 if (inp == NULL) {
418 if (log_in_vain && tiflags & TH_SYN) {
419 char buf[4*sizeof "123"];
420
421 strcpy(buf, inet_ntoa(ti->ti_dst));
422 log(LOG_INFO,
423 "Connection attempt to TCP %s:%d from %s:%d\n",
424 buf, ntohs(ti->ti_dport), inet_ntoa(ti->ti_src),
425 ntohs(ti->ti_sport));
426 }
427 #ifdef ICMP_BANDLIM
428 if (badport_bandlim(1) < 0)
429 goto drop;
430 #endif
431 goto dropwithreset;
432 }
433 tp = intotcpcb(inp);
434 if (tp == 0)
435 goto maybedropwithreset;
436 if (tp->t_state == TCPS_CLOSED)
437 goto drop;
438
439 /* Unscale the window into a 32-bit value. */
440 if ((tiflags & TH_SYN) == 0)
441 tiwin = ti->ti_win << tp->snd_scale;
442 else
443 tiwin = ti->ti_win;
444
445 so = inp->inp_socket;
446 if (so->so_options & (SO_DEBUG|SO_ACCEPTCONN)) {
447 #ifdef TCPDEBUG
448 if (so->so_options & SO_DEBUG) {
449 ostate = tp->t_state;
450 tcp_saveti = *ti;
451 }
452 #endif
453 if (so->so_options & SO_ACCEPTCONN) {
454 register struct tcpcb *tp0 = tp;
455 struct socket *so2;
456 if ((tiflags & (TH_RST|TH_ACK|TH_SYN)) != TH_SYN) {
457 /*
458 * Note: dropwithreset makes sure we don't
459 * send a RST in response to a RST.
460 */
461 if (tiflags & TH_ACK) {
462 tcpstat.tcps_badsyn++;
463 goto maybedropwithreset;
464 }
465 goto drop;
466 }
467 so2 = sonewconn(so, 0);
468 if (so2 == 0) {
469 tcpstat.tcps_listendrop++;
470 so2 = sodropablereq(so);
471 if (so2) {
472 tcp_drop(sototcpcb(so2), ETIMEDOUT);
473 so2 = sonewconn(so, 0);
474 }
475 if (!so2)
476 goto drop;
477 }
478 so = so2;
479 /*
480 * This is ugly, but ....
481 *
482 * Mark socket as temporary until we're
483 * committed to keeping it. The code at
484 * ``drop'' and ``dropwithreset'' check the
485 * flag dropsocket to see if the temporary
486 * socket created here should be discarded.
487 * We mark the socket as discardable until
488 * we're committed to it below in TCPS_LISTEN.
489 */
490 dropsocket++;
491 inp = (struct inpcb *)so->so_pcb;
492 inp->inp_laddr = ti->ti_dst;
493 inp->inp_lport = ti->ti_dport;
494 if (in_pcbinshash(inp) != 0) {
495 /*
496 * Undo the assignments above if we failed to put
497 * the PCB on the hash lists.
498 */
499 inp->inp_laddr.s_addr = INADDR_ANY;
500 inp->inp_lport = 0;
501 goto drop;
502 }
503 inp->inp_options = ip_srcroute();
504 tp = intotcpcb(inp);
505 tp->t_state = TCPS_LISTEN;
506 tp->t_flags |= tp0->t_flags & (TF_NOPUSH|TF_NOOPT);
507
508 /* Compute proper scaling value from buffer space */
509 while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
510 TCP_MAXWIN << tp->request_r_scale < so->so_rcv.sb_hiwat)
511 tp->request_r_scale++;
512 }
513 }
514
515 /*
516 * Segment received on connection.
517 * Reset idle time and keep-alive timer.
518 */
519 tp->t_idle = 0;
520 if (TCPS_HAVEESTABLISHED(tp->t_state))
521 tp->t_timer[TCPT_KEEP] = tcp_keepidle;
522
523 /*
524 * Process options if not in LISTEN state,
525 * else do it below (after getting remote address).
526 */
527 if (tp->t_state != TCPS_LISTEN)
528 tcp_dooptions(tp, optp, optlen, ti, &to);
529
530 /*
531 * Header prediction: check for the two common cases
532 * of a uni-directional data xfer. If the packet has
533 * no control flags, is in-sequence, the window didn't
534 * change and we're not retransmitting, it's a
535 * candidate. If the length is zero and the ack moved
536 * forward, we're the sender side of the xfer. Just
537 * free the data acked & wake any higher level process
538 * that was blocked waiting for space. If the length
539 * is non-zero and the ack didn't move, we're the
540 * receiver side. If we're getting packets in-order
541 * (the reassembly queue is empty), add the data to
542 * the socket buffer and note that we need a delayed ack.
543 * Make sure that the hidden state-flags are also off.
544 * Since we check for TCPS_ESTABLISHED above, it can only
545 * be TH_NEEDSYN.
546 */
547 if (tp->t_state == TCPS_ESTABLISHED &&
548 (tiflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
549 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
550 ((to.to_flag & TOF_TS) == 0 ||
551 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) &&
552 /*
553 * Using the CC option is compulsory if once started:
554 * the segment is OK if no T/TCP was negotiated or
555 * if the segment has a CC option equal to CCrecv
556 */
557 ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) != (TF_REQ_CC|TF_RCVD_CC) ||
558 ((to.to_flag & TOF_CC) != 0 && to.to_cc == tp->cc_recv)) &&
559 ti->ti_seq == tp->rcv_nxt &&
560 tiwin && tiwin == tp->snd_wnd &&
561 tp->snd_nxt == tp->snd_max) {
562
563 /*
564 * If last ACK falls within this segment's sequence numbers,
565 * record the timestamp.
566 * NOTE that the test is modified according to the latest
567 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
568 */
569 if ((to.to_flag & TOF_TS) != 0 &&
570 SEQ_LEQ(ti->ti_seq, tp->last_ack_sent)) {
571 tp->ts_recent_age = tcp_now;
572 tp->ts_recent = to.to_tsval;
573 }
574
575 if (ti->ti_len == 0) {
576 if (SEQ_GT(ti->ti_ack, tp->snd_una) &&
577 SEQ_LEQ(ti->ti_ack, tp->snd_max) &&
578 tp->snd_cwnd >= tp->snd_wnd &&
579 tp->t_dupacks < tcprexmtthresh) {
580 /*
581 * this is a pure ack for outstanding data.
582 */
583 ++tcpstat.tcps_predack;
584 if ((to.to_flag & TOF_TS) != 0)
585 tcp_xmit_timer(tp,
586 tcp_now - to.to_tsecr + 1);
587 else if (tp->t_rtt &&
588 SEQ_GT(ti->ti_ack, tp->t_rtseq))
589 tcp_xmit_timer(tp, tp->t_rtt);
590 acked = ti->ti_ack - tp->snd_una;
591 tcpstat.tcps_rcvackpack++;
592 tcpstat.tcps_rcvackbyte += acked;
593 sbdrop(&so->so_snd, acked);
594 tp->snd_una = ti->ti_ack;
595 m_freem(m);
596
597 /*
598 * If all outstanding data are acked, stop
599 * retransmit timer, otherwise restart timer
600 * using current (possibly backed-off) value.
601 * If process is waiting for space,
602 * wakeup/selwakeup/signal. If data
603 * are ready to send, let tcp_output
604 * decide between more output or persist.
605 */
606 if (tp->snd_una == tp->snd_max)
607 tp->t_timer[TCPT_REXMT] = 0;
608 else if (tp->t_timer[TCPT_PERSIST] == 0)
609 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
610
611 sowwakeup(so);
612 if (so->so_snd.sb_cc)
613 (void) tcp_output(tp);
614 return;
615 }
616 } else if (ti->ti_ack == tp->snd_una &&
617 tp->t_segq == NULL &&
618 ti->ti_len <= sbspace(&so->so_rcv)) {
619 /*
620 * this is a pure, in-sequence data packet
621 * with nothing on the reassembly queue and
622 * we have enough buffer space to take it.
623 */
624 ++tcpstat.tcps_preddat;
625 tp->rcv_nxt += ti->ti_len;
626 tcpstat.tcps_rcvpack++;
627 tcpstat.tcps_rcvbyte += ti->ti_len;
628 /*
629 * Add data to socket buffer.
630 */
631 sbappend(&so->so_rcv, m);
632 sorwakeup(so);
633 if (tcp_delack_enabled) {
634 tp->t_flags |= TF_DELACK;
635 } else {
636 tp->t_flags |= TF_ACKNOW;
637 tcp_output(tp);
638 }
639 return;
640 }
641 }
642
643 /*
644 * Calculate amount of space in receive window,
645 * and then do TCP input processing.
646 * Receive window is amount of space in rcv queue,
647 * but not less than advertised window.
648 */
649 { int win;
650
651 win = sbspace(&so->so_rcv);
652 if (win < 0)
653 win = 0;
654 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
655 }
656
657 switch (tp->t_state) {
658
659 /*
660 * If the state is LISTEN then ignore segment if it contains an RST.
