1 /*-
2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
3 * The Regents of the University of California. All rights reserved.
4 * Copyright (c) 2007-2008,2010
5 * Swinburne University of Technology, Melbourne, Australia.
6 * Copyright (c) 2009-2010 Lawrence Stewart <lstewart@freebsd.org>
7 * Copyright (c) 2010 The FreeBSD Foundation
8 * Copyright (c) 2010-2011 Juniper Networks, Inc.
9 * All rights reserved.
10 *
11 * Portions of this software were developed at the Centre for Advanced Internet
12 * Architectures, Swinburne University of Technology, by Lawrence Stewart,
13 * James Healy and David Hayes, made possible in part by a grant from the Cisco
14 * University Research Program Fund at Community Foundation Silicon Valley.
15 *
16 * Portions of this software were developed at the Centre for Advanced
17 * Internet Architectures, Swinburne University of Technology, Melbourne,
18 * Australia by David Hayes under sponsorship from the FreeBSD Foundation.
19 *
20 * Portions of this software were developed by Robert N. M. Watson under
21 * contract to Juniper Networks, Inc.
22 *
23 * Redistribution and use in source and binary forms, with or without
24 * modification, are permitted provided that the following conditions
25 * are met:
26 * 1. Redistributions of source code must retain the above copyright
27 * notice, this list of conditions and the following disclaimer.
28 * 2. Redistributions in binary form must reproduce the above copyright
29 * notice, this list of conditions and the following disclaimer in the
30 * documentation and/or other materials provided with the distribution.
31 * 4. Neither the name of the University nor the names of its contributors
32 * may be used to endorse or promote products derived from this software
33 * without specific prior written permission.
34 *
35 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
36 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
37 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
38 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
39 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
40 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
41 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
42 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
43 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
44 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
45 * SUCH DAMAGE.
46 *
47 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95
48 */
49
50 #include <sys/cdefs.h>
51 __FBSDID("$FreeBSD: releng/10.3/sys/netinet/tcp_input.c 293710 2016-01-11 23:34:29Z hiren $");
52
53 #include "opt_ipfw.h" /* for ipfw_fwd */
54 #include "opt_inet.h"
55 #include "opt_inet6.h"
56 #include "opt_ipsec.h"
57 #include "opt_kdtrace.h"
58 #include "opt_tcpdebug.h"
59
60 #include <sys/param.h>
61 #include <sys/kernel.h>
62 #include <sys/hhook.h>
63 #include <sys/malloc.h>
64 #include <sys/mbuf.h>
65 #include <sys/proc.h> /* for proc0 declaration */
66 #include <sys/protosw.h>
67 #include <sys/sdt.h>
68 #include <sys/signalvar.h>
69 #include <sys/socket.h>
70 #include <sys/socketvar.h>
71 #include <sys/sysctl.h>
72 #include <sys/syslog.h>
73 #include <sys/systm.h>
74
75 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
76
77 #include <vm/uma.h>
78
79 #include <net/if.h>
80 #include <net/route.h>
81 #include <net/vnet.h>
82
83 #define TCPSTATES /* for logging */
84
85 #include <netinet/cc.h>
86 #include <netinet/in.h>
87 #include <netinet/in_kdtrace.h>
88 #include <netinet/in_pcb.h>
89 #include <netinet/in_systm.h>
90 #include <netinet/in_var.h>
91 #include <netinet/ip.h>
92 #include <netinet/ip_icmp.h> /* required for icmp_var.h */
93 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */
94 #include <netinet/ip_var.h>
95 #include <netinet/ip_options.h>
96 #include <netinet/ip6.h>
97 #include <netinet/icmp6.h>
98 #include <netinet6/in6_pcb.h>
99 #include <netinet6/ip6_var.h>
100 #include <netinet6/nd6.h>
101 #ifdef TCP_RFC7413
102 #include <netinet/tcp_fastopen.h>
103 #endif
104 #include <netinet/tcp_fsm.h>
105 #include <netinet/tcp_seq.h>
106 #include <netinet/tcp_timer.h>
107 #include <netinet/tcp_var.h>
108 #include <netinet6/tcp6_var.h>
109 #include <netinet/tcpip.h>
110 #include <netinet/tcp_syncache.h>
111 #ifdef TCPDEBUG
112 #include <netinet/tcp_debug.h>
113 #endif /* TCPDEBUG */
114 #ifdef TCP_OFFLOAD
115 #include <netinet/tcp_offload.h>
116 #endif
117
118 #ifdef IPSEC
119 #include <netipsec/ipsec.h>
120 #include <netipsec/ipsec6.h>
121 #endif /*IPSEC*/
122
123 #include <machine/in_cksum.h>
124
125 #include <security/mac/mac_framework.h>
126
127 const int tcprexmtthresh = 3;
128
129 int tcp_log_in_vain = 0;
130 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
131 &tcp_log_in_vain, 0,
132 "Log all incoming TCP segments to closed ports");
133
134 VNET_DEFINE(int, blackhole) = 0;
135 #define V_blackhole VNET(blackhole)
136 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW,
137 &VNET_NAME(blackhole), 0,
138 "Do not send RST on segments to closed ports");
139
140 VNET_DEFINE(int, tcp_delack_enabled) = 1;
141 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW,
142 &VNET_NAME(tcp_delack_enabled), 0,
143 "Delay ACK to try and piggyback it onto a data packet");
144
145 VNET_DEFINE(int, drop_synfin) = 0;
146 #define V_drop_synfin VNET(drop_synfin)
147 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW,
148 &VNET_NAME(drop_synfin), 0,
149 "Drop TCP packets with SYN+FIN set");
150
151 VNET_DEFINE(int, tcp_do_rfc6675_pipe) = 0;
152 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_pipe, CTLFLAG_VNET | CTLFLAG_RW,
153 &VNET_NAME(tcp_do_rfc6675_pipe), 0,
154 "Use calculated pipe/in-flight bytes per RFC 6675");
155
156 VNET_DEFINE(int, tcp_do_rfc3042) = 1;
157 #define V_tcp_do_rfc3042 VNET(tcp_do_rfc3042)
158 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_RW,
159 &VNET_NAME(tcp_do_rfc3042), 0,
160 "Enable RFC 3042 (Limited Transmit)");
161
162 VNET_DEFINE(int, tcp_do_rfc3390) = 1;
163 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW,
164 &VNET_NAME(tcp_do_rfc3390), 0,
165 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
166
167 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, experimental, CTLFLAG_RW, 0,
168 "Experimental TCP extensions");
169
170 VNET_DEFINE(int, tcp_do_initcwnd10) = 1;
171 SYSCTL_VNET_INT(_net_inet_tcp_experimental, OID_AUTO, initcwnd10, CTLFLAG_RW,
172 &VNET_NAME(tcp_do_initcwnd10), 0,
173 "Enable RFC 6928 (Increasing initial CWND to 10)");
174
175 VNET_DEFINE(int, tcp_do_rfc3465) = 1;
176 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_RW,
177 &VNET_NAME(tcp_do_rfc3465), 0,
178 "Enable RFC 3465 (Appropriate Byte Counting)");
179
180 VNET_DEFINE(int, tcp_abc_l_var) = 2;
181 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_RW,
182 &VNET_NAME(tcp_abc_l_var), 2,
183 "Cap the max cwnd increment during slow-start to this number of segments");
184
185 static SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, CTLFLAG_RW, 0, "TCP ECN");
186
187 VNET_DEFINE(int, tcp_do_ecn) = 0;
188 SYSCTL_VNET_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_RW,
189 &VNET_NAME(tcp_do_ecn), 0,
190 "TCP ECN support");
191
192 VNET_DEFINE(int, tcp_ecn_maxretries) = 1;
193 SYSCTL_VNET_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_RW,
194 &VNET_NAME(tcp_ecn_maxretries), 0,
195 "Max retries before giving up on ECN");
196
197 VNET_DEFINE(int, tcp_insecure_rst) = 0;
198 #define V_tcp_insecure_rst VNET(tcp_insecure_rst)
199 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_RW,
200 &VNET_NAME(tcp_insecure_rst), 0,
201 "Follow the old (insecure) criteria for accepting RST packets");
202
203 VNET_DEFINE(int, tcp_recvspace) = 1024*64;
204 #define V_tcp_recvspace VNET(tcp_recvspace)
205 SYSCTL_VNET_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
206 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size");
207
208 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1;
209 #define V_tcp_do_autorcvbuf VNET(tcp_do_autorcvbuf)
210 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_RW,
211 &VNET_NAME(tcp_do_autorcvbuf), 0,
212 "Enable automatic receive buffer sizing");
213
214 VNET_DEFINE(int, tcp_autorcvbuf_inc) = 16*1024;
215 #define V_tcp_autorcvbuf_inc VNET(tcp_autorcvbuf_inc)
216 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_RW,
217 &VNET_NAME(tcp_autorcvbuf_inc), 0,
218 "Incrementor step size of automatic receive buffer");
219
220 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024;
221 #define V_tcp_autorcvbuf_max VNET(tcp_autorcvbuf_max)
222 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_RW,
223 &VNET_NAME(tcp_autorcvbuf_max), 0,
224 "Max size of automatic receive buffer");
225
226 VNET_DEFINE(struct inpcbhead, tcb);
227 #define tcb6 tcb /* for KAME src sync over BSD*'s */
228 VNET_DEFINE(struct inpcbinfo, tcbinfo);
229
230 static void tcp_dooptions(struct tcpopt *, u_char *, int, int);
231 static void tcp_do_segment(struct mbuf *, struct tcphdr *,
232 struct socket *, struct tcpcb *, int, int, uint8_t,
233 int);
234 static void tcp_dropwithreset(struct mbuf *, struct tcphdr *,
235 struct tcpcb *, int, int);
236 static void tcp_pulloutofband(struct socket *,
237 struct tcphdr *, struct mbuf *, int);
238 static void tcp_xmit_timer(struct tcpcb *, int);
239 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *);
240 static void inline cc_ack_received(struct tcpcb *tp, struct tcphdr *th,
241 uint16_t type);
242 static void inline cc_conn_init(struct tcpcb *tp);
243 static void inline cc_post_recovery(struct tcpcb *tp, struct tcphdr *th);
244 static void inline hhook_run_tcp_est_in(struct tcpcb *tp,
245 struct tcphdr *th, struct tcpopt *to);
246
247 /*
248 * TCP statistics are stored in an "array" of counter(9)s.
249 */
250 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat);
251 VNET_PCPUSTAT_SYSINIT(tcpstat);
252 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat,
253 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
254
255 #ifdef VIMAGE
256 VNET_PCPUSTAT_SYSUNINIT(tcpstat);
257 #endif /* VIMAGE */
258 /*
259 * Kernel module interface for updating tcpstat. The argument is an index
260 * into tcpstat treated as an array.
261 */
262 void
263 kmod_tcpstat_inc(int statnum)
264 {
265
266 counter_u64_add(VNET(tcpstat)[statnum], 1);
267 }
268
269 /*
270 * Wrapper for the TCP established input helper hook.
271 */
272 static void inline
273 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to)
274 {
275 struct tcp_hhook_data hhook_data;
276
277 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) {
278 hhook_data.tp = tp;
279 hhook_data.th = th;
280 hhook_data.to = to;
281
282 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data,
283 tp->osd);
284 }
285 }
286
287 /*
288 * CC wrapper hook functions
289 */
290 static void inline
291 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t type)
292 {
293 INP_WLOCK_ASSERT(tp->t_inpcb);
294
295 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th);
296 if (tp->snd_cwnd <= tp->snd_wnd)
297 tp->ccv->flags |= CCF_CWND_LIMITED;
298 else
299 tp->ccv->flags &= ~CCF_CWND_LIMITED;
300
301 if (type == CC_ACK) {
302 if (tp->snd_cwnd > tp->snd_ssthresh) {
303 tp->t_bytes_acked += min(tp->ccv->bytes_this_ack,
304 V_tcp_abc_l_var * tp->t_maxseg);
305 if (tp->t_bytes_acked >= tp->snd_cwnd) {
306 tp->t_bytes_acked -= tp->snd_cwnd;
307 tp->ccv->flags |= CCF_ABC_SENTAWND;
308 }
309 } else {
310 tp->ccv->flags &= ~CCF_ABC_SENTAWND;
311 tp->t_bytes_acked = 0;
312 }
313 }
314
315 if (CC_ALGO(tp)->ack_received != NULL) {
316 /* XXXLAS: Find a way to live without this */
317 tp->ccv->curack = th->th_ack;
318 CC_ALGO(tp)->ack_received(tp->ccv, type);
319 }
320 }
321
322 static void inline
323 cc_conn_init(struct tcpcb *tp)
324 {
325 struct hc_metrics_lite metrics;
326 struct inpcb *inp = tp->t_inpcb;
327 int rtt;
328
329 INP_WLOCK_ASSERT(tp->t_inpcb);
330
331 tcp_hc_get(&inp->inp_inc, &metrics);
332
333 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) {
334 tp->t_srtt = rtt;
335 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
336 TCPSTAT_INC(tcps_usedrtt);
337 if (metrics.rmx_rttvar) {
338 tp->t_rttvar = metrics.rmx_rttvar;
339 TCPSTAT_INC(tcps_usedrttvar);
340 } else {
341 /* default variation is +- 1 rtt */
342 tp->t_rttvar =
343 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
344 }
345 TCPT_RANGESET(tp->t_rxtcur,
346 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
347 tp->t_rttmin, TCPTV_REXMTMAX);
348 }
349 if (metrics.rmx_ssthresh) {
350 /*
351 * There's some sort of gateway or interface
352 * buffer limit on the path. Use this to set
353 * the slow start threshhold, but set the
354 * threshold to no less than 2*mss.
355 */
356 tp->snd_ssthresh = max(2 * tp->t_maxseg, metrics.rmx_ssthresh);
357 TCPSTAT_INC(tcps_usedssthresh);
358 }
359
360 /*
361 * Set the initial slow-start flight size.
362 *
363 * RFC5681 Section 3.1 specifies the default conservative values.
364 * RFC3390 specifies slightly more aggressive values.
365 * RFC6928 increases it to ten segments.
366 *
367 * If a SYN or SYN/ACK was lost and retransmitted, we have to
368 * reduce the initial CWND to one segment as congestion is likely
369 * requiring us to be cautious.
370 */
371 if (tp->snd_cwnd == 1)
372 tp->snd_cwnd = tp->t_maxseg; /* SYN(-ACK) lost */
373 else if (V_tcp_do_initcwnd10)
374 tp->snd_cwnd = min(10 * tp->t_maxseg,
375 max(2 * tp->t_maxseg, 14600));
376 else if (V_tcp_do_rfc3390)
377 tp->snd_cwnd = min(4 * tp->t_maxseg,
378 max(2 * tp->t_maxseg, 4380));
379 else {
380 /* Per RFC5681 Section 3.1 */
381 if (tp->t_maxseg > 2190)
382 tp->snd_cwnd = 2 * tp->t_maxseg;
383 else if (tp->t_maxseg > 1095)
384 tp->snd_cwnd = 3 * tp->t_maxseg;
385 else
386 tp->snd_cwnd = 4 * tp->t_maxseg;
387 }
388
389 if (CC_ALGO(tp)->conn_init != NULL)
390 CC_ALGO(tp)->conn_init(tp->ccv);
391 }
392
393 void inline
394 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type)
395 {
396 INP_WLOCK_ASSERT(tp->t_inpcb);
397
398 switch(type) {
399 case CC_NDUPACK:
400 if (!IN_FASTRECOVERY(tp->t_flags)) {
401 tp->snd_recover = tp->snd_max;
402 if (tp->t_flags & TF_ECN_PERMIT)
403 tp->t_flags |= TF_ECN_SND_CWR;
404 }
405 break;
406 case CC_ECN:
407 if (!IN_CONGRECOVERY(tp->t_flags)) {
408 TCPSTAT_INC(tcps_ecn_rcwnd);
409 tp->snd_recover = tp->snd_max;
410 if (tp->t_flags & TF_ECN_PERMIT)
411 tp->t_flags |= TF_ECN_SND_CWR;
412 }
413 break;
414 case CC_RTO:
415 tp->t_dupacks = 0;
416 tp->t_bytes_acked = 0;
417 EXIT_RECOVERY(tp->t_flags);
418 tp->snd_ssthresh = max(2, min(tp->snd_wnd, tp->snd_cwnd) / 2 /
419 tp->t_maxseg) * tp->t_maxseg;
420 tp->snd_cwnd = tp->t_maxseg;
421 break;
422 case CC_RTO_ERR:
423 TCPSTAT_INC(tcps_sndrexmitbad);
424 /* RTO was unnecessary, so reset everything. */
425 tp->snd_cwnd = tp->snd_cwnd_prev;
426 tp->snd_ssthresh = tp->snd_ssthresh_prev;
427 tp->snd_recover = tp->snd_recover_prev;
428 if (tp->t_flags & TF_WASFRECOVERY)
429 ENTER_FASTRECOVERY(tp->t_flags);
430 if (tp->t_flags & TF_WASCRECOVERY)
431 ENTER_CONGRECOVERY(tp->t_flags);
432 tp->snd_nxt = tp->snd_max;
433 tp->t_flags &= ~TF_PREVVALID;
434 tp->t_badrxtwin = 0;
435 break;
436 }
437
438 if (CC_ALGO(tp)->cong_signal != NULL) {
439 if (th != NULL)
440 tp->ccv->curack = th->th_ack;
441 CC_ALGO(tp)->cong_signal(tp->ccv, type);
442 }
443 }
444
445 static void inline
446 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th)
447 {
448 INP_WLOCK_ASSERT(tp->t_inpcb);
449
450 /* XXXLAS: KASSERT that we're in recovery? */
451
452 if (CC_ALGO(tp)->post_recovery != NULL) {
453 tp->ccv->curack = th->th_ack;
454 CC_ALGO(tp)->post_recovery(tp->ccv);
455 }
456 /* XXXLAS: EXIT_RECOVERY ? */
457 tp->t_bytes_acked = 0;
458 }
459
460 #ifdef TCP_SIGNATURE
461 static inline int
462 tcp_signature_verify_input(struct mbuf *m, int off0, int tlen, int optlen,
463 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag)
464 {
465 int ret;
466
467 tcp_fields_to_net(th);
468 ret = tcp_signature_verify(m, off0, tlen, optlen, to, th, tcpbflag);
469 tcp_fields_to_host(th);
470 return (ret);
471 }
472 #endif
473
474 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
475 #ifdef INET6
476 #define ND6_HINT(tp) \
477 do { \
478 if ((tp) && (tp)->t_inpcb && \
479 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \
480 nd6_nud_hint(NULL, NULL, 0); \
481 } while (0)
482 #else
483 #define ND6_HINT(tp)
484 #endif
485
486 /*
487 * Indicate whether this ack should be delayed. We can delay the ack if
488 * - there is no delayed ack timer in progress and
489 * - our last ack wasn't a 0-sized window. We never want to delay
490 * the ack that opens up a 0-sized window and
491 * - delayed acks are enabled or
492 * - this is a half-synchronized T/TCP connection.
