1 /*
2 * Copyright (c) 2002, 2003, 2004 Jeffrey M. Hsu. All rights reserved.
3 * Copyright (c) 2002, 2003, 2004 The DragonFly Project. All rights reserved.
4 *
5 * This code is derived from software contributed to The DragonFly Project
6 * by Jeffrey M. Hsu.
7 *
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
10 * are met:
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.
19 *
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
31 * SUCH DAMAGE.
32 */
33
34 /*
35 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
36 * The Regents of the University of California. All rights reserved.
37 *
38 * Redistribution and use in source and binary forms, with or without
39 * modification, are permitted provided that the following conditions
40 * are met:
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. Neither the name of the University nor the names of its contributors
47 * may be used to endorse or promote products derived from this software
48 * without specific prior written permission.
49 *
50 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
51 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
52 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
53 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
54 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
55 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
56 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
57 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
58 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
59 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
60 * SUCH DAMAGE.
61 *
62 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95
63 * $FreeBSD: src/sys/netinet/tcp_input.c,v 1.107.2.38 2003/05/21 04:46:41 cjc Exp $
64 */
65
66 #include "opt_inet.h"
67 #include "opt_inet6.h"
68 #include "opt_tcpdebug.h"
69 #include "opt_tcp_input.h"
70
71 #include <sys/param.h>
72 #include <sys/systm.h>
73 #include <sys/kernel.h>
74 #include <sys/sysctl.h>
75 #include <sys/malloc.h>
76 #include <sys/mbuf.h>
77 #include <sys/proc.h> /* for proc0 declaration */
78 #include <sys/protosw.h>
79 #include <sys/socket.h>
80 #include <sys/socketvar.h>
81 #include <sys/syslog.h>
82 #include <sys/in_cksum.h>
83
84 #include <sys/socketvar2.h>
85
86 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
87 #include <machine/stdarg.h>
88
89 #include <net/if.h>
90 #include <net/route.h>
91
92 #include <netinet/in.h>
93 #include <netinet/in_systm.h>
94 #include <netinet/ip.h>
95 #include <netinet/ip_icmp.h> /* for ICMP_BANDLIM */
96 #include <netinet/in_var.h>
97 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */
98 #include <netinet/in_pcb.h>
99 #include <netinet/ip_var.h>
100 #include <netinet/ip6.h>
101 #include <netinet/icmp6.h>
102 #include <netinet6/nd6.h>
103 #include <netinet6/ip6_var.h>
104 #include <netinet6/in6_pcb.h>
105 #include <netinet/tcp.h>
106 #include <netinet/tcp_fsm.h>
107 #include <netinet/tcp_seq.h>
108 #include <netinet/tcp_timer.h>
109 #include <netinet/tcp_timer2.h>
110 #include <netinet/tcp_var.h>
111 #include <netinet6/tcp6_var.h>
112 #include <netinet/tcpip.h>
113
114 #ifdef TCPDEBUG
115 #include <netinet/tcp_debug.h>
116
117 u_char tcp_saveipgen[40]; /* the size must be of max ip header, now IPv6 */
118 struct tcphdr tcp_savetcp;
119 #endif
120
121 /*
122 * Limit burst of new packets during SACK based fast recovery
123 * or extended limited transmit.
124 */
125 #define TCP_SACK_MAXBURST 4
126
127 MALLOC_DEFINE(M_TSEGQ, "tseg_qent", "TCP segment queue entry");
128
129 static int log_in_vain = 0;
130 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
131 &log_in_vain, 0, "Log all incoming TCP connections");
132
133 static int blackhole = 0;
134 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW,
135 &blackhole, 0, "Do not send RST when dropping refused connections");
136
137 int tcp_delack_enabled = 1;
138 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW,
139 &tcp_delack_enabled, 0,
140 "Delay ACK to try to piggyback it onto a data packet");
141
142 #ifdef TCP_DROP_SYNFIN
143 static int drop_synfin = 0;
144 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW,
145 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set");
146 #endif
147
148 static int tcp_do_limitedtransmit = 1;
149 SYSCTL_INT(_net_inet_tcp, OID_AUTO, limitedtransmit, CTLFLAG_RW,
150 &tcp_do_limitedtransmit, 0, "Enable RFC 3042 (Limited Transmit)");
151
152 static int tcp_do_early_retransmit = 1;
153 SYSCTL_INT(_net_inet_tcp, OID_AUTO, earlyretransmit, CTLFLAG_RW,
154 &tcp_do_early_retransmit, 0, "Early retransmit");
155
156 int tcp_aggregate_acks = 1;
157 SYSCTL_INT(_net_inet_tcp, OID_AUTO, aggregate_acks, CTLFLAG_RW,
158 &tcp_aggregate_acks, 0, "Aggregate built-up acks into one ack");
159
160 static int tcp_do_eifel_detect = 1;
161 SYSCTL_INT(_net_inet_tcp, OID_AUTO, eifel, CTLFLAG_RW,
162 &tcp_do_eifel_detect, 0, "Eifel detection algorithm (RFC 3522)");
163
164 static int tcp_do_abc = 1;
165 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc, CTLFLAG_RW,
166 &tcp_do_abc, 0,
167 "TCP Appropriate Byte Counting (RFC 3465)");
168
169 /*
170 * The following value actually takes range [25ms, 250ms],
171 * given that most modern systems use 1ms ~ 10ms as the unit
172 * of timestamp option.
173 */
174 static u_int tcp_paws_tolerance = 25;
175 SYSCTL_UINT(_net_inet_tcp, OID_AUTO, paws_tolerance, CTLFLAG_RW,
176 &tcp_paws_tolerance, 0, "RFC1323 PAWS tolerance");
177
178 /*
179 * Define as tunable for easy testing with SACK on and off.
180 * Warning: do not change setting in the middle of an existing active TCP flow,
181 * else strange things might happen to that flow.
182 */
183 int tcp_do_sack = 1;
184 SYSCTL_INT(_net_inet_tcp, OID_AUTO, sack, CTLFLAG_RW,
185 &tcp_do_sack, 0, "Enable SACK Algorithms");
186
187 int tcp_do_smartsack = 1;
188 SYSCTL_INT(_net_inet_tcp, OID_AUTO, smartsack, CTLFLAG_RW,
189 &tcp_do_smartsack, 0, "Enable Smart SACK Algorithms");
190
191 int tcp_do_rescuesack = 1;
192 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rescuesack, CTLFLAG_RW,
193 &tcp_do_rescuesack, 0, "Rescue retransmission for SACK");
194
195 int tcp_aggressive_rescuesack = 0;
196 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rescuesack_agg, CTLFLAG_RW,
197 &tcp_aggressive_rescuesack, 0, "Aggressive rescue retransmission for SACK");
198
199 static int tcp_force_sackrxt = 1;
200 SYSCTL_INT(_net_inet_tcp, OID_AUTO, force_sackrxt, CTLFLAG_RW,
201 &tcp_force_sackrxt, 0, "Allowed forced SACK retransmit burst");
202
203 int tcp_do_rfc6675 = 1;
204 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc6675, CTLFLAG_RW,
205 &tcp_do_rfc6675, 0, "Enable RFC6675");
206
207 int tcp_rfc6675_rxt = 0;
208 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc6675_rxt, CTLFLAG_RW,
209 &tcp_rfc6675_rxt, 0, "Enable RFC6675 retransmit");
210
211 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, reass, CTLFLAG_RW, 0,
212 "TCP Segment Reassembly Queue");
213
214 int tcp_reass_maxseg = 0;
215 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, maxsegments, CTLFLAG_RD,
216 &tcp_reass_maxseg, 0,
217 "Global maximum number of TCP Segments in Reassembly Queue");
218
219 int tcp_reass_qsize = 0;
220 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, cursegments, CTLFLAG_RD,
221 &tcp_reass_qsize, 0,
222 "Global number of TCP Segments currently in Reassembly Queue");
223
224 static int tcp_reass_overflows = 0;
225 SYSCTL_INT(_net_inet_tcp_reass, OID_AUTO, overflows, CTLFLAG_RD,
226 &tcp_reass_overflows, 0,
227 "Global number of TCP Segment Reassembly Queue Overflows");
228
229 int tcp_do_autorcvbuf = 1;
230 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_RW,
231 &tcp_do_autorcvbuf, 0, "Enable automatic receive buffer sizing");
232
233 int tcp_autorcvbuf_inc = 16*1024;
234 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_RW,
235 &tcp_autorcvbuf_inc, 0,
236 "Incrementor step size of automatic receive buffer");
237
238 int tcp_autorcvbuf_max = 2*1024*1024;
239 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_RW,
240 &tcp_autorcvbuf_max, 0, "Max size of automatic receive buffer");
241
242 int tcp_sosend_agglim = 2;
243 SYSCTL_INT(_net_inet_tcp, OID_AUTO, sosend_agglim, CTLFLAG_RW,
244 &tcp_sosend_agglim, 0, "TCP sosend mbuf aggregation limit");
245
246 int tcp_sosend_async = 1;
247 SYSCTL_INT(_net_inet_tcp, OID_AUTO, sosend_async, CTLFLAG_RW,
248 &tcp_sosend_async, 0, "TCP asynchronized pru_send");
249
250 int tcp_sosend_jcluster = 1;
251 SYSCTL_INT(_net_inet_tcp, OID_AUTO, sosend_jcluster, CTLFLAG_RW,
252 &tcp_sosend_jcluster, 0, "TCP output uses jcluster");
253
254 static int tcp_ignore_redun_dsack = 1;
255 SYSCTL_INT(_net_inet_tcp, OID_AUTO, ignore_redun_dsack, CTLFLAG_RW,
256 &tcp_ignore_redun_dsack, 0, "Ignore redundant DSACK");
257
258 static int tcp_reuseport_ext = 1;
259 SYSCTL_INT(_net_inet_tcp, OID_AUTO, reuseport_ext, CTLFLAG_RW,
260 &tcp_reuseport_ext, 0, "SO_REUSEPORT extension");
261
262 static void tcp_dooptions(struct tcpopt *, u_char *, int, boolean_t,
263 tcp_seq);
264 static void tcp_pulloutofband(struct socket *,
265 struct tcphdr *, struct mbuf *, int);
266 static int tcp_reass(struct tcpcb *, struct tcphdr *, int *,
267 struct mbuf *);
268 static void tcp_xmit_timer(struct tcpcb *, int, tcp_seq);
269 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *, int);
270 static void tcp_sack_rexmt(struct tcpcb *, boolean_t);
271 static boolean_t tcp_sack_limitedxmit(struct tcpcb *);
272 static int tcp_rmx_msl(const struct tcpcb *);
273 static void tcp_established(struct tcpcb *);
274 static boolean_t tcp_recv_dupack(struct tcpcb *, tcp_seq, u_int);
275
276 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
277 #ifdef INET6
278 #define ND6_HINT(tp) \
279 do { \
280 if ((tp) && (tp)->t_inpcb && \
281 INP_ISIPV6((tp)->t_inpcb) && \
282 (tp)->t_inpcb->in6p_route.ro_rt) \
283 nd6_nud_hint((tp)->t_inpcb->in6p_route.ro_rt, NULL, 0); \
284 } while (0)
285 #else
286 #define ND6_HINT(tp)
287 #endif
288
289 /*
290 * Indicate whether this ack should be delayed. We can delay the ack if
291 * - delayed acks are enabled and
292 * - there is no delayed ack timer in progress and
293 * - our last ack wasn't a 0-sized window. We never want to delay
294 * the ack that opens up a 0-sized window.
295 */
296 #define DELAY_ACK(tp) \
297 (tcp_delack_enabled && !tcp_callout_pending(tp, tp->tt_delack) && \
298 !(tp->t_flags & TF_RXWIN0SENT))
299
300 #define acceptable_window_update(tp, th, tiwin) \
301 (SEQ_LT(tp->snd_wl1, th->th_seq) || \
302 (tp->snd_wl1 == th->th_seq && \
303 (SEQ_LT(tp->snd_wl2, th->th_ack) || \
304 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))
305
306 #define iceildiv(n, d) (((n)+(d)-1) / (d))
307 #define need_early_retransmit(tp, ownd) \
308 (tcp_do_early_retransmit && \
309 (tcp_do_eifel_detect && (tp->t_flags & TF_RCVD_TSTMP)) && \
310 ownd < ((tp->t_rxtthresh + 1) * tp->t_maxseg) && \
311 tp->t_dupacks + 1 >= iceildiv(ownd, tp->t_maxseg) && \
312 (!TCP_DO_SACK(tp) || ownd <= tp->t_maxseg || \
313 tcp_sack_has_sacked(&tp->scb, ownd - tp->t_maxseg)))
314
315 /*
316 * Returns TRUE, if this segment can be merged with the last
317 * pending segment in the reassemble queue and this segment
318 * does not overlap with the pending segment immediately
319 * preceeding the last pending segment.
320 */
321 static __inline boolean_t
tcp_paws_canreasslast(const struct tcpcb * tp,const struct tcphdr * th,int tlen)322 tcp_paws_canreasslast(const struct tcpcb *tp, const struct tcphdr *th, int tlen)
323 {
324 const struct tseg_qent *last, *prev;
325
326 last = TAILQ_LAST(&tp->t_segq, tsegqe_head);
327 if (last == NULL)
328 return FALSE;
329
330 /* This segment comes immediately after the last pending segment */
331 if (last->tqe_th->th_seq + last->tqe_len == th->th_seq) {
332 if (last->tqe_th->th_flags & TH_FIN) {
333 /* No segments should follow segment w/ FIN */
334 return FALSE;
335 }
336 return TRUE;
337 }
338
339 if (th->th_seq + tlen != last->tqe_th->th_seq)
340 return FALSE;
341 /* This segment comes immediately before the last pending segment */
342
343 prev = TAILQ_PREV(last, tsegqe_head, tqe_q);
344 if (prev == NULL) {
345 /*
346 * No pending preceeding segment, we assume this segment
347 * could be reassembled.
348 */
349 return TRUE;
350 }
351
352 /* This segment does not overlap with the preceeding segment */
353 if (SEQ_GEQ(th->th_seq, prev->tqe_th->th_seq + prev->tqe_len))
354 return TRUE;
355
356 return FALSE;
357 }
358
359 static __inline void
tcp_ncr_update_rxtthresh(struct tcpcb * tp)360 tcp_ncr_update_rxtthresh(struct tcpcb *tp)
361 {
362 int old_rxtthresh = tp->t_rxtthresh;
363 uint32_t ownd = tp->snd_max - tp->snd_una;
364
365 tp->t_rxtthresh = min(tcp_ncr_rxtthresh_max,
366 max(tcprexmtthresh, ((ownd / tp->t_maxseg) >> 1)));
367 if (tp->t_rxtthresh != old_rxtthresh) {
368 tcp_sack_update_lostseq(&tp->scb, tp->snd_una,
369 tp->t_maxseg, tp->t_rxtthresh);
370 }
371 }
372
373 static int
tcp_reass(struct tcpcb * tp,struct tcphdr * th,int * tlenp,struct mbuf * m)374 tcp_reass(struct tcpcb *tp, struct tcphdr *th, int *tlenp, struct mbuf *m)
375 {
376 struct tseg_qent *q;
377 struct tseg_qent *p = NULL;
378 struct tseg_qent *te;
379 struct socket *so = tp->t_inpcb->inp_socket;
380 int flags;
381
382 /*
383 * Call with th == NULL after become established to
384 * force pre-ESTABLISHED data up to user socket.
385 */
386 if (th == NULL)
387 goto present;
388
389 /*
390 * Limit the number of segments in the reassembly queue to prevent
391 * holding on to too many segments (and thus running out of mbufs).
392 * Make sure to let the missing segment through which caused this
393 * queue. Always keep one global queue entry spare to be able to
394 * process the missing segment.
395 */
396 if (th->th_seq != tp->rcv_nxt &&
397 tcp_reass_qsize + 1 >= tcp_reass_maxseg) {
398 tcp_reass_overflows++;
399 tcpstat.tcps_rcvmemdrop++;
400 m_freem(m);
401 /* no SACK block to report */
402 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
403 return (0);
404 }
405
406 /* Allocate a new queue entry. */
407 te = kmalloc(sizeof(struct tseg_qent), M_TSEGQ, M_INTWAIT | M_NULLOK);
408 if (te == NULL) {
409 tcpstat.tcps_rcvmemdrop++;
410 m_freem(m);
411 /* no SACK block to report */
412 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
413 return (0);
414 }
415 atomic_add_int(&tcp_reass_qsize, 1);
416
417 if (th->th_flags & TH_FIN)
418 tp->t_flags |= TF_QUEDFIN;
419
420 /*
421 * Find a segment which begins after this one does.
422 */
423 TAILQ_FOREACH(q, &tp->t_segq, tqe_q) {
424 if (SEQ_GT(q->tqe_th->th_seq, th->th_seq))
425 break;
426 p = q;
427 }
428
429 /*
430 * If there is a preceding segment, it may provide some of
431 * our data already. If so, drop the data from the incoming
432 * segment. If it provides all of our data, drop us.
433 */
434 if (p != NULL) {
435 tcp_seq_diff_t i;
436
437 /* conversion to int (in i) handles seq wraparound */
438 i = p->tqe_th->th_seq + p->tqe_len - th->th_seq;
439 if (i > 0) { /* overlaps preceding segment */
440 tp->sack_flags |=
441 (TSACK_F_DUPSEG | TSACK_F_ENCLOSESEG);
442 /* enclosing block starts w/ preceding segment */
443 tp->encloseblk.rblk_start = p->tqe_th->th_seq;
444 if (i >= *tlenp) {
445 if (th->th_flags & TH_FIN)
446 p->tqe_th->th_flags |= TH_FIN;
447
448 /* preceding encloses incoming segment */
449 tp->encloseblk.rblk_end = TCP_SACK_BLKEND(
450 p->tqe_th->th_seq + p->tqe_len,
451 p->tqe_th->th_flags);
452 tcpstat.tcps_rcvduppack++;
453 tcpstat.tcps_rcvdupbyte += *tlenp;
454 m_freem(m);
455 kfree(te, M_TSEGQ);
456 atomic_add_int(&tcp_reass_qsize, -1);
457 /*
458 * Try to present any queued data
459 * at the left window edge to the user.