661 * If the segment contains an ACK then it is bad and send a RST.
662 * If it does not contain a SYN then it is not interesting; drop it.
663 * If it is from this socket, drop it, it must be forged.
664 * Don't bother responding if the destination was a broadcast.
665 * Otherwise initialize tp->rcv_nxt, and tp->irs, select an initial
666 * tp->iss, and send a segment:
667 * <SEQ=ISS><ACK=RCV_NXT><CTL=SYN,ACK>
668 * Also initialize tp->snd_nxt to tp->iss+1 and tp->snd_una to tp->iss.
669 * Fill in remote peer address fields if not previously specified.
670 * Enter SYN_RECEIVED state, and process any other fields of this
671 * segment in this state.
672 */
673 case TCPS_LISTEN: {
674 register struct sockaddr_in *sin;
675
676 if (tiflags & TH_RST)
677 goto drop;
678 if (tiflags & TH_ACK)
679 goto maybedropwithreset;
680 if ((tiflags & TH_SYN) == 0)
681 goto drop;
682 if ((ti->ti_dport == ti->ti_sport) &&
683 (ti->ti_dst.s_addr == ti->ti_src.s_addr))
684 goto drop;
685 /*
686 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
687 * in_broadcast() should never return true on a received
688 * packet with M_BCAST not set.
689 */
690 if (m->m_flags & (M_BCAST|M_MCAST) ||
691 IN_MULTICAST(ntohl(ti->ti_src.s_addr)) ||
692 IN_MULTICAST(ntohl(ti->ti_dst.s_addr)))
693 goto drop;
694 MALLOC(sin, struct sockaddr_in *, sizeof *sin, M_SONAME,
695 M_NOWAIT);
696 if (sin == NULL)
697 goto drop;
698 sin->sin_family = AF_INET;
699 sin->sin_len = sizeof(*sin);
700 sin->sin_addr = ti->ti_src;
701 sin->sin_port = ti->ti_sport;
702 bzero((caddr_t)sin->sin_zero, sizeof(sin->sin_zero));
703 laddr = inp->inp_laddr;
704 if (inp->inp_laddr.s_addr == INADDR_ANY)
705 inp->inp_laddr = ti->ti_dst;
706 if (in_pcbconnect(inp, (struct sockaddr *)sin, &proc0)) {
707 inp->inp_laddr = laddr;
708 FREE(sin, M_SONAME);
709 goto drop;
710 }
711 FREE(sin, M_SONAME);
712 tp->t_template = tcp_template(tp);
713 if (tp->t_template == 0) {
714 tp = tcp_drop(tp, ENOBUFS);
715 dropsocket = 0; /* socket is already gone */
716 goto drop;
717 }
718 if ((taop = tcp_gettaocache(inp)) == NULL) {
719 taop = &tao_noncached;
720 bzero(taop, sizeof(*taop));
721 }
722 tcp_dooptions(tp, optp, optlen, ti, &to);
723 if (iss)
724 tp->iss = iss;
725 else {
726 #ifdef TCP_COMPAT_42
727 tcp_iss += TCP_ISSINCR/2;
728 tp->iss = tcp_iss;
729 #else
730 tp->iss = tcp_rndiss_next();
731 #endif /* TCP_COMPAT_42 */
732 }
733 tp->irs = ti->ti_seq;
734 tcp_sendseqinit(tp);
735 tcp_rcvseqinit(tp);
736 /*
737 * Initialization of the tcpcb for transaction;
738 * set SND.WND = SEG.WND,
739 * initialize CCsend and CCrecv.
740 */
741 tp->snd_wnd = tiwin; /* initial send-window */
742 tp->cc_send = CC_INC(tcp_ccgen);
743 tp->cc_recv = to.to_cc;
744 /*
745 * Perform TAO test on incoming CC (SEG.CC) option, if any.
746 * - compare SEG.CC against cached CC from the same host,
747 * if any.
748 * - if SEG.CC > chached value, SYN must be new and is accepted
749 * immediately: save new CC in the cache, mark the socket
750 * connected, enter ESTABLISHED state, turn on flag to
751 * send a SYN in the next segment.
752 * A virtual advertised window is set in rcv_adv to
753 * initialize SWS prevention. Then enter normal segment
754 * processing: drop SYN, process data and FIN.
755 * - otherwise do a normal 3-way handshake.
756 */
757 if ((to.to_flag & TOF_CC) != 0) {
758 if (((tp->t_flags & TF_NOPUSH) != 0) &&
759 taop->tao_cc != 0 && CC_GT(to.to_cc, taop->tao_cc)) {
760
761 taop->tao_cc = to.to_cc;
762 tp->t_state = TCPS_ESTABLISHED;
763
764 /*
765 * If there is a FIN, or if there is data and the
766 * connection is local, then delay SYN,ACK(SYN) in
767 * the hope of piggy-backing it on a response
768 * segment. Otherwise must send ACK now in case
769 * the other side is slow starting.
770 */
771 if (tcp_delack_enabled && ((tiflags & TH_FIN) || (ti->ti_len != 0 &&
772 in_localaddr(inp->inp_faddr))))
773 tp->t_flags |= (TF_DELACK | TF_NEEDSYN);
774 else
775 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
776
777 /*
778 * Limit the `virtual advertised window' to TCP_MAXWIN
779 * here. Even if we requested window scaling, it will
780 * become effective only later when our SYN is acked.
781 */
782 tp->rcv_adv += min(tp->rcv_wnd, TCP_MAXWIN);
783 tcpstat.tcps_connects++;
784 soisconnected(so);
785 tp->t_timer[TCPT_KEEP] = tcp_keepinit;
786 dropsocket = 0; /* committed to socket */
787 tcpstat.tcps_accepts++;
788 goto trimthenstep6;
789 }
790 /* else do standard 3-way handshake */
791 } else {
792 /*
793 * No CC option, but maybe CC.NEW:
794 * invalidate cached value.
795 */
796 taop->tao_cc = 0;
797 }
798 /*
799 * TAO test failed or there was no CC option,
800 * do a standard 3-way handshake.
801 */
802 tp->t_flags |= TF_ACKNOW;
803 tp->t_state = TCPS_SYN_RECEIVED;
804 tp->t_timer[TCPT_KEEP] = tcp_keepinit;
805 dropsocket = 0; /* committed to socket */
806 tcpstat.tcps_accepts++;
807 goto trimthenstep6;
808 }
809
810 /*
811 * If the state is SYN_RECEIVED:
812 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
813 */
814 case TCPS_SYN_RECEIVED:
815 if ((tiflags & TH_ACK) &&
816 (SEQ_LEQ(ti->ti_ack, tp->snd_una) ||
817 SEQ_GT(ti->ti_ack, tp->snd_max)))
818 goto maybedropwithreset;
819 break;
820
821 /*
822 * If the state is SYN_SENT:
823 * if seg contains an ACK, but not for our SYN, drop the input.
824 * if seg contains a RST, then drop the connection.
825 * if seg does not contain SYN, then drop it.
826 * Otherwise this is an acceptable SYN segment
827 * initialize tp->rcv_nxt and tp->irs
828 * if seg contains ack then advance tp->snd_una
829 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
830 * arrange for segment to be acked (eventually)
831 * continue processing rest of data/controls, beginning with URG
832 */
833 case TCPS_SYN_SENT:
834 if ((taop = tcp_gettaocache(inp)) == NULL) {
835 taop = &tao_noncached;
836 bzero(taop, sizeof(*taop));
837 }
838
839 if ((tiflags & TH_ACK) &&
840 (SEQ_LEQ(ti->ti_ack, tp->iss) ||
841 SEQ_GT(ti->ti_ack, tp->snd_max))) {
842 /*
843 * If we have a cached CCsent for the remote host,
844 * hence we haven't just crashed and restarted,
845 * do not send a RST. This may be a retransmission
846 * from the other side after our earlier ACK was lost.
847 * Our new SYN, when it arrives, will serve as the
848 * needed ACK.
849 */
850 if (taop->tao_ccsent != 0)
851 goto drop;
852 else
853 goto dropwithreset;
854 }
855 if (tiflags & TH_RST) {
856 if (tiflags & TH_ACK)
857 tp = tcp_drop(tp, ECONNREFUSED);
858 goto drop;
859 }
860 if ((tiflags & TH_SYN) == 0)
861 goto drop;
862 tp->snd_wnd = ti->ti_win; /* initial send window */
863 tp->cc_recv = to.to_cc; /* foreign CC */
864
865 tp->irs = ti->ti_seq;
866 tcp_rcvseqinit(tp);
867 if (tiflags & TH_ACK) {
868 /*
869 * Our SYN was acked. If segment contains CC.ECHO
870 * option, check it to make sure this segment really
871 * matches our SYN. If not, just drop it as old
872 * duplicate, but send an RST if we're still playing
873 * by the old rules. If no CC.ECHO option, make sure
874 * we don't get fooled into using T/TCP.