493 * - the segment size is not larger than the MSS and LRO wasn't used
494 * for this segment.
495 */
496 #define DELAY_ACK(tp, tlen) \
497 ((!tcp_timer_active(tp, TT_DELACK) && \
498 (tp->t_flags & TF_RXWIN0SENT) == 0) && \
499 (tlen <= tp->t_maxopd) && \
500 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
501
502 /*
503 * TCP input handling is split into multiple parts:
504 * tcp6_input is a thin wrapper around tcp_input for the extended
505 * ip6_protox[] call format in ip6_input
506 * tcp_input handles primary segment validation, inpcb lookup and
507 * SYN processing on listen sockets
508 * tcp_do_segment processes the ACK and text of the segment for
509 * establishing, established and closing connections
510 */
511 #ifdef INET6
512 int
513 tcp6_input(struct mbuf **mp, int *offp, int proto)
514 {
515 struct mbuf *m = *mp;
516 struct in6_ifaddr *ia6;
517
518 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
519
520 /*
521 * draft-itojun-ipv6-tcp-to-anycast
522 * better place to put this in?
523 */
524 ia6 = ip6_getdstifaddr(m);
525 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
526 struct ip6_hdr *ip6;
527
528 ifa_free(&ia6->ia_ifa);
529 ip6 = mtod(m, struct ip6_hdr *);
530 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
531 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
532 return IPPROTO_DONE;
533 }
534 if (ia6)
535 ifa_free(&ia6->ia_ifa);
536
537 tcp_input(m, *offp);
538 return IPPROTO_DONE;
539 }
540 #endif /* INET6 */
541
542 void
543 tcp_input(struct mbuf *m, int off0)
544 {
545 struct tcphdr *th = NULL;
546 struct ip *ip = NULL;
547 struct inpcb *inp = NULL;
548 struct tcpcb *tp = NULL;
549 struct socket *so = NULL;
550 u_char *optp = NULL;
551 int optlen = 0;
552 #ifdef INET
553 int len;
554 #endif
555 int tlen = 0, off;
556 int drop_hdrlen;
557 int thflags;
558 int rstreason = 0; /* For badport_bandlim accounting purposes */
559 #ifdef TCP_SIGNATURE
560 uint8_t sig_checked = 0;
561 #endif
562 uint8_t iptos = 0;
563 struct m_tag *fwd_tag = NULL;
564 #ifdef INET6
565 struct ip6_hdr *ip6 = NULL;
566 int isipv6;
567 #else
568 const void *ip6 = NULL;
569 #endif /* INET6 */
570 struct tcpopt to; /* options in this segment */
571 char *s = NULL; /* address and port logging */
572 int ti_locked;
573 #define TI_UNLOCKED 1
574 #define TI_WLOCKED 2
575
576 #ifdef TCPDEBUG
577 /*
578 * The size of tcp_saveipgen must be the size of the max ip header,
579 * now IPv6.
580 */
581 u_char tcp_saveipgen[IP6_HDR_LEN];
582 struct tcphdr tcp_savetcp;
583 short ostate = 0;
584 #endif
585
586 #ifdef INET6
587 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
588 #endif
589
590 to.to_flags = 0;
591 TCPSTAT_INC(tcps_rcvtotal);
592
593 #ifdef INET6
594 if (isipv6) {
595 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */
596
597 if (m->m_len < (sizeof(*ip6) + sizeof(*th))) {
598 m = m_pullup(m, sizeof(*ip6) + sizeof(*th));
599 if (m == NULL) {
600 TCPSTAT_INC(tcps_rcvshort);
601 return;
602 }
603 }
604
605 ip6 = mtod(m, struct ip6_hdr *);
606 th = (struct tcphdr *)((caddr_t)ip6 + off0);
607 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
608 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) {
609 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
610 th->th_sum = m->m_pkthdr.csum_data;
611 else
612 th->th_sum = in6_cksum_pseudo(ip6, tlen,
613 IPPROTO_TCP, m->m_pkthdr.csum_data);
614 th->th_sum ^= 0xffff;
615 } else
616 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen);
617 if (th->th_sum) {
618 TCPSTAT_INC(tcps_rcvbadsum);
619 goto drop;
620 }
621
622 /*
623 * Be proactive about unspecified IPv6 address in source.
624 * As we use all-zero to indicate unbounded/unconnected pcb,
625 * unspecified IPv6 address can be used to confuse us.
626 *
627 * Note that packets with unspecified IPv6 destination is
628 * already dropped in ip6_input.
629 */
630 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
631 /* XXX stat */
632 goto drop;
633 }
634 }
635 #endif
636 #if defined(INET) && defined(INET6)
637 else
638 #endif
639 #ifdef INET
640 {
641 /*
642 * Get IP and TCP header together in first mbuf.
643 * Note: IP leaves IP header in first mbuf.
644 */
645 if (off0 > sizeof (struct ip)) {
646 ip_stripoptions(m);
647 off0 = sizeof(struct ip);
648 }
649 if (m->m_len < sizeof (struct tcpiphdr)) {
650 if ((m = m_pullup(m, sizeof (struct tcpiphdr)))
651 == NULL) {
652 TCPSTAT_INC(tcps_rcvshort);
653 return;
654 }
655 }
656 ip = mtod(m, struct ip *);
657 th = (struct tcphdr *)((caddr_t)ip + off0);
658 tlen = ntohs(ip->ip_len) - off0;
659
660 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
661 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
662 th->th_sum = m->m_pkthdr.csum_data;
663 else
664 th->th_sum = in_pseudo(ip->ip_src.s_addr,
665 ip->ip_dst.s_addr,
666 htonl(m->m_pkthdr.csum_data + tlen +
667 IPPROTO_TCP));
668 th->th_sum ^= 0xffff;
669 } else {
670 struct ipovly *ipov = (struct ipovly *)ip;
671
672 /*
673 * Checksum extended TCP header and data.
674 */
675 len = off0 + tlen;
676 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
677 ipov->ih_len = htons(tlen);
678 th->th_sum = in_cksum(m, len);
679 /* Reset length for SDT probes. */
680 ip->ip_len = htons(tlen + off0);
681 }
682
683 if (th->th_sum) {
684 TCPSTAT_INC(tcps_rcvbadsum);
685 goto drop;
686 }
687 /* Re-initialization for later version check */
688 ip->ip_v = IPVERSION;
689 }
690 #endif /* INET */
691
692 #ifdef INET6
693 if (isipv6)
694 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
695 #endif
696 #if defined(INET) && defined(INET6)
697 else
698 #endif
699 #ifdef INET
700 iptos = ip->ip_tos;
701 #endif
702
703 /*
704 * Check that TCP offset makes sense,
705 * pull out TCP options and adjust length. XXX
706 */
707 off = th->th_off << 2;
708 if (off < sizeof (struct tcphdr) || off > tlen) {
709 TCPSTAT_INC(tcps_rcvbadoff);
710 goto drop;
711 }
712 tlen -= off; /* tlen is used instead of ti->ti_len */
713 if (off > sizeof (struct tcphdr)) {
714 #ifdef INET6
715 if (isipv6) {
716 IP6_EXTHDR_CHECK(m, off0, off, );
717 ip6 = mtod(m, struct ip6_hdr *);
718 th = (struct tcphdr *)((caddr_t)ip6 + off0);
719 }
720 #endif
721 #if defined(INET) && defined(INET6)
722 else
723 #endif
724 #ifdef INET
725 {
726 if (m->m_len < sizeof(struct ip) + off) {
727 if ((m = m_pullup(m, sizeof (struct ip) + off))
728 == NULL) {
729 TCPSTAT_INC(tcps_rcvshort);
730 return;
731 }
732 ip = mtod(m, struct ip *);
733 th = (struct tcphdr *)((caddr_t)ip + off0);
734 }
735 }
736 #endif
737 optlen = off - sizeof (struct tcphdr);
738 optp = (u_char *)(th + 1);
739 }
740 thflags = th->th_flags;
741
742 /*
743 * Convert TCP protocol specific fields to host format.
744 */
745 tcp_fields_to_host(th);
746
747 /*
748 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options.
749 */
750 drop_hdrlen = off0 + off;
751
752 /*
753 * Locate pcb for segment; if we're likely to add or remove a
754 * connection then first acquire pcbinfo lock. There are two cases
755 * where we might discover later we need a write lock despite the
756 * flags: ACKs moving a connection out of the syncache, and ACKs for
757 * a connection in TIMEWAIT.
758 */
759 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0) {
760 INP_INFO_WLOCK(&V_tcbinfo);
761 ti_locked = TI_WLOCKED;
762 } else
763 ti_locked = TI_UNLOCKED;
764
765 /*
766 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
767 */
768 if (
769 #ifdef INET6
770 (isipv6 && (m->m_flags & M_IP6_NEXTHOP))
771 #ifdef INET
772 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP))
773 #endif
774 #endif
775 #if defined(INET) && !defined(INET6)
776 (m->m_flags & M_IP_NEXTHOP)
777 #endif
778 )
779 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
780
781 findpcb:
782 #ifdef INVARIANTS
783 if (ti_locked == TI_WLOCKED) {
784 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
785 } else {
786 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
787 }
788 #endif
789 #ifdef INET6
790 if (isipv6 && fwd_tag != NULL) {
791 struct sockaddr_in6 *next_hop6;
792
793 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1);
794 /*
795 * Transparently forwarded. Pretend to be the destination.
796 * Already got one like this?
797 */
798 inp = in6_pcblookup_mbuf(&V_tcbinfo,
799 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport,
800 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif, m);
801 if (!inp) {
802 /*
803 * It's new. Try to find the ambushing socket.
804 * Because we've rewritten the destination address,
805 * any hardware-generated hash is ignored.
806 */
807 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src,
808 th->th_sport, &next_hop6->sin6_addr,
809 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) :
810 th->th_dport, INPLOOKUP_WILDCARD |
811 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
812 }
813 } else if (isipv6) {
814 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src,
815 th->th_sport, &ip6->ip6_dst, th->th_dport,
816 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
817 m->m_pkthdr.rcvif, m);
818 }
819 #endif /* INET6 */
820 #if defined(INET6) && defined(INET)
821 else
822 #endif
823 #ifdef INET
824 if (fwd_tag != NULL) {
825 struct sockaddr_in *next_hop;
826
827 next_hop = (struct sockaddr_in *)(fwd_tag+1);
828 /*
829 * Transparently forwarded. Pretend to be the destination.
830 * already got one like this?
831 */
832 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport,
833 ip->ip_dst, th->th_dport, INPLOOKUP_WLOCKPCB,
834 m->m_pkthdr.rcvif, m);
835 if (!inp) {
836 /*
837 * It's new. Try to find the ambushing socket.
838 * Because we've rewritten the destination address,
839 * any hardware-generated hash is ignored.
840 */
841 inp = in_pcblookup(&V_tcbinfo, ip->ip_src,
842 th->th_sport, next_hop->sin_addr,
843 next_hop->sin_port ? ntohs(next_hop->sin_port) :
844 th->th_dport, INPLOOKUP_WILDCARD |
845 INPLOOKUP_WLOCKPCB, m->m_pkthdr.rcvif);
846 }
847 } else
848 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src,
849 th->th_sport, ip->ip_dst, th->th_dport,
850 INPLOOKUP_WILDCARD | INPLOOKUP_WLOCKPCB,
851 m->m_pkthdr.rcvif, m);
852 #endif /* INET */
853
854 /*
855 * If the INPCB does not exist then all data in the incoming
856 * segment is discarded and an appropriate RST is sent back.
857 * XXX MRT Send RST using which routing table?
858 */
859 if (inp == NULL) {
860 /*
861 * Log communication attempts to ports that are not
862 * in use.
863 */
864 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) ||
865 tcp_log_in_vain == 2) {
866 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6)))
867 log(LOG_INFO, "%s; %s: Connection attempt "
868 "to closed port\n", s, __func__);
869 }
870 /*
871 * When blackholing do not respond with a RST but
872 * completely ignore the segment and drop it.
873 */
874 if ((V_blackhole == 1 && (thflags & TH_SYN)) ||
875 V_blackhole == 2)
876 goto dropunlock;
877
878 rstreason = BANDLIM_RST_CLOSEDPORT;
879 goto dropwithreset;
880 }
881 INP_WLOCK_ASSERT(inp);
882 if ((inp->inp_flowtype == M_HASHTYPE_NONE) &&
883 (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) &&
884 ((inp->inp_socket == NULL) ||
885 (inp->inp_socket->so_options & SO_ACCEPTCONN) == 0)) {
886 inp->inp_flowid = m->m_pkthdr.flowid;
887 inp->inp_flowtype = M_HASHTYPE_GET(m);
888 }
889 #ifdef IPSEC
890 #ifdef INET6
891 if (isipv6 && ipsec6_in_reject(m, inp)) {
892 IPSEC6STAT_INC(ips_in_polvio);
893 goto dropunlock;
894 } else
895 #endif /* INET6 */
896 if (ipsec4_in_reject(m, inp) != 0) {
897 IPSECSTAT_INC(ips_in_polvio);
898 goto dropunlock;
899 }
900 #endif /* IPSEC */
901
902 /*
903 * Check the minimum TTL for socket.
904 */
905 if (inp->inp_ip_minttl != 0) {
906 #ifdef INET6
907 if (isipv6 && inp->inp_ip_minttl > ip6->ip6_hlim)
908 goto dropunlock;
909 else
910 #endif
911 if (inp->inp_ip_minttl > ip->ip_ttl)
912 goto dropunlock;
913 }
914
915 /*
916 * A previous connection in TIMEWAIT state is supposed to catch stray
917 * or duplicate segments arriving late. If this segment was a
918 * legitimate new connection attempt, the old INPCB gets removed and
919 * we can try again to find a listening socket.
920 *
921 * At this point, due to earlier optimism, we may hold only an inpcb
922 * lock, and not the inpcbinfo write lock. If so, we need to try to
923 * acquire it, or if that fails, acquire a reference on the inpcb,
924 * drop all locks, acquire a global write lock, and then re-acquire
925 * the inpcb lock. We may at that point discover that another thread
926 * has tried to free the inpcb, in which case we need to loop back
927 * and try to find a new inpcb to deliver to.
928 *
929 * XXXRW: It may be time to rethink timewait locking.
930 */
931 relocked:
932 if (inp->inp_flags & INP_TIMEWAIT) {
933 if (ti_locked == TI_UNLOCKED) {
934 if (INP_INFO_TRY_WLOCK(&V_tcbinfo) == 0) {
935 in_pcbref(inp);
936 INP_WUNLOCK(inp);
937 INP_INFO_WLOCK(&V_tcbinfo);
938 ti_locked = TI_WLOCKED;
939 INP_WLOCK(inp);
940 if (in_pcbrele_wlocked(inp)) {
941 inp = NULL;
942 goto findpcb;
943 }
944 } else
945 ti_locked = TI_WLOCKED;
946 }
947 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
948
949 if (thflags & TH_SYN)
950 tcp_dooptions(&to, optp, optlen, TO_SYN);
951 /*
952 * NB: tcp_twcheck unlocks the INP and frees the mbuf.
953 */
954 if (tcp_twcheck(inp, &to, th, m, tlen))
955 goto findpcb;
956 INP_INFO_WUNLOCK(&V_tcbinfo);
957 return;
958 }
959 /*
960 * The TCPCB may no longer exist if the connection is winding
961 * down or it is in the CLOSED state. Either way we drop the
962 * segment and send an appropriate response.