460 * This is needed after the 3-WHS
461 * completes.
462 */
463 goto present; /* ??? */
464 }
465 m_adj(m, i);
466 *tlenp -= i;
467 th->th_seq += i;
468 /* incoming segment end is enclosing block end */
469 tp->encloseblk.rblk_end = TCP_SACK_BLKEND(
470 th->th_seq + *tlenp, th->th_flags);
471 /* trim end of reported D-SACK block */
472 tp->reportblk.rblk_end = th->th_seq;
473 }
474 }
475 tcpstat.tcps_rcvoopack++;
476 tcpstat.tcps_rcvoobyte += *tlenp;
477
478 /*
479 * While we overlap succeeding segments trim them or,
480 * if they are completely covered, dequeue them.
481 */
482 while (q) {
483 tcp_seq_diff_t i = (th->th_seq + *tlenp) - q->tqe_th->th_seq;
484 tcp_seq qend = q->tqe_th->th_seq + q->tqe_len;
485 tcp_seq qend_sack = TCP_SACK_BLKEND(qend, q->tqe_th->th_flags);
486 struct tseg_qent *nq;
487
488 if (i <= 0)
489 break;
490 if (!(tp->sack_flags & TSACK_F_DUPSEG)) {
491 /* first time through */
492 tp->sack_flags |= (TSACK_F_DUPSEG | TSACK_F_ENCLOSESEG);
493 tp->encloseblk = tp->reportblk;
494 /* report trailing duplicate D-SACK segment */
495 tp->reportblk.rblk_start = q->tqe_th->th_seq;
496 }
497 if ((tp->sack_flags & TSACK_F_ENCLOSESEG) &&
498 SEQ_GT(qend_sack, tp->encloseblk.rblk_end)) {
499 /* extend enclosing block if one exists */
500 tp->encloseblk.rblk_end = qend_sack;
501 }
502 if (i < q->tqe_len) {
503 q->tqe_th->th_seq += i;
504 q->tqe_len -= i;
505 m_adj(q->tqe_m, i);
506 break;
507 }
508
509 if (q->tqe_th->th_flags & TH_FIN)
510 th->th_flags |= TH_FIN;
511
512 nq = TAILQ_NEXT(q, tqe_q);
513 TAILQ_REMOVE(&tp->t_segq, q, tqe_q);
514 m_freem(q->tqe_m);
515 kfree(q, M_TSEGQ);
516 atomic_add_int(&tcp_reass_qsize, -1);
517 q = nq;
518 }
519
520 /* Insert the new segment queue entry into place. */
521 te->tqe_m = m;
522 te->tqe_th = th;
523 te->tqe_len = *tlenp;
524
525 /* check if can coalesce with following segment */
526 if (q != NULL && (th->th_seq + *tlenp == q->tqe_th->th_seq)) {
527 tcp_seq tend_sack;
528
529 te->tqe_len += q->tqe_len;
530 if (q->tqe_th->th_flags & TH_FIN)
531 te->tqe_th->th_flags |= TH_FIN;
532 tend_sack = TCP_SACK_BLKEND(te->tqe_th->th_seq + te->tqe_len,
533 te->tqe_th->th_flags);
534
535 m_cat(te->tqe_m, q->tqe_m);
536 tp->encloseblk.rblk_end = tend_sack;
537 /*
538 * When not reporting a duplicate segment, use
539 * the larger enclosing block as the SACK block.
540 */
541 if (!(tp->sack_flags & TSACK_F_DUPSEG))
542 tp->reportblk.rblk_end = tend_sack;
543 TAILQ_REMOVE(&tp->t_segq, q, tqe_q);
544 kfree(q, M_TSEGQ);
545 atomic_add_int(&tcp_reass_qsize, -1);
546 }
547
548 if (p == NULL) {
549 TAILQ_INSERT_HEAD(&tp->t_segq, te, tqe_q);
550 } else {
551 /* check if can coalesce with preceding segment */
552 if (p->tqe_th->th_seq + p->tqe_len == th->th_seq) {
553 if (te->tqe_th->th_flags & TH_FIN)
554 p->tqe_th->th_flags |= TH_FIN;
555 p->tqe_len += te->tqe_len;
556 m_cat(p->tqe_m, te->tqe_m);
557 tp->encloseblk.rblk_start = p->tqe_th->th_seq;
558 /*
559 * When not reporting a duplicate segment, use
560 * the larger enclosing block as the SACK block.
561 */
562 if (!(tp->sack_flags & TSACK_F_DUPSEG))
563 tp->reportblk.rblk_start = p->tqe_th->th_seq;
564 kfree(te, M_TSEGQ);
565 atomic_add_int(&tcp_reass_qsize, -1);
566 } else {
567 TAILQ_INSERT_AFTER(&tp->t_segq, p, te, tqe_q);
568 }
569 }
570
571 present:
572 /*
573 * Present data to user, advancing rcv_nxt through
574 * completed sequence space.
575 */
576 if (!TCPS_HAVEESTABLISHED(tp->t_state))
577 return (0);
578 q = TAILQ_FIRST(&tp->t_segq);
579 if (q == NULL || q->tqe_th->th_seq != tp->rcv_nxt)
580 return (0);
581 tp->rcv_nxt += q->tqe_len;
582 if (!(tp->sack_flags & TSACK_F_DUPSEG)) {
583 /* no SACK block to report since ACK advanced */
584 tp->reportblk.rblk_start = tp->reportblk.rblk_end;
585 }
586 /* no enclosing block to report since ACK advanced */
587 tp->sack_flags &= ~TSACK_F_ENCLOSESEG;
588 flags = q->tqe_th->th_flags & TH_FIN;
589 TAILQ_REMOVE(&tp->t_segq, q, tqe_q);
590 KASSERT(TAILQ_EMPTY(&tp->t_segq) ||
591 TAILQ_FIRST(&tp->t_segq)->tqe_th->th_seq != tp->rcv_nxt,
592 ("segment not coalesced"));
593 if (so->so_state & SS_CANTRCVMORE) {
594 m_freem(q->tqe_m);
595 } else {
596 lwkt_gettoken(&so->so_rcv.ssb_token);
597 ssb_appendstream(&so->so_rcv, q->tqe_m);
598 lwkt_reltoken(&so->so_rcv.ssb_token);
599 }
600 kfree(q, M_TSEGQ);
601 atomic_add_int(&tcp_reass_qsize, -1);
602 ND6_HINT(tp);
603 sorwakeup(so);
604 return (flags);
605 }
606
607 /*
608 * TCP input routine, follows pages 65-76 of the
609 * protocol specification dated September, 1981 very closely.
610 */
611 #ifdef INET6
612 int
tcp6_input(struct mbuf ** mp,int * offp,int proto)613 tcp6_input(struct mbuf **mp, int *offp, int proto)
614 {
615 struct mbuf *m = *mp;
616 struct in6_ifaddr *ia6;
617
618 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
619
620 /*
621 * draft-itojun-ipv6-tcp-to-anycast
622 * better place to put this in?
623 */
624 ia6 = ip6_getdstifaddr(m);
625 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
626 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
627 offsetof(struct ip6_hdr, ip6_dst));
628 return (IPPROTO_DONE);
629 }
630
631 tcp_input(mp, offp, proto);
632 return (IPPROTO_DONE);
633 }
634 #endif
635
636 int
tcp_input(struct mbuf ** mp,int * offp,int proto)637 tcp_input(struct mbuf **mp, int *offp, int proto)
638 {
639 int off0;
640 struct tcphdr *th;
641 struct ip *ip = NULL;
642 struct ipovly *ipov;
643 struct inpcb *inp = NULL;
644 u_char *optp = NULL;
645 int optlen = 0;
646 int tlen, off;
647 int len = 0;
648 int drop_hdrlen;
649 struct tcpcb *tp = NULL;
650 int thflags;
651 struct socket *so = NULL;
652 int todrop, acked;
653 boolean_t ourfinisacked, needoutput = FALSE, delayed_dupack = FALSE;
654 tcp_seq th_dupack = 0; /* XXX gcc warning */
655 u_int to_flags = 0; /* XXX gcc warning */
656 u_long tiwin;
657 int recvwin;
658 struct tcpopt to; /* options in this segment */
659 struct sockaddr_in *next_hop = NULL;
660 int rstreason; /* For badport_bandlim accounting purposes */
661 int cpu;
662 struct ip6_hdr *ip6 = NULL;
663 struct mbuf *m;
664 #ifdef INET6
665 boolean_t isipv6;
666 #else
667 const boolean_t isipv6 = FALSE;
668 #endif
669 #ifdef TCPDEBUG
670 short ostate = 0;
671 #endif
672
673 off0 = *offp;
674 m = *mp;
675 *mp = NULL;
676
677 tcpstat.tcps_rcvtotal++;
678
679 if (m->m_pkthdr.fw_flags & IPFORWARD_MBUF_TAGGED) {
680 struct m_tag *mtag;
681
682 mtag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
683 KKASSERT(mtag != NULL);
684 next_hop = m_tag_data(mtag);
685 }
686
687 #ifdef INET6
688 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? TRUE : FALSE;
689 #endif
690
691 if (isipv6) {
692 /* IP6_EXTHDR_CHECK() is already done at tcp6_input() */
693 ip6 = mtod(m, struct ip6_hdr *);
694 tlen = (sizeof *ip6) + ntohs(ip6->ip6_plen) - off0;
695 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) {
696 tcpstat.tcps_rcvbadsum++;
697 goto drop;
698 }
699 th = (struct tcphdr *)((caddr_t)ip6 + off0);
700
701 /*
702 * Be proactive about unspecified IPv6 address in source.
703 * As we use all-zero to indicate unbounded/unconnected pcb,
704 * unspecified IPv6 address can be used to confuse us.
705 *
706 * Note that packets with unspecified IPv6 destination is
707 * already dropped in ip6_input.
708 */
709 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
710 /* XXX stat */
711 goto drop;
712 }
713 } else {
714 /*
715 * Get IP and TCP header together in first mbuf.
716 * Note: IP leaves IP header in first mbuf.
717 */
718 if (off0 > sizeof(struct ip)) {
719 ip_stripoptions(m);
720 off0 = sizeof(struct ip);
721 }
722 /* already checked and pulled up in ip_demux() */
723 KASSERT(m->m_len >= sizeof(struct tcpiphdr),
724 ("TCP header not in one mbuf: m->m_len %d", m->m_len));
725 ip = mtod(m, struct ip *);
726 ipov = (struct ipovly *)ip;
727 th = (struct tcphdr *)((caddr_t)ip + off0);
728 tlen = ntohs(ip->ip_len) - off0;
729
730 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
731 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
732 th->th_sum = m->m_pkthdr.csum_data;
733 else
734 th->th_sum = in_pseudo(ip->ip_src.s_addr,
735 ip->ip_dst.s_addr,
736 htonl(m->m_pkthdr.csum_data +
737 ntohs(ip->ip_len) +
738 IPPROTO_TCP));
739 th->th_sum ^= 0xffff;
740 } else {
741 /*
742 * Checksum extended TCP header and data.
743 */
744 len = sizeof(struct ip) + tlen;
745 bzero(ipov->ih_x1, sizeof ipov->ih_x1);
746 ipov->ih_len = (u_short)tlen;
747 ipov->ih_len = htons(ipov->ih_len);
748 th->th_sum = in_cksum(m, len);
749 }
750 if (th->th_sum) {
751 tcpstat.tcps_rcvbadsum++;
752 goto drop;
753 }
754 #ifdef INET6
755 /* Re-initialization for later version check */
756 ip->ip_v = IPVERSION;
757 #endif
758 }
759
760 /*
761 * Check that TCP offset makes sense,
762 * pull out TCP options and adjust length. XXX
763 */
764 off = th->th_off << 2;
765 /* already checked and pulled up in ip_demux() */
766 KASSERT(off >= sizeof(struct tcphdr) && off <= tlen,
767 ("bad TCP data offset %d (tlen %d)", off, tlen));
768 tlen -= off; /* tlen is used instead of ti->ti_len */
769 if (off > sizeof(struct tcphdr)) {
770 if (isipv6) {
771 IP6_EXTHDR_CHECK(m, off0, off, IPPROTO_DONE);
772 ip6 = mtod(m, struct ip6_hdr *);
773 th = (struct tcphdr *)((caddr_t)ip6 + off0);
774 } else {
775 /* already pulled up in ip_demux() */
776 KASSERT(m->m_len >= sizeof(struct ip) + off,
777 ("TCP header and options not in one mbuf: "
778 "m_len %d, off %d", m->m_len, off));
779 }
780 optlen = off - sizeof(struct tcphdr);
781 optp = (u_char *)(th + 1);
782 }
783 thflags = th->th_flags;
784
785 #ifdef TCP_DROP_SYNFIN
786 /*
787 * If the drop_synfin option is enabled, drop all packets with
788 * both the SYN and FIN bits set. This prevents e.g. nmap from
789 * identifying the TCP/IP stack.
790 *
791 * This is a violation of the TCP specification.
792 */
793 if (drop_synfin && (thflags & (TH_SYN | TH_FIN)) == (TH_SYN | TH_FIN))
794 goto drop;
795 #endif
796
797 /*
798 * Convert TCP protocol specific fields to host format.
799 */
800 th->th_seq = ntohl(th->th_seq);
801 th->th_ack = ntohl(th->th_ack);
802 th->th_win = ntohs(th->th_win);
803 th->th_urp = ntohs(th->th_urp);
804
805 /*
806 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options,
807 * until after ip6_savecontrol() is called and before other functions
808 * which don't want those proto headers.
809 * Because ip6_savecontrol() is going to parse the mbuf to
810 * search for data to be passed up to user-land, it wants mbuf
811 * parameters to be unchanged.
812 * XXX: the call of ip6_savecontrol() has been obsoleted based on
813 * latest version of the advanced API (20020110).
814 */
815 drop_hdrlen = off0 + off;
816
817 /*
818 * Locate pcb for segment.
819 */
820 findpcb:
821 /* IPFIREWALL_FORWARD section */
822 if (next_hop != NULL && !isipv6) { /* IPv6 support is not there yet */
823 /*
824 * Transparently forwarded. Pretend to be the destination.
825 * already got one like this?
826 */
827 cpu = mycpu->gd_cpuid;
828 inp = in_pcblookup_hash(&tcbinfo[cpu],
829 ip->ip_src, th->th_sport,
830 ip->ip_dst, th->th_dport,
831 0, m->m_pkthdr.rcvif);
832 if (!inp) {
833 /*
834 * It's new. Try to find the ambushing socket.
835 */
836
837 /*
838 * The rest of the ipfw code stores the port in
839 * host order. XXX
840 * (The IP address is still in network order.)
841 */
842 in_port_t dport = next_hop->sin_port ?
843 htons(next_hop->sin_port) :
844 th->th_dport;
845
846 cpu = tcp_addrcpu(ip->ip_src.s_addr, th->th_sport,
847 next_hop->sin_addr.s_addr, dport);
848 inp = in_pcblookup_hash(&tcbinfo[cpu],
849 ip->ip_src, th->th_sport,
850 next_hop->sin_addr, dport,
851 1, m->m_pkthdr.rcvif);
852 }
853 } else {
854 if (isipv6) {
855 inp = in6_pcblookup_hash(&tcbinfo[0],
856 &ip6->ip6_src, th->th_sport,
857 &ip6->ip6_dst, th->th_dport,
858 1, m->m_pkthdr.rcvif);
859 } else {
860 cpu = mycpu->gd_cpuid;
861 inp = in_pcblookup_pkthash(&tcbinfo[cpu],
862 ip->ip_src, th->th_sport,
863 ip->ip_dst, th->th_dport,
864 1, m->m_pkthdr.rcvif,
865 tcp_reuseport_ext ? m : NULL);
866 }
867 }
868
869 /*
870 * If the state is CLOSED (i.e., TCB does not exist) then
871 * all data in the incoming segment is discarded.
872 * If the TCB exists but is in CLOSED state, it is embryonic,
873 * but should either do a listen or a connect soon.
874 */
875 if (inp == NULL) {
876 if (log_in_vain) {
877 #ifdef INET6
878 char dbuf[INET6_ADDRSTRLEN+2], sbuf[INET6_ADDRSTRLEN+2];
879 #else
880 char dbuf[INET_ADDRSTRLEN], sbuf[INET_ADDRSTRLEN];
881 #endif
882 if (isipv6) {
883 strcpy(dbuf, "[");
884 strcat(dbuf, ip6_sprintf(&ip6->ip6_dst));
885 strcat(dbuf, "]");
886 strcpy(sbuf, "[");
887 strcat(sbuf, ip6_sprintf(&ip6->ip6_src));
888 strcat(sbuf, "]");
889 } else {
890 kinet_ntoa(ip->ip_dst, dbuf);
891 kinet_ntoa(ip->ip_src, sbuf);
892 }
893 switch (log_in_vain) {
894 case 1:
895 if (!(thflags & TH_SYN))
896 break;
897 case 2:
898 log(LOG_INFO,
899 "Connection attempt to TCP %s:%d "
900 "from %s:%d flags:0x%02x\n",
901 dbuf, ntohs(th->th_dport), sbuf,
902 ntohs(th->th_sport), thflags);
903 break;
904 default:
905 break;
906 }
907 }
908 if (blackhole) {
909 switch (blackhole) {
910 case 1:
911 if (thflags & TH_SYN)
912 goto drop;
913 break;
914 case 2:
915 goto drop;
916 default:
917 goto drop;
918 }
919 }
920 rstreason = BANDLIM_RST_CLOSEDPORT;
921 goto dropwithreset;
922 }
923
924 /* Check the minimum TTL for socket. */
925 #ifdef INET6
926 if ((isipv6 ? ip6->ip6_hlim : ip->ip_ttl) < inp->inp_ip_minttl)
927 goto drop;
928 #endif
929
930 tp = intotcpcb(inp);
931 KASSERT(tp != NULL, ("tcp_input: tp is NULL"));
932 if (tp->t_state <= TCPS_CLOSED)
933 goto drop;
934
935 so = inp->inp_socket;
936
937 #ifdef TCPDEBUG
938 if (so->so_options & SO_DEBUG) {
939 ostate = tp->t_state;
940 if (isipv6)
941 bcopy(ip6, tcp_saveipgen, sizeof(*ip6));
942 else
943 bcopy(ip, tcp_saveipgen, sizeof(*ip));
944 tcp_savetcp = *th;
945 }
946 #endif
947
948 bzero(&to, sizeof to);
949
950 if (so->so_options & SO_ACCEPTCONN) {
951 struct in_conninfo inc;
952
953 #ifdef INET6
954 inc.inc_isipv6 = (isipv6 == TRUE);
955 #endif
956 if (isipv6) {
957 inc.inc6_faddr = ip6->ip6_src;
958 inc.inc6_laddr = ip6->ip6_dst;
959 inc.inc6_route.ro_rt = NULL; /* XXX */
960 } else {
961 inc.inc_faddr = ip->ip_src;
962 inc.inc_laddr = ip->ip_dst;
963 inc.inc_route.ro_rt = NULL; /* XXX */
964 }
965 inc.inc_fport = th->th_sport;
966 inc.inc_lport = th->th_dport;
967
968 /*
969 * If the state is LISTEN then ignore segment if it contains
970 * a RST. If the segment contains an ACK then it is bad and
971 * send a RST. If it does not contain a SYN then it is not
972 * interesting; drop it.
