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 * All rights reserved.
9 *
10 * Portions of this software were developed at the Centre for Advanced Internet
11 * Architectures, Swinburne University of Technology, by Lawrence Stewart,
12 * James Healy and David Hayes, made possible in part by a grant from the Cisco
13 * University Research Program Fund at Community Foundation Silicon Valley.
14 *
15 * Portions of this software were developed at the Centre for Advanced
16 * Internet Architectures, Swinburne University of Technology, Melbourne,
17 * Australia by David Hayes under sponsorship from the FreeBSD Foundation.
18 *
19 * Redistribution and use in source and binary forms, with or without
20 * modification, are permitted provided that the following conditions
21 * are met:
22 * 1. Redistributions of source code must retain the above copyright
23 * notice, this list of conditions and the following disclaimer.
24 * 2. Redistributions in binary form must reproduce the above copyright
25 * notice, this list of conditions and the following disclaimer in the
26 * documentation and/or other materials provided with the distribution.
27 * 4. Neither the name of the University nor the names of its contributors
28 * may be used to endorse or promote products derived from this software
29 * without specific prior written permission.
30 *
31 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
32 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
33 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
34 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
35 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
36 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
37 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
38 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
39 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
40 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
41 * SUCH DAMAGE.
42 *
43 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95
44 */
45
46 #include <sys/cdefs.h>
47 __FBSDID("$FreeBSD: releng/8.3/sys/netinet/tcp_input.c 236953 2012-06-12 12:10:10Z bz $");
48
49 #include "opt_ipfw.h" /* for ipfw_fwd */
50 #include "opt_inet.h"
51 #include "opt_inet6.h"
52 #include "opt_ipsec.h"
53 #include "opt_tcpdebug.h"
54
55 #include <sys/param.h>
56 #include <sys/kernel.h>
57 #include <sys/hhook.h>
58 #include <sys/malloc.h>
59 #include <sys/mbuf.h>
60 #include <sys/proc.h> /* for proc0 declaration */
61 #include <sys/protosw.h>
62 #include <sys/signalvar.h>
63 #include <sys/socket.h>
64 #include <sys/socketvar.h>
65 #include <sys/sysctl.h>
66 #include <sys/syslog.h>
67 #include <sys/systm.h>
68
69 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
70
71 #include <vm/uma.h>
72
73 #include <net/if.h>
74 #include <net/route.h>
75 #include <net/vnet.h>
76
77 #define TCPSTATES /* for logging */
78
79 #include <netinet/cc.h>
80 #include <netinet/in.h>
81 #include <netinet/in_pcb.h>
82 #include <netinet/in_systm.h>
83 #include <netinet/in_var.h>
84 #include <netinet/ip.h>
85 #include <netinet/ip_icmp.h> /* required for icmp_var.h */
86 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */
87 #include <netinet/ip_var.h>
88 #include <netinet/ip_options.h>
89 #include <netinet/ip6.h>
90 #include <netinet/icmp6.h>
91 #include <netinet6/in6_pcb.h>
92 #include <netinet6/ip6_var.h>
93 #include <netinet6/nd6.h>
94 #include <netinet/tcp_fsm.h>
95 #include <netinet/tcp_seq.h>
96 #include <netinet/tcp_timer.h>
97 #include <netinet/tcp_var.h>
98 #include <netinet6/tcp6_var.h>
99 #include <netinet/tcpip.h>
100 #include <netinet/tcp_syncache.h>
101 #ifdef TCPDEBUG
102 #include <netinet/tcp_debug.h>
103 #endif /* TCPDEBUG */
104
105 #ifdef IPSEC
106 #include <netipsec/ipsec.h>
107 #include <netipsec/ipsec6.h>
108 #endif /*IPSEC*/
109
110 #include <machine/in_cksum.h>
111
112 #include <security/mac/mac_framework.h>
113