963 */
964 tp = intotcpcb(inp);
965 if (tp == NULL || tp->t_state == TCPS_CLOSED) {
966 rstreason = BANDLIM_RST_CLOSEDPORT;
967 goto dropwithreset;
968 }
969
970 #ifdef TCP_OFFLOAD
971 if (tp->t_flags & TF_TOE) {
972 tcp_offload_input(tp, m);
973 m = NULL; /* consumed by the TOE driver */
974 goto dropunlock;
975 }
976 #endif
977
978 /*
979 * We've identified a valid inpcb, but it could be that we need an
980 * inpcbinfo write lock but don't hold it. In this case, attempt to
981 * acquire using the same strategy as the TIMEWAIT case above. If we
982 * relock, we have to jump back to 'relocked' as the connection might
983 * now be in TIMEWAIT.
984 */
985 #ifdef INVARIANTS
986 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0)
987 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
988 #endif
989 if (tp->t_state != TCPS_ESTABLISHED) {
990 if (ti_locked == TI_UNLOCKED) {
991 if (INP_INFO_TRY_WLOCK(&V_tcbinfo) == 0) {
992 in_pcbref(inp);
993 INP_WUNLOCK(inp);
994 INP_INFO_WLOCK(&V_tcbinfo);
995 ti_locked = TI_WLOCKED;
996 INP_WLOCK(inp);
997 if (in_pcbrele_wlocked(inp)) {
998 inp = NULL;
999 goto findpcb;
1000 }
1001 goto relocked;
1002 } else
1003 ti_locked = TI_WLOCKED;
1004 }
1005 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1006 }
1007
1008 #ifdef MAC
1009 INP_WLOCK_ASSERT(inp);
1010 if (mac_inpcb_check_deliver(inp, m))
1011 goto dropunlock;
1012 #endif
1013 so = inp->inp_socket;
1014 KASSERT(so != NULL, ("%s: so == NULL", __func__));
1015 #ifdef TCPDEBUG
1016 if (so->so_options & SO_DEBUG) {
1017 ostate = tp->t_state;
1018 #ifdef INET6
1019 if (isipv6) {
1020 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
1021 } else
1022 #endif
1023 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
1024 tcp_savetcp = *th;
1025 }
1026 #endif /* TCPDEBUG */
1027 /*
1028 * When the socket is accepting connections (the INPCB is in LISTEN
1029 * state) we look into the SYN cache if this is a new connection
1030 * attempt or the completion of a previous one. Because listen
1031 * sockets are never in TCPS_ESTABLISHED, the V_tcbinfo lock will be
1032 * held in this case.
1033 */
1034 if (so->so_options & SO_ACCEPTCONN) {
1035 struct in_conninfo inc;
1036
1037 KASSERT(tp->t_state == TCPS_LISTEN, ("%s: so accepting but "
1038 "tp not listening", __func__));
1039 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1040
1041 bzero(&inc, sizeof(inc));
1042 #ifdef INET6
1043 if (isipv6) {
1044 inc.inc_flags |= INC_ISIPV6;
1045 inc.inc6_faddr = ip6->ip6_src;
1046 inc.inc6_laddr = ip6->ip6_dst;
1047 } else
1048 #endif
1049 {
1050 inc.inc_faddr = ip->ip_src;
1051 inc.inc_laddr = ip->ip_dst;
1052 }
1053 inc.inc_fport = th->th_sport;
1054 inc.inc_lport = th->th_dport;
1055 inc.inc_fibnum = so->so_fibnum;
1056
1057 /*
1058 * Check for an existing connection attempt in syncache if
1059 * the flag is only ACK. A successful lookup creates a new
1060 * socket appended to the listen queue in SYN_RECEIVED state.
1061 */
1062 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
1063 /*
1064 * Parse the TCP options here because
1065 * syncookies need access to the reflected
1066 * timestamp.
1067 */
1068 tcp_dooptions(&to, optp, optlen, 0);
1069 /*
1070 * NB: syncache_expand() doesn't unlock
1071 * inp and tcpinfo locks.
1072 */
1073 if (!syncache_expand(&inc, &to, th, &so, m)) {
1074 /*
1075 * No syncache entry or ACK was not
1076 * for our SYN/ACK. Send a RST.
1077 * NB: syncache did its own logging
1078 * of the failure cause.
1079 */
1080 rstreason = BANDLIM_RST_OPENPORT;
1081 goto dropwithreset;
1082 }
1083 #ifdef TCP_RFC7413
1084 new_tfo_socket:
1085 #endif
1086 if (so == NULL) {
1087 /*
1088 * We completed the 3-way handshake
1089 * but could not allocate a socket
1090 * either due to memory shortage,
1091 * listen queue length limits or
1092 * global socket limits. Send RST
1093 * or wait and have the remote end
1094 * retransmit the ACK for another
1095 * try.
1096 */
1097 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1098 log(LOG_DEBUG, "%s; %s: Listen socket: "
1099 "Socket allocation failed due to "
1100 "limits or memory shortage, %s\n",
1101 s, __func__,
1102 V_tcp_sc_rst_sock_fail ?
1103 "sending RST" : "try again");
1104 if (V_tcp_sc_rst_sock_fail) {
1105 rstreason = BANDLIM_UNLIMITED;
1106 goto dropwithreset;
1107 } else
1108 goto dropunlock;
1109 }
1110 /*
1111 * Socket is created in state SYN_RECEIVED.
1112 * Unlock the listen socket, lock the newly
1113 * created socket and update the tp variable.
1114 */
1115 INP_WUNLOCK(inp); /* listen socket */
1116 inp = sotoinpcb(so);
1117 INP_WLOCK(inp); /* new connection */
1118 tp = intotcpcb(inp);
1119 KASSERT(tp->t_state == TCPS_SYN_RECEIVED,
1120 ("%s: ", __func__));
1121 #ifdef TCP_SIGNATURE
1122 if (sig_checked == 0) {
1123 tcp_dooptions(&to, optp, optlen,
1124 (thflags & TH_SYN) ? TO_SYN : 0);
1125 if (!tcp_signature_verify_input(m, off0, tlen,
1126 optlen, &to, th, tp->t_flags)) {
1127
1128 /*
1129 * In SYN_SENT state if it receives an
1130 * RST, it is allowed for further
1131 * processing.
1132 */
1133 if ((thflags & TH_RST) == 0 ||
1134 (tp->t_state == TCPS_SYN_SENT) == 0)
1135 goto dropunlock;
1136 }
1137 sig_checked = 1;
1138 }
1139 #endif
1140
1141 /*
1142 * Process the segment and the data it
1143 * contains. tcp_do_segment() consumes
1144 * the mbuf chain and unlocks the inpcb.
1145 */
1146 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen,
1147 iptos, ti_locked);
1148 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1149 return;
1150 }
1151 /*
1152 * Segment flag validation for new connection attempts:
1153 *
1154 * Our (SYN|ACK) response was rejected.
1155 * Check with syncache and remove entry to prevent
1156 * retransmits.
1157 *
1158 * NB: syncache_chkrst does its own logging of failure
1159 * causes.
1160 */
1161 if (thflags & TH_RST) {
1162 syncache_chkrst(&inc, th);
1163 goto dropunlock;
1164 }
1165 /*
1166 * We can't do anything without SYN.
1167 */
1168 if ((thflags & TH_SYN) == 0) {
1169 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1170 log(LOG_DEBUG, "%s; %s: Listen socket: "
1171 "SYN is missing, segment ignored\n",
1172 s, __func__);
1173 TCPSTAT_INC(tcps_badsyn);
1174 goto dropunlock;
1175 }
1176 /*
1177 * (SYN|ACK) is bogus on a listen socket.
1178 */
1179 if (thflags & TH_ACK) {
1180 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1181 log(LOG_DEBUG, "%s; %s: Listen socket: "
1182 "SYN|ACK invalid, segment rejected\n",
1183 s, __func__);
1184 syncache_badack(&inc); /* XXX: Not needed! */
1185 TCPSTAT_INC(tcps_badsyn);
1186 rstreason = BANDLIM_RST_OPENPORT;
1187 goto dropwithreset;
1188 }
1189 /*
1190 * If the drop_synfin option is enabled, drop all
1191 * segments with both the SYN and FIN bits set.
1192 * This prevents e.g. nmap from identifying the
1193 * TCP/IP stack.
1194 * XXX: Poor reasoning. nmap has other methods
1195 * and is constantly refining its stack detection
1196 * strategies.
1197 * XXX: This is a violation of the TCP specification
1198 * and was used by RFC1644.
1199 */
1200 if ((thflags & TH_FIN) && V_drop_synfin) {
1201 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1202 log(LOG_DEBUG, "%s; %s: Listen socket: "
1203 "SYN|FIN segment ignored (based on "
1204 "sysctl setting)\n", s, __func__);
1205 TCPSTAT_INC(tcps_badsyn);
1206 goto dropunlock;
1207 }
1208 /*
1209 * Segment's flags are (SYN) or (SYN|FIN).
1210 *
1211 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored
1212 * as they do not affect the state of the TCP FSM.
1213 * The data pointed to by TH_URG and th_urp is ignored.
1214 */
1215 KASSERT((thflags & (TH_RST|TH_ACK)) == 0,
1216 ("%s: Listen socket: TH_RST or TH_ACK set", __func__));
1217 KASSERT(thflags & (TH_SYN),
1218 ("%s: Listen socket: TH_SYN not set", __func__));
1219 #ifdef INET6
1220 /*
1221 * If deprecated address is forbidden,
1222 * we do not accept SYN to deprecated interface
1223 * address to prevent any new inbound connection from
1224 * getting established.
1225 * When we do not accept SYN, we send a TCP RST,
1226 * with deprecated source address (instead of dropping
1227 * it). We compromise it as it is much better for peer
1228 * to send a RST, and RST will be the final packet
1229 * for the exchange.
1230 *
1231 * If we do not forbid deprecated addresses, we accept
1232 * the SYN packet. RFC2462 does not suggest dropping
1233 * SYN in this case.
1234 * If we decipher RFC2462 5.5.4, it says like this:
1235 * 1. use of deprecated addr with existing
1236 * communication is okay - "SHOULD continue to be
1237 * used"
1238 * 2. use of it with new communication:
1239 * (2a) "SHOULD NOT be used if alternate address
1240 * with sufficient scope is available"
1241 * (2b) nothing mentioned otherwise.
1242 * Here we fall into (2b) case as we have no choice in
1243 * our source address selection - we must obey the peer.
1244 *
1245 * The wording in RFC2462 is confusing, and there are
1246 * multiple description text for deprecated address
1247 * handling - worse, they are not exactly the same.
1248 * I believe 5.5.4 is the best one, so we follow 5.5.4.
1249 */
1250 if (isipv6 && !V_ip6_use_deprecated) {
1251 struct in6_ifaddr *ia6;
1252
1253 ia6 = ip6_getdstifaddr(m);
1254 if (ia6 != NULL &&
1255 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1256 ifa_free(&ia6->ia_ifa);
1257 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1258 log(LOG_DEBUG, "%s; %s: Listen socket: "
1259 "Connection attempt to deprecated "
1260 "IPv6 address rejected\n",
1261 s, __func__);
1262 rstreason = BANDLIM_RST_OPENPORT;
1263 goto dropwithreset;
1264 }
1265 if (ia6)
1266 ifa_free(&ia6->ia_ifa);
1267 }
1268 #endif /* INET6 */
1269 /*
1270 * Basic sanity checks on incoming SYN requests:
1271 * Don't respond if the destination is a link layer
1272 * broadcast according to RFC1122 4.2.3.10, p. 104.
1273 * If it is from this socket it must be forged.
1274 * Don't respond if the source or destination is a
1275 * global or subnet broad- or multicast address.
1276 * Note that it is quite possible to receive unicast
1277 * link-layer packets with a broadcast IP address. Use
1278 * in_broadcast() to find them.
1279 */
1280 if (m->m_flags & (M_BCAST|M_MCAST)) {
1281 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1282 log(LOG_DEBUG, "%s; %s: Listen socket: "
1283 "Connection attempt from broad- or multicast "
1284 "link layer address ignored\n", s, __func__);
1285 goto dropunlock;
1286 }
1287 #ifdef INET6
1288 if (isipv6) {
1289 if (th->th_dport == th->th_sport &&
1290 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) {
1291 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1292 log(LOG_DEBUG, "%s; %s: Listen socket: "
1293 "Connection attempt to/from self "
1294 "ignored\n", s, __func__);
1295 goto dropunlock;
1296 }
1297 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1298 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) {
1299 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1300 log(LOG_DEBUG, "%s; %s: Listen socket: "
1301 "Connection attempt from/to multicast "
1302 "address ignored\n", s, __func__);
1303 goto dropunlock;
1304 }
1305 }
1306 #endif
1307 #if defined(INET) && defined(INET6)
1308 else
1309 #endif
1310 #ifdef INET
1311 {
1312 if (th->th_dport == th->th_sport &&
1313 ip->ip_dst.s_addr == ip->ip_src.s_addr) {
1314 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1315 log(LOG_DEBUG, "%s; %s: Listen socket: "
1316 "Connection attempt from/to self "
1317 "ignored\n", s, __func__);
1318 goto dropunlock;
1319 }
1320 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1321 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1322 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1323 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
1324 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1325 log(LOG_DEBUG, "%s; %s: Listen socket: "
1326 "Connection attempt from/to broad- "
1327 "or multicast address ignored\n",
1328 s, __func__);
1329 goto dropunlock;
1330 }
1331 }
1332 #endif
1333 /*
1334 * SYN appears to be valid. Create compressed TCP state
1335 * for syncache.
1336 */
1337 #ifdef TCPDEBUG
1338 if (so->so_options & SO_DEBUG)
1339 tcp_trace(TA_INPUT, ostate, tp,
1340 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1341 #endif
1342 tcp_dooptions(&to, optp, optlen, TO_SYN);
1343 #ifdef TCP_RFC7413
1344 if (syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL))
1345 goto new_tfo_socket;
1346 #else
1347 syncache_add(&inc, &to, th, inp, &so, m, NULL, NULL);
1348 #endif
1349 /*
1350 * Entry added to syncache and mbuf consumed.
1351 * Everything already unlocked by syncache_add().
1352 */
1353 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1354 return;
1355 } else if (tp->t_state == TCPS_LISTEN) {
1356 /*
1357 * When a listen socket is torn down the SO_ACCEPTCONN
1358 * flag is removed first while connections are drained
1359 * from the accept queue in a unlock/lock cycle of the
1360 * ACCEPT_LOCK, opening a race condition allowing a SYN
1361 * attempt go through unhandled.
1362 */
1363 goto dropunlock;
1364 }
1365
1366 #ifdef TCP_SIGNATURE
1367 if (sig_checked == 0) {
1368 tcp_dooptions(&to, optp, optlen,
1369 (thflags & TH_SYN) ? TO_SYN : 0);
1370 if (!tcp_signature_verify_input(m, off0, tlen, optlen, &to,
1371 th, tp->t_flags)) {
1372
1373 /*
1374 * In SYN_SENT state if it receives an RST, it is
1375 * allowed for further processing.
1376 */
1377 if ((thflags & TH_RST) == 0 ||
1378 (tp->t_state == TCPS_SYN_SENT) == 0)
1379 goto dropunlock;
1380 }
1381 sig_checked = 1;
1382 }
1383 #endif
1384
1385 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th);
1386
1387 /*
1388 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later
1389 * state. tcp_do_segment() always consumes the mbuf chain, unlocks
1390 * the inpcb, and unlocks pcbinfo.
1391 */
1392 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos, ti_locked);
1393 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1394 return;
1395
1396 dropwithreset:
1397 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th);
1398
1399 if (ti_locked == TI_WLOCKED) {
1400 INP_INFO_WUNLOCK(&V_tcbinfo);
1401 ti_locked = TI_UNLOCKED;
1402 }
1403 #ifdef INVARIANTS
1404 else {
1405 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropwithreset "
1406 "ti_locked: %d", __func__, ti_locked));
1407 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1408 }
1409 #endif
1410
1411 if (inp != NULL) {
1412 tcp_dropwithreset(m, th, tp, tlen, rstreason);
1413 INP_WUNLOCK(inp);
1414 } else
1415 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
1416 m = NULL; /* mbuf chain got consumed. */
1417 goto drop;
1418
1419 dropunlock:
1420 if (m != NULL)
1421 TCP_PROBE5(receive, NULL, tp, mtod(m, const char *), tp, th);
1422
1423 if (ti_locked == TI_WLOCKED) {
1424 INP_INFO_WUNLOCK(&V_tcbinfo);
1425 ti_locked = TI_UNLOCKED;
1426 }
1427 #ifdef INVARIANTS
1428 else {
1429 KASSERT(ti_locked == TI_UNLOCKED, ("%s: dropunlock "
1430 "ti_locked: %d", __func__, ti_locked));
1431 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1432 }
1433 #endif
1434
1435 if (inp != NULL)
1436 INP_WUNLOCK(inp);
1437
1438 drop:
1439 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1440 if (s != NULL)
1441 free(s, M_TCPLOG);
1442 if (m != NULL)
1443 m_freem(m);
1444 }
1445
1446 static void
1447 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so,
1448 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos,
1449 int ti_locked)
1450 {
1451 int thflags, acked, ourfinisacked, needoutput = 0, sack_changed;
1452 int rstreason, todrop, win;
1453 u_long tiwin;
1454 char *s;
1455 struct in_conninfo *inc;
1456 struct mbuf *mfree;
1457 struct tcpopt to;
1458 int tfo_syn;
1459
1460 #ifdef TCPDEBUG
1461 /*
1462 * The size of tcp_saveipgen must be the size of the max ip header,
1463 * now IPv6.