875 */
876 if (to.to_flag & TOF_CCECHO) {
877 if (tp->cc_send != to.to_ccecho)
878 if (taop->tao_ccsent != 0)
879 goto drop;
880 else
881 goto dropwithreset;
882 } else
883 tp->t_flags &= ~TF_RCVD_CC;
884 tcpstat.tcps_connects++;
885 soisconnected(so);
886 /* Do window scaling on this connection? */
887 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
888 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
889 tp->snd_scale = tp->requested_s_scale;
890 tp->rcv_scale = tp->request_r_scale;
891 }
892 /* Segment is acceptable, update cache if undefined. */
893 if (taop->tao_ccsent == 0)
894 taop->tao_ccsent = to.to_ccecho;
895
896 tp->rcv_adv += tp->rcv_wnd;
897 tp->snd_una++; /* SYN is acked */
898 /*
899 * If there's data, delay ACK; if there's also a FIN
900 * ACKNOW will be turned on later.
901 */
902 if (tcp_delack_enabled && ti->ti_len != 0)
903 tp->t_flags |= TF_DELACK;
904 else
905 tp->t_flags |= TF_ACKNOW;
906 /*
907 * Received <SYN,ACK> in SYN_SENT[*] state.
908 * Transitions:
909 * SYN_SENT --> ESTABLISHED
910 * SYN_SENT* --> FIN_WAIT_1
911 */
912 if (tp->t_flags & TF_NEEDFIN) {
913 tp->t_state = TCPS_FIN_WAIT_1;
914 tp->t_flags &= ~TF_NEEDFIN;
915 tiflags &= ~TH_SYN;
916 } else {
917 tp->t_state = TCPS_ESTABLISHED;
918 tp->t_timer[TCPT_KEEP] = tcp_keepidle;
919 }
920 } else {
921 /*
922 * Received initial SYN in SYN-SENT[*] state => simul-
923 * taneous open. If segment contains CC option and there is
924 * a cached CC, apply TAO test; if it succeeds, connection is
925 * half-synchronized. Otherwise, do 3-way handshake:
926 * SYN-SENT -> SYN-RECEIVED
927 * SYN-SENT* -> SYN-RECEIVED*
928 * If there was no CC option, clear cached CC value.
929 */
930 tp->t_flags |= TF_ACKNOW;
931 tp->t_timer[TCPT_REXMT] = 0;
932 if (to.to_flag & TOF_CC) {
933 if (taop->tao_cc != 0 &&
934 CC_GT(to.to_cc, taop->tao_cc)) {
935 /*
936 * update cache and make transition:
937 * SYN-SENT -> ESTABLISHED*
938 * SYN-SENT* -> FIN-WAIT-1*
939 */
940 taop->tao_cc = to.to_cc;
941 if (tp->t_flags & TF_NEEDFIN) {
942 tp->t_state = TCPS_FIN_WAIT_1;
943 tp->t_flags &= ~TF_NEEDFIN;
944 } else {
945 tp->t_state = TCPS_ESTABLISHED;
946 tp->t_timer[TCPT_KEEP] = tcp_keepidle;
947 }
948 tp->t_flags |= TF_NEEDSYN;
949 } else
950 tp->t_state = TCPS_SYN_RECEIVED;
951 } else {
952 /* CC.NEW or no option => invalidate cache */
953 taop->tao_cc = 0;
954 tp->t_state = TCPS_SYN_RECEIVED;
955 }
956 }
957
958 trimthenstep6:
959 /*
960 * Advance ti->ti_seq to correspond to first data byte.
961 * If data, trim to stay within window,
962 * dropping FIN if necessary.
963 */
964 ti->ti_seq++;
965 if (ti->ti_len > tp->rcv_wnd) {
966 todrop = ti->ti_len - tp->rcv_wnd;
967 m_adj(m, -todrop);
968 ti->ti_len = tp->rcv_wnd;
969 tiflags &= ~TH_FIN;
970 tcpstat.tcps_rcvpackafterwin++;
971 tcpstat.tcps_rcvbyteafterwin += todrop;
972 }
973 tp->snd_wl1 = ti->ti_seq - 1;
974 tp->rcv_up = ti->ti_seq;
975 /*
976 * Client side of transaction: already sent SYN and data.
977 * If the remote host used T/TCP to validate the SYN,
978 * our data will be ACK'd; if so, enter normal data segment
979 * processing in the middle of step 5, ack processing.
980 * Otherwise, goto step 6.
981 */
982 if (tiflags & TH_ACK)
983 goto process_ACK;
984 goto step6;
985 /*
986 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
987 * if segment contains a SYN and CC [not CC.NEW] option:
988 * if state == TIME_WAIT and connection duration > MSL,
989 * drop packet and send RST;
990 *
991 * if SEG.CC > CCrecv then is new SYN, and can implicitly
992 * ack the FIN (and data) in retransmission queue.
993 * Complete close and delete TCPCB. Then reprocess
994 * segment, hoping to find new TCPCB in LISTEN state;
995 *
996 * else must be old SYN; drop it.
997 * else do normal processing.
998 */
999 case TCPS_LAST_ACK:
1000 case TCPS_CLOSING:
1001 case TCPS_TIME_WAIT:
1002 if ((tiflags & TH_SYN) &&
1003 (to.to_flag & TOF_CC) && tp->cc_recv != 0) {
1004 if (tp->t_state == TCPS_TIME_WAIT &&
1005 tp->t_duration > TCPTV_MSL)
1006 goto dropwithreset;
1007 if (CC_GT(to.to_cc, tp->cc_recv)) {
1008 tp = tcp_close(tp);
1009 goto findpcb;
1010 }
1011 else
1012 goto drop;
1013 }
1014 break; /* continue normal processing */
1015 }
1016
1017 /*
1018 * States other than LISTEN or SYN_SENT.
1019 * First check the RST flag and sequence number since reset segments
1020 * are exempt from the timestamp and connection count tests. This
1021 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
1022 * below which allowed reset segments in half the sequence space
1023 * to fall though and be processed (which gives forged reset
1024 * segments with a random sequence number a 50 percent chance of
1025 * killing a connection).
1026 * Then check timestamp, if present.
1027 * Then check the connection count, if present.
1028 * Then check that at least some bytes of segment are within
1029 * receive window. If segment begins before rcv_nxt,
1030 * drop leading data (and SYN); if nothing left, just ack.
1031 *
1032 *
1033 * If the RST bit is set, check the sequence number to see
1034 * if this is a valid reset segment.
1035 * RFC 793 page 37:
1036 * In all states except SYN-SENT, all reset (RST) segments
1037 * are validated by checking their SEQ-fields. A reset is
1038 * valid if its sequence number is in the window.
1039 * Note: this does not take into account delayed ACKs, so
1040 * we should test against last_ack_sent instead of rcv_nxt.
1041 * Also, it does not make sense to allow reset segments with
1042 * sequence numbers greater than last_ack_sent to be processed
1043 * since these sequence numbers are just the acknowledgement
1044 * numbers in our outgoing packets being echoed back at us,
1045 * and these acknowledgement numbers are monotonically
1046 * increasing.
1047 * If we have multiple segments in flight, the intial reset
1048 * segment sequence numbers will be to the left of last_ack_sent,
1049 * but they will eventually catch up.
1050 * In any case, it never made sense to trim reset segments to
1051 * fit the receive window since RFC 1122 says:
1052 * 4.2.2.12 RST Segment: RFC-793 Section 3.4
1053 *
1054 * A TCP SHOULD allow a received RST segment to include data.
1055 *
1056 * DISCUSSION
1057 * It has been suggested that a RST segment could contain
1058 * ASCII text that encoded and explained the cause of the
1059 * RST. No standard has yet been established for such
1060 * data.
1061 *
1062 * If the reset segment passes the sequence number test examine
1063 * the state:
1064 * SYN_RECEIVED STATE:
1065 * If passive open, return to LISTEN state.
1066 * If active open, inform user that connection was refused.
1067 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT2, CLOSE_WAIT STATES:
1068 * Inform user that connection was reset, and close tcb.
1069 * CLOSING, LAST_ACK, TIME_WAIT STATES
1070 * Close the tcb.
1071 * TIME_WAIT state:
1072 * Drop the segment - see Stevens, vol. 2, p. 964 and
1073 * RFC 1337.
1074 */
1075 if (tiflags & TH_RST) {
1076 if (tp->last_ack_sent == ti->ti_seq) {
1077 switch (tp->t_state) {
1078
1079 case TCPS_SYN_RECEIVED:
1080 so->so_error = ECONNREFUSED;
1081 goto close;
1082
1083 case TCPS_ESTABLISHED:
1084 case TCPS_FIN_WAIT_1:
1085 case TCPS_FIN_WAIT_2:
1086 case TCPS_CLOSE_WAIT:
1087 so->so_error = ECONNRESET;
1088 close:
1089 tp->t_state = TCPS_CLOSED;
1090 tcpstat.tcps_drops++;
1091 tp = tcp_close(tp);
1092 break;
1093
1094 case TCPS_CLOSING:
1095 case TCPS_LAST_ACK:
1096 tp = tcp_close(tp);
1097 break;
1098
1099 case TCPS_TIME_WAIT:
1100 break;
1101 }
1102 }
1103 goto drop;
1104 }
1105
1106 /*
1107 * RFC 1323 PAWS: If we have a timestamp reply on this segment
1108 * and it's less than ts_recent, drop it.