114 const int tcprexmtthresh = 3;
115
116 VNET_DEFINE(struct tcpstat, tcpstat);
117 SYSCTL_VNET_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW,
118 &VNET_NAME(tcpstat), tcpstat,
119 "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
120
121 int tcp_log_in_vain = 0;
122 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
123 &tcp_log_in_vain, 0,
124 "Log all incoming TCP segments to closed ports");
125
126 VNET_DEFINE(int, blackhole) = 0;
127 #define V_blackhole VNET(blackhole)
128 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW,
129 &VNET_NAME(blackhole), 0,
130 "Do not send RST on segments to closed ports");
131
132 VNET_DEFINE(int, tcp_delack_enabled) = 1;
133 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW,
134 &VNET_NAME(tcp_delack_enabled), 0,
135 "Delay ACK to try and piggyback it onto a data packet");
136
137 VNET_DEFINE(int, drop_synfin) = 0;
138 #define V_drop_synfin VNET(drop_synfin)
139 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW,
140 &VNET_NAME(drop_synfin), 0,
141 "Drop TCP packets with SYN+FIN set");
142
143 VNET_DEFINE(int, tcp_do_rfc3042) = 1;
144 #define V_tcp_do_rfc3042 VNET(tcp_do_rfc3042)
145 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_RW,
146 &VNET_NAME(tcp_do_rfc3042), 0,
147 "Enable RFC 3042 (Limited Transmit)");
148
149 VNET_DEFINE(int, tcp_do_rfc3390) = 1;
150 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW,
151 &VNET_NAME(tcp_do_rfc3390), 0,
152 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
153
154 VNET_DEFINE(int, tcp_do_rfc3465) = 1;
155 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_RW,
156 &VNET_NAME(tcp_do_rfc3465), 0,
157 "Enable RFC 3465 (Appropriate Byte Counting)");
158
159 VNET_DEFINE(int, tcp_abc_l_var) = 2;
160 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_RW,
161 &VNET_NAME(tcp_abc_l_var), 2,
162 "Cap the max cwnd increment during slow-start to this number of segments");
163
164 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, CTLFLAG_RW, 0, "TCP ECN");
165
166 VNET_DEFINE(int, tcp_do_ecn) = 0;
167 SYSCTL_VNET_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_RW,
168 &VNET_NAME(tcp_do_ecn), 0,
169 "TCP ECN support");
170
171 VNET_DEFINE(int, tcp_ecn_maxretries) = 1;
172 SYSCTL_VNET_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_RW,
173 &VNET_NAME(tcp_ecn_maxretries), 0,
174 "Max retries before giving up on ECN");
175
176 VNET_DEFINE(int, tcp_insecure_rst) = 0;
177 #define V_tcp_insecure_rst VNET(tcp_insecure_rst)
178 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_RW,
179 &VNET_NAME(tcp_insecure_rst), 0,
180 "Follow the old (insecure) criteria for accepting RST packets");
181
182 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1;
183 #define V_tcp_do_autorcvbuf VNET(tcp_do_autorcvbuf)
184 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_RW,
185 &VNET_NAME(tcp_do_autorcvbuf), 0,
186 "Enable automatic receive buffer sizing");
187
188 VNET_DEFINE(int, tcp_autorcvbuf_inc) = 16*1024;
189 #define V_tcp_autorcvbuf_inc VNET(tcp_autorcvbuf_inc)
190 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_RW,
191 &VNET_NAME(tcp_autorcvbuf_inc), 0,
192 "Incrementor step size of automatic receive buffer");
193
194 VNET_DEFINE(int, tcp_autorcvbuf_max) = 256*1024;
195 #define V_tcp_autorcvbuf_max VNET(tcp_autorcvbuf_max)
196 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_RW,
197 &VNET_NAME(tcp_autorcvbuf_max), 0,
198 "Max size of automatic receive buffer");
199
200 int tcp_read_locking = 1;
201 SYSCTL_INT(_net_inet_tcp, OID_AUTO, read_locking, CTLFLAG_RW,
202 &tcp_read_locking, 0, "Enable read locking strategy");
203
204 VNET_DEFINE(struct inpcbhead, tcb);
205 #define tcb6 tcb /* for KAME src sync over BSD*'s */
206 VNET_DEFINE(struct inpcbinfo, tcbinfo);
207
208 static void tcp_dooptions(struct tcpopt *, u_char *, int, int);
209 static void tcp_do_segment(struct mbuf *, struct tcphdr *,
210 struct socket *, struct tcpcb *, int, int, uint8_t,