973 *
974 * If the state is SYN_RECEIVED (syncache) and seg contains
975 * an ACK, but not for our SYN/ACK, send a RST. If the seg
976 * contains a RST, check the sequence number to see if it
977 * is a valid reset segment.
978 */
979 if ((thflags & (TH_RST | TH_ACK | TH_SYN)) != TH_SYN) {
980 if ((thflags & (TH_RST | TH_ACK | TH_SYN)) == TH_ACK) {
981 if (!syncache_expand(&inc, th, &so, m)) {
982 /*
983 * No syncache entry, or ACK was not
984 * for our SYN/ACK. Send a RST.
985 */
986 tcpstat.tcps_badsyn++;
987 rstreason = BANDLIM_RST_OPENPORT;
988 goto dropwithreset;
989 }
990
991 /*
992 * Could not complete 3-way handshake,
993 * connection is being closed down, and
994 * syncache will free mbuf.
995 */
996 if (so == NULL)
997 return(IPPROTO_DONE);
998
999 /*
1000 * We must be in the correct protocol thread
1001 * for this connection.
1002 */
1003 KKASSERT(so->so_port == &curthread->td_msgport);
1004
1005 /*
1006 * Socket is created in state SYN_RECEIVED.
1007 * Continue processing segment.
1008 */
1009 inp = so->so_pcb;
1010 tp = intotcpcb(inp);
1011 /*
1012 * This is what would have happened in
1013 * tcp_output() when the SYN,ACK was sent.
1014 */
1015 tp->snd_up = tp->snd_una;
1016 tp->snd_max = tp->snd_nxt = tp->iss + 1;
1017 tp->last_ack_sent = tp->rcv_nxt;
1018
1019 goto after_listen;
1020 }
1021 if (thflags & TH_RST) {
1022 syncache_chkrst(&inc, th);
1023 goto drop;
1024 }
1025 if (thflags & TH_ACK) {
1026 syncache_badack(&inc);
1027 tcpstat.tcps_badsyn++;
1028 rstreason = BANDLIM_RST_OPENPORT;
1029 goto dropwithreset;
1030 }
1031 goto drop;
1032 }
1033
1034 /*
1035 * Segment's flags are (SYN) or (SYN | FIN).
1036 */
1037 #ifdef INET6
1038 /*
1039 * If deprecated address is forbidden,
1040 * we do not accept SYN to deprecated interface
1041 * address to prevent any new inbound connection from
1042 * getting established.
1043 * When we do not accept SYN, we send a TCP RST,
1044 * with deprecated source address (instead of dropping
1045 * it). We compromise it as it is much better for peer
1046 * to send a RST, and RST will be the final packet
1047 * for the exchange.
1048 *
1049 * If we do not forbid deprecated addresses, we accept
1050 * the SYN packet. RFC2462 does not suggest dropping
1051 * SYN in this case.
1052 * If we decipher RFC2462 5.5.4, it says like this:
1053 * 1. use of deprecated addr with existing
1054 * communication is okay - "SHOULD continue to be
1055 * used"
1056 * 2. use of it with new communication:
1057 * (2a) "SHOULD NOT be used if alternate address
1058 * with sufficient scope is available"
1059 * (2b) nothing mentioned otherwise.
1060 * Here we fall into (2b) case as we have no choice in
1061 * our source address selection - we must obey the peer.
1062 *
1063 * The wording in RFC2462 is confusing, and there are
1064 * multiple description text for deprecated address
1065 * handling - worse, they are not exactly the same.
1066 * I believe 5.5.4 is the best one, so we follow 5.5.4.
1067 */
1068 if (isipv6 && !ip6_use_deprecated) {
1069 struct in6_ifaddr *ia6;
1070
1071 if ((ia6 = ip6_getdstifaddr(m)) &&
1072 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1073 tp = NULL;
1074 rstreason = BANDLIM_RST_OPENPORT;
1075 goto dropwithreset;
1076 }
1077 }
1078 #endif
1079 /*
1080 * If it is from this socket, drop it, it must be forged.
1081 * Don't bother responding if the destination was a broadcast.
1082 */
1083 if (th->th_dport == th->th_sport) {
1084 if (isipv6) {
1085 if (IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
1086 &ip6->ip6_src))
1087 goto drop;
1088 } else {
1089 if (ip->ip_dst.s_addr == ip->ip_src.s_addr)
1090 goto drop;
1091 }
1092 }
1093 /*
1094 * RFC1122 4.2.3.10, p. 104: discard bcast/mcast SYN
1095 *
1096 * Note that it is quite possible to receive unicast
1097 * link-layer packets with a broadcast IP address. Use
1098 * in_broadcast() to find them.
1099 */
1100 if (m->m_flags & (M_BCAST | M_MCAST))
1101 goto drop;
1102 if (isipv6) {
1103 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1104 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
1105 goto drop;
1106 } else {
1107 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1108 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1109 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1110 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
1111 goto drop;
1112 }
1113 /*
1114 * SYN appears to be valid; create compressed TCP state
1115 * for syncache.
1116 */
1117 if (so->so_qlen <= so->so_qlimit) {
1118 tcp_dooptions(&to, optp, optlen, TRUE, th->th_ack);
1119 if (!syncache_add(&inc, &to, th, so, m))
1120 goto drop;
1121
1122 /*
1123 * Entry added to syncache, mbuf used to
1124 * send SYN,ACK packet.
1125 */
1126 return(IPPROTO_DONE);
1127 }
1128 goto drop;
1129 }
1130
1131 after_listen:
1132 /*
1133 * Should not happen - syncache should pick up these connections.
1134 *
1135 * Once we are past handling listen sockets we must be in the
1136 * correct protocol processing thread.
1137 */
1138 KASSERT(tp->t_state != TCPS_LISTEN, ("tcp_input: TCPS_LISTEN state"));
1139 KKASSERT(so->so_port == &curthread->td_msgport);
1140
1141 /* Unscale the window into a 32-bit value. */
1142 if (!(thflags & TH_SYN))
1143 tiwin = th->th_win << tp->snd_scale;
1144 else
1145 tiwin = th->th_win;
1146
1147 /*
1148 * This is the second part of the MSS DoS prevention code (after
1149 * minmss on the sending side) and it deals with too many too small
1150 * tcp packets in a too short timeframe (1 second).
1151 *
1152 * XXX Removed. This code was crap. It does not scale to network
1153 * speed, and default values break NFS. Gone.
1154 */
1155 /* REMOVED */
1156
1157 /*
1158 * Segment received on connection.
1159 *
1160 * Reset idle time and keep-alive timer. Don't waste time if less
1161 * then a second has elapsed.
1162 */
1163 if ((int)(ticks - tp->t_rcvtime) > hz)
1164 tcp_timer_keep_activity(tp, thflags);
1165
1166 /*
1167 * Process options.
1168 * XXX this is tradtitional behavior, may need to be cleaned up.
1169 */
1170 tcp_dooptions(&to, optp, optlen, (thflags & TH_SYN) != 0, th->th_ack);
1171 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1172 if ((to.to_flags & TOF_SCALE) && (tp->t_flags & TF_REQ_SCALE)) {
1173 tp->t_flags |= TF_RCVD_SCALE;
1174 tp->snd_scale = to.to_requested_s_scale;
1175 }
1176
1177 /*
1178 * Initial send window; will be updated upon next ACK
1179 */
1180 tp->snd_wnd = th->th_win;
1181
1182 if (to.to_flags & TOF_TS) {
1183 tp->t_flags |= TF_RCVD_TSTMP;
1184 tp->ts_recent = to.to_tsval;
1185 tp->ts_recent_age = ticks;
1186 }
1187 if (!(to.to_flags & TOF_MSS))
1188 to.to_mss = 0;
1189 tcp_rmx_init(tp, to.to_mss);
1190 /*
1191 * Only set the TF_SACK_PERMITTED per-connection flag
1192 * if we got a SACK_PERMITTED option from the other side
1193 * and the global tcp_do_sack variable is true.
1194 */
1195 if (tcp_do_sack && (to.to_flags & TOF_SACK_PERMITTED))
1196 tp->t_flags |= TF_SACK_PERMITTED;
1197 }
1198
1199 /*
1200 * Header prediction: check for the two common cases
1201 * of a uni-directional data xfer. If the packet has
1202 * no control flags, is in-sequence, the window didn't
1203 * change and we're not retransmitting, it's a
1204 * candidate. If the length is zero and the ack moved
1205 * forward, we're the sender side of the xfer. Just
1206 * free the data acked & wake any higher level process
1207 * that was blocked waiting for space. If the length
1208 * is non-zero and the ack didn't move, we're the
1209 * receiver side. If we're getting packets in-order
1210 * (the reassembly queue is empty), add the data to
1211 * the socket buffer and note that we need a delayed ack.
1212 * Make sure that the hidden state-flags are also off.
1213 * Since we check for TCPS_ESTABLISHED above, it can only
1214 * be TH_NEEDSYN.
1215 */
1216 if (tp->t_state == TCPS_ESTABLISHED &&
1217 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1218 !(tp->t_flags & (TF_NEEDSYN | TF_NEEDFIN)) &&
1219 (!(to.to_flags & TOF_TS) ||
1220 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) &&
1221 th->th_seq == tp->rcv_nxt &&
1222 tp->snd_nxt == tp->snd_max) {
1223
1224 /*
1225 * If last ACK falls within this segment's sequence numbers,
1226 * record the timestamp.
1227 * NOTE that the test is modified according to the latest
1228 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1229 */
1230 if ((to.to_flags & TOF_TS) &&
1231 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1232 tp->ts_recent_age = ticks;
1233 tp->ts_recent = to.to_tsval;
1234 }
1235
1236 if (tlen == 0) {
1237 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1238 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1239 tp->snd_cwnd >= tp->snd_wnd &&
1240 !IN_FASTRECOVERY(tp)) {
1241 /*
1242 * This is a pure ack for outstanding data.
1243 */
1244 ++tcpstat.tcps_predack;
1245 /*
1246 * "bad retransmit" recovery
1247 *
1248 * If Eifel detection applies, then
1249 * it is deterministic, so use it
1250 * unconditionally over the old heuristic.
1251 * Otherwise, fall back to the old heuristic.
1252 */
1253 if (tcp_do_eifel_detect &&
1254 (to.to_flags & TOF_TS) && to.to_tsecr &&
1255 (tp->rxt_flags & TRXT_F_FIRSTACCACK)) {
1256 /* Eifel detection applicable. */
1257 if (to.to_tsecr < tp->t_rexmtTS) {
1258 tcp_revert_congestion_state(tp);
1259 ++tcpstat.tcps_eifeldetected;
1260 if (tp->t_rxtshift != 1 ||
1261 ticks >= tp->t_badrxtwin)
1262 ++tcpstat.tcps_rttcantdetect;
1263 }
1264 } else if (tp->t_rxtshift == 1 &&
1265 ticks < tp->t_badrxtwin) {
1266 tcp_revert_congestion_state(tp);
1267 ++tcpstat.tcps_rttdetected;
1268 }
1269 tp->rxt_flags &= ~(TRXT_F_FIRSTACCACK |
1270 TRXT_F_FASTREXMT | TRXT_F_EARLYREXMT);
1271 /*
1272 * Recalculate the retransmit timer / rtt.
1273 *
1274 * Some machines (certain windows boxes)
1275 * send broken timestamp replies during the
1276 * SYN+ACK phase, ignore timestamps of 0.
1277 */
1278 if ((to.to_flags & TOF_TS) && to.to_tsecr) {
1279 tcp_xmit_timer(tp,
1280 ticks - to.to_tsecr + 1,
1281 th->th_ack);
1282 } else if (tp->t_rtttime &&
1283 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1284 tcp_xmit_timer(tp,
1285 ticks - tp->t_rtttime + 1,
1286 th->th_ack);
1287 }
1288 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1289 acked = th->th_ack - tp->snd_una;
1290 tcpstat.tcps_rcvackpack++;
1291 tcpstat.tcps_rcvackbyte += acked;
1292 sbdrop(&so->so_snd.sb, acked);
1293 tp->snd_recover = th->th_ack - 1;
1294 tp->snd_una = th->th_ack;
1295 tp->t_dupacks = 0;
1296 /*
1297 * Update window information.
1298 */
1299 if (tiwin != tp->snd_wnd &&
1300 acceptable_window_update(tp, th, tiwin)) {
1301 /* keep track of pure window updates */
1302 if (tp->snd_wl2 == th->th_ack &&
1303 tiwin > tp->snd_wnd)
1304 tcpstat.tcps_rcvwinupd++;
1305 tp->snd_wnd = tiwin;
1306 tp->snd_wl1 = th->th_seq;
1307 tp->snd_wl2 = th->th_ack;
1308 if (tp->snd_wnd > tp->max_sndwnd)
1309 tp->max_sndwnd = tp->snd_wnd;
1310 }
1311 m_freem(m);
1312 ND6_HINT(tp); /* some progress has been done */
1313 /*
1314 * If all outstanding data are acked, stop
1315 * retransmit timer, otherwise restart timer
1316 * using current (possibly backed-off) value.
1317 * If process is waiting for space,
1318 * wakeup/selwakeup/signal. If data
1319 * are ready to send, let tcp_output
1320 * decide between more output or persist.
1321 */
1322 if (tp->snd_una == tp->snd_max) {
1323 tcp_callout_stop(tp, tp->tt_rexmt);
1324 } else if (!tcp_callout_active(tp,
1325 tp->tt_persist)) {
1326 tcp_callout_reset(tp, tp->tt_rexmt,
1327 tp->t_rxtcur, tcp_timer_rexmt);
1328 }
1329 sowwakeup(so);
1330 if (so->so_snd.ssb_cc > 0 &&
1331 !tcp_output_pending(tp))
1332 tcp_output_fair(tp);
1333 return(IPPROTO_DONE);
1334 }
1335 } else if (tiwin == tp->snd_wnd &&
1336 th->th_ack == tp->snd_una &&
1337 TAILQ_EMPTY(&tp->t_segq) &&
1338 tlen <= ssb_space(&so->so_rcv)) {
1339 u_long newsize = 0; /* automatic sockbuf scaling */
1340 /*
1341 * This is a pure, in-sequence data packet
1342 * with nothing on the reassembly queue and
1343 * we have enough buffer space to take it.
1344 */
1345 ++tcpstat.tcps_preddat;
1346 tp->rcv_nxt += tlen;
1347 tcpstat.tcps_rcvpack++;
1348 tcpstat.tcps_rcvbyte += tlen;
1349 ND6_HINT(tp); /* some progress has been done */
1350 /*
1351 * Automatic sizing of receive socket buffer. Often the send
1352 * buffer size is not optimally adjusted to the actual network
1353 * conditions at hand (delay bandwidth product). Setting the
1354 * buffer size too small limits throughput on links with high
1355 * bandwidth and high delay (eg. trans-continental/oceanic links).
1356 *
1357 * On the receive side the socket buffer memory is only rarely
1358 * used to any significant extent. This allows us to be much
1359 * more aggressive in scaling the receive socket buffer. For
1360 * the case that the buffer space is actually used to a large
1361 * extent and we run out of kernel memory we can simply drop
1362 * the new segments; TCP on the sender will just retransmit it
1363 * later. Setting the buffer size too big may only consume too
1364 * much kernel memory if the application doesn't read() from
1365 * the socket or packet loss or reordering makes use of the
1366 * reassembly queue.
1367 *
1368 * The criteria to step up the receive buffer one notch are:
1369 * 1. the number of bytes received during the time it takes
1370 * one timestamp to be reflected back to us (the RTT);
1371 * 2. received bytes per RTT is within seven eighth of the
1372 * current socket buffer size;
1373 * 3. receive buffer size has not hit maximal automatic size;
1374 *
1375 * This algorithm does one step per RTT at most and only if
1376 * we receive a bulk stream w/o packet losses or reorderings.
1377 * Shrinking the buffer during idle times is not necessary as
1378 * it doesn't consume any memory when idle.
1379 *
1380 * TODO: Only step up if the application is actually serving
1381 * the buffer to better manage the socket buffer resources.
1382 */
1383 if (tcp_do_autorcvbuf &&
1384 to.to_tsecr &&
1385 (so->so_rcv.ssb_flags & SSB_AUTOSIZE)) {
1386 if (to.to_tsecr > tp->rfbuf_ts &&
1387 to.to_tsecr - tp->rfbuf_ts < hz) {
1388 if (tp->rfbuf_cnt >
1389 (so->so_rcv.ssb_hiwat / 8 * 7) &&
1390 so->so_rcv.ssb_hiwat <
1391 tcp_autorcvbuf_max) {
1392 newsize =
1393 ulmin(so->so_rcv.ssb_hiwat +
1394 tcp_autorcvbuf_inc,
1395 tcp_autorcvbuf_max);
1396 }
1397 /* Start over with next RTT. */
1398 tp->rfbuf_ts = 0;
1399 tp->rfbuf_cnt = 0;
1400 } else
1401 tp->rfbuf_cnt += tlen; /* add up */
1402 }
1403 /*
1404 * Add data to socket buffer.