1464 */
1465 u_char tcp_saveipgen[IP6_HDR_LEN];
1466 struct tcphdr tcp_savetcp;
1467 short ostate = 0;
1468 #endif
1469 thflags = th->th_flags;
1470 inc = &tp->t_inpcb->inp_inc;
1471 tp->sackhint.last_sack_ack = 0;
1472 sack_changed = 0;
1473
1474 /*
1475 * If this is either a state-changing packet or current state isn't
1476 * established, we require a write lock on tcbinfo. Otherwise, we
1477 * allow the tcbinfo to be in either alocked or unlocked, as the
1478 * caller may have unnecessarily acquired a write lock due to a race.
1479 */
1480 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 ||
1481 tp->t_state != TCPS_ESTABLISHED) {
1482 KASSERT(ti_locked == TI_WLOCKED, ("%s ti_locked %d for "
1483 "SYN/FIN/RST/!EST", __func__, ti_locked));
1484 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1485 } else {
1486 #ifdef INVARIANTS
1487 if (ti_locked == TI_WLOCKED)
1488 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1489 else {
1490 KASSERT(ti_locked == TI_UNLOCKED, ("%s: EST "
1491 "ti_locked: %d", __func__, ti_locked));
1492 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1493 }
1494 #endif
1495 }
1496 INP_WLOCK_ASSERT(tp->t_inpcb);
1497 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN",
1498 __func__));
1499 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT",
1500 __func__));
1501
1502 /*
1503 * Segment received on connection.
1504 * Reset idle time and keep-alive timer.
1505 * XXX: This should be done after segment
1506 * validation to ignore broken/spoofed segs.
1507 */
1508 tp->t_rcvtime = ticks;
1509 if (TCPS_HAVEESTABLISHED(tp->t_state))
1510 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
1511
1512 /*
1513 * Unscale the window into a 32-bit value.
1514 * For the SYN_SENT state the scale is zero.
1515 */
1516 tiwin = th->th_win << tp->snd_scale;
1517
1518 /*
1519 * TCP ECN processing.
1520 */
1521 if (tp->t_flags & TF_ECN_PERMIT) {
1522 if (thflags & TH_CWR)
1523 tp->t_flags &= ~TF_ECN_SND_ECE;
1524 switch (iptos & IPTOS_ECN_MASK) {
1525 case IPTOS_ECN_CE:
1526 tp->t_flags |= TF_ECN_SND_ECE;
1527 TCPSTAT_INC(tcps_ecn_ce);
1528 break;
1529 case IPTOS_ECN_ECT0:
1530 TCPSTAT_INC(tcps_ecn_ect0);
1531 break;
1532 case IPTOS_ECN_ECT1:
1533 TCPSTAT_INC(tcps_ecn_ect1);
1534 break;
1535 }
1536 /* Congestion experienced. */
1537 if (thflags & TH_ECE) {
1538 cc_cong_signal(tp, th, CC_ECN);
1539 }
1540 }
1541
1542 /*
1543 * Parse options on any incoming segment.
1544 */
1545 tcp_dooptions(&to, (u_char *)(th + 1),
1546 (th->th_off << 2) - sizeof(struct tcphdr),
1547 (thflags & TH_SYN) ? TO_SYN : 0);
1548
1549 /*
1550 * If echoed timestamp is later than the current time,
1551 * fall back to non RFC1323 RTT calculation. Normalize
1552 * timestamp if syncookies were used when this connection
1553 * was established.
1554 */
1555 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1556 to.to_tsecr -= tp->ts_offset;
1557 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks()))
1558 to.to_tsecr = 0;
1559 }
1560 /*
1561 * If timestamps were negotiated during SYN/ACK they should
1562 * appear on every segment during this session and vice versa.
1563 */
1564 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) {
1565 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1566 log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1567 "no action\n", s, __func__);
1568 free(s, M_TCPLOG);
1569 }
1570 }
1571 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) {
1572 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1573 log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
1574 "no action\n", s, __func__);
1575 free(s, M_TCPLOG);
1576 }
1577 }
1578
1579 /*
1580 * Process options only when we get SYN/ACK back. The SYN case
1581 * for incoming connections is handled in tcp_syncache.
1582 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
1583 * or <SYN,ACK>) segment itself is never scaled.
1584 * XXX this is traditional behavior, may need to be cleaned up.
1585 */
1586 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1587 if ((to.to_flags & TOF_SCALE) &&
1588 (tp->t_flags & TF_REQ_SCALE)) {
1589 tp->t_flags |= TF_RCVD_SCALE;
1590 tp->snd_scale = to.to_wscale;
1591 }
1592 /*
1593 * Initial send window. It will be updated with
1594 * the next incoming segment to the scaled value.
1595 */
1596 tp->snd_wnd = th->th_win;
1597 if (to.to_flags & TOF_TS) {
1598 tp->t_flags |= TF_RCVD_TSTMP;
1599 tp->ts_recent = to.to_tsval;
1600 tp->ts_recent_age = tcp_ts_getticks();
1601 }
1602 if (to.to_flags & TOF_MSS)
1603 tcp_mss(tp, to.to_mss);
1604 if ((tp->t_flags & TF_SACK_PERMIT) &&
1605 (to.to_flags & TOF_SACKPERM) == 0)
1606 tp->t_flags &= ~TF_SACK_PERMIT;
1607 }
1608
1609 /*
1610 * Header prediction: check for the two common cases
1611 * of a uni-directional data xfer. If the packet has
1612 * no control flags, is in-sequence, the window didn't
1613 * change and we're not retransmitting, it's a
1614 * candidate. If the length is zero and the ack moved
1615 * forward, we're the sender side of the xfer. Just
1616 * free the data acked & wake any higher level process
1617 * that was blocked waiting for space. If the length
1618 * is non-zero and the ack didn't move, we're the
1619 * receiver side. If we're getting packets in-order
1620 * (the reassembly queue is empty), add the data to
1621 * the socket buffer and note that we need a delayed ack.
1622 * Make sure that the hidden state-flags are also off.
1623 * Since we check for TCPS_ESTABLISHED first, it can only
1624 * be TH_NEEDSYN.
1625 */
1626 if (tp->t_state == TCPS_ESTABLISHED &&
1627 th->th_seq == tp->rcv_nxt &&
1628 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1629 tp->snd_nxt == tp->snd_max &&
1630 tiwin && tiwin == tp->snd_wnd &&
1631 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1632 LIST_EMPTY(&tp->t_segq) &&
1633 ((to.to_flags & TOF_TS) == 0 ||
1634 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) {
1635
1636 /*
1637 * If last ACK falls within this segment's sequence numbers,
1638 * record the timestamp.
1639 * NOTE that the test is modified according to the latest
1640 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1641 */
1642 if ((to.to_flags & TOF_TS) != 0 &&
1643 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1644 tp->ts_recent_age = tcp_ts_getticks();
1645 tp->ts_recent = to.to_tsval;
1646 }
1647
1648 if (tlen == 0) {
1649 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1650 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1651 !IN_RECOVERY(tp->t_flags) &&
1652 (to.to_flags & TOF_SACK) == 0 &&
1653 TAILQ_EMPTY(&tp->snd_holes)) {
1654 /*
1655 * This is a pure ack for outstanding data.
1656 */
1657 if (ti_locked == TI_WLOCKED)
1658 INP_INFO_WUNLOCK(&V_tcbinfo);
1659 ti_locked = TI_UNLOCKED;
1660
1661 TCPSTAT_INC(tcps_predack);
1662
1663 /*
1664 * "bad retransmit" recovery.
1665 */
1666 if (tp->t_rxtshift == 1 &&
1667 tp->t_flags & TF_PREVVALID &&
1668 (int)(ticks - tp->t_badrxtwin) < 0) {
1669 cc_cong_signal(tp, th, CC_RTO_ERR);
1670 }
1671
1672 /*
1673 * Recalculate the transmit timer / rtt.
1674 *
1675 * Some boxes send broken timestamp replies
1676 * during the SYN+ACK phase, ignore
1677 * timestamps of 0 or we could calculate a
1678 * huge RTT and blow up the retransmit timer.
1679 */
1680 if ((to.to_flags & TOF_TS) != 0 &&
1681 to.to_tsecr) {
1682 u_int t;
1683
1684 t = tcp_ts_getticks() - to.to_tsecr;
1685 if (!tp->t_rttlow || tp->t_rttlow > t)
1686 tp->t_rttlow = t;
1687 tcp_xmit_timer(tp,
1688 TCP_TS_TO_TICKS(t) + 1);
1689 } else if (tp->t_rtttime &&
1690 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1691 if (!tp->t_rttlow ||
1692 tp->t_rttlow > ticks - tp->t_rtttime)
1693 tp->t_rttlow = ticks - tp->t_rtttime;
1694 tcp_xmit_timer(tp,
1695 ticks - tp->t_rtttime);
1696 }
1697 acked = BYTES_THIS_ACK(tp, th);
1698
1699 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
1700 hhook_run_tcp_est_in(tp, th, &to);
1701
1702 TCPSTAT_INC(tcps_rcvackpack);
1703 TCPSTAT_ADD(tcps_rcvackbyte, acked);
1704 sbdrop(&so->so_snd, acked);
1705 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1706 SEQ_LEQ(th->th_ack, tp->snd_recover))
1707 tp->snd_recover = th->th_ack - 1;
1708
1709 /*
1710 * Let the congestion control algorithm update
1711 * congestion control related information. This
1712 * typically means increasing the congestion
1713 * window.
1714 */
1715 cc_ack_received(tp, th, CC_ACK);
1716
1717 tp->snd_una = th->th_ack;
1718 /*
1719 * Pull snd_wl2 up to prevent seq wrap relative
1720 * to th_ack.
1721 */
1722 tp->snd_wl2 = th->th_ack;
1723 tp->t_dupacks = 0;
1724 m_freem(m);
1725 ND6_HINT(tp); /* Some progress has been made. */
1726
1727 /*
1728 * If all outstanding data are acked, stop
1729 * retransmit timer, otherwise restart timer
1730 * using current (possibly backed-off) value.
1731 * If process is waiting for space,
1732 * wakeup/selwakeup/signal. If data
1733 * are ready to send, let tcp_output
1734 * decide between more output or persist.
1735 */
1736 #ifdef TCPDEBUG
1737 if (so->so_options & SO_DEBUG)
1738 tcp_trace(TA_INPUT, ostate, tp,
1739 (void *)tcp_saveipgen,
1740 &tcp_savetcp, 0);
1741 #endif
1742 if (tp->snd_una == tp->snd_max)
1743 tcp_timer_activate(tp, TT_REXMT, 0);
1744 else if (!tcp_timer_active(tp, TT_PERSIST))
1745 tcp_timer_activate(tp, TT_REXMT,
1746 tp->t_rxtcur);
1747 sowwakeup(so);
1748 if (so->so_snd.sb_cc)
1749 (void) tcp_output(tp);
1750 goto check_delack;
1751 }
1752 } else if (th->th_ack == tp->snd_una &&
1753 tlen <= sbspace(&so->so_rcv)) {
1754 int newsize = 0; /* automatic sockbuf scaling */
1755
1756 /*
1757 * This is a pure, in-sequence data packet with
1758 * nothing on the reassembly queue and we have enough
1759 * buffer space to take it.
1760 */
1761 if (ti_locked == TI_WLOCKED)
1762 INP_INFO_WUNLOCK(&V_tcbinfo);
1763 ti_locked = TI_UNLOCKED;
1764
1765 /* Clean receiver SACK report if present */
1766 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks)
1767 tcp_clean_sackreport(tp);
1768 TCPSTAT_INC(tcps_preddat);
1769 tp->rcv_nxt += tlen;
1770 /*
1771 * Pull snd_wl1 up to prevent seq wrap relative to
1772 * th_seq.
1773 */
1774 tp->snd_wl1 = th->th_seq;
1775 /*
1776 * Pull rcv_up up to prevent seq wrap relative to
1777 * rcv_nxt.
1778 */
1779 tp->rcv_up = tp->rcv_nxt;
1780 TCPSTAT_INC(tcps_rcvpack);
1781 TCPSTAT_ADD(tcps_rcvbyte, tlen);
1782 ND6_HINT(tp); /* Some progress has been made */
1783 #ifdef TCPDEBUG
1784 if (so->so_options & SO_DEBUG)
1785 tcp_trace(TA_INPUT, ostate, tp,
1786 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1787 #endif
1788 /*
1789 * Automatic sizing of receive socket buffer. Often the send
1790 * buffer size is not optimally adjusted to the actual network
1791 * conditions at hand (delay bandwidth product). Setting the
1792 * buffer size too small limits throughput on links with high
1793 * bandwidth and high delay (eg. trans-continental/oceanic links).
1794 *
1795 * On the receive side the socket buffer memory is only rarely
1796 * used to any significant extent. This allows us to be much
1797 * more aggressive in scaling the receive socket buffer. For
1798 * the case that the buffer space is actually used to a large
1799 * extent and we run out of kernel memory we can simply drop
1800 * the new segments; TCP on the sender will just retransmit it
1801 * later. Setting the buffer size too big may only consume too
1802 * much kernel memory if the application doesn't read() from
1803 * the socket or packet loss or reordering makes use of the
1804 * reassembly queue.
1805 *
1806 * The criteria to step up the receive buffer one notch are:
1807 * 1. the number of bytes received during the time it takes
1808 * one timestamp to be reflected back to us (the RTT);
1809 * 2. received bytes per RTT is within seven eighth of the
1810 * current socket buffer size;
1811 * 3. receive buffer size has not hit maximal automatic size;
1812 *
1813 * This algorithm does one step per RTT at most and only if
1814 * we receive a bulk stream w/o packet losses or reorderings.
1815 * Shrinking the buffer during idle times is not necessary as
1816 * it doesn't consume any memory when idle.
1817 *
1818 * TODO: Only step up if the application is actually serving
1819 * the buffer to better manage the socket buffer resources.
1820 */
1821 if (V_tcp_do_autorcvbuf &&
1822 (to.to_flags & TOF_TS) &&
1823 to.to_tsecr &&
1824 (so->so_rcv.sb_flags & SB_AUTOSIZE)) {
1825 if (TSTMP_GT(to.to_tsecr, tp->rfbuf_ts) &&
1826 to.to_tsecr - tp->rfbuf_ts < hz) {
1827 if (tp->rfbuf_cnt >
1828 (so->so_rcv.sb_hiwat / 8 * 7) &&
1829 so->so_rcv.sb_hiwat <
1830 V_tcp_autorcvbuf_max) {
1831 newsize =
1832 min(so->so_rcv.sb_hiwat +
1833 V_tcp_autorcvbuf_inc,
1834 V_tcp_autorcvbuf_max);
1835 }
1836 /* Start over with next RTT. */
1837 tp->rfbuf_ts = 0;
1838 tp->rfbuf_cnt = 0;
1839 } else
1840 tp->rfbuf_cnt += tlen; /* add up */
1841 }
1842
1843 /* Add data to socket buffer. */
1844 SOCKBUF_LOCK(&so->so_rcv);
1845 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1846 m_freem(m);
1847 } else {
1848 /*
1849 * Set new socket buffer size.
1850 * Give up when limit is reached.
1851 */
1852 if (newsize)
1853 if (!sbreserve_locked(&so->so_rcv,
1854 newsize, so, NULL))
1855 so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1856 m_adj(m, drop_hdrlen); /* delayed header drop */
1857 sbappendstream_locked(&so->so_rcv, m);
1858 }
1859 /* NB: sorwakeup_locked() does an implicit unlock. */
1860 sorwakeup_locked(so);
1861 if (DELAY_ACK(tp, tlen)) {
1862 tp->t_flags |= TF_DELACK;
1863 } else {
1864 tp->t_flags |= TF_ACKNOW;
1865 tcp_output(tp);
1866 }
1867 goto check_delack;
1868 }
1869 }
1870
1871 /*
1872 * Calculate amount of space in receive window,
1873 * and then do TCP input processing.
1874 * Receive window is amount of space in rcv queue,
1875 * but not less than advertised window.
1876 */
1877 win = sbspace(&so->so_rcv);
1878 if (win < 0)
1879 win = 0;
1880 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1881
1882 /* Reset receive buffer auto scaling when not in bulk receive mode. */
1883 tp->rfbuf_ts = 0;
1884 tp->rfbuf_cnt = 0;
1885
1886 switch (tp->t_state) {
1887
1888 /*
1889 * If the state is SYN_RECEIVED:
1890 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1891 */
1892 case TCPS_SYN_RECEIVED:
1893 if ((thflags & TH_ACK) &&
1894 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1895 SEQ_GT(th->th_ack, tp->snd_max))) {
1896 rstreason = BANDLIM_RST_OPENPORT;
1897 goto dropwithreset;
1898 }
1899 #ifdef TCP_RFC7413
1900 if (tp->t_flags & TF_FASTOPEN) {
1901 /*
1902 * When a TFO connection is in SYN_RECEIVED, the
1903 * only valid packets are the initial SYN, a
1904 * retransmit/copy of the initial SYN (possibly with
1905 * a subset of the original data), a valid ACK, a
1906 * FIN, or a RST.