211 int);
212 static void tcp_dropwithreset(struct mbuf *, struct tcphdr *,
213 struct tcpcb *, int, int);
214 static void tcp_pulloutofband(struct socket *,
215 struct tcphdr *, struct mbuf *, int);
216 static void tcp_xmit_timer(struct tcpcb *, int);
217 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *);
218 static void inline cc_ack_received(struct tcpcb *tp, struct tcphdr *th,
219 uint16_t type);
220 static void inline cc_conn_init(struct tcpcb *tp);
221 static void inline cc_post_recovery(struct tcpcb *tp, struct tcphdr *th);
222 static void inline tcp_fields_to_host(struct tcphdr *);
223 static void inline hhook_run_tcp_est_in(struct tcpcb *tp,
224 struct tcphdr *th, struct tcpopt *to);
225 #ifdef TCP_SIGNATURE
226 static void inline tcp_fields_to_net(struct tcphdr *);
227 static int inline tcp_signature_verify_input(struct mbuf *, int, int,
228 int, struct tcpopt *, struct tcphdr *, u_int);
229 #endif
230
231 /*
232 * Kernel module interface for updating tcpstat. The argument is an index
233 * into tcpstat treated as an array of u_long. While this encodes the
234 * general layout of tcpstat into the caller, it doesn't encode its location,
235 * so that future changes to add, for example, per-CPU stats support won't
236 * cause binary compatibility problems for kernel modules.
237 */
238 void
239 kmod_tcpstat_inc(int statnum)
240 {
241
242 (*((u_long *)&V_tcpstat + statnum))++;
243 }
244
245 /*
246 * Wrapper for the TCP established input helper hook.
247 */
248 static void inline
249 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to)
250 {
251 struct tcp_hhook_data hhook_data;
252
253 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) {
254 hhook_data.tp = tp;
255 hhook_data.th = th;
256 hhook_data.to = to;
257
258 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data,
259 tp->osd);
260 }
261 }
262
263 /*
264 * CC wrapper hook functions
265 */
266 static void inline
267 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t type)
268 {
269 INP_WLOCK_ASSERT(tp->t_inpcb);
270
271 tp->ccv->bytes_this_ack = BYTES_THIS_ACK(tp, th);
272 if (tp->snd_cwnd == min(tp->snd_cwnd, tp->snd_wnd))
273 tp->ccv->flags |= CCF_CWND_LIMITED;
274 else
275 tp->ccv->flags &= ~CCF_CWND_LIMITED;
276
277 if (type == CC_ACK) {
278 if (tp->snd_cwnd > tp->snd_ssthresh) {
279 tp->t_bytes_acked += min(tp->ccv->bytes_this_ack,
280 V_tcp_abc_l_var * tp->t_maxseg);
281 if (tp->t_bytes_acked >= tp->snd_cwnd) {
282 tp->t_bytes_acked -= tp->snd_cwnd;
283 tp->ccv->flags |= CCF_ABC_SENTAWND;
284 }
285 } else {
286 tp->ccv->flags &= ~CCF_ABC_SENTAWND;
287 tp->t_bytes_acked = 0;
288 }
289 }
290
291 if (CC_ALGO(tp)->ack_received != NULL) {
292 /* XXXLAS: Find a way to live without this */
293 tp->ccv->curack = th->th_ack;
294 CC_ALGO(tp)->ack_received(tp->ccv, type);
295 }
296 }
297
298 static void inline
299 cc_conn_init(struct tcpcb *tp)
300 {
301 struct hc_metrics_lite metrics;
302 struct inpcb *inp = tp->t_inpcb;
303 int rtt;
304 #ifdef INET6
305 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
306 #endif
307
308 INP_WLOCK_ASSERT(tp->t_inpcb);
309
310 tcp_hc_get(&inp->inp_inc, &metrics);
311
312 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) {
313 tp->t_srtt = rtt;
314 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
315 TCPSTAT_INC(tcps_usedrtt);
316 if (metrics.rmx_rttvar) {
317 tp->t_rttvar = metrics.rmx_rttvar;
318 TCPSTAT_INC(tcps_usedrttvar);
319 } else {
320 /* default variation is +- 1 rtt */
321 tp->t_rttvar =
322 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
323 }
324 TCPT_RANGESET(tp->t_rxtcur,
325 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
326 tp->t_rttmin, TCPTV_REXMTMAX);
327 }
328 if (metrics.rmx_ssthresh) {
329 /*
330 * There's some sort of gateway or interface
331 * buffer limit on the path. Use this to set
332 * the slow start threshhold, but set the
333 * threshold to no less than 2*mss.