1405 */
1406 if (so->so_state & SS_CANTRCVMORE) {
1407 m_freem(m);
1408 } else {
1409 /*
1410 * Set new socket buffer size, give up when
1411 * limit is reached.
1412 *
1413 * Adjusting the size can mess up ACK
1414 * sequencing when pure window updates are
1415 * being avoided (which is the default),
1416 * so force an ack.
1417 */
1418 lwkt_gettoken(&so->so_rcv.ssb_token);
1419 if (newsize) {
1420 tp->t_flags |= TF_RXRESIZED;
1421 if (!ssb_reserve(&so->so_rcv, newsize,
1422 so, NULL)) {
1423 atomic_clear_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE);
1424 }
1425 if (newsize >=
1426 (TCP_MAXWIN << tp->rcv_scale)) {
1427 atomic_clear_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE);
1428 }
1429 }
1430 m_adj(m, drop_hdrlen); /* delayed header drop */
1431 ssb_appendstream(&so->so_rcv, m);
1432 lwkt_reltoken(&so->so_rcv.ssb_token);
1433 }
1434 sorwakeup(so);
1435 /*
1436 * This code is responsible for most of the ACKs
1437 * the TCP stack sends back after receiving a data
1438 * packet. Note that the DELAY_ACK check fails if
1439 * the delack timer is already running, which results
1440 * in an ack being sent every other packet (which is
1441 * what we want).
1442 *
1443 * We then further aggregate acks by not actually
1444 * sending one until the protocol thread has completed
1445 * processing the current backlog of packets. This
1446 * does not delay the ack any further, but allows us
1447 * to take advantage of the packet aggregation that
1448 * high speed NICs do (usually blocks of 8-10 packets)
1449 * to send a single ack rather then four or five acks,
1450 * greatly reducing the ack rate, the return channel
1451 * bandwidth, and the protocol overhead on both ends.
1452 *
1453 * Since this also has the effect of slowing down
1454 * the exponential slow-start ramp-up, systems with
1455 * very large bandwidth-delay products might want
1456 * to turn the feature off.
1457 */
1458 if (DELAY_ACK(tp)) {
1459 tcp_callout_reset(tp, tp->tt_delack,
1460 tcp_delacktime, tcp_timer_delack);
1461 } else if (tcp_aggregate_acks) {
1462 tp->t_flags |= TF_ACKNOW;
1463 if (!(tp->t_flags & TF_ONOUTPUTQ)) {
1464 tp->t_flags |= TF_ONOUTPUTQ;
1465 tp->tt_cpu = mycpu->gd_cpuid;
1466 TAILQ_INSERT_TAIL(
1467 &tcpcbackq[tp->tt_cpu].head,
1468 tp, t_outputq);
1469 }
1470 } else {
1471 tp->t_flags |= TF_ACKNOW;
1472 tcp_output(tp);
1473 }
1474 return(IPPROTO_DONE);
1475 }
1476 }
1477
1478 /*
1479 * Calculate amount of space in receive window,
1480 * and then do TCP input processing.
1481 * Receive window is amount of space in rcv queue,
1482 * but not less than advertised window.
1483 */
1484 recvwin = ssb_space(&so->so_rcv);
1485 if (recvwin < 0)
1486 recvwin = 0;
1487 tp->rcv_wnd = imax(recvwin, (int)(tp->rcv_adv - tp->rcv_nxt));
1488
1489 /* Reset receive buffer auto scaling when not in bulk receive mode. */
1490 tp->rfbuf_ts = 0;
1491 tp->rfbuf_cnt = 0;
1492
1493 switch (tp->t_state) {
1494 /*
1495 * If the state is SYN_RECEIVED:
1496 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1497 */
1498 case TCPS_SYN_RECEIVED:
1499 if ((thflags & TH_ACK) &&
1500 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1501 SEQ_GT(th->th_ack, tp->snd_max))) {
1502 rstreason = BANDLIM_RST_OPENPORT;
1503 goto dropwithreset;
1504 }
1505 break;
1506
1507 /*
1508 * If the state is SYN_SENT:
1509 * if seg contains an ACK, but not for our SYN, drop the input.
1510 * if seg contains a RST, then drop the connection.
1511 * if seg does not contain SYN, then drop it.
1512 * Otherwise this is an acceptable SYN segment
1513 * initialize tp->rcv_nxt and tp->irs
1514 * if seg contains ack then advance tp->snd_una
1515 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1516 * arrange for segment to be acked (eventually)
1517 * continue processing rest of data/controls, beginning with URG
1518 */
1519 case TCPS_SYN_SENT:
1520 if ((thflags & TH_ACK) &&
1521 (SEQ_LEQ(th->th_ack, tp->iss) ||
1522 SEQ_GT(th->th_ack, tp->snd_max))) {
1523 rstreason = BANDLIM_UNLIMITED;
1524 goto dropwithreset;
1525 }
1526 if (thflags & TH_RST) {
1527 if (thflags & TH_ACK)
1528 tp = tcp_drop(tp, ECONNREFUSED);
1529 goto drop;
1530 }
1531 if (!(thflags & TH_SYN))
1532 goto drop;
1533
1534 tp->irs = th->th_seq;
1535 tcp_rcvseqinit(tp);
1536 if (thflags & TH_ACK) {
1537 /* Our SYN was acked. */
1538 tcpstat.tcps_connects++;
1539 soisconnected(so);
1540 /* Do window scaling on this connection? */
1541 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
1542 (TF_RCVD_SCALE | TF_REQ_SCALE))
1543 tp->rcv_scale = tp->request_r_scale;
1544 tp->rcv_adv += tp->rcv_wnd;
1545 tp->snd_una++; /* SYN is acked */
1546 tcp_callout_stop(tp, tp->tt_rexmt);
1547 /*
1548 * If there's data, delay ACK; if there's also a FIN
1549 * ACKNOW will be turned on later.
1550 */
1551 if (DELAY_ACK(tp) && tlen != 0) {
1552 tcp_callout_reset(tp, tp->tt_delack,
1553 tcp_delacktime, tcp_timer_delack);
1554 } else {
1555 tp->t_flags |= TF_ACKNOW;
1556 }
1557 /*
1558 * Received <SYN,ACK> in SYN_SENT[*] state.
1559 * Transitions:
1560 * SYN_SENT --> ESTABLISHED
1561 * SYN_SENT* --> FIN_WAIT_1
1562 */
1563 tp->t_starttime = ticks;
1564 if (tp->t_flags & TF_NEEDFIN) {
1565 TCP_STATE_CHANGE(tp, TCPS_FIN_WAIT_1);
1566 tp->t_flags &= ~TF_NEEDFIN;
1567 thflags &= ~TH_SYN;
1568 } else {
1569 tcp_established(tp);
1570 }
1571 } else {
1572 /*
1573 * Received initial SYN in SYN-SENT[*] state =>
1574 * simultaneous open.
1575 * Do 3-way handshake:
1576 * SYN-SENT -> SYN-RECEIVED
1577 * SYN-SENT* -> SYN-RECEIVED*
1578 */
1579 tp->t_flags |= TF_ACKNOW;
1580 tcp_callout_stop(tp, tp->tt_rexmt);
1581 TCP_STATE_CHANGE(tp, TCPS_SYN_RECEIVED);
1582 }
1583
1584 /*
1585 * Advance th->th_seq to correspond to first data byte.
1586 * If data, trim to stay within window,
1587 * dropping FIN if necessary.
1588 */
1589 th->th_seq++;
1590 if (tlen > tp->rcv_wnd) {
1591 todrop = tlen - tp->rcv_wnd;
1592 m_adj(m, -todrop);
1593 tlen = tp->rcv_wnd;
1594 thflags &= ~TH_FIN;
1595 tcpstat.tcps_rcvpackafterwin++;
1596 tcpstat.tcps_rcvbyteafterwin += todrop;
1597 }
1598 tp->snd_wl1 = th->th_seq - 1;
1599 tp->rcv_up = th->th_seq;
1600 /*
1601 * Client side of transaction: already sent SYN and data.
1602 * If the remote host used T/TCP to validate the SYN,
1603 * our data will be ACK'd; if so, enter normal data segment
1604 * processing in the middle of step 5, ack processing.
1605 * Otherwise, goto step 6.
1606 */
1607 if (thflags & TH_ACK)
1608 goto process_ACK;
1609
1610 goto step6;
1611
1612 /*
1613 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
1614 * do normal processing (we no longer bother with T/TCP).
1615 */
1616 case TCPS_LAST_ACK:
1617 case TCPS_CLOSING:
1618 case TCPS_TIME_WAIT:
1619 /*
1620 * Continue normal processing
1621 */
1622 break;
1623 }
1624
1625 /*
1626 * States other than LISTEN or SYN_SENT.
1627 * First check the RST flag and sequence number since reset segments
1628 * are exempt from the timestamp and connection count tests. This
1629 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
1630 * below which allowed reset segments in half the sequence space
1631 * to fall though and be processed (which gives forged reset
1632 * segments with a random sequence number a 50 percent chance of
1633 * killing a connection).
1634 * Then check timestamp, if present.
1635 * Then check the connection count, if present.
1636 * Then check that at least some bytes of segment are within
1637 * receive window. If segment begins before rcv_nxt,
1638 * drop leading data (and SYN); if nothing left, just ack.
1639 *
1640 *
1641 * If the RST bit is set, check the sequence number to see
1642 * if this is a valid reset segment.
1643 * RFC 793 page 37:
1644 * In all states except SYN-SENT, all reset (RST) segments
1645 * are validated by checking their SEQ-fields. A reset is
1646 * valid if its sequence number is in the window.
1647 * Note: this does not take into account delayed ACKs, so
1648 * we should test against last_ack_sent instead of rcv_nxt.
1649 * The sequence number in the reset segment is normally an
1650 * echo of our outgoing acknowledgement numbers, but some hosts
1651 * send a reset with the sequence number at the rightmost edge
1652 * of our receive window, and we have to handle this case.
1653 * If we have multiple segments in flight, the intial reset
1654 * segment sequence numbers will be to the left of last_ack_sent,
1655 * but they will eventually catch up.
1656 * In any case, it never made sense to trim reset segments to
1657 * fit the receive window since RFC 1122 says:
1658 * 4.2.2.12 RST Segment: RFC-793 Section 3.4
1659 *
1660 * A TCP SHOULD allow a received RST segment to include data.
1661 *
1662 * DISCUSSION
1663 * It has been suggested that a RST segment could contain
1664 * ASCII text that encoded and explained the cause of the
1665 * RST. No standard has yet been established for such
1666 * data.
1667 *
1668 * If the reset segment passes the sequence number test examine
1669 * the state:
1670 * SYN_RECEIVED STATE:
1671 * If passive open, return to LISTEN state.
1672 * If active open, inform user that connection was refused.
1673 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
1674 * Inform user that connection was reset, and close tcb.
1675 * CLOSING, LAST_ACK STATES:
1676 * Close the tcb.
1677 * TIME_WAIT STATE:
1678 * Drop the segment - see Stevens, vol. 2, p. 964 and
1679 * RFC 1337.
1680 */
1681 if (thflags & TH_RST) {
1682 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
1683 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
1684 switch (tp->t_state) {
1685
1686 case TCPS_SYN_RECEIVED:
1687 so->so_error = ECONNREFUSED;
1688 goto close;
1689
1690 case TCPS_ESTABLISHED:
1691 case TCPS_FIN_WAIT_1:
1692 case TCPS_FIN_WAIT_2:
1693 case TCPS_CLOSE_WAIT:
1694 so->so_error = ECONNRESET;
1695 close:
1696 TCP_STATE_CHANGE(tp, TCPS_CLOSED);
1697 tcpstat.tcps_drops++;
1698 tp = tcp_close(tp);
1699 break;
1700
1701 case TCPS_CLOSING:
1702 case TCPS_LAST_ACK:
1703 tp = tcp_close(tp);
1704 break;
1705
1706 case TCPS_TIME_WAIT:
1707 break;
1708 }
1709 }
1710 goto drop;
1711 }
1712
1713 /*
1714 * Allow a new connection to replace an existing connection that is
1715 * in the TIME_WAIT state if the new connection's SYN seq is different
1716 * from tp->irs.
1717 */
1718 if ((thflags & TH_SYN) &&
1719 tp->t_state == TCPS_TIME_WAIT &&
1720 th->th_seq != tp->irs) {
1721 tp = tcp_close(tp);
1722 goto findpcb;
1723 }
1724
1725 /*
1726 * RFC 1323 PAWS: If we have a timestamp reply on this segment
1727 * and it's less than ts_recent, drop it.
1728 */
1729 if ((to.to_flags & TOF_TS) && tp->ts_recent != 0 &&
1730 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
1731 /* Check to see if ts_recent is over 24 days old. */
1732 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1733 /*
1734 * Invalidate ts_recent. If this segment updates
1735 * ts_recent, the age will be reset later and ts_recent
1736 * will get a valid value. If it does not, setting
1737 * ts_recent to zero will at least satisfy the
1738 * requirement that zero be placed in the timestamp
1739 * echo reply when ts_recent isn't valid. The
1740 * age isn't reset until we get a valid ts_recent
1741 * because we don't want out-of-order segments to be
1742 * dropped when ts_recent is old.
1743 */
1744 tp->ts_recent = 0;
1745 } else if (tcp_paws_tolerance && tlen != 0 &&
1746 tp->t_state == TCPS_ESTABLISHED &&
1747 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK&&
1748 !(tp->t_flags & (TF_NEEDSYN | TF_NEEDFIN)) &&
1749 th->th_ack == tp->snd_una &&
1750 tiwin == tp->snd_wnd &&
1751 TSTMP_GEQ(to.to_tsval + tcp_paws_tolerance, tp->ts_recent)&&
1752 (th->th_seq == tp->rcv_nxt ||
1753 (SEQ_GT(th->th_seq, tp->rcv_nxt) &&
1754 tcp_paws_canreasslast(tp, th, tlen)))) {
1755 /*
1756 * This tends to prevent valid new segments from being
1757 * dropped by the reordered segments sent by the fast
1758 * retransmission algorithm on the sending side, i.e.
1759 * the fast retransmitted segment w/ larger timestamp
1760 * arrives earlier than the previously sent new segments
1761 * w/ smaller timestamp.
1762 *
1763 * If following conditions are met, the segment is
1764 * accepted:
1765 * - The segment contains data
1766 * - The connection is established
1767 * - The header does not contain important flags
1768 * - SYN or FIN is not needed
1769 * - It does not acknowledge new data
1770 * - Receive window is not changed
1771 * - The timestamp is within "acceptable" range
1772 * - The new segment is what we are expecting or
1773 * the new segment could be merged w/ the last
1774 * pending segment on the reassemble queue
1775 */
1776 tcpstat.tcps_pawsaccept++;
1777 tcpstat.tcps_pawsdrop++;
1778 } else {
1779 tcpstat.tcps_rcvduppack++;
1780 tcpstat.tcps_rcvdupbyte += tlen;
1781 tcpstat.tcps_pawsdrop++;
1782 if (tlen)
1783 goto dropafterack;
1784 goto drop;
1785 }
1786 }
1787
1788 /*
1789 * In the SYN-RECEIVED state, validate that the packet belongs to
1790 * this connection before trimming the data to fit the receive
1791 * window. Check the sequence number versus IRS since we know
1792 * the sequence numbers haven't wrapped. This is a partial fix
1793 * for the "LAND" DoS attack.
1794 */
1795 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
1796 rstreason = BANDLIM_RST_OPENPORT;
1797 goto dropwithreset;
1798 }
1799
1800 todrop = tp->rcv_nxt - th->th_seq;
1801 if (todrop > 0) {
1802 if (TCP_DO_SACK(tp)) {
1803 /* Report duplicate segment at head of packet. */
1804 tp->reportblk.rblk_start = th->th_seq;
1805 tp->reportblk.rblk_end = TCP_SACK_BLKEND(
1806 th->th_seq + tlen, thflags);
1807 if (SEQ_GT(tp->reportblk.rblk_end, tp->rcv_nxt))
1808 tp->reportblk.rblk_end = tp->rcv_nxt;
1809 tp->sack_flags |= (TSACK_F_DUPSEG | TSACK_F_SACKLEFT);
1810 tp->t_flags |= TF_ACKNOW;
1811 }
1812 if (thflags & TH_SYN) {
1813 thflags &= ~TH_SYN;
1814 th->th_seq++;
1815 if (th->th_urp > 1)
1816 th->th_urp--;
1817 else
1818 thflags &= ~TH_URG;
1819 todrop--;
1820 }
1821 /*
1822 * Following if statement from Stevens, vol. 2, p. 960.
1823 */
1824 if (todrop > tlen ||
1825 (todrop == tlen && !(thflags & TH_FIN))) {
1826 /*
1827 * Any valid FIN must be to the left of the window.
1828 * At this point the FIN must be a duplicate or out
1829 * of sequence; drop it.
1830 */
1831 thflags &= ~TH_FIN;
1832
1833 /*
1834 * Send an ACK to resynchronize and drop any data.
1835 * But keep on processing for RST or ACK.
1836 */
1837 tp->t_flags |= TF_ACKNOW;
1838 todrop = tlen;
1839 tcpstat.tcps_rcvduppack++;
1840 tcpstat.tcps_rcvdupbyte += todrop;
1841 } else {
1842 tcpstat.tcps_rcvpartduppack++;
1843 tcpstat.tcps_rcvpartdupbyte += todrop;
1844 }
1845 drop_hdrlen += todrop; /* drop from the top afterwards */
1846 th->th_seq += todrop;
1847 tlen -= todrop;
1848 if (th->th_urp > todrop)
1849 th->th_urp -= todrop;
1850 else {
1851 thflags &= ~TH_URG;
1852 th->th_urp = 0;
1853 }
1854 }
1855
1856 /*
1857 * If new data are received on a connection after the
1858 * user processes are gone, then RST the other end.