1109 */
1110 if ((to.to_flag & TOF_TS) != 0 && tp->ts_recent &&
1111 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
1112
1113 /* Check to see if ts_recent is over 24 days old. */
1114 if ((int)(tcp_now - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1115 /*
1116 * Invalidate ts_recent. If this segment updates
1117 * ts_recent, the age will be reset later and ts_recent
1118 * will get a valid value. If it does not, setting
1119 * ts_recent to zero will at least satisfy the
1120 * requirement that zero be placed in the timestamp
1121 * echo reply when ts_recent isn't valid. The
1122 * age isn't reset until we get a valid ts_recent
1123 * because we don't want out-of-order segments to be
1124 * dropped when ts_recent is old.
1125 */
1126 tp->ts_recent = 0;
1127 } else {
1128 tcpstat.tcps_rcvduppack++;
1129 tcpstat.tcps_rcvdupbyte += ti->ti_len;
1130 tcpstat.tcps_pawsdrop++;
1131 goto dropafterack;
1132 }
1133 }
1134
1135 /*
1136 * T/TCP mechanism
1137 * If T/TCP was negotiated and the segment doesn't have CC,
1138 * or if its CC is wrong then drop the segment.
1139 * RST segments do not have to comply with this.
1140 */
1141 if ((tp->t_flags & (TF_REQ_CC|TF_RCVD_CC)) == (TF_REQ_CC|TF_RCVD_CC) &&
1142 ((to.to_flag & TOF_CC) == 0 || tp->cc_recv != to.to_cc))
1143 goto dropafterack;
1144
1145 /*
1146 * In the SYN-RECEIVED state, validate that the packet belongs to
1147 * this connection before trimming the data to fit the receive
1148 * window. Check the sequence number versus IRS since we know
1149 * the sequence numbers haven't wrapped. This is a partial fix
1150 * for the "LAND" DoS attack.
1151 */
1152 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(ti->ti_seq, tp->irs))
1153 goto dropwithreset;
1154
1155 todrop = tp->rcv_nxt - ti->ti_seq;
1156 if (todrop > 0) {
1157 if (tiflags & TH_SYN) {
1158 tiflags &= ~TH_SYN;
1159 ti->ti_seq++;
1160 if (ti->ti_urp > 1)
1161 ti->ti_urp--;
1162 else
1163 tiflags &= ~TH_URG;
1164 todrop--;
1165 }
1166 /*
1167 * Following if statement from Stevens, vol. 2, p. 960.
1168 */
1169 if (todrop > ti->ti_len
1170 || (todrop == ti->ti_len && (tiflags & TH_FIN) == 0)) {
1171 /*
1172 * Any valid FIN must be to the left of the window.
1173 * At this point the FIN must be a duplicate or out
1174 * of sequence; drop it.
1175 */
1176 tiflags &= ~TH_FIN;
1177
1178 /*
1179 * Send an ACK to resynchronize and drop any data.
1180 * But keep on processing for RST or ACK.
1181 */
1182 tp->t_flags |= TF_ACKNOW;
1183 todrop = ti->ti_len;
1184 tcpstat.tcps_rcvduppack++;
1185 tcpstat.tcps_rcvdupbyte += todrop;
1186 } else {
1187 tcpstat.tcps_rcvpartduppack++;
1188 tcpstat.tcps_rcvpartdupbyte += todrop;
1189 }
1190 m_adj(m, todrop);
1191 ti->ti_seq += todrop;
1192 ti->ti_len -= todrop;
1193 if (ti->ti_urp > todrop)
1194 ti->ti_urp -= todrop;
1195 else {
1196 tiflags &= ~TH_URG;
1197 ti->ti_urp = 0;
1198 }
1199 }
1200
1201 /*
1202 * If new data are received on a connection after the
1203 * user processes are gone, then RST the other end.
1204 */
1205 if ((so->so_state & SS_NOFDREF) &&
1206 tp->t_state > TCPS_CLOSE_WAIT && ti->ti_len) {
1207 tp = tcp_close(tp);
1208 tcpstat.tcps_rcvafterclose++;
1209 goto dropwithreset;
1210 }
1211
1212 /*
1213 * If segment ends after window, drop trailing data
1214 * (and PUSH and FIN); if nothing left, just ACK.
1215 */
1216 todrop = (ti->ti_seq+ti->ti_len) - (tp->rcv_nxt+tp->rcv_wnd);
1217 if (todrop > 0) {
1218 tcpstat.tcps_rcvpackafterwin++;
1219 if (todrop >= ti->ti_len) {
1220 tcpstat.tcps_rcvbyteafterwin += ti->ti_len;
1221 /*
1222 * If a new connection request is received
1223 * while in TIME_WAIT, drop the old connection
1224 * and start over if the sequence numbers
1225 * are above the previous ones.
1226 */
1227 if (tiflags & TH_SYN &&
1228 tp->t_state == TCPS_TIME_WAIT &&
1229 SEQ_GT(ti->ti_seq, tp->rcv_nxt)) {
1230 #ifdef TCP_COMPAT_42
1231 iss = tp->snd_nxt + TCP_ISSINCR;
1232 #else
1233 iss = tcp_rndiss_next();
1234 #endif /* TCP_COMPAT_42 */
1235 tp = tcp_close(tp);
1236 goto findpcb;
1237 }
1238 /*
1239 * If window is closed can only take segments at
1240 * window edge, and have to drop data and PUSH from
1241 * incoming segments. Continue processing, but
1242 * remember to ack. Otherwise, drop segment
1243 * and ack.
1244 */
1245 if (tp->rcv_wnd == 0 && ti->ti_seq == tp->rcv_nxt) {
1246 tp->t_flags |= TF_ACKNOW;
1247 tcpstat.tcps_rcvwinprobe++;
1248 } else
1249 goto dropafterack;
1250 } else
1251 tcpstat.tcps_rcvbyteafterwin += todrop;
1252 m_adj(m, -todrop);
1253 ti->ti_len -= todrop;
1254 tiflags &= ~(TH_PUSH|TH_FIN);
1255 }
1256
1257 /*
1258 * If last ACK falls within this segment's sequence numbers,
1259 * record its timestamp.
1260 * NOTE that the test is modified according to the latest
1261 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1262 */
1263 if ((to.to_flag & TOF_TS) != 0 &&
1264 SEQ_LEQ(ti->ti_seq, tp->last_ack_sent)) {
1265 tp->ts_recent_age = tcp_now;
1266 tp->ts_recent = to.to_tsval;
1267 }
1268
1269 /*
1270 * If a SYN is in the window, then this is an
1271 * error and we send an RST and drop the connection.
1272 */
1273 if (tiflags & TH_SYN) {
1274 tp = tcp_drop(tp, ECONNRESET);
1275 goto dropwithreset;
1276 }
1277
1278 /*
1279 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
1280 * flag is on (half-synchronized state), then queue data for
1281 * later processing; else drop segment and return.
1282 */
1283 if ((tiflags & TH_ACK) == 0) {
1284 if (tp->t_state == TCPS_SYN_RECEIVED ||
1285 (tp->t_flags & TF_NEEDSYN))
1286 goto step6;
1287 else
1288 goto drop;
1289 }
1290
1291 /*
1292 * Ack processing.
1293 */
1294 switch (tp->t_state) {
1295
1296 /*
1297 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
1298 * ESTABLISHED state and continue processing.
1299 * The ACK was checked above.
1300 */
1301 case TCPS_SYN_RECEIVED:
1302
1303 tcpstat.tcps_connects++;
1304 soisconnected(so);
1305 /* Do window scaling? */
1306 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1307 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1308 tp->snd_scale = tp->requested_s_scale;
1309 tp->rcv_scale = tp->request_r_scale;
1310 }
1311 /*
1312 * Upon successful completion of 3-way handshake,
1313 * update cache.CC if it was undefined, pass any queued
1314 * data to the user, and advance state appropriately.
1315 */
1316 if ((taop = tcp_gettaocache(inp)) != NULL &&
1317 taop->tao_cc == 0)
1318 taop->tao_cc = tp->cc_recv;
1319
1320 /*
1321 * Make transitions:
1322 * SYN-RECEIVED -> ESTABLISHED
1323 * SYN-RECEIVED* -> FIN-WAIT-1
1324 */
1325 if (tp->t_flags & TF_NEEDFIN) {
1326 tp->t_state = TCPS_FIN_WAIT_1;
1327 tp->t_flags &= ~TF_NEEDFIN;
1328 } else {
1329 tp->t_state = TCPS_ESTABLISHED;
1330 tp->t_timer[TCPT_KEEP] = tcp_keepidle;
1331 }
1332 /*
1333 * If segment contains data or ACK, will call tcp_reass()
1334 * later; if not, do so now to pass queued data to user.