1907 */
1908 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) {
1909 rstreason = BANDLIM_RST_OPENPORT;
1910 goto dropwithreset;
1911 } else if (thflags & TH_SYN) {
1912 /* non-initial SYN is ignored */
1913 if ((tcp_timer_active(tp, TT_DELACK) ||
1914 tcp_timer_active(tp, TT_REXMT)))
1915 goto drop;
1916 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) {
1917 goto drop;
1918 }
1919 }
1920 #endif
1921 break;
1922
1923 /*
1924 * If the state is SYN_SENT:
1925 * if seg contains an ACK, but not for our SYN, drop the input.
1926 * if seg contains a RST, then drop the connection.
1927 * if seg does not contain SYN, then drop it.
1928 * Otherwise this is an acceptable SYN segment
1929 * initialize tp->rcv_nxt and tp->irs
1930 * if seg contains ack then advance tp->snd_una
1931 * if seg contains an ECE and ECN support is enabled, the stream
1932 * is ECN capable.
1933 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1934 * arrange for segment to be acked (eventually)
1935 * continue processing rest of data/controls, beginning with URG
1936 */
1937 case TCPS_SYN_SENT:
1938 if ((thflags & TH_ACK) &&
1939 (SEQ_LEQ(th->th_ack, tp->iss) ||
1940 SEQ_GT(th->th_ack, tp->snd_max))) {
1941 rstreason = BANDLIM_UNLIMITED;
1942 goto dropwithreset;
1943 }
1944 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) {
1945 TCP_PROBE5(connect__refused, NULL, tp,
1946 mtod(m, const char *), tp, th);
1947 tp = tcp_drop(tp, ECONNREFUSED);
1948 }
1949 if (thflags & TH_RST)
1950 goto drop;
1951 if (!(thflags & TH_SYN))
1952 goto drop;
1953
1954 tp->irs = th->th_seq;
1955 tcp_rcvseqinit(tp);
1956 if (thflags & TH_ACK) {
1957 TCPSTAT_INC(tcps_connects);
1958 soisconnected(so);
1959 #ifdef MAC
1960 mac_socketpeer_set_from_mbuf(m, so);
1961 #endif
1962 /* Do window scaling on this connection? */
1963 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1964 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1965 tp->rcv_scale = tp->request_r_scale;
1966 }
1967 tp->rcv_adv += imin(tp->rcv_wnd,
1968 TCP_MAXWIN << tp->rcv_scale);
1969 tp->snd_una++; /* SYN is acked */
1970 /*
1971 * If there's data, delay ACK; if there's also a FIN
1972 * ACKNOW will be turned on later.
1973 */
1974 if (DELAY_ACK(tp, tlen) && tlen != 0)
1975 tcp_timer_activate(tp, TT_DELACK,
1976 tcp_delacktime);
1977 else
1978 tp->t_flags |= TF_ACKNOW;
1979
1980 if ((thflags & TH_ECE) && V_tcp_do_ecn) {
1981 tp->t_flags |= TF_ECN_PERMIT;
1982 TCPSTAT_INC(tcps_ecn_shs);
1983 }
1984
1985 /*
1986 * Received <SYN,ACK> in SYN_SENT[*] state.
1987 * Transitions:
1988 * SYN_SENT --> ESTABLISHED
1989 * SYN_SENT* --> FIN_WAIT_1
1990 */
1991 tp->t_starttime = ticks;
1992 if (tp->t_flags & TF_NEEDFIN) {
1993 tcp_state_change(tp, TCPS_FIN_WAIT_1);
1994 tp->t_flags &= ~TF_NEEDFIN;
1995 thflags &= ~TH_SYN;
1996 } else {
1997 tcp_state_change(tp, TCPS_ESTABLISHED);
1998 TCP_PROBE5(connect__established, NULL, tp,
1999 mtod(m, const char *), tp, th);
2000 cc_conn_init(tp);
2001 tcp_timer_activate(tp, TT_KEEP,
2002 TP_KEEPIDLE(tp));
2003 }
2004 } else {
2005 /*
2006 * Received initial SYN in SYN-SENT[*] state =>
2007 * simultaneous open. If segment contains CC option
2008 * and there is a cached CC, apply TAO test.
2009 * If it succeeds, connection is * half-synchronized.
2010 * Otherwise, do 3-way handshake:
2011 * SYN-SENT -> SYN-RECEIVED
2012 * SYN-SENT* -> SYN-RECEIVED*
2013 * If there was no CC option, clear cached CC value.
2014 */
2015 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
2016 tcp_timer_activate(tp, TT_REXMT, 0);
2017 tcp_state_change(tp, TCPS_SYN_RECEIVED);
2018 }
2019
2020 KASSERT(ti_locked == TI_WLOCKED, ("%s: trimthenstep6: "
2021 "ti_locked %d", __func__, ti_locked));
2022 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2023 INP_WLOCK_ASSERT(tp->t_inpcb);
2024
2025 /*
2026 * Advance th->th_seq to correspond to first data byte.
2027 * If data, trim to stay within window,
2028 * dropping FIN if necessary.
2029 */
2030 th->th_seq++;
2031 if (tlen > tp->rcv_wnd) {
2032 todrop = tlen - tp->rcv_wnd;
2033 m_adj(m, -todrop);
2034 tlen = tp->rcv_wnd;
2035 thflags &= ~TH_FIN;
2036 TCPSTAT_INC(tcps_rcvpackafterwin);
2037 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2038 }
2039 tp->snd_wl1 = th->th_seq - 1;
2040 tp->rcv_up = th->th_seq;
2041 /*
2042 * Client side of transaction: already sent SYN and data.
2043 * If the remote host used T/TCP to validate the SYN,
2044 * our data will be ACK'd; if so, enter normal data segment
2045 * processing in the middle of step 5, ack processing.
2046 * Otherwise, goto step 6.
2047 */
2048 if (thflags & TH_ACK)
2049 goto process_ACK;
2050
2051 goto step6;
2052
2053 /*
2054 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
2055 * do normal processing.
2056 *
2057 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later.
2058 */
2059 case TCPS_LAST_ACK:
2060 case TCPS_CLOSING:
2061 break; /* continue normal processing */
2062 }
2063
2064 /*
2065 * States other than LISTEN or SYN_SENT.
2066 * First check the RST flag and sequence number since reset segments
2067 * are exempt from the timestamp and connection count tests. This
2068 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
2069 * below which allowed reset segments in half the sequence space
2070 * to fall though and be processed (which gives forged reset
2071 * segments with a random sequence number a 50 percent chance of
2072 * killing a connection).
2073 * Then check timestamp, if present.
2074 * Then check the connection count, if present.
2075 * Then check that at least some bytes of segment are within
2076 * receive window. If segment begins before rcv_nxt,
2077 * drop leading data (and SYN); if nothing left, just ack.
2078 *
2079 *
2080 * If the RST bit is set, check the sequence number to see
2081 * if this is a valid reset segment.
2082 * RFC 793 page 37:
2083 * In all states except SYN-SENT, all reset (RST) segments
2084 * are validated by checking their SEQ-fields. A reset is
2085 * valid if its sequence number is in the window.
2086 * Note: this does not take into account delayed ACKs, so
2087 * we should test against last_ack_sent instead of rcv_nxt.
2088 * The sequence number in the reset segment is normally an
2089 * echo of our outgoing acknowlegement numbers, but some hosts
2090 * send a reset with the sequence number at the rightmost edge
2091 * of our receive window, and we have to handle this case.
2092 * Note 2: Paul Watson's paper "Slipping in the Window" has shown
2093 * that brute force RST attacks are possible. To combat this,
2094 * we use a much stricter check while in the ESTABLISHED state,
2095 * only accepting RSTs where the sequence number is equal to
2096 * last_ack_sent. In all other states (the states in which a
2097 * RST is more likely), the more permissive check is used.
2098 * If we have multiple segments in flight, the initial reset
2099 * segment sequence numbers will be to the left of last_ack_sent,
2100 * but they will eventually catch up.
2101 * In any case, it never made sense to trim reset segments to
2102 * fit the receive window since RFC 1122 says:
2103 * 4.2.2.12 RST Segment: RFC-793 Section 3.4
2104 *
2105 * A TCP SHOULD allow a received RST segment to include data.
2106 *
2107 * DISCUSSION
2108 * It has been suggested that a RST segment could contain
2109 * ASCII text that encoded and explained the cause of the
2110 * RST. No standard has yet been established for such
2111 * data.
2112 *
2113 * If the reset segment passes the sequence number test examine
2114 * the state:
2115 * SYN_RECEIVED STATE:
2116 * If passive open, return to LISTEN state.
2117 * If active open, inform user that connection was refused.
2118 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
2119 * Inform user that connection was reset, and close tcb.
2120 * CLOSING, LAST_ACK STATES:
2121 * Close the tcb.
2122 * TIME_WAIT STATE:
2123 * Drop the segment - see Stevens, vol. 2, p. 964 and
2124 * RFC 1337.
2125 */
2126 if (thflags & TH_RST) {
2127 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) &&
2128 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
2129 switch (tp->t_state) {
2130
2131 case TCPS_SYN_RECEIVED:
2132 so->so_error = ECONNREFUSED;
2133 goto close;
2134
2135 case TCPS_ESTABLISHED:
2136 if (V_tcp_insecure_rst == 0 &&
2137 !(SEQ_GEQ(th->th_seq, tp->rcv_nxt - 1) &&
2138 SEQ_LEQ(th->th_seq, tp->rcv_nxt + 1)) &&
2139 !(SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) &&
2140 SEQ_LEQ(th->th_seq, tp->last_ack_sent + 1))) {
2141 TCPSTAT_INC(tcps_badrst);
2142 goto drop;
2143 }
2144 /* FALLTHROUGH */
2145 case TCPS_FIN_WAIT_1:
2146 case TCPS_FIN_WAIT_2:
2147 case TCPS_CLOSE_WAIT:
2148 so->so_error = ECONNRESET;
2149 close:
2150 KASSERT(ti_locked == TI_WLOCKED,
2151 ("tcp_do_segment: TH_RST 1 ti_locked %d",
2152 ti_locked));
2153 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2154
2155 tcp_state_change(tp, TCPS_CLOSED);
2156 TCPSTAT_INC(tcps_drops);
2157 tp = tcp_close(tp);
2158 break;
2159
2160 case TCPS_CLOSING:
2161 case TCPS_LAST_ACK:
2162 KASSERT(ti_locked == TI_WLOCKED,
2163 ("tcp_do_segment: TH_RST 2 ti_locked %d",
2164 ti_locked));
2165 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2166
2167 tp = tcp_close(tp);
2168 break;
2169 }
2170 }
2171 goto drop;
2172 }
2173
2174 /*
2175 * RFC 1323 PAWS: If we have a timestamp reply on this segment
2176 * and it's less than ts_recent, drop it.
2177 */
2178 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
2179 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
2180
2181 /* Check to see if ts_recent is over 24 days old. */
2182 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) {
2183 /*
2184 * Invalidate ts_recent. If this segment updates
2185 * ts_recent, the age will be reset later and ts_recent
2186 * will get a valid value. If it does not, setting
2187 * ts_recent to zero will at least satisfy the
2188 * requirement that zero be placed in the timestamp
2189 * echo reply when ts_recent isn't valid. The
2190 * age isn't reset until we get a valid ts_recent
2191 * because we don't want out-of-order segments to be
2192 * dropped when ts_recent is old.
2193 */
2194 tp->ts_recent = 0;
2195 } else {
2196 TCPSTAT_INC(tcps_rcvduppack);
2197 TCPSTAT_ADD(tcps_rcvdupbyte, tlen);
2198 TCPSTAT_INC(tcps_pawsdrop);
2199 if (tlen)
2200 goto dropafterack;
2201 goto drop;
2202 }
2203 }
2204
2205 /*
2206 * In the SYN-RECEIVED state, validate that the packet belongs to
2207 * this connection before trimming the data to fit the receive
2208 * window. Check the sequence number versus IRS since we know
2209 * the sequence numbers haven't wrapped. This is a partial fix
2210 * for the "LAND" DoS attack.
2211 */
2212 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
2213 rstreason = BANDLIM_RST_OPENPORT;
2214 goto dropwithreset;
2215 }
2216
2217 todrop = tp->rcv_nxt - th->th_seq;
2218 if (todrop > 0) {
2219 if (thflags & TH_SYN) {
2220 thflags &= ~TH_SYN;
2221 th->th_seq++;
2222 if (th->th_urp > 1)
2223 th->th_urp--;
2224 else
2225 thflags &= ~TH_URG;
2226 todrop--;
2227 }
2228 /*
2229 * Following if statement from Stevens, vol. 2, p. 960.
2230 */
2231 if (todrop > tlen
2232 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
2233 /*
2234 * Any valid FIN must be to the left of the window.
2235 * At this point the FIN must be a duplicate or out
2236 * of sequence; drop it.
2237 */
2238 thflags &= ~TH_FIN;
2239
2240 /*
2241 * Send an ACK to resynchronize and drop any data.
2242 * But keep on processing for RST or ACK.
2243 */
2244 tp->t_flags |= TF_ACKNOW;
2245 todrop = tlen;
2246 TCPSTAT_INC(tcps_rcvduppack);
2247 TCPSTAT_ADD(tcps_rcvdupbyte, todrop);
2248 } else {
2249 TCPSTAT_INC(tcps_rcvpartduppack);
2250 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop);
2251 }
2252 drop_hdrlen += todrop; /* drop from the top afterwards */
2253 th->th_seq += todrop;
2254 tlen -= todrop;
2255 if (th->th_urp > todrop)
2256 th->th_urp -= todrop;
2257 else {
2258 thflags &= ~TH_URG;
2259 th->th_urp = 0;
2260 }
2261 }
2262
2263 /*
2264 * If new data are received on a connection after the
2265 * user processes are gone, then RST the other end.
2266 */
2267 if ((so->so_state & SS_NOFDREF) &&
2268 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
2269 KASSERT(ti_locked == TI_WLOCKED, ("%s: SS_NOFDEREF && "
2270 "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked));
2271 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2272
2273 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
2274 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data "
2275 "after socket was closed, "
2276 "sending RST and removing tcpcb\n",
2277 s, __func__, tcpstates[tp->t_state], tlen);
2278 free(s, M_TCPLOG);
2279 }
2280 tp = tcp_close(tp);
2281 TCPSTAT_INC(tcps_rcvafterclose);
2282 rstreason = BANDLIM_UNLIMITED;
2283 goto dropwithreset;
2284 }
2285
2286 /*
2287 * If segment ends after window, drop trailing data
2288 * (and PUSH and FIN); if nothing left, just ACK.
2289 */
2290 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
2291 if (todrop > 0) {
2292 TCPSTAT_INC(tcps_rcvpackafterwin);
2293 if (todrop >= tlen) {
2294 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen);
2295 /*
2296 * If window is closed can only take segments at
2297 * window edge, and have to drop data and PUSH from
2298 * incoming segments. Continue processing, but
2299 * remember to ack. Otherwise, drop segment
2300 * and ack.
2301 */
2302 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2303 tp->t_flags |= TF_ACKNOW;
2304 TCPSTAT_INC(tcps_rcvwinprobe);
2305 } else
2306 goto dropafterack;
2307 } else
2308 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2309 m_adj(m, -todrop);
2310 tlen -= todrop;
2311 thflags &= ~(TH_PUSH|TH_FIN);
2312 }
2313
2314 /*
2315 * If last ACK falls within this segment's sequence numbers,
2316 * record its timestamp.
2317 * NOTE:
2318 * 1) That the test incorporates suggestions from the latest
2319 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
2320 * 2) That updating only on newer timestamps interferes with
2321 * our earlier PAWS tests, so this check should be solely
2322 * predicated on the sequence space of this segment.
2323 * 3) That we modify the segment boundary check to be
2324 * Last.ACK.Sent <= SEG.SEQ + SEG.Len
2325 * instead of RFC1323's
2326 * Last.ACK.Sent < SEG.SEQ + SEG.Len,
2327 * This modified check allows us to overcome RFC1323's
2328 * limitations as described in Stevens TCP/IP Illustrated
2329 * Vol. 2 p.869. In such cases, we can still calculate the
2330 * RTT correctly when RCV.NXT == Last.ACK.Sent.
2331 */
2332 if ((to.to_flags & TOF_TS) != 0 &&
2333 SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2334 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
2335 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
2336 tp->ts_recent_age = tcp_ts_getticks();
2337 tp->ts_recent = to.to_tsval;
2338 }
2339
2340 /*
2341 * If a SYN is in the window, then this is an
2342 * error and we send an RST and drop the connection.
2343 */
2344 if (thflags & TH_SYN) {
2345 KASSERT(ti_locked == TI_WLOCKED,
2346 ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked));
2347 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2348
2349 tp = tcp_drop(tp, ECONNRESET);
2350 rstreason = BANDLIM_UNLIMITED;
2351 goto drop;
2352 }
2353
2354 /*
2355 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
2356 * flag is on (half-synchronized state), then queue data for
2357 * later processing; else drop segment and return.
2358 */
2359 if ((thflags & TH_ACK) == 0) {
2360 if (tp->t_state == TCPS_SYN_RECEIVED ||
2361 (tp->t_flags & TF_NEEDSYN)) {
2362 #ifdef TCP_RFC7413
2363 if (tp->t_state == TCPS_SYN_RECEIVED &&
2364 tp->t_flags & TF_FASTOPEN) {
2365 tp->snd_wnd = tiwin;
2366 cc_conn_init(tp);
2367 }
2368 #endif
2369 goto step6;
2370 } else if (tp->t_flags & TF_ACKNOW)
2371 goto dropafterack;
2372 else
2373 goto drop;
2374 }
2375
2376 /*
2377 * Ack processing.