334 */
335 tp->snd_ssthresh = max(2 * tp->t_maxseg, metrics.rmx_ssthresh);
336 TCPSTAT_INC(tcps_usedssthresh);
337 }
338
339 /*
340 * Set the slow-start flight size depending on whether this
341 * is a local network or not.
342 *
343 * Extend this so we cache the cwnd too and retrieve it here.
344 * Make cwnd even bigger than RFC3390 suggests but only if we
345 * have previous experience with the remote host. Be careful
346 * not make cwnd bigger than remote receive window or our own
347 * send socket buffer. Maybe put some additional upper bound
348 * on the retrieved cwnd. Should do incremental updates to
349 * hostcache when cwnd collapses so next connection doesn't
350 * overloads the path again.
351 *
352 * XXXAO: Initializing the CWND from the hostcache is broken
353 * and in its current form not RFC conformant. It is disabled
354 * until fixed or removed entirely.
355 *
356 * RFC3390 says only do this if SYN or SYN/ACK didn't got lost.
357 * We currently check only in syncache_socket for that.
358 */
359 /* #define TCP_METRICS_CWND */
360 #ifdef TCP_METRICS_CWND
361 if (metrics.rmx_cwnd)
362 tp->snd_cwnd = max(tp->t_maxseg, min(metrics.rmx_cwnd / 2,
363 min(tp->snd_wnd, so->so_snd.sb_hiwat)));
364 else
365 #endif
366 if (V_tcp_do_rfc3390)
367 tp->snd_cwnd = min(4 * tp->t_maxseg,
368 max(2 * tp->t_maxseg, 4380));
369 #ifdef INET6
370 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) ||
371 (!isipv6 && in_localaddr(inp->inp_faddr)))
372 #else
373 else if (in_localaddr(inp->inp_faddr))
374 #endif
375 tp->snd_cwnd = tp->t_maxseg * V_ss_fltsz_local;
376 else
377 tp->snd_cwnd = tp->t_maxseg * V_ss_fltsz;
378
379 if (CC_ALGO(tp)->conn_init != NULL)
380 CC_ALGO(tp)->conn_init(tp->ccv);
381 }
382
383 void inline
384 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type)
385 {
386 INP_WLOCK_ASSERT(tp->t_inpcb);
387
388 switch(type) {
389 case CC_NDUPACK:
390 if (!IN_FASTRECOVERY(tp->t_flags)) {
391 tp->snd_recover = tp->snd_max;
392 if (tp->t_flags & TF_ECN_PERMIT)
393 tp->t_flags |= TF_ECN_SND_CWR;
394 }
395 break;
396 case CC_ECN:
397 if (!IN_CONGRECOVERY(tp->t_flags)) {
398 TCPSTAT_INC(tcps_ecn_rcwnd);
399 tp->snd_recover = tp->snd_max;
400 if (tp->t_flags & TF_ECN_PERMIT)
401 tp->t_flags |= TF_ECN_SND_CWR;
402 }
403 break;
404 case CC_RTO:
405 tp->t_dupacks = 0;
406 tp->t_bytes_acked = 0;
407 EXIT_RECOVERY(tp->t_flags);
408 tp->snd_ssthresh = max(2, min(tp->snd_wnd, tp->snd_cwnd) / 2 /
409 tp->t_maxseg) * tp->t_maxseg;
410 tp->snd_cwnd = tp->t_maxseg;
411 break;