1335 */
1336 if (ti->ti_len == 0 && (tiflags & TH_FIN) == 0)
1337 (void) tcp_reass(tp, (struct tcpiphdr *)0,
1338 (struct mbuf *)0);
1339 tp->snd_wl1 = ti->ti_seq - 1;
1340 /* fall into ... */
1341
1342 /*
1343 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1344 * ACKs. If the ack is in the range
1345 * tp->snd_una < ti->ti_ack <= tp->snd_max
1346 * then advance tp->snd_una to ti->ti_ack and drop
1347 * data from the retransmission queue. If this ACK reflects
1348 * more up to date window information we update our window information.
1349 */
1350 case TCPS_ESTABLISHED:
1351 case TCPS_FIN_WAIT_1:
1352 case TCPS_FIN_WAIT_2:
1353 case TCPS_CLOSE_WAIT:
1354 case TCPS_CLOSING:
1355 case TCPS_LAST_ACK:
1356 case TCPS_TIME_WAIT:
1357
1358 if (SEQ_LEQ(ti->ti_ack, tp->snd_una)) {
1359 if (ti->ti_len == 0 && tiwin == tp->snd_wnd) {
1360 tcpstat.tcps_rcvdupack++;
1361 /*
1362 * If we have outstanding data (other than
1363 * a window probe), this is a completely
1364 * duplicate ack (ie, window info didn't
1365 * change), the ack is the biggest we've
1366 * seen and we've seen exactly our rexmt
1367 * threshhold of them, assume a packet
1368 * has been dropped and retransmit it.
1369 * Kludge snd_nxt & the congestion
1370 * window so we send only this one
1371 * packet.
1372 *
1373 * We know we're losing at the current
1374 * window size so do congestion avoidance
1375 * (set ssthresh to half the current window
1376 * and pull our congestion window back to
1377 * the new ssthresh).
1378 *
1379 * Dup acks mean that packets have left the
1380 * network (they're now cached at the receiver)
1381 * so bump cwnd by the amount in the receiver
1382 * to keep a constant cwnd packets in the
1383 * network.
1384 */
1385 if (tp->t_timer[TCPT_REXMT] == 0 ||
1386 ti->ti_ack != tp->snd_una)
1387 tp->t_dupacks = 0;
1388 else if (++tp->t_dupacks == tcprexmtthresh) {
1389 tcp_seq onxt = tp->snd_nxt;
1390 u_int win =
1391 min(tp->snd_wnd, tp->snd_cwnd) / 2 /
1392 tp->t_maxseg;
1393
1394 if (win < 2)
1395 win = 2;
1396 tp->snd_ssthresh = win * tp->t_maxseg;
1397 tp->t_timer[TCPT_REXMT] = 0;
1398 tp->t_rtt = 0;
1399 tp->snd_nxt = ti->ti_ack;
1400 tp->snd_cwnd = tp->t_maxseg;
1401 (void) tcp_output(tp);
1402 tp->snd_cwnd = tp->snd_ssthresh +
1403 tp->t_maxseg * tp->t_dupacks;
1404 if (SEQ_GT(onxt, tp->snd_nxt))
1405 tp->snd_nxt = onxt;
1406 goto drop;
1407 } else if (tp->t_dupacks > tcprexmtthresh) {
1408 tp->snd_cwnd += tp->t_maxseg;
1409 (void) tcp_output(tp);
1410 goto drop;
1411 }
1412 } else
1413 tp->t_dupacks = 0;
1414 break;
1415 }
1416 /*
1417 * If the congestion window was inflated to account
1418 * for the other side's cached packets, retract it.
1419 */
1420 if (tp->t_dupacks >= tcprexmtthresh &&
1421 tp->snd_cwnd > tp->snd_ssthresh)
1422 tp->snd_cwnd = tp->snd_ssthresh;
1423 tp->t_dupacks = 0;
1424 if (SEQ_GT(ti->ti_ack, tp->snd_max)) {
1425 tcpstat.tcps_rcvacktoomuch++;
1426 goto dropafterack;
1427 }
1428 /*
1429 * If we reach this point, ACK is not a duplicate,
1430 * i.e., it ACKs something we sent.
1431 */
1432 if (tp->t_flags & TF_NEEDSYN) {
1433 /*
1434 * T/TCP: Connection was half-synchronized, and our
1435 * SYN has been ACK'd (so connection is now fully
1436 * synchronized). Go to non-starred state,
1437 * increment snd_una for ACK of SYN, and check if
1438 * we can do window scaling.
1439 */
1440 tp->t_flags &= ~TF_NEEDSYN;
1441 tp->snd_una++;
1442 /* Do window scaling? */
1443 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1444 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1445 tp->snd_scale = tp->requested_s_scale;
1446 tp->rcv_scale = tp->request_r_scale;
1447 }
1448 }
1449
1450 process_ACK:
1451 acked = ti->ti_ack - tp->snd_una;
1452 tcpstat.tcps_rcvackpack++;
1453 tcpstat.tcps_rcvackbyte += acked;
1454
1455 /*
1456 * If we have a timestamp reply, update smoothed
1457 * round trip time. If no timestamp is present but
1458 * transmit timer is running and timed sequence
1459 * number was acked, update smoothed round trip time.
1460 * Since we now have an rtt measurement, cancel the
1461 * timer backoff (cf., Phil Karn's retransmit alg.).
1462 * Recompute the initial retransmit timer.
1463 */
1464 if (to.to_flag & TOF_TS)
1465 tcp_xmit_timer(tp, tcp_now - to.to_tsecr + 1);
1466 else if (tp->t_rtt && SEQ_GT(ti->ti_ack, tp->t_rtseq))
1467 tcp_xmit_timer(tp,tp->t_rtt);
1468
1469 /*
1470 * If all outstanding data is acked, stop retransmit
1471 * timer and remember to restart (more output or persist).
1472 * If there is more data to be acked, restart retransmit
1473 * timer, using current (possibly backed-off) value.
1474 */
1475 if (ti->ti_ack == tp->snd_max) {
1476 tp->t_timer[TCPT_REXMT] = 0;
1477 needoutput = 1;
1478 } else if (tp->t_timer[TCPT_PERSIST] == 0)
1479 tp->t_timer[TCPT_REXMT] = tp->t_rxtcur;
1480
1481 /*
1482 * If no data (only SYN) was ACK'd,
1483 * skip rest of ACK processing.
1484 */
1485 if (acked == 0)
1486 goto step6;
1487
1488 /*
1489 * When new data is acked, open the congestion window.
1490 * If the window gives us less than ssthresh packets
1491 * in flight, open exponentially (maxseg per packet).
1492 * Otherwise open linearly: maxseg per window
1493 * (maxseg^2 / cwnd per packet).
1494 */
1495 {
1496 register u_int cw = tp->snd_cwnd;
1497 register u_int incr = tp->t_maxseg;
1498
1499 if (cw > tp->snd_ssthresh)
1500 incr = incr * incr / cw;
1501 tp->snd_cwnd = min(cw + incr, TCP_MAXWIN<<tp->snd_scale);
1502 }
1503 if (acked > so->so_snd.sb_cc) {
1504 tp->snd_wnd -= so->so_snd.sb_cc;
1505 sbdrop(&so->so_snd, (int)so->so_snd.sb_cc);
1506 ourfinisacked = 1;
1507 } else {
1508 sbdrop(&so->so_snd, acked);
1509 tp->snd_wnd -= acked;
1510 ourfinisacked = 0;
1511 }
1512 sowwakeup(so);
1513 tp->snd_una = ti->ti_ack;
1514 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
1515 tp->snd_nxt = tp->snd_una;
1516
1517 switch (tp->t_state) {
1518
1519 /*
1520 * In FIN_WAIT_1 STATE in addition to the processing
1521 * for the ESTABLISHED state if our FIN is now acknowledged
1522 * then enter FIN_WAIT_2.
1523 */
1524 case TCPS_FIN_WAIT_1:
1525 if (ourfinisacked) {
1526 /*
1527 * If we can't receive any more
1528 * data, then closing user can proceed.
1529 * Starting the timer is contrary to the
1530 * specification, but if we don't get a FIN
1531 * we'll hang forever.
1532 */
1533 if (so->so_state & SS_CANTRCVMORE) {
1534 soisdisconnected(so);
1535 tp->t_timer[TCPT_2MSL] = tcp_maxidle;
1536 }
1537 tp->t_state = TCPS_FIN_WAIT_2;
1538 }
1539 break;
1540
1541 /*
1542 * In CLOSING STATE in addition to the processing for
1543 * the ESTABLISHED state if the ACK acknowledges our FIN
1544 * then enter the TIME-WAIT state, otherwise ignore
1545 * the segment.