1859 */
1860 if ((so->so_state & SS_NOFDREF) &&
1861 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1862 tp = tcp_close(tp);
1863 tcpstat.tcps_rcvafterclose++;
1864 rstreason = BANDLIM_UNLIMITED;
1865 goto dropwithreset;
1866 }
1867
1868 /*
1869 * If segment ends after window, drop trailing data
1870 * (and PUSH and FIN); if nothing left, just ACK.
1871 */
1872 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
1873 if (todrop > 0) {
1874 tcpstat.tcps_rcvpackafterwin++;
1875 if (todrop >= tlen) {
1876 tcpstat.tcps_rcvbyteafterwin += tlen;
1877
1878 /*
1879 * If window is closed can only take segments at
1880 * window edge, and have to drop data and PUSH from
1881 * incoming segments. Continue processing, but
1882 * remember to ack. Otherwise, drop segment
1883 * and ack.
1884 */
1885 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1886 tp->t_flags |= TF_ACKNOW;
1887 tcpstat.tcps_rcvwinprobe++;
1888 } else
1889 goto dropafterack;
1890 } else
1891 tcpstat.tcps_rcvbyteafterwin += todrop;
1892 m_adj(m, -todrop);
1893 tlen -= todrop;
1894 thflags &= ~(TH_PUSH | TH_FIN);
1895 }
1896
1897 /*
1898 * If last ACK falls within this segment's sequence numbers,
1899 * record its timestamp.
1900 * NOTE:
1901 * 1) That the test incorporates suggestions from the latest
1902 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1903 * 2) That updating only on newer timestamps interferes with
1904 * our earlier PAWS tests, so this check should be solely
1905 * predicated on the sequence space of this segment.
1906 * 3) That we modify the segment boundary check to be
1907 * Last.ACK.Sent <= SEG.SEQ + SEG.LEN
1908 * instead of RFC1323's
1909 * Last.ACK.Sent < SEG.SEQ + SEG.LEN,
1910 * This modified check allows us to overcome RFC1323's
1911 * limitations as described in Stevens TCP/IP Illustrated
1912 * Vol. 2 p.869. In such cases, we can still calculate the
1913 * RTT correctly when RCV.NXT == Last.ACK.Sent.
1914 */
1915 if ((to.to_flags & TOF_TS) && SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
1916 SEQ_LEQ(tp->last_ack_sent, (th->th_seq + tlen
1917 + ((thflags & TH_SYN) != 0)
1918 + ((thflags & TH_FIN) != 0)))) {
1919 tp->ts_recent_age = ticks;
1920 tp->ts_recent = to.to_tsval;
1921 }
1922
1923 /*
1924 * If a SYN is in the window, then this is an
1925 * error and we send an RST and drop the connection.
1926 */
1927 if (thflags & TH_SYN) {
1928 tp = tcp_drop(tp, ECONNRESET);
1929 rstreason = BANDLIM_UNLIMITED;
1930 goto dropwithreset;
1931 }
1932
1933 /*
1934 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
1935 * flag is on (half-synchronized state), then queue data for
1936 * later processing; else drop segment and return.
1937 */
1938 if (!(thflags & TH_ACK)) {
1939 if (tp->t_state == TCPS_SYN_RECEIVED ||
1940 (tp->t_flags & TF_NEEDSYN))
1941 goto step6;
1942 else
1943 goto drop;
1944 }
1945
1946 /*
1947 * Ack processing.
1948 */
1949 switch (tp->t_state) {
1950 /*
1951 * In SYN_RECEIVED state, the ACK acknowledges our SYN, so enter
1952 * ESTABLISHED state and continue processing.
1953 * The ACK was checked above.
1954 */
1955 case TCPS_SYN_RECEIVED:
1956
1957 tcpstat.tcps_connects++;
1958 soisconnected(so);
1959 /* Do window scaling? */
1960 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
1961 (TF_RCVD_SCALE | TF_REQ_SCALE))
1962 tp->rcv_scale = tp->request_r_scale;
1963 /*
1964 * Make transitions:
1965 * SYN-RECEIVED -> ESTABLISHED
1966 * SYN-RECEIVED* -> FIN-WAIT-1
1967 */
1968 tp->t_starttime = ticks;
1969 if (tp->t_flags & TF_NEEDFIN) {
1970 TCP_STATE_CHANGE(tp, TCPS_FIN_WAIT_1);
1971 tp->t_flags &= ~TF_NEEDFIN;
1972 } else {
1973 tcp_established(tp);
1974 }
1975 /*
1976 * If segment contains data or ACK, will call tcp_reass()
1977 * later; if not, do so now to pass queued data to user.
1978 */
1979 if (tlen == 0 && !(thflags & TH_FIN))
1980 tcp_reass(tp, NULL, NULL, NULL);
1981 /* fall into ... */
1982
1983 /*
1984 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1985 * ACKs. If the ack is in the range
1986 * tp->snd_una < th->th_ack <= tp->snd_max
1987 * then advance tp->snd_una to th->th_ack and drop
1988 * data from the retransmission queue. If this ACK reflects
1989 * more up to date window information we update our window information.
1990 */
1991 case TCPS_ESTABLISHED:
1992 case TCPS_FIN_WAIT_1:
1993 case TCPS_FIN_WAIT_2:
1994 case TCPS_CLOSE_WAIT:
1995 case TCPS_CLOSING:
1996 case TCPS_LAST_ACK:
1997 case TCPS_TIME_WAIT:
1998
1999 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2000 boolean_t maynotdup = FALSE;
2001
2002 if (TCP_DO_SACK(tp))
2003 tcp_sack_update_scoreboard(tp, &to);
2004
2005 if (tlen != 0 || tiwin != tp->snd_wnd ||
2006 ((thflags & TH_FIN) && !(tp->t_flags & TF_SAWFIN)))
2007 maynotdup = TRUE;
2008
2009 if (!tcp_callout_active(tp, tp->tt_rexmt) ||
2010 th->th_ack != tp->snd_una) {
2011 if (!maynotdup)
2012 tcpstat.tcps_rcvdupack++;
2013 tp->t_dupacks = 0;
2014 break;
2015 }
2016
2017 #define DELAY_DUPACK \
2018 do { \
2019 delayed_dupack = TRUE; \
2020 th_dupack = th->th_ack; \
2021 to_flags = to.to_flags; \
2022 } while (0)
2023 if (maynotdup) {
2024 if (!tcp_do_rfc6675 ||
2025 !TCP_DO_SACK(tp) ||
2026 (to.to_flags &
2027 (TOF_SACK | TOF_SACK_REDUNDANT))
2028 != TOF_SACK) {
2029 tp->t_dupacks = 0;
2030 } else {
2031 DELAY_DUPACK;
2032 }
2033 break;
2034 }
2035 if ((thflags & TH_FIN) && !(tp->t_flags & TF_QUEDFIN)) {
2036 /*
2037 * This could happen, if the reassemable
2038 * queue overflew or was drained. Don't
2039 * drop this FIN here; defer the duplicated
2040 * ACK processing until this FIN gets queued.
2041 */
2042 DELAY_DUPACK;
2043 break;
2044 }
2045 #undef DELAY_DUPACK
2046
2047 if (tcp_recv_dupack(tp, th->th_ack, to.to_flags))
2048 goto drop;
2049 else
2050 break;
2051 }
2052
2053 KASSERT(SEQ_GT(th->th_ack, tp->snd_una), ("th_ack <= snd_una"));
2054 tp->t_dupacks = 0;
2055 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2056 /*
2057 * Detected optimistic ACK attack.
2058 * Force slow-start to de-synchronize attack.
2059 */
2060 tp->snd_cwnd = tp->t_maxseg;
2061 tp->snd_wacked = 0;
2062
2063 tcpstat.tcps_rcvacktoomuch++;
2064 goto dropafterack;
2065 }
2066 /*
2067 * If we reach this point, ACK is not a duplicate,
2068 * i.e., it ACKs something we sent.
2069 */
2070 if (tp->t_flags & TF_NEEDSYN) {
2071 /*
2072 * T/TCP: Connection was half-synchronized, and our
2073 * SYN has been ACK'd (so connection is now fully
2074 * synchronized). Go to non-starred state,
2075 * increment snd_una for ACK of SYN, and check if
2076 * we can do window scaling.
2077 */
2078 tp->t_flags &= ~TF_NEEDSYN;
2079 tp->snd_una++;
2080 /* Do window scaling? */
2081 if ((tp->t_flags & (TF_RCVD_SCALE | TF_REQ_SCALE)) ==
2082 (TF_RCVD_SCALE | TF_REQ_SCALE))
2083 tp->rcv_scale = tp->request_r_scale;
2084 }
2085
2086 process_ACK:
2087 acked = th->th_ack - tp->snd_una;
2088 tcpstat.tcps_rcvackpack++;
2089 tcpstat.tcps_rcvackbyte += acked;
2090
2091 if (tcp_do_eifel_detect && acked > 0 &&
2092 (to.to_flags & TOF_TS) && (to.to_tsecr != 0) &&
2093 (tp->rxt_flags & TRXT_F_FIRSTACCACK)) {
2094 /* Eifel detection applicable. */
2095 if (to.to_tsecr < tp->t_rexmtTS) {
2096 ++tcpstat.tcps_eifeldetected;
2097 tcp_revert_congestion_state(tp);
2098 if (tp->t_rxtshift != 1 ||
2099 ticks >= tp->t_badrxtwin)
2100 ++tcpstat.tcps_rttcantdetect;
2101 }
2102 } else if (tp->t_rxtshift == 1 && ticks < tp->t_badrxtwin) {
2103 /*
2104 * If we just performed our first retransmit,
2105 * and the ACK arrives within our recovery window,
2106 * then it was a mistake to do the retransmit
2107 * in the first place. Recover our original cwnd
2108 * and ssthresh, and proceed to transmit where we
2109 * left off.
2110 */
2111 tcp_revert_congestion_state(tp);
2112 ++tcpstat.tcps_rttdetected;
2113 }
2114
2115 /*
2116 * If we have a timestamp reply, update smoothed
2117 * round trip time. If no timestamp is present but
2118 * transmit timer is running and timed sequence
2119 * number was acked, update smoothed round trip time.
2120 * Since we now have an rtt measurement, cancel the
2121 * timer backoff (cf., Phil Karn's retransmit alg.).
2122 * Recompute the initial retransmit timer.
2123 *
2124 * Some machines (certain windows boxes) send broken
2125 * timestamp replies during the SYN+ACK phase, ignore
2126 * timestamps of 0.
2127 */
2128 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0))
2129 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1,
2130 th->th_ack);
2131 else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq))
2132 tcp_xmit_timer(tp, ticks - tp->t_rtttime + 1,
2133 th->th_ack);
2134 tcp_xmit_bandwidth_limit(tp, th->th_ack);
2135
2136 /*
2137 * If no data (only SYN) was ACK'd,
2138 * skip rest of ACK processing.
2139 */
2140 if (acked == 0)
2141 goto step6;
2142
2143 /* Stop looking for an acceptable ACK since one was received. */
2144 tp->rxt_flags &= ~(TRXT_F_FIRSTACCACK |
2145 TRXT_F_FASTREXMT | TRXT_F_EARLYREXMT);
2146
2147 if (acked > so->so_snd.ssb_cc) {
2148 tp->snd_wnd -= so->so_snd.ssb_cc;
2149 sbdrop(&so->so_snd.sb, (int)so->so_snd.ssb_cc);
2150 ourfinisacked = TRUE;
2151 } else {
2152 sbdrop(&so->so_snd.sb, acked);
2153 tp->snd_wnd -= acked;
2154 ourfinisacked = FALSE;
2155 }
2156 sowwakeup(so);
2157
2158 /*
2159 * Update window information.
2160 */
2161 if (acceptable_window_update(tp, th, tiwin)) {
2162 /* keep track of pure window updates */
2163 if (tlen == 0 && tp->snd_wl2 == th->th_ack &&
2164 tiwin > tp->snd_wnd)
2165 tcpstat.tcps_rcvwinupd++;
2166 tp->snd_wnd = tiwin;
2167 tp->snd_wl1 = th->th_seq;
2168 tp->snd_wl2 = th->th_ack;
2169 if (tp->snd_wnd > tp->max_sndwnd)
2170 tp->max_sndwnd = tp->snd_wnd;
2171 needoutput = TRUE;
2172 }
2173
2174 tp->snd_una = th->th_ack;
2175 if (TCP_DO_SACK(tp))
2176 tcp_sack_update_scoreboard(tp, &to);
2177 if (IN_FASTRECOVERY(tp)) {
2178 if (SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2179 EXIT_FASTRECOVERY(tp);
2180 needoutput = TRUE;
2181 /*
2182 * If the congestion window was inflated
2183 * to account for the other side's
2184 * cached packets, retract it.
2185 */
2186 if (!TCP_DO_SACK(tp))
2187 tp->snd_cwnd = tp->snd_ssthresh;
2188
2189 /*
2190 * Window inflation should have left us
2191 * with approximately snd_ssthresh outstanding
2192 * data. But, in case we would be inclined
2193 * to send a burst, better do it using
2194 * slow start.
2195 */
2196 if (SEQ_GT(th->th_ack + tp->snd_cwnd,
2197 tp->snd_max + 2 * tp->t_maxseg))
2198 tp->snd_cwnd =
2199 (tp->snd_max - tp->snd_una) +
2200 2 * tp->t_maxseg;
2201
2202 tp->snd_wacked = 0;
2203 } else {
2204 if (TCP_DO_SACK(tp)) {
2205 tp->snd_max_rexmt = tp->snd_max;
2206 tcp_sack_rexmt(tp,
2207 tp->snd_una == tp->rexmt_high);
2208 } else {
2209 tcp_newreno_partial_ack(tp, th, acked);
2210 }
2211 needoutput = FALSE;
2212 }
2213 } else {
2214 /*
2215 * Open the congestion window. When in slow-start,
2216 * open exponentially: maxseg per packet. Otherwise,
2217 * open linearly: maxseg per window.
2218 */
2219 if (tp->snd_cwnd <= tp->snd_ssthresh) {
2220 u_int abc_sslimit =
2221 (SEQ_LT(tp->snd_nxt, tp->snd_max) ?
2222 tp->t_maxseg : 2 * tp->t_maxseg);
2223
2224 /* slow-start */
2225 tp->snd_cwnd += tcp_do_abc ?
2226 min(acked, abc_sslimit) : tp->t_maxseg;
2227 } else {
2228 /* linear increase */
2229 tp->snd_wacked += tcp_do_abc ? acked :
2230 tp->t_maxseg;
2231 if (tp->snd_wacked >= tp->snd_cwnd) {
2232 tp->snd_wacked -= tp->snd_cwnd;
2233 tp->snd_cwnd += tp->t_maxseg;
2234 }
2235 }
2236 tp->snd_cwnd = min(tp->snd_cwnd,
2237 TCP_MAXWIN << tp->snd_scale);
2238 tp->snd_recover = th->th_ack - 1;
2239 }
2240 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2241 tp->snd_nxt = tp->snd_una;
2242
2243 /*
2244 * If all outstanding data is acked, stop retransmit
2245 * timer and remember to restart (more output or persist).
2246 * If there is more data to be acked, restart retransmit
2247 * timer, using current (possibly backed-off) value.
2248 */
2249 if (th->th_ack == tp->snd_max) {
2250 tcp_callout_stop(tp, tp->tt_rexmt);
2251 needoutput = TRUE;
2252 } else if (!tcp_callout_active(tp, tp->tt_persist)) {
2253 tcp_callout_reset(tp, tp->tt_rexmt, tp->t_rxtcur,
2254 tcp_timer_rexmt);
2255 }
2256
2257 switch (tp->t_state) {
2258 /*
2259 * In FIN_WAIT_1 STATE in addition to the processing
2260 * for the ESTABLISHED state if our FIN is now acknowledged
2261 * then enter FIN_WAIT_2.
2262 */
2263 case TCPS_FIN_WAIT_1:
2264 if (ourfinisacked) {
2265 /*
2266 * If we can't receive any more
2267 * data, then closing user can proceed.
2268 * Starting the timer is contrary to the
2269 * specification, but if we don't get a FIN
2270 * we'll hang forever.
2271 */
2272 if (so->so_state & SS_CANTRCVMORE) {
2273 soisdisconnected(so);
2274 tcp_callout_reset(tp, tp->tt_2msl,
2275 tp->t_maxidle, tcp_timer_2msl);
2276 }
2277 TCP_STATE_CHANGE(tp, TCPS_FIN_WAIT_2);
2278 }
2279 break;
2280
2281 /*
2282 * In CLOSING STATE in addition to the processing for
2283 * the ESTABLISHED state if the ACK acknowledges our FIN
2284 * then enter the TIME-WAIT state, otherwise ignore
2285 * the segment.
2286 */
2287 case TCPS_CLOSING:
2288 if (ourfinisacked) {
2289 TCP_STATE_CHANGE(tp, TCPS_TIME_WAIT);
2290 tcp_canceltimers(tp);
2291 tcp_callout_reset(tp, tp->tt_2msl,
2292 2 * tcp_rmx_msl(tp),
2293 tcp_timer_2msl);
2294 soisdisconnected(so);
2295 }
2296 break;
2297
2298 /*
2299 * In LAST_ACK, we may still be waiting for data to drain
2300 * and/or to be acked, as well as for the ack of our FIN.
2301 * If our FIN is now acknowledged, delete the TCB,
2302 * enter the closed state and return.
2303 */
2304 case TCPS_LAST_ACK:
2305 if (ourfinisacked) {
2306 tp = tcp_close(tp);
2307 goto drop;
2308 }
2309 break;
2310
2311 /*
2312 * In TIME_WAIT state the only thing that should arrive
2313 * is a retransmission of the remote FIN. Acknowledge
2314 * it and restart the finack timer.
