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