1546 */
1547 case TCPS_CLOSING:
1548 if (ourfinisacked) {
1549 tp->t_state = TCPS_TIME_WAIT;
1550 tcp_canceltimers(tp);
1551 /* Shorten TIME_WAIT [RFC-1644, p.28] */
1552 if (tp->cc_recv != 0 &&
1553 tp->t_duration < TCPTV_MSL)
1554 tp->t_timer[TCPT_2MSL] =
1555 tp->t_rxtcur * TCPTV_TWTRUNC;
1556 else
1557 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1558 soisdisconnected(so);
1559 }
1560 break;
1561
1562 /*
1563 * In LAST_ACK, we may still be waiting for data to drain
1564 * and/or to be acked, as well as for the ack of our FIN.
1565 * If our FIN is now acknowledged, delete the TCB,
1566 * enter the closed state and return.
1567 */
1568 case TCPS_LAST_ACK:
1569 if (ourfinisacked) {
1570 tp = tcp_close(tp);
1571 goto drop;
1572 }
1573 break;
1574
1575 /*
1576 * In TIME_WAIT state the only thing that should arrive
1577 * is a retransmission of the remote FIN. Acknowledge
1578 * it and restart the finack timer.
1579 */
1580 case TCPS_TIME_WAIT:
1581 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1582 goto dropafterack;
1583 }
1584 }
1585
1586 step6:
1587 /*
1588 * Update window information.
1589 * Don't look at window if no ACK: TAC's send garbage on first SYN.
1590 */
1591 if ((tiflags & TH_ACK) &&
1592 (SEQ_LT(tp->snd_wl1, ti->ti_seq) ||
1593 (tp->snd_wl1 == ti->ti_seq && (SEQ_LT(tp->snd_wl2, ti->ti_ack) ||
1594 (tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd))))) {
1595 /* keep track of pure window updates */
1596 if (ti->ti_len == 0 &&
1597 tp->snd_wl2 == ti->ti_ack && tiwin > tp->snd_wnd)
1598 tcpstat.tcps_rcvwinupd++;
1599 tp->snd_wnd = tiwin;
1600 tp->snd_wl1 = ti->ti_seq;
1601 tp->snd_wl2 = ti->ti_ack;
1602 if (tp->snd_wnd > tp->max_sndwnd)
1603 tp->max_sndwnd = tp->snd_wnd;
1604 needoutput = 1;
1605 }
1606
1607 /*
1608 * Process segments with URG.
1609 */
1610 if ((tiflags & TH_URG) && ti->ti_urp &&
1611 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
1612 /*
1613 * This is a kludge, but if we receive and accept
1614 * random urgent pointers, we'll crash in
1615 * soreceive. It's hard to imagine someone
1616 * actually wanting to send this much urgent data.
1617 */
1618 if (ti->ti_urp + so->so_rcv.sb_cc > sb_max) {
1619 ti->ti_urp = 0; /* XXX */
1620 tiflags &= ~TH_URG; /* XXX */
1621 goto dodata; /* XXX */
1622 }
1623 /*
1624 * If this segment advances the known urgent pointer,
1625 * then mark the data stream. This should not happen
1626 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
1627 * a FIN has been received from the remote side.
1628 * In these states we ignore the URG.
1629 *
1630 * According to RFC961 (Assigned Protocols),
1631 * the urgent pointer points to the last octet
1632 * of urgent data. We continue, however,
1633 * to consider it to indicate the first octet
1634 * of data past the urgent section as the original
1635 * spec states (in one of two places).
1636 */
1637 if (SEQ_GT(ti->ti_seq+ti->ti_urp, tp->rcv_up)) {
1638 tp->rcv_up = ti->ti_seq + ti->ti_urp;
1639 so->so_oobmark = so->so_rcv.sb_cc +
1640 (tp->rcv_up - tp->rcv_nxt) - 1;
1641 if (so->so_oobmark == 0)
1642 so->so_state |= SS_RCVATMARK;
1643 sohasoutofband(so);
1644 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
1645 }
1646 /*
1647 * Remove out of band data so doesn't get presented to user.
1648 * This can happen independent of advancing the URG pointer,
1649 * but if two URG's are pending at once, some out-of-band
1650 * data may creep in... ick.
1651 */
1652 if (ti->ti_urp <= (u_long)ti->ti_len
1653 #ifdef SO_OOBINLINE
1654 && (so->so_options & SO_OOBINLINE) == 0
1655 #endif
1656 )
1657 tcp_pulloutofband(so, ti, m);
1658 } else
1659 /*
1660 * If no out of band data is expected,
1661 * pull receive urgent pointer along
1662 * with the receive window.
1663 */
1664 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
1665 tp->rcv_up = tp->rcv_nxt;
1666 dodata: /* XXX */
1667
1668 /*
1669 * Process the segment text, merging it into the TCP sequencing queue,
1670 * and arranging for acknowledgment of receipt if necessary.
1671 * This process logically involves adjusting tp->rcv_wnd as data
1672 * is presented to the user (this happens in tcp_usrreq.c,
1673 * case PRU_RCVD). If a FIN has already been received on this
1674 * connection then we just ignore the text.
1675 */
1676 if ((ti->ti_len || (tiflags&TH_FIN)) &&
1677 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
1678 TCP_REASS(tp, ti, m, so, tiflags);
1679 /*
1680 * Note the amount of data that peer has sent into
1681 * our window, in order to estimate the sender's
1682 * buffer size.
1683 */
1684 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
1685 } else {
1686 m_freem(m);
1687 tiflags &= ~TH_FIN;
1688 }
1689
1690 /*
1691 * If FIN is received ACK the FIN and let the user know
1692 * that the connection is closing.
1693 */
1694 if (tiflags & TH_FIN) {
1695 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
1696 socantrcvmore(so);
1697 /*
1698 * If connection is half-synchronized
1699 * (ie NEEDSYN flag on) then delay ACK,
1700 * so it may be piggybacked when SYN is sent.
1701 * Otherwise, since we received a FIN then no
1702 * more input can be expected, send ACK now.
1703 */
1704 if (tcp_delack_enabled && (tp->t_flags & TF_NEEDSYN))
1705 tp->t_flags |= TF_DELACK;
1706 else
1707 tp->t_flags |= TF_ACKNOW;
1708 tp->rcv_nxt++;
1709 }
1710 switch (tp->t_state) {
1711
1712 /*
1713 * In SYN_RECEIVED and ESTABLISHED STATES
1714 * enter the CLOSE_WAIT state.
1715 */
1716 case TCPS_SYN_RECEIVED:
1717 case TCPS_ESTABLISHED:
1718 tp->t_state = TCPS_CLOSE_WAIT;
1719 break;
1720
1721 /*
1722 * If still in FIN_WAIT_1 STATE FIN has not been acked so
1723 * enter the CLOSING state.
1724 */
1725 case TCPS_FIN_WAIT_1:
1726 tp->t_state = TCPS_CLOSING;
1727 break;
1728
1729 /*
1730 * In FIN_WAIT_2 state enter the TIME_WAIT state,
1731 * starting the time-wait timer, turning off the other
1732 * standard timers.
1733 */
1734 case TCPS_FIN_WAIT_2:
1735 tp->t_state = TCPS_TIME_WAIT;
1736 tcp_canceltimers(tp);
1737 /* Shorten TIME_WAIT [RFC-1644, p.28] */
1738 if (tp->cc_recv != 0 &&
1739 tp->t_duration < TCPTV_MSL) {
1740 tp->t_timer[TCPT_2MSL] =
1741 tp->t_rxtcur * TCPTV_TWTRUNC;
1742 /* For transaction client, force ACK now. */
1743 tp->t_flags |= TF_ACKNOW;
1744 }
1745 else
1746 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1747 soisdisconnected(so);
1748 break;
1749
1750 /*
1751 * In TIME_WAIT state restart the 2 MSL time_wait timer.
1752 */
1753 case TCPS_TIME_WAIT:
1754 tp->t_timer[TCPT_2MSL] = 2 * TCPTV_MSL;
1755 break;
1756 }
1757 }
1758 #ifdef TCPDEBUG
1759 if (so->so_options & SO_DEBUG)
1760 tcp_trace(TA_INPUT, ostate, tp, &tcp_saveti, 0);
1761 #endif
1762
1763 /*
1764 * Return any desired output.
1765 */
1766 if (needoutput || (tp->t_flags & TF_ACKNOW))
1767 (void) tcp_output(tp);
1768 return;
1769
1770 dropafterack:
1771 /*
1772 * Generate an ACK dropping incoming segment if it occupies
1773 * sequence space, where the ACK reflects our state.
1774 *
1775 * We can now skip the test for the RST flag since all
1776 * paths to this code happen after packets containing
1777 * RST have been dropped.
1778 *
1779 * In the SYN-RECEIVED state, don't send an ACK unless the
1780 * segment we received passes the SYN-RECEIVED ACK test.
1781 * If it fails send a RST. This breaks the loop in the
1782 * "LAND" DoS attack, and also prevents an ACK storm
1783 * between two listening ports that have been sent forged
1784 * SYN segments, each with the source address of the other.