412 case CC_RTO_ERR:
413 TCPSTAT_INC(tcps_sndrexmitbad);
414 /* RTO was unnecessary, so reset everything. */
415 tp->snd_cwnd = tp->snd_cwnd_prev;
416 tp->snd_ssthresh = tp->snd_ssthresh_prev;
417 tp->snd_recover = tp->snd_recover_prev;
418 if (tp->t_flags & TF_WASFRECOVERY)
419 ENTER_FASTRECOVERY(tp->t_flags);
420 if (tp->t_flags & TF_WASCRECOVERY)
421 ENTER_CONGRECOVERY(tp->t_flags);
422 tp->snd_nxt = tp->snd_max;
423 tp->t_flags &= ~TF_PREVVALID;
424 tp->t_badrxtwin = 0;
425 break;
426 }
427
428 if (CC_ALGO(tp)->cong_signal != NULL) {
429 if (th != NULL)
430 tp->ccv->curack = th->th_ack;
431 CC_ALGO(tp)->cong_signal(tp->ccv, type);
432 }
433 }
434
435 static void inline
436 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th)
437 {
438 INP_WLOCK_ASSERT(tp->t_inpcb);
439
440 /* XXXLAS: KASSERT that we're in recovery? */
441
442 if (CC_ALGO(tp)->post_recovery != NULL) {
443 tp->ccv->curack = th->th_ack;
444 CC_ALGO(tp)->post_recovery(tp->ccv);
445 }
446 /* XXXLAS: EXIT_RECOVERY ? */
447 tp->t_bytes_acked = 0;
448 }
449
450 static inline void
451 tcp_fields_to_host(struct tcphdr *th)
452 {
453
454 th->th_seq = ntohl(th->th_seq);
455 th->th_ack = ntohl(th->th_ack);
456 th->th_win = ntohs(th->th_win);
457 th->th_urp = ntohs(th->th_urp);
458 }
459
460 #ifdef TCP_SIGNATURE
461 static inline void
462 tcp_fields_to_net(struct tcphdr *th)
463 {
464
465 th->th_seq = htonl(th->th_seq);
466 th->th_ack = htonl(th->th_ack);
467 th->th_win = htons(th->th_win);
468 th->th_urp = htons(th->th_urp);
469 }
470
471 static inline int
472 tcp_signature_verify_input(struct mbuf *m, int off0, int tlen, int optlen,
473 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag)
474 {
475 int ret;
476
477 tcp_fields_to_net(th);
478 ret = tcp_signature_verify(m, off0, tlen, optlen, to, th, tcpbflag);
479 tcp_fields_to_host(th);
480 return (ret);
481 }
482 #endif
483
484 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
485 #ifdef INET6
486 #define ND6_HINT(tp) \
487 do { \
488 if ((tp) && (tp)->t_inpcb && \
489 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \
490 nd6_nud_hint(NULL, NULL, 0); \
491 } while (0)
492 #else
493 #define ND6_HINT(tp)
494 #endif
495
496 /*
497 * Indicate whether this ack should be delayed. We can delay the ack if
498 * - there is no delayed ack timer in progress and
499 * - our last ack wasn't a 0-sized window. We never want to delay
500 * the ack that opens up a 0-sized window and
501 * - delayed acks are enabled or
502 * - this is a half-synchronized T/TCP connection.