2378 */
2379 switch (tp->t_state) {
2380
2381 /*
2382 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
2383 * ESTABLISHED state and continue processing.
2384 * The ACK was checked above.
2385 */
2386 case TCPS_SYN_RECEIVED:
2387
2388 TCPSTAT_INC(tcps_connects);
2389 soisconnected(so);
2390 /* Do window scaling? */
2391 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2392 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2393 tp->rcv_scale = tp->request_r_scale;
2394 tp->snd_wnd = tiwin;
2395 }
2396 /*
2397 * Make transitions:
2398 * SYN-RECEIVED -> ESTABLISHED
2399 * SYN-RECEIVED* -> FIN-WAIT-1
2400 */
2401 tp->t_starttime = ticks;
2402 if (tp->t_flags & TF_NEEDFIN) {
2403 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2404 tp->t_flags &= ~TF_NEEDFIN;
2405 } else {
2406 tcp_state_change(tp, TCPS_ESTABLISHED);
2407 TCP_PROBE5(accept__established, NULL, tp,
2408 mtod(m, const char *), tp, th);
2409 #ifdef TCP_RFC7413
2410 if (tp->t_tfo_pending) {
2411 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2412 tp->t_tfo_pending = NULL;
2413
2414 /*
2415 * Account for the ACK of our SYN prior to
2416 * regular ACK processing below.
2417 */
2418 tp->snd_una++;
2419 }
2420 /*
2421 * TFO connections call cc_conn_init() during SYN
2422 * processing. Calling it again here for such
2423 * connections is not harmless as it would undo the
2424 * snd_cwnd reduction that occurs when a TFO SYN|ACK
2425 * is retransmitted.
2426 */
2427 if (!(tp->t_flags & TF_FASTOPEN))
2428 #endif
2429 cc_conn_init(tp);
2430 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
2431 }
2432 /*
2433 * If segment contains data or ACK, will call tcp_reass()
2434 * later; if not, do so now to pass queued data to user.
2435 */
2436 if (tlen == 0 && (thflags & TH_FIN) == 0)
2437 (void) tcp_reass(tp, (struct tcphdr *)0, 0,
2438 (struct mbuf *)0);
2439 tp->snd_wl1 = th->th_seq - 1;
2440 /* FALLTHROUGH */
2441
2442 /*
2443 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2444 * ACKs. If the ack is in the range
2445 * tp->snd_una < th->th_ack <= tp->snd_max
2446 * then advance tp->snd_una to th->th_ack and drop
2447 * data from the retransmission queue. If this ACK reflects
2448 * more up to date window information we update our window information.
2449 */
2450 case TCPS_ESTABLISHED:
2451 case TCPS_FIN_WAIT_1:
2452 case TCPS_FIN_WAIT_2:
2453 case TCPS_CLOSE_WAIT:
2454 case TCPS_CLOSING:
2455 case TCPS_LAST_ACK:
2456 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2457 TCPSTAT_INC(tcps_rcvacktoomuch);
2458 goto dropafterack;
2459 }
2460 if ((tp->t_flags & TF_SACK_PERMIT) &&
2461 ((to.to_flags & TOF_SACK) ||
2462 !TAILQ_EMPTY(&tp->snd_holes)))
2463 sack_changed = tcp_sack_doack(tp, &to, th->th_ack);
2464 else
2465 /*
2466 * Reset the value so that previous (valid) value
2467 * from the last ack with SACK doesn't get used.
2468 */
2469 tp->sackhint.sacked_bytes = 0;
2470
2471 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
2472 hhook_run_tcp_est_in(tp, th, &to);
2473
2474 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2475 if (tlen == 0 &&
2476 (tiwin == tp->snd_wnd ||
2477 (tp->t_flags & TF_SACK_PERMIT))) {
2478 TCPSTAT_INC(tcps_rcvdupack);
2479 /*
2480 * If we have outstanding data (other than
2481 * a window probe), this is a completely
2482 * duplicate ack (ie, window info didn't
2483 * change), the ack is the biggest we've
2484 * seen and we've seen exactly our rexmt
2485 * threshhold of them, assume a packet
2486 * has been dropped and retransmit it.
2487 * Kludge snd_nxt & the congestion
2488 * window so we send only this one
2489 * packet.
2490 *
2491 * We know we're losing at the current
2492 * window size so do congestion avoidance
2493 * (set ssthresh to half the current window
2494 * and pull our congestion window back to
2495 * the new ssthresh).
2496 *
2497 * Dup acks mean that packets have left the
2498 * network (they're now cached at the receiver)
2499 * so bump cwnd by the amount in the receiver
2500 * to keep a constant cwnd packets in the
2501 * network.
2502 *
2503 * When using TCP ECN, notify the peer that
2504 * we reduced the cwnd.
2505 */
2506 /*
2507 * Following 2 kinds of acks should not affect
2508 * dupack counting:
2509 * 1) Old acks
2510 * 2) Acks with SACK but without any new SACK
2511 * information in them. These could result from
2512 * any anomaly in the network like a switch
2513 * duplicating packets or a possible DoS attack.
2514 */
2515 if (th->th_ack != tp->snd_una ||
2516 ((tp->t_flags & TF_SACK_PERMIT) &&
2517 !sack_changed))
2518 break;
2519 else if (!tcp_timer_active(tp, TT_REXMT))
2520 tp->t_dupacks = 0;
2521 else if (++tp->t_dupacks > tcprexmtthresh ||
2522 IN_FASTRECOVERY(tp->t_flags)) {
2523 cc_ack_received(tp, th, CC_DUPACK);
2524 if ((tp->t_flags & TF_SACK_PERMIT) &&
2525 IN_FASTRECOVERY(tp->t_flags)) {
2526 int awnd;
2527
2528 /*
2529 * Compute the amount of data in flight first.
2530 * We can inject new data into the pipe iff
2531 * we have less than 1/2 the original window's
2532 * worth of data in flight.
2533 */
2534 if (V_tcp_do_rfc6675_pipe)
2535 awnd = tcp_compute_pipe(tp);
2536 else
2537 awnd = (tp->snd_nxt - tp->snd_fack) +
2538 tp->sackhint.sack_bytes_rexmit;
2539
2540 if (awnd < tp->snd_ssthresh) {
2541 tp->snd_cwnd += tp->t_maxseg;
2542 if (tp->snd_cwnd > tp->snd_ssthresh)
2543 tp->snd_cwnd = tp->snd_ssthresh;
2544 }
2545 } else
2546 tp->snd_cwnd += tp->t_maxseg;
2547 if ((thflags & TH_FIN) &&
2548 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2549 /*
2550 * If its a fin we need to process
2551 * it to avoid a race where both
2552 * sides enter FIN-WAIT and send FIN|ACK
2553 * at the same time.
2554 */
2555 break;
2556 }
2557 (void) tcp_output(tp);
2558 goto drop;
2559 } else if (tp->t_dupacks == tcprexmtthresh) {
2560 tcp_seq onxt = tp->snd_nxt;
2561
2562 /*
2563 * If we're doing sack, check to
2564 * see if we're already in sack
2565 * recovery. If we're not doing sack,
2566 * check to see if we're in newreno
2567 * recovery.
2568 */
2569 if (tp->t_flags & TF_SACK_PERMIT) {
2570 if (IN_FASTRECOVERY(tp->t_flags)) {
2571 tp->t_dupacks = 0;
2572 break;
2573 }
2574 } else {
2575 if (SEQ_LEQ(th->th_ack,
2576 tp->snd_recover)) {
2577 tp->t_dupacks = 0;
2578 break;
2579 }
2580 }
2581 /* Congestion signal before ack. */
2582 cc_cong_signal(tp, th, CC_NDUPACK);
2583 cc_ack_received(tp, th, CC_DUPACK);
2584 tcp_timer_activate(tp, TT_REXMT, 0);
2585 tp->t_rtttime = 0;
2586 if (tp->t_flags & TF_SACK_PERMIT) {
2587 TCPSTAT_INC(
2588 tcps_sack_recovery_episode);
2589 tp->sack_newdata = tp->snd_nxt;
2590 tp->snd_cwnd = tp->t_maxseg;
2591 (void) tcp_output(tp);
2592 goto drop;
2593 }
2594 tp->snd_nxt = th->th_ack;
2595 tp->snd_cwnd = tp->t_maxseg;
2596 if ((thflags & TH_FIN) &&
2597 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2598 /*
2599 * If its a fin we need to process
2600 * it to avoid a race where both
2601 * sides enter FIN-WAIT and send FIN|ACK
2602 * at the same time.
2603 */
2604 break;
2605 }
2606 (void) tcp_output(tp);
2607 KASSERT(tp->snd_limited <= 2,
2608 ("%s: tp->snd_limited too big",
2609 __func__));
2610 tp->snd_cwnd = tp->snd_ssthresh +
2611 tp->t_maxseg *
2612 (tp->t_dupacks - tp->snd_limited);
2613 if (SEQ_GT(onxt, tp->snd_nxt))
2614 tp->snd_nxt = onxt;
2615 goto drop;
2616 } else if (V_tcp_do_rfc3042) {
2617 /*
2618 * Process first and second duplicate
2619 * ACKs. Each indicates a segment
2620 * leaving the network, creating room
2621 * for more. Make sure we can send a
2622 * packet on reception of each duplicate
2623 * ACK by increasing snd_cwnd by one
2624 * segment. Restore the original
2625 * snd_cwnd after packet transmission.
2626 */
2627 cc_ack_received(tp, th, CC_DUPACK);
2628 u_long oldcwnd = tp->snd_cwnd;
2629 tcp_seq oldsndmax = tp->snd_max;
2630 u_int sent;
2631 int avail;
2632
2633 KASSERT(tp->t_dupacks == 1 ||
2634 tp->t_dupacks == 2,
2635 ("%s: dupacks not 1 or 2",
2636 __func__));
2637 if (tp->t_dupacks == 1)
2638 tp->snd_limited = 0;
2639 tp->snd_cwnd =
2640 (tp->snd_nxt - tp->snd_una) +
2641 (tp->t_dupacks - tp->snd_limited) *
2642 tp->t_maxseg;
2643 if ((thflags & TH_FIN) &&
2644 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2645 /*
2646 * If its a fin we need to process
2647 * it to avoid a race where both
2648 * sides enter FIN-WAIT and send FIN|ACK
2649 * at the same time.
2650 */
2651 break;
2652 }
2653 /*
2654 * Only call tcp_output when there
2655 * is new data available to be sent.
2656 * Otherwise we would send pure ACKs.
2657 */
2658 SOCKBUF_LOCK(&so->so_snd);
2659 avail = so->so_snd.sb_cc -
2660 (tp->snd_nxt - tp->snd_una);
2661 SOCKBUF_UNLOCK(&so->so_snd);
2662 if (avail > 0)
2663 (void) tcp_output(tp);
2664 sent = tp->snd_max - oldsndmax;
2665 if (sent > tp->t_maxseg) {
2666 KASSERT((tp->t_dupacks == 2 &&
2667 tp->snd_limited == 0) ||
2668 (sent == tp->t_maxseg + 1 &&
2669 tp->t_flags & TF_SENTFIN),
2670 ("%s: sent too much",
2671 __func__));
2672 tp->snd_limited = 2;
2673 } else if (sent > 0)
2674 ++tp->snd_limited;
2675 tp->snd_cwnd = oldcwnd;
2676 goto drop;
2677 }
2678 }
2679 break;
2680 } else {
2681 /*
2682 * This ack is advancing the left edge, reset the
2683 * counter.
2684 */
2685 tp->t_dupacks = 0;
2686 /*
2687 * If this ack also has new SACK info, increment the
2688 * counter as per rfc6675.
2689 */
2690 if ((tp->t_flags & TF_SACK_PERMIT) && sack_changed)
2691 tp->t_dupacks++;
2692 }
2693
2694 KASSERT(SEQ_GT(th->th_ack, tp->snd_una),
2695 ("%s: th_ack <= snd_una", __func__));
2696
2697 /*
2698 * If the congestion window was inflated to account
2699 * for the other side's cached packets, retract it.
2700 */
2701 if (IN_FASTRECOVERY(tp->t_flags)) {
2702 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2703 if (tp->t_flags & TF_SACK_PERMIT)
2704 tcp_sack_partialack(tp, th);
2705 else
2706 tcp_newreno_partial_ack(tp, th);
2707 } else
2708 cc_post_recovery(tp, th);
2709 }
2710 /*
2711 * If we reach this point, ACK is not a duplicate,
2712 * i.e., it ACKs something we sent.
2713 */
2714 if (tp->t_flags & TF_NEEDSYN) {
2715 /*
2716 * T/TCP: Connection was half-synchronized, and our
2717 * SYN has been ACK'd (so connection is now fully
2718 * synchronized). Go to non-starred state,
2719 * increment snd_una for ACK of SYN, and check if
2720 * we can do window scaling.
2721 */
2722 tp->t_flags &= ~TF_NEEDSYN;
2723 tp->snd_una++;
2724 /* Do window scaling? */
2725 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2726 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2727 tp->rcv_scale = tp->request_r_scale;
2728 /* Send window already scaled. */
2729 }
2730 }
2731
2732 process_ACK:
2733 INP_WLOCK_ASSERT(tp->t_inpcb);
2734
2735 acked = BYTES_THIS_ACK(tp, th);
2736 TCPSTAT_INC(tcps_rcvackpack);
2737 TCPSTAT_ADD(tcps_rcvackbyte, acked);
2738
2739 /*
2740 * If we just performed our first retransmit, and the ACK
2741 * arrives within our recovery window, then it was a mistake
2742 * to do the retransmit in the first place. Recover our
2743 * original cwnd and ssthresh, and proceed to transmit where
2744 * we left off.
2745 */
2746 if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID &&
2747 (int)(ticks - tp->t_badrxtwin) < 0)
2748 cc_cong_signal(tp, th, CC_RTO_ERR);
2749
2750 /*
2751 * If we have a timestamp reply, update smoothed
2752 * round trip time. If no timestamp is present but
2753 * transmit timer is running and timed sequence
2754 * number was acked, update smoothed round trip time.
2755 * Since we now have an rtt measurement, cancel the
2756 * timer backoff (cf., Phil Karn's retransmit alg.).
2757 * Recompute the initial retransmit timer.
2758 *
2759 * Some boxes send broken timestamp replies
2760 * during the SYN+ACK phase, ignore
2761 * timestamps of 0 or we could calculate a
2762 * huge RTT and blow up the retransmit timer.
2763 */
2764 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) {
2765 u_int t;
2766
2767 t = tcp_ts_getticks() - to.to_tsecr;
2768 if (!tp->t_rttlow || tp->t_rttlow > t)
2769 tp->t_rttlow = t;
2770 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1);
2771 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2772 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2773 tp->t_rttlow = ticks - tp->t_rtttime;
2774 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2775 }
2776
2777 /*
2778 * If all outstanding data is acked, stop retransmit
2779 * timer and remember to restart (more output or persist).
2780 * If there is more data to be acked, restart retransmit
2781 * timer, using current (possibly backed-off) value.
2782 */
2783 if (th->th_ack == tp->snd_max) {
2784 tcp_timer_activate(tp, TT_REXMT, 0);
2785 needoutput = 1;
2786 } else if (!tcp_timer_active(tp, TT_PERSIST))
2787 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
2788
2789 /*
2790 * If no data (only SYN) was ACK'd,
2791 * skip rest of ACK processing.
2792 */
2793 if (acked == 0)
2794 goto step6;
2795
2796 /*
2797 * Let the congestion control algorithm update congestion
2798 * control related information. This typically means increasing
2799 * the congestion window.
2800 */
2801 cc_ack_received(tp, th, CC_ACK);
2802
2803 SOCKBUF_LOCK(&so->so_snd);
2804 if (acked > so->so_snd.sb_cc) {
2805 tp->snd_wnd -= so->so_snd.sb_cc;
2806 mfree = sbcut_locked(&so->so_snd,
2807 (int)so->so_snd.sb_cc);
2808 ourfinisacked = 1;
2809 } else {
2810 mfree = sbcut_locked(&so->so_snd, acked);
2811 tp->snd_wnd -= acked;
2812 ourfinisacked = 0;
2813 }
2814 /* NB: sowwakeup_locked() does an implicit unlock. */
2815 sowwakeup_locked(so);
2816 m_freem(mfree);
2817 /* Detect una wraparound. */
2818 if (!IN_RECOVERY(tp->t_flags) &&
2819 SEQ_GT(tp->snd_una, tp->snd_recover) &&
2820 SEQ_LEQ(th->th_ack, tp->snd_recover))
2821 tp->snd_recover = th->th_ack - 1;
2822 /* XXXLAS: Can this be moved up into cc_post_recovery? */
2823 if (IN_RECOVERY(tp->t_flags) &&
2824 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2825 EXIT_RECOVERY(tp->t_flags);
2826 }
2827 tp->snd_una = th->th_ack;
2828 if (tp->t_flags & TF_SACK_PERMIT) {
2829 if (SEQ_GT(tp->snd_una, tp->snd_recover))
2830 tp->snd_recover = tp->snd_una;
2831 }
2832 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2833 tp->snd_nxt = tp->snd_una;
2834
2835 switch (tp->t_state) {
2836
2837 /*
2838 * In FIN_WAIT_1 STATE in addition to the processing
2839 * for the ESTABLISHED state if our FIN is now acknowledged
2840 * then enter FIN_WAIT_2.