1785 */
1786 if (tp->t_state == TCPS_SYN_RECEIVED && (tiflags & TH_ACK) &&
1787 (SEQ_GT(tp->snd_una, ti->ti_ack) ||
1788 SEQ_GT(ti->ti_ack, tp->snd_max)) )
1789 goto maybedropwithreset;
1790 #ifdef TCPDEBUG
1791 if (so->so_options & SO_DEBUG)
1792 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0);
1793 #endif
1794 m_freem(m);
1795 tp->t_flags |= TF_ACKNOW;
1796 (void) tcp_output(tp);
1797 return;
1798
1799 /*
1800 * Conditionally drop with reset or just drop depending on whether
1801 * we think we are under attack or not.
1802 */
1803 maybedropwithreset:
1804 #ifdef ICMP_BANDLIM
1805 if (badport_bandlim(1) < 0)
1806 goto drop;
1807 #endif
1808 /* fall through */
1809 dropwithreset:
1810 #ifdef TCP_RESTRICT_RST
1811 if (restrict_rst)
1812 goto drop;
1813 #endif
1814 /*
1815 * Generate a RST, dropping incoming segment.
1816 * Make ACK acceptable to originator of segment.
1817 * Don't bother to respond if destination was broadcast/multicast.
1818 */
1819 if ((tiflags & TH_RST) ||
1820 m->m_flags & (M_BCAST|M_MCAST) ||
1821 IN_MULTICAST(ntohl(ti->ti_src.s_addr)) ||
1822 IN_MULTICAST(ntohl(ti->ti_dst.s_addr)))
1823 goto drop;
1824 #ifdef TCPDEBUG
1825 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
1826 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0);
1827 #endif
1828 if (tiflags & TH_ACK)
1829 tcp_respond(tp, ti, m, (tcp_seq)0, ti->ti_ack, TH_RST);
1830 else {
1831 if (tiflags & TH_SYN)
1832 ti->ti_len++;
1833 tcp_respond(tp, ti, m, ti->ti_seq+ti->ti_len, (tcp_seq)0,
1834 TH_RST|TH_ACK);
1835 }
1836 /* destroy temporarily created socket */
1837 if (dropsocket)
1838 (void) soabort(so);
1839 return;
1840
1841 drop:
1842 /*
1843 * Drop space held by incoming segment and return.
1844 */
1845 #ifdef TCPDEBUG
1846 if (tp == 0 || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
1847 tcp_trace(TA_DROP, ostate, tp, &tcp_saveti, 0);
1848 #endif
1849 m_freem(m);
1850 /* destroy temporarily created socket */
1851 if (dropsocket)
1852 (void) soabort(so);
1853 return;
1854 }
1855
1856 static void
1857 tcp_dooptions(tp, cp, cnt, ti, to)
1858 struct tcpcb *tp;
1859 u_char *cp;
1860 int cnt;
1861 struct tcpiphdr *ti;
1862 struct tcpopt *to;
1863 {
1864 u_short mss = 0;
1865 int opt, optlen;
1866
1867 for (; cnt > 0; cnt -= optlen, cp += optlen) {
1868 opt = cp[0];
1869 if (opt == TCPOPT_EOL)
1870 break;
1871 if (opt == TCPOPT_NOP)
1872 optlen = 1;
1873 else {
1874 optlen = cp[1];
1875 if (optlen <= 0)
1876 break;
1877 }
1878 switch (opt) {
1879
1880 default:
1881 continue;
1882
1883 case TCPOPT_MAXSEG:
1884 if (optlen != TCPOLEN_MAXSEG)
1885 continue;
1886 if (!(ti->ti_flags & TH_SYN))
1887 continue;
1888 bcopy((char *) cp + 2, (char *) &mss, sizeof(mss));
1889 NTOHS(mss);
1890 break;
1891
1892 case TCPOPT_WINDOW:
1893 if (optlen != TCPOLEN_WINDOW)
1894 continue;
1895 if (!(ti->ti_flags & TH_SYN))
1896 continue;
1897 tp->t_flags |= TF_RCVD_SCALE;
1898 tp->requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
1899 break;
1900
1901 case TCPOPT_TIMESTAMP:
1902 if (optlen != TCPOLEN_TIMESTAMP)
1903 continue;
1904 to->to_flag |= TOF_TS;
1905 bcopy((char *)cp + 2,
1906 (char *)&to->to_tsval, sizeof(to->to_tsval));
1907 NTOHL(to->to_tsval);
1908 bcopy((char *)cp + 6,
1909 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
1910 NTOHL(to->to_tsecr);
1911
1912 /*
1913 * A timestamp received in a SYN makes
1914 * it ok to send timestamp requests and replies.
1915 */
1916 if (ti->ti_flags & TH_SYN) {
1917 tp->t_flags |= TF_RCVD_TSTMP;
1918 tp->ts_recent = to->to_tsval;
1919 tp->ts_recent_age = tcp_now;
1920 }
1921 break;
1922 case TCPOPT_CC:
1923 if (optlen != TCPOLEN_CC)
1924 continue;
1925 to->to_flag |= TOF_CC;
1926 bcopy((char *)cp + 2,
1927 (char *)&to->to_cc, sizeof(to->to_cc));
1928 NTOHL(to->to_cc);
1929 /*
1930 * A CC or CC.new option received in a SYN makes
1931 * it ok to send CC in subsequent segments.
1932 */
1933 if (ti->ti_flags & TH_SYN)
1934 tp->t_flags |= TF_RCVD_CC;
1935 break;
1936 case TCPOPT_CCNEW:
1937 if (optlen != TCPOLEN_CC)
1938 continue;
1939 if (!(ti->ti_flags & TH_SYN))
1940 continue;
1941 to->to_flag |= TOF_CCNEW;
1942 bcopy((char *)cp + 2,
1943 (char *)&to->to_cc, sizeof(to->to_cc));
1944 NTOHL(to->to_cc);
1945 /*
1946 * A CC or CC.new option received in a SYN makes
1947 * it ok to send CC in subsequent segments.
1948 */
1949 tp->t_flags |= TF_RCVD_CC;
1950 break;
1951 case TCPOPT_CCECHO:
1952 if (optlen != TCPOLEN_CC)
1953 continue;
1954 if (!(ti->ti_flags & TH_SYN))
1955 continue;
1956 to->to_flag |= TOF_CCECHO;
1957 bcopy((char *)cp + 2,
1958 (char *)&to->to_ccecho, sizeof(to->to_ccecho));
1959 NTOHL(to->to_ccecho);
1960 break;
1961 }
1962 }
1963 if (ti->ti_flags & TH_SYN)
1964 tcp_mss(tp, mss); /* sets t_maxseg */
1965 }
1966
1967 /*
1968 * Pull out of band byte out of a segment so
1969 * it doesn't appear in the user's data queue.
1970 * It is still reflected in the segment length for
1971 * sequencing purposes.
1972 */
1973 static void
1974 tcp_pulloutofband(so, ti, m)
1975 struct socket *so;
1976 struct tcpiphdr *ti;
1977 register struct mbuf *m;
1978 {
1979 int cnt = ti->ti_urp - 1;
1980
1981 while (cnt >= 0) {
1982 if (m->m_len > cnt) {
1983 char *cp = mtod(m, caddr_t) + cnt;
1984 struct tcpcb *tp = sototcpcb(so);
1985
1986 tp->t_iobc = *cp;
1987 tp->t_oobflags |= TCPOOB_HAVEDATA;
1988 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
1989 m->m_len--;
1990 return;
1991 }
1992 cnt -= m->m_len;
1993 m = m->m_next;
1994 if (m == 0)
1995 break;
1996 }
1997 panic("tcp_pulloutofband");
1998 }
1999
2000 /*
2001 * Collect new round-trip time estimate
2002 * and update averages and current timeout.
2003 */
2004 static void
2005 tcp_xmit_timer(tp, rtt)
2006 register struct tcpcb *tp;
2007 short rtt;
2008 {
2009 register int delta;
2010
2011 tcpstat.tcps_rttupdated++;
2012 tp->t_rttupdated++;
2013 if (tp->t_srtt != 0) {
2014 /*
2015 * srtt is stored as fixed point with 5 bits after the
2016 * binary point (i.e., scaled by 8). The following magic
2017 * is equivalent to the smoothing algorithm in rfc793 with
2018 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2019 * point). Adjust rtt to origin 0.
2020 */
2021 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
2022 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
2023
2024 if ((tp->t_srtt += delta) <= 0)
2025 tp->t_srtt = 1;
2026
2027 /*
2028 * We accumulate a smoothed rtt variance (actually, a
2029 * smoothed mean difference), then set the retransmit
2030 * timer to smoothed rtt + 4 times the smoothed variance.
2031 * rttvar is stored as fixed point with 4 bits after the
2032 * binary point (scaled by 16). The following is
2033 * equivalent to rfc793 smoothing with an alpha of .75
2034 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
2035 * rfc793's wired-in beta.
2036 */
2037 if (delta < 0)
2038 delta = -delta;
2039 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
2040 if ((tp->t_rttvar += delta) <= 0)
2041 tp->t_rttvar = 1;
2042 } else {
2043 /*
2044 * No rtt measurement yet - use the unsmoothed rtt.