2315 */
2316 case TCPS_TIME_WAIT:
2317 tcp_callout_reset(tp, tp->tt_2msl, 2 * tcp_rmx_msl(tp),
2318 tcp_timer_2msl);
2319 goto dropafterack;
2320 }
2321 }
2322
2323 step6:
2324 /*
2325 * Update window information.
2326 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2327 */
2328 if ((thflags & TH_ACK) &&
2329 acceptable_window_update(tp, th, tiwin)) {
2330 /* keep track of pure window updates */
2331 if (tlen == 0 && tp->snd_wl2 == th->th_ack &&
2332 tiwin > tp->snd_wnd)
2333 tcpstat.tcps_rcvwinupd++;
2334 tp->snd_wnd = tiwin;
2335 tp->snd_wl1 = th->th_seq;
2336 tp->snd_wl2 = th->th_ack;
2337 if (tp->snd_wnd > tp->max_sndwnd)
2338 tp->max_sndwnd = tp->snd_wnd;
2339 needoutput = TRUE;
2340 }
2341
2342 /*
2343 * Process segments with URG.
2344 */
2345 if ((thflags & TH_URG) && th->th_urp &&
2346 !TCPS_HAVERCVDFIN(tp->t_state)) {
2347 /*
2348 * This is a kludge, but if we receive and accept
2349 * random urgent pointers, we'll crash in
2350 * soreceive. It's hard to imagine someone
2351 * actually wanting to send this much urgent data.
2352 */
2353 if (th->th_urp + so->so_rcv.ssb_cc > sb_max) {
2354 th->th_urp = 0; /* XXX */
2355 thflags &= ~TH_URG; /* XXX */
2356 goto dodata; /* XXX */
2357 }
2358 /*
2359 * If this segment advances the known urgent pointer,
2360 * then mark the data stream. This should not happen
2361 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2362 * a FIN has been received from the remote side.
2363 * In these states we ignore the URG.
2364 *
2365 * According to RFC961 (Assigned Protocols),
2366 * the urgent pointer points to the last octet
2367 * of urgent data. We continue, however,
2368 * to consider it to indicate the first octet
2369 * of data past the urgent section as the original
2370 * spec states (in one of two places).
2371 */
2372 if (SEQ_GT(th->th_seq + th->th_urp, tp->rcv_up)) {
2373 tp->rcv_up = th->th_seq + th->th_urp;
2374 so->so_oobmark = so->so_rcv.ssb_cc +
2375 (tp->rcv_up - tp->rcv_nxt) - 1;
2376 if (so->so_oobmark == 0)
2377 sosetstate(so, SS_RCVATMARK);
2378 sohasoutofband(so);
2379 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2380 }
2381 /*
2382 * Remove out of band data so doesn't get presented to user.
2383 * This can happen independent of advancing the URG pointer,
2384 * but if two URG's are pending at once, some out-of-band
2385 * data may creep in... ick.
2386 */
2387 if (th->th_urp <= (u_long)tlen &&
2388 !(so->so_options & SO_OOBINLINE)) {
2389 /* hdr drop is delayed */
2390 tcp_pulloutofband(so, th, m, drop_hdrlen);
2391 }
2392 } else {
2393 /*
2394 * If no out of band data is expected,
2395 * pull receive urgent pointer along
2396 * with the receive window.
2397 */
2398 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2399 tp->rcv_up = tp->rcv_nxt;
2400 }
2401
2402 dodata: /* XXX */
2403 /*
2404 * Process the segment text, merging it into the TCP sequencing queue,
2405 * and arranging for acknowledgment of receipt if necessary.
2406 * This process logically involves adjusting tp->rcv_wnd as data
2407 * is presented to the user (this happens in tcp_usrreq.c,
2408 * case PRU_RCVD). If a FIN has already been received on this
2409 * connection then we just ignore the text.
2410 */
2411 if ((tlen || (thflags & TH_FIN)) && !TCPS_HAVERCVDFIN(tp->t_state)) {
2412 if (thflags & TH_FIN)
2413 tp->t_flags |= TF_SAWFIN;
2414 m_adj(m, drop_hdrlen); /* delayed header drop */
2415 /*
2416 * Insert segment which includes th into TCP reassembly queue
2417 * with control block tp. Set thflags to whether reassembly now
2418 * includes a segment with FIN. This handles the common case
2419 * inline (segment is the next to be received on an established
2420 * connection, and the queue is empty), avoiding linkage into
2421 * and removal from the queue and repetition of various
2422 * conversions.
2423 * Set DELACK for segments received in order, but ack
2424 * immediately when segments are out of order (so
2425 * fast retransmit can work).
2426 */
2427 if (th->th_seq == tp->rcv_nxt &&
2428 TAILQ_EMPTY(&tp->t_segq) &&
2429 TCPS_HAVEESTABLISHED(tp->t_state)) {
2430 if (thflags & TH_FIN)
2431 tp->t_flags |= TF_QUEDFIN;
2432 if (DELAY_ACK(tp)) {
2433 tcp_callout_reset(tp, tp->tt_delack,
2434 tcp_delacktime, tcp_timer_delack);
2435 } else {
2436 tp->t_flags |= TF_ACKNOW;
2437 }
2438 tp->rcv_nxt += tlen;
2439 thflags = th->th_flags & TH_FIN;
2440 tcpstat.tcps_rcvpack++;
2441 tcpstat.tcps_rcvbyte += tlen;
2442 ND6_HINT(tp);
2443 if (so->so_state & SS_CANTRCVMORE) {
2444 m_freem(m);
2445 } else {
2446 lwkt_gettoken(&so->so_rcv.ssb_token);
2447 ssb_appendstream(&so->so_rcv, m);
2448 lwkt_reltoken(&so->so_rcv.ssb_token);
2449 }
2450 sorwakeup(so);
2451 } else {
2452 if (!(tp->sack_flags & TSACK_F_DUPSEG)) {
2453 /* Initialize SACK report block. */
2454 tp->reportblk.rblk_start = th->th_seq;
2455 tp->reportblk.rblk_end = TCP_SACK_BLKEND(
2456 th->th_seq + tlen, thflags);
2457 }
2458 thflags = tcp_reass(tp, th, &tlen, m);
2459 tp->t_flags |= TF_ACKNOW;
2460 }
2461
2462 /*
2463 * Note the amount of data that peer has sent into
2464 * our window, in order to estimate the sender's
2465 * buffer size.
2466 */
2467 len = so->so_rcv.ssb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2468 } else {
2469 m_freem(m);
2470 thflags &= ~TH_FIN;
2471 }
2472
2473 /*
2474 * If FIN is received ACK the FIN and let the user know
2475 * that the connection is closing.
2476 */
2477 if (thflags & TH_FIN) {
2478 if (!TCPS_HAVERCVDFIN(tp->t_state)) {
2479 socantrcvmore(so);
2480 /*
2481 * If connection is half-synchronized
2482 * (ie NEEDSYN flag on) then delay ACK,
2483 * so it may be piggybacked when SYN is sent.
2484 * Otherwise, since we received a FIN then no
2485 * more input can be expected, send ACK now.
2486 */
2487 if (DELAY_ACK(tp) && (tp->t_flags & TF_NEEDSYN)) {
2488 tcp_callout_reset(tp, tp->tt_delack,
2489 tcp_delacktime, tcp_timer_delack);
2490 } else {
2491 tp->t_flags |= TF_ACKNOW;
2492 }
2493 tp->rcv_nxt++;
2494 }
2495
2496 switch (tp->t_state) {
2497 /*
2498 * In SYN_RECEIVED and ESTABLISHED STATES
2499 * enter the CLOSE_WAIT state.
2500 */
2501 case TCPS_SYN_RECEIVED:
2502 tp->t_starttime = ticks;
2503 /*FALLTHROUGH*/
2504 case TCPS_ESTABLISHED:
2505 TCP_STATE_CHANGE(tp, TCPS_CLOSE_WAIT);
2506 break;
2507
2508 /*
2509 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2510 * enter the CLOSING state.
2511 */
2512 case TCPS_FIN_WAIT_1:
2513 TCP_STATE_CHANGE(tp, TCPS_CLOSING);
2514 break;
2515
2516 /*
2517 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2518 * starting the time-wait timer, turning off the other
2519 * standard timers.
2520 */
2521 case TCPS_FIN_WAIT_2:
2522 TCP_STATE_CHANGE(tp, TCPS_TIME_WAIT);
2523 tcp_canceltimers(tp);
2524 tcp_callout_reset(tp, tp->tt_2msl, 2 * tcp_rmx_msl(tp),
2525 tcp_timer_2msl);
2526 soisdisconnected(so);
2527 break;
2528
2529 /*
2530 * In TIME_WAIT state restart the 2 MSL time_wait timer.
2531 */
2532 case TCPS_TIME_WAIT:
2533 tcp_callout_reset(tp, tp->tt_2msl, 2 * tcp_rmx_msl(tp),
2534 tcp_timer_2msl);
2535 break;
2536 }
2537 }
2538
2539 #ifdef TCPDEBUG
2540 if (so->so_options & SO_DEBUG)
2541 tcp_trace(TA_INPUT, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2542 #endif
2543
2544 /*
2545 * Delayed duplicated ACK processing
2546 */
2547 if (delayed_dupack && tcp_recv_dupack(tp, th_dupack, to_flags))
2548 needoutput = FALSE;
2549
2550 /*
2551 * Return any desired output.
2552 */
2553 if ((tp->t_flags & TF_ACKNOW) ||
2554 (needoutput && tcp_sack_report_needed(tp))) {
2555 tcp_output_cancel(tp);
2556 tcp_output_fair(tp);
2557 } else if (needoutput && !tcp_output_pending(tp)) {
2558 tcp_output_fair(tp);
2559 }
2560 tcp_sack_report_cleanup(tp);
2561 return(IPPROTO_DONE);
2562
2563 dropafterack:
2564 /*
2565 * Generate an ACK dropping incoming segment if it occupies
2566 * sequence space, where the ACK reflects our state.
2567 *
2568 * We can now skip the test for the RST flag since all
2569 * paths to this code happen after packets containing
2570 * RST have been dropped.
2571 *
2572 * In the SYN-RECEIVED state, don't send an ACK unless the
2573 * segment we received passes the SYN-RECEIVED ACK test.
2574 * If it fails send a RST. This breaks the loop in the
2575 * "LAND" DoS attack, and also prevents an ACK storm
2576 * between two listening ports that have been sent forged
2577 * SYN segments, each with the source address of the other.
2578 */
2579 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
2580 (SEQ_GT(tp->snd_una, th->th_ack) ||
2581 SEQ_GT(th->th_ack, tp->snd_max)) ) {
2582 rstreason = BANDLIM_RST_OPENPORT;
2583 goto dropwithreset;
2584 }
2585 #ifdef TCPDEBUG
2586 if (so->so_options & SO_DEBUG)
2587 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2588 #endif
2589 m_freem(m);
2590 tp->t_flags |= TF_ACKNOW;
2591 tcp_output(tp);
2592 tcp_sack_report_cleanup(tp);
2593 return(IPPROTO_DONE);
2594
2595 dropwithreset:
2596 /*
2597 * Generate a RST, dropping incoming segment.
2598 * Make ACK acceptable to originator of segment.
2599 * Don't bother to respond if destination was broadcast/multicast.
2600 */
2601 if ((thflags & TH_RST) || m->m_flags & (M_BCAST | M_MCAST))
2602 goto drop;
2603 if (isipv6) {
2604 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
2605 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
2606 goto drop;
2607 } else {
2608 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
2609 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
2610 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
2611 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2612 goto drop;
2613 }
2614 /* IPv6 anycast check is done at tcp6_input() */
2615
2616 /*
2617 * Perform bandwidth limiting.
2618 */
2619 #ifdef ICMP_BANDLIM
2620 if (badport_bandlim(rstreason) < 0)
2621 goto drop;
2622 #endif
2623
2624 #ifdef TCPDEBUG
2625 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2626 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2627 #endif
2628 if (thflags & TH_ACK)
2629 /* mtod() below is safe as long as hdr dropping is delayed */
2630 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0, th->th_ack,
2631 TH_RST);
2632 else {
2633 if (thflags & TH_SYN)
2634 tlen++;
2635 /* mtod() below is safe as long as hdr dropping is delayed */
2636 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq + tlen,
2637 (tcp_seq)0, TH_RST | TH_ACK);
2638 }
2639 if (tp != NULL)
2640 tcp_sack_report_cleanup(tp);
2641 return(IPPROTO_DONE);
2642
2643 drop:
2644 /*
2645 * Drop space held by incoming segment and return.
2646 */
2647 #ifdef TCPDEBUG
2648 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2649 tcp_trace(TA_DROP, ostate, tp, tcp_saveipgen, &tcp_savetcp, 0);
2650 #endif
2651 m_freem(m);
2652 if (tp != NULL)
2653 tcp_sack_report_cleanup(tp);
2654 return(IPPROTO_DONE);
2655 }
2656
2657 /*
2658 * Parse TCP options and place in tcpopt.
2659 */
2660 static void
tcp_dooptions(struct tcpopt * to,u_char * cp,int cnt,boolean_t is_syn,tcp_seq ack)2661 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, boolean_t is_syn,
2662 tcp_seq ack)
2663 {
2664 int opt, optlen, i;
2665
2666 to->to_flags = 0;
2667 for (; cnt > 0; cnt -= optlen, cp += optlen) {
2668 opt = cp[0];
2669 if (opt == TCPOPT_EOL)
2670 break;
2671 if (opt == TCPOPT_NOP)
2672 optlen = 1;
2673 else {
2674 if (cnt < 2)
2675 break;
2676 optlen = cp[1];
2677 if (optlen < 2 || optlen > cnt)
2678 break;
2679 }
2680 switch (opt) {
2681 case TCPOPT_MAXSEG:
2682 if (optlen != TCPOLEN_MAXSEG)
2683 continue;
2684 if (!is_syn)
2685 continue;
2686 to->to_flags |= TOF_MSS;
2687 bcopy(cp + 2, &to->to_mss, sizeof to->to_mss);
2688 to->to_mss = ntohs(to->to_mss);
2689 break;
2690 case TCPOPT_WINDOW:
2691 if (optlen != TCPOLEN_WINDOW)
2692 continue;
2693 if (!is_syn)
2694 continue;
2695 to->to_flags |= TOF_SCALE;
2696 to->to_requested_s_scale = min(cp[2], TCP_MAX_WINSHIFT);
2697 break;
2698 case TCPOPT_TIMESTAMP:
2699 if (optlen != TCPOLEN_TIMESTAMP)
2700 continue;
2701 to->to_flags |= TOF_TS;
2702 bcopy(cp + 2, &to->to_tsval, sizeof to->to_tsval);
2703 to->to_tsval = ntohl(to->to_tsval);
2704 bcopy(cp + 6, &to->to_tsecr, sizeof to->to_tsecr);
2705 to->to_tsecr = ntohl(to->to_tsecr);
2706 /*
2707 * If echoed timestamp is later than the current time,
2708 * fall back to non RFC1323 RTT calculation.
2709 */
2710 if (to->to_tsecr != 0 && TSTMP_GT(to->to_tsecr, ticks))
2711 to->to_tsecr = 0;
2712 break;
2713 case TCPOPT_SACK_PERMITTED:
2714 if (optlen != TCPOLEN_SACK_PERMITTED)
2715 continue;
2716 if (!is_syn)
2717 continue;
2718 to->to_flags |= TOF_SACK_PERMITTED;
2719 break;
2720 case TCPOPT_SACK:
2721 if ((optlen - 2) & 0x07) /* not multiple of 8 */
2722 continue;
2723 to->to_nsackblocks = (optlen - 2) / 8;
2724 to->to_sackblocks = (struct raw_sackblock *) (cp + 2);
2725 to->to_flags |= TOF_SACK;
2726 for (i = 0; i < to->to_nsackblocks; i++) {
2727 struct raw_sackblock *r = &to->to_sackblocks[i];
2728
2729 r->rblk_start = ntohl(r->rblk_start);
2730 r->rblk_end = ntohl(r->rblk_end);
2731
2732 if (SEQ_LEQ(r->rblk_end, r->rblk_start)) {
2733 /*
2734 * Invalid SACK block; discard all
2735 * SACK blocks
2736 */
2737 tcpstat.tcps_rcvbadsackopt++;
2738 to->to_nsackblocks = 0;
2739 to->to_sackblocks = NULL;
2740 to->to_flags &= ~TOF_SACK;
2741 break;
2742 }
2743 }
2744 if ((to->to_flags & TOF_SACK) &&
2745 tcp_sack_ndsack_blocks(to->to_sackblocks,
2746 to->to_nsackblocks, ack))
2747 to->to_flags |= TOF_DSACK;
2748 break;
2749 #ifdef TCP_SIGNATURE
2750 /*
2751 * XXX In order to reply to a host which has set the
2752 * TCP_SIGNATURE option in its initial SYN, we have to
2753 * record the fact that the option was observed here
2754 * for the syncache code to perform the correct response.
2755 */
2756 case TCPOPT_SIGNATURE:
2757 if (optlen != TCPOLEN_SIGNATURE)
2758 continue;
2759 to->to_flags |= (TOF_SIGNATURE | TOF_SIGLEN);
2760 break;
2761 #endif /* TCP_SIGNATURE */
2762 default:
2763 continue;
2764 }
2765 }
2766 }
2767
2768 /*
2769 * Pull out of band byte out of a segment so
2770 * it doesn't appear in the user's data queue.
2771 * It is still reflected in the segment length for
2772 * sequencing purposes.
2773 * "off" is the delayed to be dropped hdrlen.
2774 */
2775 static void
tcp_pulloutofband(struct socket * so,struct tcphdr * th,struct mbuf * m,int off)2776 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m, int off)
2777 {
2778 int cnt = off + th->th_urp - 1;
2779
2780 while (cnt >= 0) {
2781 if (m->m_len > cnt) {
2782 char *cp = mtod(m, caddr_t) + cnt;
2783 struct tcpcb *tp = sototcpcb(so);
2784
2785 tp->t_iobc = *cp;
2786 tp->t_oobflags |= TCPOOB_HAVEDATA;
2787 bcopy(cp + 1, cp, m->m_len - cnt - 1);
2788 m->m_len--;
2789 if (m->m_flags & M_PKTHDR)
2790 m->m_pkthdr.len--;
2791 return;
2792 }
2793 cnt -= m->m_len;
2794 m = m->m_next;
2795 if (m == NULL)
2796 break;
2797 }
2798 panic("tcp_pulloutofband");
2799 }
2800
2801 /*
2802 * Collect new round-trip time estimate and update averages and current
2803 * timeout.