2841 */
2842 case TCPS_FIN_WAIT_1:
2843 if (ourfinisacked) {
2844 /*
2845 * If we can't receive any more
2846 * data, then closing user can proceed.
2847 * Starting the timer is contrary to the
2848 * specification, but if we don't get a FIN
2849 * we'll hang forever.
2850 *
2851 * XXXjl:
2852 * we should release the tp also, and use a
2853 * compressed state.
2854 */
2855 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2856 soisdisconnected(so);
2857 tcp_timer_activate(tp, TT_2MSL,
2858 (tcp_fast_finwait2_recycle ?
2859 tcp_finwait2_timeout :
2860 TP_MAXIDLE(tp)));
2861 }
2862 tcp_state_change(tp, TCPS_FIN_WAIT_2);
2863 }
2864 break;
2865
2866 /*
2867 * In CLOSING STATE in addition to the processing for
2868 * the ESTABLISHED state if the ACK acknowledges our FIN
2869 * then enter the TIME-WAIT state, otherwise ignore
2870 * the segment.
2871 */
2872 case TCPS_CLOSING:
2873 if (ourfinisacked) {
2874 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2875 tcp_twstart(tp);
2876 INP_INFO_WUNLOCK(&V_tcbinfo);
2877 m_freem(m);
2878 return;
2879 }
2880 break;
2881
2882 /*
2883 * In LAST_ACK, we may still be waiting for data to drain
2884 * and/or to be acked, as well as for the ack of our FIN.
2885 * If our FIN is now acknowledged, delete the TCB,
2886 * enter the closed state and return.
2887 */
2888 case TCPS_LAST_ACK:
2889 if (ourfinisacked) {
2890 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2891 tp = tcp_close(tp);
2892 goto drop;
2893 }
2894 break;
2895 }
2896 }
2897
2898 step6:
2899 INP_WLOCK_ASSERT(tp->t_inpcb);
2900
2901 /*
2902 * Update window information.
2903 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2904 */
2905 if ((thflags & TH_ACK) &&
2906 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2907 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2908 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2909 /* keep track of pure window updates */
2910 if (tlen == 0 &&
2911 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2912 TCPSTAT_INC(tcps_rcvwinupd);
2913 tp->snd_wnd = tiwin;
2914 tp->snd_wl1 = th->th_seq;
2915 tp->snd_wl2 = th->th_ack;
2916 if (tp->snd_wnd > tp->max_sndwnd)
2917 tp->max_sndwnd = tp->snd_wnd;
2918 needoutput = 1;
2919 }
2920
2921 /*
2922 * Process segments with URG.
2923 */
2924 if ((thflags & TH_URG) && th->th_urp &&
2925 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2926 /*
2927 * This is a kludge, but if we receive and accept
2928 * random urgent pointers, we'll crash in
2929 * soreceive. It's hard to imagine someone
2930 * actually wanting to send this much urgent data.
2931 */
2932 SOCKBUF_LOCK(&so->so_rcv);
2933 if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2934 th->th_urp = 0; /* XXX */
2935 thflags &= ~TH_URG; /* XXX */
2936 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */
2937 goto dodata; /* XXX */
2938 }
2939 /*
2940 * If this segment advances the known urgent pointer,
2941 * then mark the data stream. This should not happen
2942 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2943 * a FIN has been received from the remote side.
2944 * In these states we ignore the URG.
2945 *
2946 * According to RFC961 (Assigned Protocols),
2947 * the urgent pointer points to the last octet
2948 * of urgent data. We continue, however,
2949 * to consider it to indicate the first octet
2950 * of data past the urgent section as the original
2951 * spec states (in one of two places).
2952 */
2953 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2954 tp->rcv_up = th->th_seq + th->th_urp;
2955 so->so_oobmark = so->so_rcv.sb_cc +
2956 (tp->rcv_up - tp->rcv_nxt) - 1;
2957 if (so->so_oobmark == 0)
2958 so->so_rcv.sb_state |= SBS_RCVATMARK;
2959 sohasoutofband(so);
2960 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2961 }
2962 SOCKBUF_UNLOCK(&so->so_rcv);
2963 /*
2964 * Remove out of band data so doesn't get presented to user.
2965 * This can happen independent of advancing the URG pointer,
2966 * but if two URG's are pending at once, some out-of-band
2967 * data may creep in... ick.
2968 */
2969 if (th->th_urp <= (u_long)tlen &&
2970 !(so->so_options & SO_OOBINLINE)) {
2971 /* hdr drop is delayed */
2972 tcp_pulloutofband(so, th, m, drop_hdrlen);
2973 }
2974 } else {
2975 /*
2976 * If no out of band data is expected,
2977 * pull receive urgent pointer along
2978 * with the receive window.
2979 */
2980 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2981 tp->rcv_up = tp->rcv_nxt;
2982 }
2983 dodata: /* XXX */
2984 INP_WLOCK_ASSERT(tp->t_inpcb);
2985
2986 /*
2987 * Process the segment text, merging it into the TCP sequencing queue,
2988 * and arranging for acknowledgment of receipt if necessary.
2989 * This process logically involves adjusting tp->rcv_wnd as data
2990 * is presented to the user (this happens in tcp_usrreq.c,
2991 * case PRU_RCVD). If a FIN has already been received on this
2992 * connection then we just ignore the text.
2993 */
2994 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) &&
2995 (tp->t_flags & TF_FASTOPEN));
2996 if ((tlen || (thflags & TH_FIN) || tfo_syn) &&
2997 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2998 tcp_seq save_start = th->th_seq;
2999
3000 m_adj(m, drop_hdrlen); /* delayed header drop */
3001 /*
3002 * Insert segment which includes th into TCP reassembly queue
3003 * with control block tp. Set thflags to whether reassembly now
3004 * includes a segment with FIN. This handles the common case
3005 * inline (segment is the next to be received on an established
3006 * connection, and the queue is empty), avoiding linkage into
3007 * and removal from the queue and repetition of various
3008 * conversions.
3009 * Set DELACK for segments received in order, but ack
3010 * immediately when segments are out of order (so
3011 * fast retransmit can work).
3012 */
3013 if (th->th_seq == tp->rcv_nxt &&
3014 LIST_EMPTY(&tp->t_segq) &&
3015 (TCPS_HAVEESTABLISHED(tp->t_state) ||
3016 tfo_syn)) {
3017 if (DELAY_ACK(tp, tlen) || tfo_syn)
3018 tp->t_flags |= TF_DELACK;
3019 else
3020 tp->t_flags |= TF_ACKNOW;
3021 tp->rcv_nxt += tlen;
3022 thflags = th->th_flags & TH_FIN;
3023 TCPSTAT_INC(tcps_rcvpack);
3024 TCPSTAT_ADD(tcps_rcvbyte, tlen);
3025 ND6_HINT(tp);
3026 SOCKBUF_LOCK(&so->so_rcv);
3027 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
3028 m_freem(m);
3029 else
3030 sbappendstream_locked(&so->so_rcv, m);
3031 /* NB: sorwakeup_locked() does an implicit unlock. */
3032 sorwakeup_locked(so);
3033 } else {
3034 /*
3035 * XXX: Due to the header drop above "th" is
3036 * theoretically invalid by now. Fortunately
3037 * m_adj() doesn't actually frees any mbufs
3038 * when trimming from the head.
3039 */
3040 thflags = tcp_reass(tp, th, &tlen, m);
3041 tp->t_flags |= TF_ACKNOW;
3042 }
3043 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT))
3044 tcp_update_sack_list(tp, save_start, save_start + tlen);
3045 #if 0
3046 /*
3047 * Note the amount of data that peer has sent into
3048 * our window, in order to estimate the sender's
3049 * buffer size.
3050 * XXX: Unused.
3051 */
3052 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
3053 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
3054 else
3055 len = so->so_rcv.sb_hiwat;
3056 #endif
3057 } else {
3058 m_freem(m);
3059 thflags &= ~TH_FIN;
3060 }
3061
3062 /*
3063 * If FIN is received ACK the FIN and let the user know
3064 * that the connection is closing.
3065 */
3066 if (thflags & TH_FIN) {
3067 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3068 socantrcvmore(so);
3069 /*
3070 * If connection is half-synchronized
3071 * (ie NEEDSYN flag on) then delay ACK,
3072 * so it may be piggybacked when SYN is sent.
3073 * Otherwise, since we received a FIN then no
3074 * more input can be expected, send ACK now.
3075 */
3076 if (tp->t_flags & TF_NEEDSYN)
3077 tp->t_flags |= TF_DELACK;
3078 else
3079 tp->t_flags |= TF_ACKNOW;
3080 tp->rcv_nxt++;
3081 }
3082 switch (tp->t_state) {
3083
3084 /*
3085 * In SYN_RECEIVED and ESTABLISHED STATES
3086 * enter the CLOSE_WAIT state.
3087 */
3088 case TCPS_SYN_RECEIVED:
3089 tp->t_starttime = ticks;
3090 /* FALLTHROUGH */
3091 case TCPS_ESTABLISHED:
3092 tcp_state_change(tp, TCPS_CLOSE_WAIT);
3093 break;
3094
3095 /*
3096 * If still in FIN_WAIT_1 STATE FIN has not been acked so
3097 * enter the CLOSING state.
3098 */
3099 case TCPS_FIN_WAIT_1:
3100 tcp_state_change(tp, TCPS_CLOSING);
3101 break;
3102
3103 /*
3104 * In FIN_WAIT_2 state enter the TIME_WAIT state,
3105 * starting the time-wait timer, turning off the other
3106 * standard timers.
3107 */
3108 case TCPS_FIN_WAIT_2:
3109 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
3110 KASSERT(ti_locked == TI_WLOCKED, ("%s: dodata "
3111 "TCP_FIN_WAIT_2 ti_locked: %d", __func__,
3112 ti_locked));
3113
3114 tcp_twstart(tp);
3115 INP_INFO_WUNLOCK(&V_tcbinfo);
3116 return;
3117 }
3118 }
3119 if (ti_locked == TI_WLOCKED)
3120 INP_INFO_WUNLOCK(&V_tcbinfo);
3121 ti_locked = TI_UNLOCKED;
3122
3123 #ifdef TCPDEBUG
3124 if (so->so_options & SO_DEBUG)
3125 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
3126 &tcp_savetcp, 0);
3127 #endif
3128
3129 /*
3130 * Return any desired output.
3131 */
3132 if (needoutput || (tp->t_flags & TF_ACKNOW))
3133 (void) tcp_output(tp);
3134
3135 check_delack:
3136 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d",
3137 __func__, ti_locked));
3138 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
3139 INP_WLOCK_ASSERT(tp->t_inpcb);
3140
3141 if (tp->t_flags & TF_DELACK) {
3142 tp->t_flags &= ~TF_DELACK;
3143 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
3144 }
3145 INP_WUNLOCK(tp->t_inpcb);
3146 return;
3147
3148 dropafterack:
3149 /*
3150 * Generate an ACK dropping incoming segment if it occupies
3151 * sequence space, where the ACK reflects our state.
3152 *
3153 * We can now skip the test for the RST flag since all
3154 * paths to this code happen after packets containing
3155 * RST have been dropped.
3156 *
3157 * In the SYN-RECEIVED state, don't send an ACK unless the
3158 * segment we received passes the SYN-RECEIVED ACK test.
3159 * If it fails send a RST. This breaks the loop in the
3160 * "LAND" DoS attack, and also prevents an ACK storm
3161 * between two listening ports that have been sent forged
3162 * SYN segments, each with the source address of the other.
3163 */
3164 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
3165 (SEQ_GT(tp->snd_una, th->th_ack) ||
3166 SEQ_GT(th->th_ack, tp->snd_max)) ) {
3167 rstreason = BANDLIM_RST_OPENPORT;
3168 goto dropwithreset;
3169 }
3170 #ifdef TCPDEBUG
3171 if (so->so_options & SO_DEBUG)
3172 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3173 &tcp_savetcp, 0);
3174 #endif
3175 if (ti_locked == TI_WLOCKED)
3176 INP_INFO_WUNLOCK(&V_tcbinfo);
3177 ti_locked = TI_UNLOCKED;
3178
3179 tp->t_flags |= TF_ACKNOW;
3180 (void) tcp_output(tp);
3181 INP_WUNLOCK(tp->t_inpcb);
3182 m_freem(m);
3183 return;
3184
3185 dropwithreset:
3186 if (ti_locked == TI_WLOCKED)
3187 INP_INFO_WUNLOCK(&V_tcbinfo);
3188 ti_locked = TI_UNLOCKED;
3189
3190 if (tp != NULL) {
3191 tcp_dropwithreset(m, th, tp, tlen, rstreason);
3192 INP_WUNLOCK(tp->t_inpcb);
3193 } else
3194 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
3195 return;
3196
3197 drop:
3198 if (ti_locked == TI_WLOCKED) {
3199 INP_INFO_WUNLOCK(&V_tcbinfo);
3200 ti_locked = TI_UNLOCKED;
3201 }
3202 #ifdef INVARIANTS
3203 else
3204 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
3205 #endif
3206
3207 /*
3208 * Drop space held by incoming segment and return.
3209 */
3210 #ifdef TCPDEBUG
3211 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
3212 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
3213 &tcp_savetcp, 0);
3214 #endif
3215 if (tp != NULL)
3216 INP_WUNLOCK(tp->t_inpcb);
3217 m_freem(m);
3218 }
3219
3220 /*
3221 * Issue RST and make ACK acceptable to originator of segment.
3222 * The mbuf must still include the original packet header.
3223 * tp may be NULL.
3224 */
3225 static void
3226 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
3227 int tlen, int rstreason)
3228 {
3229 #ifdef INET
3230 struct ip *ip;
3231 #endif
3232 #ifdef INET6
3233 struct ip6_hdr *ip6;
3234 #endif
3235
3236 if (tp != NULL) {
3237 INP_WLOCK_ASSERT(tp->t_inpcb);
3238 }
3239
3240 /* Don't bother if destination was broadcast/multicast. */
3241 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
3242 goto drop;
3243 #ifdef INET6
3244 if (mtod(m, struct ip *)->ip_v == 6) {
3245 ip6 = mtod(m, struct ip6_hdr *);
3246 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3247 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3248 goto drop;
3249 /* IPv6 anycast check is done at tcp6_input() */
3250 }
3251 #endif
3252 #if defined(INET) && defined(INET6)
3253 else
3254 #endif
3255 #ifdef INET
3256 {
3257 ip = mtod(m, struct ip *);
3258 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3259 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3260 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3261 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3262 goto drop;
3263 }
3264 #endif
3265
3266 /* Perform bandwidth limiting. */
3267 if (badport_bandlim(rstreason) < 0)
3268 goto drop;
3269
3270 /* tcp_respond consumes the mbuf chain. */
3271 if (th->th_flags & TH_ACK) {
3272 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0,
3273 th->th_ack, TH_RST);
3274 } else {
3275 if (th->th_flags & TH_SYN)
3276 tlen++;
3277 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
3278 (tcp_seq)0, TH_RST|TH_ACK);
3279 }
3280 return;
3281 drop:
3282 m_freem(m);
3283 }
3284
3285 /*
3286 * Parse TCP options and place in tcpopt.
3287 */
3288 static void
3289 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags)
3290 {
3291 int opt, optlen;
3292
3293 to->to_flags = 0;
3294 for (; cnt > 0; cnt -= optlen, cp += optlen) {
3295 opt = cp[0];
3296 if (opt == TCPOPT_EOL)
3297 break;
3298 if (opt == TCPOPT_NOP)
3299 optlen = 1;
3300 else {
3301 if (cnt < 2)
3302 break;
3303 optlen = cp[1];
3304 if (optlen < 2 || optlen > cnt)
3305 break;
3306 }
3307 switch (opt) {
3308 case TCPOPT_MAXSEG:
3309 if (optlen != TCPOLEN_MAXSEG)
3310 continue;
3311 if (!(flags & TO_SYN))
3312 continue;
3313 to->to_flags |= TOF_MSS;
3314 bcopy((char *)cp + 2,
3315 (char *)&to->to_mss, sizeof(to->to_mss));
3316 to->to_mss = ntohs(to->to_mss);
3317 break;
3318 case TCPOPT_WINDOW:
3319 if (optlen != TCPOLEN_WINDOW)
3320 continue;
3321 if (!(flags & TO_SYN))
3322 continue;
3323 to->to_flags |= TOF_SCALE;
3324 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT);
3325 break;
3326 case TCPOPT_TIMESTAMP:
3327 if (optlen != TCPOLEN_TIMESTAMP)
3328 continue;
3329 to->to_flags |= TOF_TS;
3330 bcopy((char *)cp + 2,
3331 (char *)&to->to_tsval, sizeof(to->to_tsval));
3332 to->to_tsval = ntohl(to->to_tsval);
3333 bcopy((char *)cp + 6,
3334 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
3335 to->to_tsecr = ntohl(to->to_tsecr);
3336 break;
3337 #ifdef TCP_SIGNATURE
3338 /*
3339 * XXX In order to reply to a host which has set the
3340 * TCP_SIGNATURE option in its initial SYN, we have to
3341 * record the fact that the option was observed here
3342 * for the syncache code to perform the correct response.