2045 * Set the variance to half the rtt (so our first
2046 * retransmit happens at 3*rtt).
2047 */
2048 tp->t_srtt = rtt << TCP_RTT_SHIFT;
2049 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
2050 }
2051 tp->t_rtt = 0;
2052 tp->t_rxtshift = 0;
2053
2054 /*
2055 * the retransmit should happen at rtt + 4 * rttvar.
2056 * Because of the way we do the smoothing, srtt and rttvar
2057 * will each average +1/2 tick of bias. When we compute
2058 * the retransmit timer, we want 1/2 tick of rounding and
2059 * 1 extra tick because of +-1/2 tick uncertainty in the
2060 * firing of the timer. The bias will give us exactly the
2061 * 1.5 tick we need. But, because the bias is
2062 * statistical, we have to test that we don't drop below
2063 * the minimum feasible timer (which is 2 ticks).
2064 */
2065 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
2066 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
2067
2068 /*
2069 * We received an ack for a packet that wasn't retransmitted;
2070 * it is probably safe to discard any error indications we've
2071 * received recently. This isn't quite right, but close enough
2072 * for now (a route might have failed after we sent a segment,
2073 * and the return path might not be symmetrical).
2074 */
2075 tp->t_softerror = 0;
2076 }
2077
2078 /*
2079 * Determine a reasonable value for maxseg size.
2080 * If the route is known, check route for mtu.
2081 * If none, use an mss that can be handled on the outgoing
2082 * interface without forcing IP to fragment; if bigger than
2083 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2084 * to utilize large mbufs. If no route is found, route has no mtu,
2085 * or the destination isn't local, use a default, hopefully conservative
2086 * size (usually 512 or the default IP max size, but no more than the mtu
2087 * of the interface), as we can't discover anything about intervening
2088 * gateways or networks. We also initialize the congestion/slow start
2089 * window to be a single segment if the destination isn't local.
2090 * While looking at the routing entry, we also initialize other path-dependent
2091 * parameters from pre-set or cached values in the routing entry.
2092 *
2093 * Also take into account the space needed for options that we
2094 * send regularly. Make maxseg shorter by that amount to assure
2095 * that we can send maxseg amount of data even when the options
2096 * are present. Store the upper limit of the length of options plus
2097 * data in maxopd.
2098 *
2099 * NOTE that this routine is only called when we process an incoming
2100 * segment, for outgoing segments only tcp_mssopt is called.
2101 *
2102 * In case of T/TCP, we call this routine during implicit connection
2103 * setup as well (offer = -1), to initialize maxseg from the cached
2104 * MSS of our peer.
2105 */
2106 void
2107 tcp_mss(tp, offer)
2108 struct tcpcb *tp;
2109 int offer;
2110 {
2111 register struct rtentry *rt;
2112 struct ifnet *ifp;
2113 register int rtt, mss;
2114 u_long bufsize;
2115 struct inpcb *inp;
2116 struct socket *so;
2117 struct rmxp_tao *taop;
2118 int origoffer = offer;
2119
2120 inp = tp->t_inpcb;
2121 if ((rt = tcp_rtlookup(inp)) == NULL) {
2122 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt;
2123 return;
2124 }
2125 ifp = rt->rt_ifp;
2126 so = inp->inp_socket;
2127
2128 taop = rmx_taop(rt->rt_rmx);
2129 /*
2130 * Offer == -1 means that we didn't receive SYN yet,
2131 * use cached value in that case;
2132 */
2133 if (offer == -1)
2134 offer = taop->tao_mssopt;
2135 /*
2136 * Offer == 0 means that there was no MSS on the SYN segment,
2137 * in this case we use tcp_mssdflt.
2138 */
2139 if (offer == 0)
2140 offer = tcp_mssdflt;
2141 else
2142 /*
2143 * Sanity check: make sure that maxopd will be large
2144 * enough to allow some data on segments even is the
2145 * all the option space is used (40bytes). Otherwise
2146 * funny things may happen in tcp_output.
2147 */
2148 offer = max(offer, 64);
2149 taop->tao_mssopt = offer;
2150
2151 /*
2152 * While we're here, check if there's an initial rtt
2153 * or rttvar. Convert from the route-table units
2154 * to scaled multiples of the slow timeout timer.
2155 */
2156 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
2157 /*
2158 * XXX the lock bit for RTT indicates that the value
2159 * is also a minimum value; this is subject to time.
2160 */
2161 if (rt->rt_rmx.rmx_locks & RTV_RTT)
2162 tp->t_rttmin = rtt / (RTM_RTTUNIT / PR_SLOWHZ);
2163 tp->t_srtt = rtt / (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTT_SCALE));
2164 tcpstat.tcps_usedrtt++;
2165 if (rt->rt_rmx.rmx_rttvar) {
2166 tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
2167 (RTM_RTTUNIT / (PR_SLOWHZ * TCP_RTTVAR_SCALE));
2168 tcpstat.tcps_usedrttvar++;
2169 } else {
2170 /* default variation is +- 1 rtt */
2171 tp->t_rttvar =
2172 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
2173 }
2174 TCPT_RANGESET(tp->t_rxtcur,
2175 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
2176 tp->t_rttmin, TCPTV_REXMTMAX);
2177 }
2178 /*
2179 * if there's an mtu associated with the route, use it
2180 */
2181 if (rt->rt_rmx.rmx_mtu)
2182 mss = rt->rt_rmx.rmx_mtu - sizeof(struct tcpiphdr);
2183 else
2184 {
2185 mss = ifp->if_mtu - sizeof(struct tcpiphdr);
2186 if (!in_localaddr(inp->inp_faddr))
2187 mss = min(mss, tcp_mssdflt);
2188 }
2189 mss = min(mss, offer);
2190 /*
2191 * maxopd stores the maximum length of data AND options
2192 * in a segment; maxseg is the amount of data in a normal
2193 * segment. We need to store this value (maxopd) apart
2194 * from maxseg, because now every segment carries options
2195 * and thus we normally have somewhat less data in segments.
2196 */
2197 tp->t_maxopd = mss;
2198
2199 /*
2200 * In case of T/TCP, origoffer==-1 indicates, that no segments
2201 * were received yet. In this case we just guess, otherwise
2202 * we do the same as before T/TCP.
2203 */
2204 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
2205 (origoffer == -1 ||
2206 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
2207 mss -= TCPOLEN_TSTAMP_APPA;
2208 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC &&
2209 (origoffer == -1 ||
2210 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC))
2211 mss -= TCPOLEN_CC_APPA;
2212
2213 #if (MCLBYTES & (MCLBYTES - 1)) == 0
2214 if (mss > MCLBYTES)
2215 mss &= ~(MCLBYTES-1);
2216 #else
2217 if (mss > MCLBYTES)
2218 mss = mss / MCLBYTES * MCLBYTES;
2219 #endif
2220 /*
2221 * If there's a pipesize, change the socket buffer
2222 * to that size. Make the socket buffers an integral
2223 * number of mss units; if the mss is larger than
2224 * the socket buffer, decrease the mss.
2225 */
2226 #ifdef RTV_SPIPE
2227 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0)
2228 #endif
2229 bufsize = so->so_snd.sb_hiwat;
2230 if (bufsize < mss)
2231 mss = bufsize;
2232 else {
2233 bufsize = roundup(bufsize, mss);
2234 if (bufsize > sb_max)
2235 bufsize = sb_max;
2236 (void)sbreserve(&so->so_snd, bufsize);
2237 }
2238 tp->t_maxseg = mss;
2239
2240 #ifdef RTV_RPIPE
2241 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0)
2242 #endif
2243 bufsize = so->so_rcv.sb_hiwat;
2244 if (bufsize > mss) {
2245 bufsize = roundup(bufsize, mss);
2246 if (bufsize > sb_max)
2247 bufsize = sb_max;
2248 (void)sbreserve(&so->so_rcv, bufsize);
2249 }
2250 /*
2251 * Don't force slow-start on local network.
2252 */
2253 if (!in_localaddr(inp->inp_faddr))
2254 tp->snd_cwnd = mss;
2255
2256 if (rt->rt_rmx.rmx_ssthresh) {
2257 /*
2258 * There's some sort of gateway or interface
2259 * buffer limit on the path. Use this to set
2260 * the slow start threshhold, but set the
2261 * threshold to no less than 2*mss.
2262 */
2263 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
2264 tcpstat.tcps_usedssthresh++;
2265 }
2266 }
2267
2268 /*
2269 * Determine the MSS option to send on an outgoing SYN.
2270 */
2271 int
2272 tcp_mssopt(tp)
2273 struct tcpcb *tp;
2274 {
2275 struct rtentry *rt;
2276
2277 rt = tcp_rtlookup(tp->t_inpcb);
2278 if (rt == NULL)
2279 return tcp_mssdflt;
2280
2281 return rt->rt_ifp->if_mtu - sizeof(struct tcpiphdr);
2282 }
Cache object: 6bd523d143759947cefff63027f9777c
|