2804 */
2805 static void
tcp_xmit_timer(struct tcpcb * tp,int rtt,tcp_seq ack)2806 tcp_xmit_timer(struct tcpcb *tp, int rtt, tcp_seq ack)
2807 {
2808 int rebaserto = 0;
2809
2810 tcpstat.tcps_rttupdated++;
2811 tp->t_rttupdated++;
2812 if ((tp->rxt_flags & TRXT_F_REBASERTO) &&
2813 SEQ_GT(ack, tp->snd_max_prev)) {
2814 #ifdef DEBUG_EIFEL_RESPONSE
2815 kprintf("srtt/rttvar, prev %d/%d, cur %d/%d, ",
2816 tp->t_srtt_prev, tp->t_rttvar_prev,
2817 tp->t_srtt, tp->t_rttvar);
2818 #endif
2819
2820 tcpstat.tcps_eifelresponse++;
2821 rebaserto = 1;
2822 tp->rxt_flags &= ~TRXT_F_REBASERTO;
2823 tp->t_srtt = max(tp->t_srtt_prev, (rtt << TCP_RTT_SHIFT));
2824 tp->t_rttvar = max(tp->t_rttvar_prev,
2825 (rtt << (TCP_RTTVAR_SHIFT - 1)));
2826 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
2827 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2828
2829 #ifdef DEBUG_EIFEL_RESPONSE
2830 kprintf("new %d/%d ", tp->t_srtt, tp->t_rttvar);
2831 #endif
2832 } else if (tp->t_srtt != 0) {
2833 int delta;
2834
2835 /*
2836 * srtt is stored as fixed point with 5 bits after the
2837 * binary point (i.e., scaled by 32). The following magic
2838 * is equivalent to the smoothing algorithm in rfc793 with
2839 * an alpha of .875 (srtt = rtt/32 + srtt*31/32 in fixed
2840 * point). Adjust rtt to origin 0.
2841 */
2842 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
2843 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
2844
2845 if ((tp->t_srtt += delta) <= 0)
2846 tp->t_srtt = 1;
2847
2848 /*
2849 * We accumulate a smoothed rtt variance (actually, a
2850 * smoothed mean difference), then set the retransmit
2851 * timer to smoothed rtt + 4 times the smoothed variance.
2852 * rttvar is stored as fixed point with 4 bits after the
2853 * binary point (scaled by 16). The following is
2854 * equivalent to rfc793 smoothing with an alpha of .75
2855 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
2856 * rfc793's wired-in beta.
2857 */
2858 if (delta < 0)
2859 delta = -delta;
2860 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
2861 if ((tp->t_rttvar += delta) <= 0)
2862 tp->t_rttvar = 1;
2863 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
2864 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2865 } else {
2866 /*
2867 * No rtt measurement yet - use the unsmoothed rtt.
2868 * Set the variance to half the rtt (so our first
2869 * retransmit happens at 3*rtt).
2870 */
2871 tp->t_srtt = rtt << TCP_RTT_SHIFT;
2872 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
2873 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2874 }
2875 tp->t_rtttime = 0;
2876 tp->t_rxtshift = 0;
2877
2878 #ifdef DEBUG_EIFEL_RESPONSE
2879 if (rebaserto) {
2880 kprintf("| rxtcur prev %d, old %d, ",
2881 tp->t_rxtcur_prev, tp->t_rxtcur);
2882 }
2883 #endif
2884
2885 /*
2886 * the retransmit should happen at rtt + 4 * rttvar.
2887 * Because of the way we do the smoothing, srtt and rttvar
2888 * will each average +1/2 tick of bias. When we compute
2889 * the retransmit timer, we want 1/2 tick of rounding and
2890 * 1 extra tick because of +-1/2 tick uncertainty in the
2891 * firing of the timer. The bias will give us exactly the
2892 * 1.5 tick we need. But, because the bias is
2893 * statistical, we have to test that we don't drop below
2894 * the minimum feasible timer (which is 2 ticks).
2895 */
2896 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
2897 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
2898
2899 if (rebaserto) {
2900 if (tp->t_rxtcur < tp->t_rxtcur_prev + tcp_eifel_rtoinc) {
2901 /*
2902 * RFC4015 requires that the new RTO is at least
2903 * 2*G (tcp_eifel_rtoinc) greater then the RTO
2904 * (t_rxtcur_prev) when the spurious retransmit
2905 * timeout happens.
2906 *
2907 * The above condition could be true, if the SRTT
2908 * and RTTVAR used to calculate t_rxtcur_prev
2909 * resulted in a value less than t_rttmin. So
2910 * simply increasing SRTT by tcp_eifel_rtoinc when
2911 * preparing for the Eifel response could not ensure
2912 * that the new RTO will be tcp_eifel_rtoinc greater
2913 * t_rxtcur_prev.
2914 */
2915 tp->t_rxtcur = tp->t_rxtcur_prev + tcp_eifel_rtoinc;
2916 }
2917 #ifdef DEBUG_EIFEL_RESPONSE
2918 kprintf("new %d\n", tp->t_rxtcur);
2919 #endif
2920 }
2921
2922 /*
2923 * We received an ack for a packet that wasn't retransmitted;
2924 * it is probably safe to discard any error indications we've
2925 * received recently. This isn't quite right, but close enough
2926 * for now (a route might have failed after we sent a segment,
2927 * and the return path might not be symmetrical).
2928 */
2929 tp->t_softerror = 0;
2930 }
2931
2932 /*
2933 * Determine a reasonable value for maxseg size.
2934 * If the route is known, check route for mtu.
2935 * If none, use an mss that can be handled on the outgoing
2936 * interface without forcing IP to fragment; if bigger than
2937 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2938 * to utilize large mbufs. If no route is found, route has no mtu,
2939 * or the destination isn't local, use a default, hopefully conservative
2940 * size (usually 512 or the default IP max size, but no more than the mtu
2941 * of the interface), as we can't discover anything about intervening
2942 * gateways or networks. We also initialize the congestion/slow start
2943 * window to be a single segment if the destination isn't local.
2944 *
2945 * Also take into account the space needed for options that we
2946 * send regularly. Make maxseg shorter by that amount to assure
2947 * that we can send maxseg amount of data even when the options
2948 * are present. Store the upper limit of the length of options plus
2949 * data in maxopd.
2950 *
2951 * NOTE that this routine is only called when we process an incoming
2952 * segment, for outgoing segments only tcp_mssopt is called.
2953 */
2954 static void
tcp_rmx_mss(struct tcpcb * tp,struct rtentry * rt,int offer)2955 tcp_rmx_mss(struct tcpcb *tp, struct rtentry *rt, int offer)
2956 {
2957 struct ifnet *ifp;
2958 int mss;
2959 u_long bufsize;
2960 struct inpcb *inp = tp->t_inpcb;
2961 struct socket *so;
2962 #ifdef INET6
2963 boolean_t isipv6 = INP_ISIPV6(inp);
2964 size_t min_protoh = isipv6 ?
2965 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
2966 sizeof(struct tcpiphdr);
2967 #else
2968 const boolean_t isipv6 = FALSE;
2969 const size_t min_protoh = sizeof(struct tcpiphdr);
2970 #endif
2971
2972 if (rt == NULL) {
2973 tp->t_maxopd = tp->t_maxseg =
2974 (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
2975 return;
2976 }
2977 ifp = rt->rt_ifp;
2978 so = inp->inp_socket;
2979
2980 /*
2981 * Offer == 0 means that there was no MSS on the SYN segment,
2982 * in this case we use either the interface mtu or tcp_mssdflt.
2983 *
2984 * An offer which is too large will be cut down later.
2985 */
2986 if (offer == 0) {
2987 if (isipv6) {
2988 if (in6_localaddr(&inp->in6p_faddr))
2989 offer = IN6_LINKMTU(rt->rt_ifp) - min_protoh;
2990 else
2991 offer = tcp_v6mssdflt;
2992 } else {
2993 if (in_localaddr(inp->inp_faddr))
2994 offer = ifp->if_mtu - min_protoh;
2995 else
2996 offer = tcp_mssdflt;
2997 }
2998 }
2999
3000 /*
3001 * Prevent DoS attack with too small MSS. Round up
3002 * to at least minmss.
3003 *
3004 * Sanity check: make sure that maxopd will be large
3005 * enough to allow some data on segments even is the
3006 * all the option space is used (40bytes). Otherwise
3007 * funny things may happen in tcp_output.
3008 */
3009 offer = max(offer, tcp_minmss);
3010 offer = max(offer, 64);
3011
3012 rt->rt_rmx.rmx_mssopt = offer;
3013
3014 /*
3015 * if there's an mtu associated with the route, use it
3016 * else, use the link mtu. Take the smaller of mss or offer
3017 * as our final mss.
3018 */
3019 if (rt->rt_rmx.rmx_mtu) {
3020 mss = rt->rt_rmx.rmx_mtu;
3021 } else {
3022 if (isipv6)
3023 mss = IN6_LINKMTU(rt->rt_ifp);
3024 else
3025 mss = ifp->if_mtu;
3026 }
3027 mss -= min_protoh;
3028 mss = min(mss, offer);
3029
3030 /*
3031 * maxopd stores the maximum length of data AND options
3032 * in a segment; maxseg is the amount of data in a normal
3033 * segment. We need to store this value (maxopd) apart
3034 * from maxseg, because now every segment carries options
3035 * and thus we normally have somewhat less data in segments.
3036 */
3037 tp->t_maxopd = mss;
3038
3039 if ((tp->t_flags & (TF_REQ_TSTMP | TF_NOOPT)) == TF_REQ_TSTMP &&
3040 ((tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
3041 mss -= TCPOLEN_TSTAMP_APPA;
3042
3043 #if (MCLBYTES & (MCLBYTES - 1)) == 0
3044 if (mss > MCLBYTES)
3045 mss &= ~(MCLBYTES-1);
3046 #else
3047 if (mss > MCLBYTES)
3048 mss = rounddown(mss, MCLBYTES);
3049 #endif
3050 /*
3051 * If there's a pipesize, change the socket buffer
3052 * to that size. Make the socket buffers an integral
3053 * number of mss units; if the mss is larger than
3054 * the socket buffer, decrease the mss.
3055 */
3056 #ifdef RTV_SPIPE
3057 if ((bufsize = rt->rt_rmx.rmx_sendpipe) == 0)
3058 #endif
3059 bufsize = so->so_snd.ssb_hiwat;
3060 if (bufsize < mss)
3061 mss = bufsize;
3062 else {
3063 bufsize = roundup(bufsize, mss);
3064 if (bufsize > sb_max)
3065 bufsize = sb_max;
3066 if (bufsize > so->so_snd.ssb_hiwat)
3067 ssb_reserve(&so->so_snd, bufsize, so, NULL);
3068 }
3069 tp->t_maxseg = mss;
3070
3071 #ifdef RTV_RPIPE
3072 if ((bufsize = rt->rt_rmx.rmx_recvpipe) == 0)
3073 #endif
3074 bufsize = so->so_rcv.ssb_hiwat;
3075 if (bufsize > mss) {
3076 bufsize = roundup(bufsize, mss);
3077 if (bufsize > sb_max)
3078 bufsize = sb_max;
3079 if (bufsize > so->so_rcv.ssb_hiwat) {
3080 lwkt_gettoken(&so->so_rcv.ssb_token);
3081 ssb_reserve(&so->so_rcv, bufsize, so, NULL);
3082 lwkt_reltoken(&so->so_rcv.ssb_token);
3083 }
3084 }
3085
3086 /*
3087 * Set the slow-start flight size
3088 *
3089 * NOTE: t_maxseg must have been configured!
3090 */
3091 tp->snd_cwnd = tcp_initial_window(tp);
3092
3093 if (rt->rt_rmx.rmx_ssthresh) {
3094 /*
3095 * There's some sort of gateway or interface
3096 * buffer limit on the path. Use this to set
3097 * the slow start threshhold, but set the
3098 * threshold to no less than 2*mss.
3099 */
3100 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
3101 tcpstat.tcps_usedssthresh++;
3102 }
3103 }
3104
3105 static void
tcp_rmx_rtt(struct tcpcb * tp,struct rtentry * rt)3106 tcp_rmx_rtt(struct tcpcb *tp, struct rtentry *rt)
3107 {
3108 int rtt;
3109
3110 if (rt == NULL)
3111 return;
3112
3113 /*
3114 * Check if there's an initial rtt or rttvar. Convert
3115 * from the route-table units to scaled multiples of
3116 * the slow timeout timer.
3117 */
3118 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
3119 /*
3120 * XXX the lock bit for RTT indicates that the value
3121 * is also a minimum value; this is subject to time.
3122 */
3123 if (rt->rt_rmx.rmx_locks & RTV_RTT)
3124 tp->t_rttmin = rtt / (RTM_RTTUNIT / hz);
3125 tp->t_srtt = rtt / (RTM_RTTUNIT / (hz * TCP_RTT_SCALE));
3126 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
3127 tcpstat.tcps_usedrtt++;
3128 if (rt->rt_rmx.rmx_rttvar) {
3129 tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
3130 (RTM_RTTUNIT / (hz * TCP_RTTVAR_SCALE));
3131 tcpstat.tcps_usedrttvar++;
3132 } else {
3133 /* default variation is +- 1 rtt */
3134 tp->t_rttvar =
3135 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
3136 }
3137 TCPT_RANGESET(tp->t_rxtcur,
3138 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
3139 tp->t_rttmin, TCPTV_REXMTMAX);
3140 }
3141 }
3142
3143 void
tcp_rmx_init(struct tcpcb * tp,int offer)3144 tcp_rmx_init(struct tcpcb *tp, int offer)
3145 {
3146 struct inpcb *inp = tp->t_inpcb;
3147 #ifdef INET6
3148 boolean_t isipv6 = INP_ISIPV6(inp);
3149 #else
3150 const boolean_t isipv6 = FALSE;
3151 #endif
3152 struct rtentry *rt;
3153
3154 if (isipv6)
3155 rt = tcp_rtlookup6(&inp->inp_inc);
3156 else
3157 rt = tcp_rtlookup(&inp->inp_inc);
3158
3159 tcp_rmx_mss(tp, rt, offer);
3160 tcp_rmx_rtt(tp, rt);
3161
3162 if (rt != NULL && !tcp_ncr_linklocal && (rt->rt_flags & RTF_LLINFO)) {
3163 /* Don't enable NCR on link-local network. */
3164 tp->t_flags &= ~TF_NCR;
3165 }
3166 }
3167
3168 /*
3169 * Determine the MSS option to send on an outgoing SYN.
3170 */
3171 int
tcp_mssopt(struct tcpcb * tp)3172 tcp_mssopt(struct tcpcb *tp)
3173 {
3174 struct rtentry *rt;
3175 #ifdef INET6
3176 boolean_t isipv6 = INP_ISIPV6(tp->t_inpcb);
3177 int min_protoh = isipv6 ?
3178 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
3179 sizeof(struct tcpiphdr);
3180 #else
3181 const boolean_t isipv6 = FALSE;
3182 const size_t min_protoh = sizeof(struct tcpiphdr);
3183 #endif
3184
3185 if (isipv6)
3186 rt = tcp_rtlookup6(&tp->t_inpcb->inp_inc);
3187 else
3188 rt = tcp_rtlookup(&tp->t_inpcb->inp_inc);
3189 if (rt == NULL)
3190 return (isipv6 ? tcp_v6mssdflt : tcp_mssdflt);
3191
3192 #ifdef INET6
3193 return ((isipv6 ? IN6_LINKMTU(rt->rt_ifp) : rt->rt_ifp->if_mtu) -
3194 min_protoh);
3195 #else
3196 return (rt->rt_ifp->if_mtu - min_protoh);
3197 #endif
3198 }
3199
3200 /*
3201 * When a partial ack arrives, force the retransmission of the
3202 * next unacknowledged segment. Do not exit Fast Recovery.
3203 *
3204 * Implement the Slow-but-Steady variant of NewReno by restarting the
3205 * the retransmission timer. Turn it off here so it can be restarted
3206 * later in tcp_output().
3207 */
3208 static void
tcp_newreno_partial_ack(struct tcpcb * tp,struct tcphdr * th,int acked)3209 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th, int acked)
3210 {
3211 tcp_seq old_snd_nxt = tp->snd_nxt;
3212 u_long ocwnd = tp->snd_cwnd;
3213
3214 tcp_callout_stop(tp, tp->tt_rexmt);
3215 tp->t_rtttime = 0;
3216 tp->snd_nxt = th->th_ack;
3217 /* Set snd_cwnd to one segment beyond acknowledged offset. */
3218 tp->snd_cwnd = tp->t_maxseg;
3219 tp->t_flags |= TF_ACKNOW;
3220 tcp_output(tp);
3221 if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
3222 tp->snd_nxt = old_snd_nxt;
3223 /* partial window deflation */
3224 if (ocwnd > acked)
3225 tp->snd_cwnd = ocwnd - acked + tp->t_maxseg;
3226 else
3227 tp->snd_cwnd = tp->t_maxseg;
3228 }
3229
3230 /*
3231 * In contrast to the Slow-but-Steady NewReno variant,
3232 * we do not reset the retransmission timer for SACK retransmissions,
3233 * except when retransmitting snd_una.
3234 */
3235 static void
tcp_sack_rexmt(struct tcpcb * tp,boolean_t force)3236 tcp_sack_rexmt(struct tcpcb *tp, boolean_t force)
3237 {
3238 tcp_seq old_snd_nxt = tp->snd_nxt;
3239 u_long ocwnd = tp->snd_cwnd;
3240 uint32_t pipe;
3241 int nseg = 0; /* consecutive new segments */
3242 int nseg_rexmt = 0; /* retransmitted segments */
3243 int maxrexmt = 0;
3244
3245 if (force) {
3246 uint32_t unsacked = tcp_sack_first_unsacked_len(tp);
3247
3248 /*
3249 * Try to fill the first hole in the receiver's
3250 * reassemble queue.