3343 */
3344 case TCPOPT_SIGNATURE:
3345 if (optlen != TCPOLEN_SIGNATURE)
3346 continue;
3347 to->to_flags |= TOF_SIGNATURE;
3348 to->to_signature = cp + 2;
3349 break;
3350 #endif
3351 case TCPOPT_SACK_PERMITTED:
3352 if (optlen != TCPOLEN_SACK_PERMITTED)
3353 continue;
3354 if (!(flags & TO_SYN))
3355 continue;
3356 if (!V_tcp_do_sack)
3357 continue;
3358 to->to_flags |= TOF_SACKPERM;
3359 break;
3360 case TCPOPT_SACK:
3361 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
3362 continue;
3363 if (flags & TO_SYN)
3364 continue;
3365 to->to_flags |= TOF_SACK;
3366 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
3367 to->to_sacks = cp + 2;
3368 TCPSTAT_INC(tcps_sack_rcv_blocks);
3369 break;
3370 #ifdef TCP_RFC7413
3371 case TCPOPT_FAST_OPEN:
3372 if ((optlen != TCPOLEN_FAST_OPEN_EMPTY) &&
3373 (optlen < TCPOLEN_FAST_OPEN_MIN) &&
3374 (optlen > TCPOLEN_FAST_OPEN_MAX))
3375 continue;
3376 if (!(flags & TO_SYN))
3377 continue;
3378 if (!V_tcp_fastopen_enabled)
3379 continue;
3380 to->to_flags |= TOF_FASTOPEN;
3381 to->to_tfo_len = optlen - 2;
3382 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL;
3383 break;
3384 #endif
3385 default:
3386 continue;
3387 }
3388 }
3389 }
3390
3391 /*
3392 * Pull out of band byte out of a segment so
3393 * it doesn't appear in the user's data queue.
3394 * It is still reflected in the segment length for
3395 * sequencing purposes.
3396 */
3397 static void
3398 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m,
3399 int off)
3400 {
3401 int cnt = off + th->th_urp - 1;
3402
3403 while (cnt >= 0) {
3404 if (m->m_len > cnt) {
3405 char *cp = mtod(m, caddr_t) + cnt;
3406 struct tcpcb *tp = sototcpcb(so);
3407
3408 INP_WLOCK_ASSERT(tp->t_inpcb);
3409
3410 tp->t_iobc = *cp;
3411 tp->t_oobflags |= TCPOOB_HAVEDATA;
3412 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3413 m->m_len--;
3414 if (m->m_flags & M_PKTHDR)
3415 m->m_pkthdr.len--;
3416 return;
3417 }
3418 cnt -= m->m_len;
3419 m = m->m_next;
3420 if (m == NULL)
3421 break;
3422 }
3423 panic("tcp_pulloutofband");
3424 }
3425
3426 /*
3427 * Collect new round-trip time estimate
3428 * and update averages and current timeout.
3429 */
3430 static void
3431 tcp_xmit_timer(struct tcpcb *tp, int rtt)
3432 {
3433 int delta;
3434
3435 INP_WLOCK_ASSERT(tp->t_inpcb);
3436
3437 TCPSTAT_INC(tcps_rttupdated);
3438 tp->t_rttupdated++;
3439 if (tp->t_srtt != 0) {
3440 /*
3441 * srtt is stored as fixed point with 5 bits after the
3442 * binary point (i.e., scaled by 8). The following magic
3443 * is equivalent to the smoothing algorithm in rfc793 with
3444 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3445 * point). Adjust rtt to origin 0.
3446 */
3447 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
3448 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
3449
3450 if ((tp->t_srtt += delta) <= 0)
3451 tp->t_srtt = 1;
3452
3453 /*
3454 * We accumulate a smoothed rtt variance (actually, a
3455 * smoothed mean difference), then set the retransmit
3456 * timer to smoothed rtt + 4 times the smoothed variance.
3457 * rttvar is stored as fixed point with 4 bits after the
3458 * binary point (scaled by 16). The following is
3459 * equivalent to rfc793 smoothing with an alpha of .75
3460 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
3461 * rfc793's wired-in beta.
3462 */
3463 if (delta < 0)
3464 delta = -delta;
3465 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
3466 if ((tp->t_rttvar += delta) <= 0)
3467 tp->t_rttvar = 1;
3468 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
3469 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3470 } else {
3471 /*
3472 * No rtt measurement yet - use the unsmoothed rtt.
3473 * Set the variance to half the rtt (so our first
3474 * retransmit happens at 3*rtt).
3475 */
3476 tp->t_srtt = rtt << TCP_RTT_SHIFT;
3477 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
3478 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3479 }
3480 tp->t_rtttime = 0;
3481 tp->t_rxtshift = 0;
3482
3483 /*
3484 * the retransmit should happen at rtt + 4 * rttvar.
3485 * Because of the way we do the smoothing, srtt and rttvar
3486 * will each average +1/2 tick of bias. When we compute
3487 * the retransmit timer, we want 1/2 tick of rounding and
3488 * 1 extra tick because of +-1/2 tick uncertainty in the
3489 * firing of the timer. The bias will give us exactly the
3490 * 1.5 tick we need. But, because the bias is
3491 * statistical, we have to test that we don't drop below
3492 * the minimum feasible timer (which is 2 ticks).
3493 */
3494 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3495 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3496
3497 /*
3498 * We received an ack for a packet that wasn't retransmitted;
3499 * it is probably safe to discard any error indications we've
3500 * received recently. This isn't quite right, but close enough
3501 * for now (a route might have failed after we sent a segment,
3502 * and the return path might not be symmetrical).
3503 */
3504 tp->t_softerror = 0;
3505 }
3506
3507 /*
3508 * Determine a reasonable value for maxseg size.
3509 * If the route is known, check route for mtu.
3510 * If none, use an mss that can be handled on the outgoing interface
3511 * without forcing IP to fragment. If no route is found, route has no mtu,
3512 * or the destination isn't local, use a default, hopefully conservative
3513 * size (usually 512 or the default IP max size, but no more than the mtu
3514 * of the interface), as we can't discover anything about intervening
3515 * gateways or networks. We also initialize the congestion/slow start
3516 * window to be a single segment if the destination isn't local.
3517 * While looking at the routing entry, we also initialize other path-dependent
3518 * parameters from pre-set or cached values in the routing entry.
3519 *
3520 * Also take into account the space needed for options that we
3521 * send regularly. Make maxseg shorter by that amount to assure
3522 * that we can send maxseg amount of data even when the options
3523 * are present. Store the upper limit of the length of options plus
3524 * data in maxopd.
3525 *
3526 * NOTE that this routine is only called when we process an incoming
3527 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS
3528 * settings are handled in tcp_mssopt().
3529 */
3530 void
3531 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer,
3532 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap)
3533 {
3534 int mss = 0;
3535 u_long maxmtu = 0;
3536 struct inpcb *inp = tp->t_inpcb;
3537 struct hc_metrics_lite metrics;
3538 int origoffer;
3539 #ifdef INET6
3540 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
3541 size_t min_protoh = isipv6 ?
3542 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
3543 sizeof (struct tcpiphdr);
3544 #else
3545 const size_t min_protoh = sizeof(struct tcpiphdr);
3546 #endif
3547
3548 INP_WLOCK_ASSERT(tp->t_inpcb);
3549
3550 if (mtuoffer != -1) {
3551 KASSERT(offer == -1, ("%s: conflict", __func__));
3552 offer = mtuoffer - min_protoh;
3553 }
3554 origoffer = offer;
3555
3556 /* Initialize. */
3557 #ifdef INET6
3558 if (isipv6) {
3559 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap);
3560 tp->t_maxopd = tp->t_maxseg = V_tcp_v6mssdflt;
3561 }
3562 #endif
3563 #if defined(INET) && defined(INET6)
3564 else
3565 #endif
3566 #ifdef INET
3567 {
3568 maxmtu = tcp_maxmtu(&inp->inp_inc, cap);
3569 tp->t_maxopd = tp->t_maxseg = V_tcp_mssdflt;
3570 }
3571 #endif
3572
3573 /*
3574 * No route to sender, stay with default mss and return.
3575 */
3576 if (maxmtu == 0) {
3577 /*
3578 * In case we return early we need to initialize metrics
3579 * to a defined state as tcp_hc_get() would do for us
3580 * if there was no cache hit.
3581 */
3582 if (metricptr != NULL)
3583 bzero(metricptr, sizeof(struct hc_metrics_lite));
3584 return;
3585 }
3586
3587 /* What have we got? */
3588 switch (offer) {
3589 case 0:
3590 /*
3591 * Offer == 0 means that there was no MSS on the SYN
3592 * segment, in this case we use tcp_mssdflt as
3593 * already assigned to t_maxopd above.
3594 */
3595 offer = tp->t_maxopd;
3596 break;
3597
3598 case -1:
3599 /*
3600 * Offer == -1 means that we didn't receive SYN yet.
3601 */
3602 /* FALLTHROUGH */
3603
3604 default:
3605 /*
3606 * Prevent DoS attack with too small MSS. Round up
3607 * to at least minmss.
3608 */
3609 offer = max(offer, V_tcp_minmss);
3610 }
3611
3612 /*
3613 * rmx information is now retrieved from tcp_hostcache.
3614 */
3615 tcp_hc_get(&inp->inp_inc, &metrics);
3616 if (metricptr != NULL)
3617 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite));
3618
3619 /*
3620 * If there's a discovered mtu int tcp hostcache, use it
3621 * else, use the link mtu.
3622 */
3623 if (metrics.rmx_mtu)
3624 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
3625 else {
3626 #ifdef INET6
3627 if (isipv6) {
3628 mss = maxmtu - min_protoh;
3629 if (!V_path_mtu_discovery &&
3630 !in6_localaddr(&inp->in6p_faddr))
3631 mss = min(mss, V_tcp_v6mssdflt);
3632 }
3633 #endif
3634 #if defined(INET) && defined(INET6)
3635 else
3636 #endif
3637 #ifdef INET
3638 {
3639 mss = maxmtu - min_protoh;
3640 if (!V_path_mtu_discovery &&
3641 !in_localaddr(inp->inp_faddr))
3642 mss = min(mss, V_tcp_mssdflt);
3643 }
3644 #endif
3645 /*
3646 * XXX - The above conditional (mss = maxmtu - min_protoh)
3647 * probably violates the TCP spec.
3648 * The problem is that, since we don't know the
3649 * other end's MSS, we are supposed to use a conservative
3650 * default. But, if we do that, then MTU discovery will
3651 * never actually take place, because the conservative
3652 * default is much less than the MTUs typically seen
3653 * on the Internet today. For the moment, we'll sweep
3654 * this under the carpet.
3655 *
3656 * The conservative default might not actually be a problem
3657 * if the only case this occurs is when sending an initial
3658 * SYN with options and data to a host we've never talked
3659 * to before. Then, they will reply with an MSS value which
3660 * will get recorded and the new parameters should get
3661 * recomputed. For Further Study.
3662 */
3663 }
3664 mss = min(mss, offer);
3665
3666 /*
3667 * Sanity check: make sure that maxopd will be large
3668 * enough to allow some data on segments even if the
3669 * all the option space is used (40bytes). Otherwise
3670 * funny things may happen in tcp_output.
3671 */
3672 mss = max(mss, 64);
3673
3674 /*
3675 * maxopd stores the maximum length of data AND options
3676 * in a segment; maxseg is the amount of data in a normal
3677 * segment. We need to store this value (maxopd) apart
3678 * from maxseg, because now every segment carries options
3679 * and thus we normally have somewhat less data in segments.
3680 */
3681 tp->t_maxopd = mss;
3682
3683 /*
3684 * origoffer==-1 indicates that no segments were received yet.
3685 * In this case we just guess.
3686 */
3687 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
3688 (origoffer == -1 ||
3689 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
3690 mss -= TCPOLEN_TSTAMP_APPA;
3691
3692 tp->t_maxseg = mss;
3693 }
3694
3695 void
3696 tcp_mss(struct tcpcb *tp, int offer)
3697 {
3698 int mss;
3699 u_long bufsize;
3700 struct inpcb *inp;
3701 struct socket *so;
3702 struct hc_metrics_lite metrics;
3703 struct tcp_ifcap cap;
3704
3705 KASSERT(tp != NULL, ("%s: tp == NULL", __func__));
3706
3707 bzero(&cap, sizeof(cap));
3708 tcp_mss_update(tp, offer, -1, &metrics, &cap);
3709
3710 mss = tp->t_maxseg;
3711 inp = tp->t_inpcb;
3712
3713 /*
3714 * If there's a pipesize, change the socket buffer to that size,
3715 * don't change if sb_hiwat is different than default (then it
3716 * has been changed on purpose with setsockopt).
3717 * Make the socket buffers an integral number of mss units;
3718 * if the mss is larger than the socket buffer, decrease the mss.
3719 */
3720 so = inp->inp_socket;
3721 SOCKBUF_LOCK(&so->so_snd);
3722 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe)
3723 bufsize = metrics.rmx_sendpipe;
3724 else
3725 bufsize = so->so_snd.sb_hiwat;
3726 if (bufsize < mss)
3727 mss = bufsize;
3728 else {
3729 bufsize = roundup(bufsize, mss);
3730 if (bufsize > sb_max)
3731 bufsize = sb_max;
3732 if (bufsize > so->so_snd.sb_hiwat)
3733 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
3734 }
3735 SOCKBUF_UNLOCK(&so->so_snd);
3736 tp->t_maxseg = mss;
3737
3738 SOCKBUF_LOCK(&so->so_rcv);
3739 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe)
3740 bufsize = metrics.rmx_recvpipe;
3741 else
3742 bufsize = so->so_rcv.sb_hiwat;
3743 if (bufsize > mss) {
3744 bufsize = roundup(bufsize, mss);
3745 if (bufsize > sb_max)
3746 bufsize = sb_max;
3747 if (bufsize > so->so_rcv.sb_hiwat)
3748 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
3749 }
3750 SOCKBUF_UNLOCK(&so->so_rcv);
3751
3752 /* Check the interface for TSO capabilities. */
3753 if (cap.ifcap & CSUM_TSO) {
3754 tp->t_flags |= TF_TSO;
3755 tp->t_tsomax = cap.tsomax;
3756 tp->t_tsomaxsegcount = cap.tsomaxsegcount;
3757 tp->t_tsomaxsegsize = cap.tsomaxsegsize;
3758 }
3759 }
3760
3761 /*
3762 * Determine the MSS option to send on an outgoing SYN.
3763 */
3764 int
3765 tcp_mssopt(struct in_conninfo *inc)
3766 {
3767 int mss = 0;
3768 u_long maxmtu = 0;
3769 u_long thcmtu = 0;
3770 size_t min_protoh;
3771
3772 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3773
3774 #ifdef INET6
3775 if (inc->inc_flags & INC_ISIPV6) {
3776 mss = V_tcp_v6mssdflt;
3777 maxmtu = tcp_maxmtu6(inc, NULL);
3778 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3779 }
3780 #endif
3781 #if defined(INET) && defined(INET6)
3782 else
3783 #endif
3784 #ifdef INET
3785 {
3786 mss = V_tcp_mssdflt;
3787 maxmtu = tcp_maxmtu(inc, NULL);
3788 min_protoh = sizeof(struct tcpiphdr);
3789 }
3790 #endif
3791 #if defined(INET6) || defined(INET)
3792 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3793 #endif
3794
3795 if (maxmtu && thcmtu)
3796 mss = min(maxmtu, thcmtu) - min_protoh;
3797 else if (maxmtu || thcmtu)
3798 mss = max(maxmtu, thcmtu) - min_protoh;
3799
3800 return (mss);
3801 }
3802
3803
3804 /*
3805 * On a partial ack arrives, force the retransmission of the
3806 * next unacknowledged segment. Do not clear tp->t_dupacks.
3807 * By setting snd_nxt to ti_ack, this forces retransmission timer to
3808 * be started again.
3809 */
3810 static void
3811 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th)
3812 {
3813 tcp_seq onxt = tp->snd_nxt;
3814 u_long ocwnd = tp->snd_cwnd;
3815
3816 INP_WLOCK_ASSERT(tp->t_inpcb);
3817
3818 tcp_timer_activate(tp, TT_REXMT, 0);
3819 tp->t_rtttime = 0;
3820 tp->snd_nxt = th->th_ack;
3821 /*
3822 * Set snd_cwnd to one segment beyond acknowledged offset.
3823 * (tp->snd_una has not yet been updated when this function is called.)
3824 */
3825 tp->snd_cwnd = tp->t_maxseg + BYTES_THIS_ACK(tp, th);
3826 tp->t_flags |= TF_ACKNOW;
3827 (void) tcp_output(tp);
3828 tp->snd_cwnd = ocwnd;
3829 if (SEQ_GT(onxt, tp->snd_nxt))
3830 tp->snd_nxt = onxt;
3831 /*
3832 * Partial window deflation. Relies on fact that tp->snd_una
3833 * not updated yet.
3834 */
3835 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th))
3836 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th);
3837 else
3838 tp->snd_cwnd = 0;
3839 tp->snd_cwnd += tp->t_maxseg;
3840 }
3841
3842 int
3843 tcp_compute_pipe(struct tcpcb *tp)
3844 {
3845 return (tp->snd_max - tp->snd_una +
3846 tp->sackhint.sack_bytes_rexmit -
3847 tp->sackhint.sacked_bytes);
3848 }
Cache object: 0074b6349189a80d0e9a189d9112e897
|