3251 */
3252 maxrexmt = howmany(unsacked, tp->t_maxseg);
3253 if (maxrexmt > tcp_force_sackrxt)
3254 maxrexmt = tcp_force_sackrxt;
3255 }
3256
3257 tp->t_rtttime = 0;
3258 pipe = tcp_sack_compute_pipe(tp);
3259 while (((tcp_seq_diff_t)(ocwnd - pipe) >= (tcp_seq_diff_t)tp->t_maxseg
3260 || (force && nseg_rexmt < maxrexmt && nseg == 0)) &&
3261 (!tcp_do_smartsack || nseg < TCP_SACK_MAXBURST)) {
3262 tcp_seq old_snd_max, old_rexmt_high, nextrexmt;
3263 uint32_t sent, seglen;
3264 boolean_t rescue;
3265 int error;
3266
3267 old_rexmt_high = tp->rexmt_high;
3268 if (!tcp_sack_nextseg(tp, &nextrexmt, &seglen, &rescue)) {
3269 tp->rexmt_high = old_rexmt_high;
3270 break;
3271 }
3272
3273 /*
3274 * If the next tranmission is a rescue retranmission,
3275 * we check whether we have already sent some data
3276 * (either new segments or retransmitted segments)
3277 * into the the network or not. Since the idea of rescue
3278 * retransmission is to sustain ACK clock, as long as
3279 * some segments are in the network, ACK clock will be
3280 * kept ticking.
3281 */
3282 if (rescue && (nseg_rexmt > 0 || nseg > 0)) {
3283 tp->rexmt_high = old_rexmt_high;
3284 break;
3285 }
3286
3287 if (nextrexmt == tp->snd_max)
3288 ++nseg;
3289 else
3290 ++nseg_rexmt;
3291 tp->snd_nxt = nextrexmt;
3292 tp->snd_cwnd = nextrexmt - tp->snd_una + seglen;
3293 old_snd_max = tp->snd_max;
3294 if (nextrexmt == tp->snd_una)
3295 tcp_callout_stop(tp, tp->tt_rexmt);
3296 tp->t_flags |= TF_XMITNOW;
3297 error = tcp_output(tp);
3298 if (error != 0) {
3299 tp->rexmt_high = old_rexmt_high;
3300 break;
3301 }
3302 sent = tp->snd_nxt - nextrexmt;
3303 if (sent <= 0) {
3304 tp->rexmt_high = old_rexmt_high;
3305 break;
3306 }
3307 pipe += sent;
3308 tcpstat.tcps_sndsackpack++;
3309 tcpstat.tcps_sndsackbyte += sent;
3310
3311 if (rescue) {
3312 tcpstat.tcps_sackrescue++;
3313 tp->rexmt_rescue = tp->snd_nxt;
3314 tp->sack_flags |= TSACK_F_SACKRESCUED;
3315 break;
3316 }
3317 if (SEQ_LT(nextrexmt, old_snd_max) &&
3318 SEQ_LT(tp->rexmt_high, tp->snd_nxt)) {
3319 tp->rexmt_high = seq_min(tp->snd_nxt, old_snd_max);
3320 if (tcp_aggressive_rescuesack &&
3321 (tp->sack_flags & TSACK_F_SACKRESCUED) &&
3322 SEQ_LT(tp->rexmt_rescue, tp->rexmt_high)) {
3323 /* Drag RescueRxt along with HighRxt */
3324 tp->rexmt_rescue = tp->rexmt_high;
3325 }
3326 }
3327 }
3328 if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
3329 tp->snd_nxt = old_snd_nxt;
3330 tp->snd_cwnd = ocwnd;
3331 }
3332
3333 /*
3334 * Return TRUE, if some new segments are sent
3335 */
3336 static boolean_t
tcp_sack_limitedxmit(struct tcpcb * tp)3337 tcp_sack_limitedxmit(struct tcpcb *tp)
3338 {
3339 tcp_seq oldsndnxt = tp->snd_nxt;
3340 tcp_seq oldsndmax = tp->snd_max;
3341 u_long ocwnd = tp->snd_cwnd;
3342 uint32_t pipe, sent;
3343 boolean_t ret = FALSE;
3344 tcp_seq_diff_t cwnd_left;
3345 tcp_seq next;
3346
3347 tp->rexmt_high = tp->snd_una - 1;
3348 pipe = tcp_sack_compute_pipe(tp);
3349 cwnd_left = (tcp_seq_diff_t)(ocwnd - pipe);
3350 if (cwnd_left < (tcp_seq_diff_t)tp->t_maxseg)
3351 return FALSE;
3352
3353 if (tcp_do_smartsack)
3354 cwnd_left = ulmin(cwnd_left, tp->t_maxseg * TCP_SACK_MAXBURST);
3355
3356 next = tp->snd_nxt = tp->snd_max;
3357 tp->snd_cwnd = tp->snd_nxt - tp->snd_una +
3358 rounddown(cwnd_left, tp->t_maxseg);
3359
3360 tp->t_flags |= TF_XMITNOW;
3361 tcp_output(tp);
3362
3363 sent = tp->snd_nxt - next;
3364 if (sent > 0) {
3365 tcpstat.tcps_sndlimited += howmany(sent, tp->t_maxseg);
3366 ret = TRUE;
3367 }
3368
3369 if (SEQ_LT(oldsndnxt, oldsndmax)) {
3370 KASSERT(SEQ_GEQ(oldsndnxt, tp->snd_una),
3371 ("snd_una moved in other threads"));
3372 tp->snd_nxt = oldsndnxt;
3373 }
3374 tp->snd_cwnd = ocwnd;
3375
3376 if (ret && TCP_DO_NCR(tp))
3377 tcp_ncr_update_rxtthresh(tp);
3378
3379 return ret;
3380 }
3381
3382 /*
3383 * Reset idle time and keep-alive timer, typically called when a valid
3384 * tcp packet is received but may also be called when FASTKEEP is set
3385 * to prevent the previous long-timeout from calculating to a drop.
3386 *
3387 * Only update t_rcvtime for non-SYN packets.
3388 *
3389 * Handle the case where one side thinks the connection is established
3390 * but the other side has, say, rebooted without cleaning out the
3391 * connection. The SYNs could be construed as an attack and wind
3392 * up ignored, but in case it isn't an attack we can validate the
3393 * connection by forcing a keepalive.
3394 */
3395 void
tcp_timer_keep_activity(struct tcpcb * tp,int thflags)3396 tcp_timer_keep_activity(struct tcpcb *tp, int thflags)
3397 {
3398 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
3399 if ((thflags & (TH_SYN | TH_ACK)) == TH_SYN) {
3400 tp->t_flags |= TF_KEEPALIVE;
3401 tcp_callout_reset(tp, tp->tt_keep, hz / 2,
3402 tcp_timer_keep);
3403 } else {
3404 tp->t_rcvtime = ticks;
3405 tp->t_flags &= ~TF_KEEPALIVE;
3406 tcp_callout_reset(tp, tp->tt_keep,
3407 tp->t_keepidle,
3408 tcp_timer_keep);
3409 }
3410 }
3411 }
3412
3413 static int
tcp_rmx_msl(const struct tcpcb * tp)3414 tcp_rmx_msl(const struct tcpcb *tp)
3415 {
3416 struct rtentry *rt;
3417 struct inpcb *inp = tp->t_inpcb;
3418 int msl;
3419 #ifdef INET6
3420 boolean_t isipv6 = INP_ISIPV6(inp);
3421 #else
3422 const boolean_t isipv6 = FALSE;
3423 #endif
3424
3425 if (isipv6)
3426 rt = tcp_rtlookup6(&inp->inp_inc);
3427 else
3428 rt = tcp_rtlookup(&inp->inp_inc);
3429 if (rt == NULL || rt->rt_rmx.rmx_msl == 0)
3430 return tcp_msl;
3431
3432 msl = (rt->rt_rmx.rmx_msl * hz) / 1000;
3433 if (msl == 0)
3434 msl = 1;
3435
3436 return msl;
3437 }
3438
3439 static void
tcp_established(struct tcpcb * tp)3440 tcp_established(struct tcpcb *tp)
3441 {
3442 TCP_STATE_CHANGE(tp, TCPS_ESTABLISHED);
3443 tcp_callout_reset(tp, tp->tt_keep, tp->t_keepidle, tcp_timer_keep);
3444
3445 if (tp->t_rxtsyn > 0) {
3446 /*
3447 * RFC6298:
3448 * "If the timer expires awaiting the ACK of a SYN segment
3449 * and the TCP implementation is using an RTO less than 3
3450 * seconds, the RTO MUST be re-initialized to 3 seconds
3451 * when data transmission begins"
3452 */
3453 if (tp->t_rxtcur < TCPTV_RTOBASE3)
3454 tp->t_rxtcur = TCPTV_RTOBASE3;
3455 }
3456 }
3457
3458 /*
3459 * Returns TRUE, if the ACK should be dropped
3460 */
3461 static boolean_t
tcp_recv_dupack(struct tcpcb * tp,tcp_seq th_ack,u_int to_flags)3462 tcp_recv_dupack(struct tcpcb *tp, tcp_seq th_ack, u_int to_flags)
3463 {
3464 boolean_t fast_sack_rexmt = TRUE;
3465
3466 tcpstat.tcps_rcvdupack++;
3467
3468 /*
3469 * We have outstanding data (other than a window probe),
3470 * this is a completely duplicate ack (ie, window info
3471 * didn't change), the ack is the biggest we've seen and
3472 * we've seen exactly our rexmt threshhold of them, so
3473 * assume a packet has been dropped and retransmit it.
3474 * Kludge snd_nxt & the congestion window so we send only
3475 * this one packet.
3476 */
3477 if (IN_FASTRECOVERY(tp)) {
3478 if (TCP_DO_SACK(tp)) {
3479 boolean_t force = FALSE;
3480
3481 if (tp->snd_una == tp->rexmt_high &&
3482 (to_flags & (TOF_SACK | TOF_SACK_REDUNDANT)) ==
3483 TOF_SACK) {
3484 /*
3485 * New segments got SACKed and
3486 * no retransmit yet.
3487 */
3488 force = TRUE;
3489 }
3490
3491 /* No artifical cwnd inflation. */
3492 tcp_sack_rexmt(tp, force);
3493 } else {
3494 /*
3495 * Dup acks mean that packets have left
3496 * the network (they're now cached at the
3497 * receiver) so bump cwnd by the amount in
3498 * the receiver to keep a constant cwnd
3499 * packets in the network.
3500 */
3501 tp->snd_cwnd += tp->t_maxseg;
3502 tcp_output(tp);
3503 }
3504 return TRUE;
3505 } else if (SEQ_LT(th_ack, tp->snd_recover)) {
3506 tp->t_dupacks = 0;
3507 return FALSE;
3508 } else if (tcp_ignore_redun_dsack && TCP_DO_SACK(tp) &&
3509 (to_flags & (TOF_DSACK | TOF_SACK_REDUNDANT)) ==
3510 (TOF_DSACK | TOF_SACK_REDUNDANT)) {
3511 /*
3512 * If the ACK carries DSACK and other SACK blocks
3513 * carry information that we have already known,
3514 * don't count this ACK as duplicate ACK. This
3515 * prevents spurious early retransmit and fast
3516 * retransmit. This also meets the requirement of
3517 * RFC3042 that new segments should not be sent if
3518 * the SACK blocks do not contain new information
3519 * (XXX we actually loosen the requirment that only
3520 * DSACK is checked here).
3521 *
3522 * This kind of ACKs are usually sent after spurious
3523 * retransmit.
3524 */
3525 /* Do nothing; don't change t_dupacks */
3526 return TRUE;
3527 } else if (tp->t_dupacks == 0 && TCP_DO_NCR(tp)) {
3528 tcp_ncr_update_rxtthresh(tp);
3529 }
3530
3531 if (++tp->t_dupacks == tp->t_rxtthresh) {
3532 tcp_seq old_snd_nxt;
3533 u_int win;
3534
3535 fastretransmit:
3536 if (tcp_do_eifel_detect && (tp->t_flags & TF_RCVD_TSTMP)) {
3537 tcp_save_congestion_state(tp);
3538 tp->rxt_flags |= TRXT_F_FASTREXMT;
3539 }
3540 /*
3541 * We know we're losing at the current window size,
3542 * so do congestion avoidance: set ssthresh to half
3543 * the current window and pull our congestion window
3544 * back to the new ssthresh.
3545 */
3546 win = min(tp->snd_wnd, tp->snd_cwnd) / 2 / tp->t_maxseg;
3547 if (win < 2)
3548 win = 2;
3549 tp->snd_ssthresh = win * tp->t_maxseg;
3550 ENTER_FASTRECOVERY(tp);
3551 tp->snd_recover = tp->snd_max;
3552 tcp_callout_stop(tp, tp->tt_rexmt);
3553 tp->t_rtttime = 0;
3554 old_snd_nxt = tp->snd_nxt;
3555 tp->snd_nxt = th_ack;
3556 if (TCP_DO_SACK(tp)) {
3557 uint32_t rxtlen;
3558
3559 rxtlen = tcp_sack_first_unsacked_len(tp);
3560 if (rxtlen > tp->t_maxseg)
3561 rxtlen = tp->t_maxseg;
3562 tp->snd_cwnd = rxtlen;
3563 } else {
3564 tp->snd_cwnd = tp->t_maxseg;
3565 }
3566 tcp_output(tp);
3567 ++tcpstat.tcps_sndfastrexmit;
3568 tp->snd_cwnd = tp->snd_ssthresh;
3569 tp->rexmt_high = tp->snd_nxt;
3570 tp->sack_flags &= ~TSACK_F_SACKRESCUED;
3571 if (SEQ_GT(old_snd_nxt, tp->snd_nxt))
3572 tp->snd_nxt = old_snd_nxt;
3573 KASSERT(tp->snd_limited <= 2, ("tp->snd_limited too big"));
3574 if (TCP_DO_SACK(tp)) {
3575 if (fast_sack_rexmt)
3576 tcp_sack_rexmt(tp, FALSE);
3577 } else {
3578 tp->snd_cwnd += tp->t_maxseg *
3579 (tp->t_dupacks - tp->snd_limited);
3580 }
3581 } else if ((tcp_do_rfc6675 && TCP_DO_SACK(tp)) || TCP_DO_NCR(tp)) {
3582 /*
3583 * The RFC6675 recommends to reduce the byte threshold,
3584 * and enter fast retransmit if IsLost(snd_una). However,
3585 * if we use IsLost(snd_una) based fast retransmit here,
3586 * segments reordering will cause spurious retransmit. So
3587 * we defer the IsLost(snd_una) based fast retransmit until
3588 * the extended limited transmit can't send any segments and
3589 * early retransmit can't be done.
3590 */
3591 if (tcp_rfc6675_rxt && tcp_do_rfc6675 &&
3592 tcp_sack_islost(&tp->scb, tp->snd_una))
3593 goto fastretransmit;
3594
3595 if (tcp_do_limitedtransmit || TCP_DO_NCR(tp)) {
3596 if (!tcp_sack_limitedxmit(tp)) {
3597 /* outstanding data */
3598 uint32_t ownd = tp->snd_max - tp->snd_una;
3599
3600 if (need_early_retransmit(tp, ownd)) {
3601 ++tcpstat.tcps_sndearlyrexmit;
3602 tp->rxt_flags |= TRXT_F_EARLYREXMT;
3603 goto fastretransmit;
3604 } else if (tcp_do_rfc6675 &&
3605 tcp_sack_islost(&tp->scb, tp->snd_una)) {
3606 fast_sack_rexmt = FALSE;
3607 goto fastretransmit;
3608 }
3609 }
3610 }
3611 } else if (tcp_do_limitedtransmit) {
3612 u_long oldcwnd = tp->snd_cwnd;
3613 tcp_seq oldsndmax = tp->snd_max;
3614 tcp_seq oldsndnxt = tp->snd_nxt;
3615 /* outstanding data */
3616 uint32_t ownd = tp->snd_max - tp->snd_una;
3617 u_int sent;
3618
3619 KASSERT(tp->t_dupacks == 1 || tp->t_dupacks == 2,
3620 ("dupacks not 1 or 2"));
3621 if (tp->t_dupacks == 1)
3622 tp->snd_limited = 0;
3623 tp->snd_nxt = tp->snd_max;
3624 tp->snd_cwnd = ownd +
3625 (tp->t_dupacks - tp->snd_limited) * tp->t_maxseg;
3626 tp->t_flags |= TF_XMITNOW;
3627 tcp_output(tp);
3628
3629 if (SEQ_LT(oldsndnxt, oldsndmax)) {
3630 KASSERT(SEQ_GEQ(oldsndnxt, tp->snd_una),
3631 ("snd_una moved in other threads"));
3632 tp->snd_nxt = oldsndnxt;
3633 }
3634 tp->snd_cwnd = oldcwnd;
3635 sent = tp->snd_max - oldsndmax;
3636 if (sent > tp->t_maxseg) {
3637 KASSERT((tp->t_dupacks == 2 && tp->snd_limited == 0) ||
3638 (sent == tp->t_maxseg + 1 &&
3639 (tp->t_flags & TF_SENTFIN)),
3640 ("sent too much"));
3641 KASSERT(sent <= tp->t_maxseg * 2,
3642 ("sent too many segments"));
3643 tp->snd_limited = 2;
3644 tcpstat.tcps_sndlimited += 2;
3645 } else if (sent > 0) {
3646 ++tp->snd_limited;
3647 ++tcpstat.tcps_sndlimited;
3648 } else if (need_early_retransmit(tp, ownd)) {
3649 ++tcpstat.tcps_sndearlyrexmit;
3650 tp->rxt_flags |= TRXT_F_EARLYREXMT;
3651 goto fastretransmit;
3652 }
3653 }
3654 return TRUE;
3655 }
3656