503 */
504 #define DELAY_ACK(tp) \
505 ((!tcp_timer_active(tp, TT_DELACK) && \
506 (tp->t_flags & TF_RXWIN0SENT) == 0) && \
507 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
508
509 /*
510 * TCP input handling is split into multiple parts:
511 * tcp6_input is a thin wrapper around tcp_input for the extended
512 * ip6_protox[] call format in ip6_input
513 * tcp_input handles primary segment validation, inpcb lookup and
514 * SYN processing on listen sockets
515 * tcp_do_segment processes the ACK and text of the segment for
516 * establishing, established and closing connections
517 */
518 #ifdef INET6
519 int
520 tcp6_input(struct mbuf **mp, int *offp, int proto)
521 {
522 struct mbuf *m = *mp;
523 struct in6_ifaddr *ia6;
524
525 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
526
527 /*
528 * draft-itojun-ipv6-tcp-to-anycast
529 * better place to put this in?
530 */
531 ia6 = ip6_getdstifaddr(m);
532 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
533 struct ip6_hdr *ip6;
534
535 ifa_free(&ia6->ia_ifa);
536 ip6 = mtod(m, struct ip6_hdr *);
537 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
538 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
539 return IPPROTO_DONE;
540 }
541 if (ia6)
542 ifa_free(&ia6->ia_ifa);
543
544 tcp_input(m, *offp);
545 return IPPROTO_DONE;
546 }
547 #endif
548
549 void
550 tcp_input(struct mbuf *m, int off0)
551 {
552 struct tcphdr *th;
553 struct ip *ip = NULL;
554 struct ipovly *ipov;
555 struct inpcb *inp = NULL;
556 struct tcpcb *tp = NULL;
557 struct socket *so = NULL;
558 u_char *optp = NULL;
559 int optlen = 0;
560 int len, tlen, off;
561 int drop_hdrlen;
562 int thflags;
563 int rstreason = 0; /* For badport_bandlim accounting purposes */
564 uint8_t iptos;
565 #ifdef TCP_SIGNATURE
566 uint8_t sig_checked = 0;
567 #endif
568 #ifdef IPFIREWALL_FORWARD
569 struct m_tag *fwd_tag;
570 #endif
571 #ifdef INET6
572 struct ip6_hdr *ip6 = NULL;
573 int isipv6;
574 #else
575 const void *ip6 = NULL;
576 const int isipv6 = 0;
577 #endif
578 struct tcpopt to; /* options in this segment */
579 char *s = NULL; /* address and port logging */
580 int ti_locked;
581 #define TI_UNLOCKED 1
582 #define TI_RLOCKED 2
583 #define TI_WLOCKED 3
584
585 #ifdef TCPDEBUG
586 /*
587 * The size of tcp_saveipgen must be the size of the max ip header,
588 * now IPv6.
589 */
590 u_char tcp_saveipgen[IP6_HDR_LEN];
591 struct tcphdr tcp_savetcp;
592 short ostate = 0;
593 #endif
594
595 #ifdef INET6
596 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
597 #endif
598
599 to.to_flags = 0;
600 TCPSTAT_INC(tcps_rcvtotal);
601
602 if (isipv6) {
603 #ifdef INET6
604 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */
605 ip6 = mtod(m, struct ip6_hdr *);
606 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
607 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) {
608 TCPSTAT_INC(tcps_rcvbadsum);
609 goto drop;
610 }
611 th = (struct tcphdr *)((caddr_t)ip6 + off0);
612
613 /*
614 * Be proactive about unspecified IPv6 address in source.
615 * As we use all-zero to indicate unbounded/unconnected pcb,
616 * unspecified IPv6 address can be used to confuse us.
617 *
618 * Note that packets with unspecified IPv6 destination is
619 * already dropped in ip6_input.
620 */
621 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
622 /* XXX stat */
623 goto drop;
624 }
625 #else
626 th = NULL; /* XXX: Avoid compiler warning. */
627 #endif
628 } else {
629 /*
630 * Get IP and TCP header together in first mbuf.
631 * Note: IP leaves IP header in first mbuf.
632 */
633 if (off0 > sizeof (struct ip)) {
634 ip_stripoptions(m, (struct mbuf *)0);
635 off0 = sizeof(struct ip);
636 }
637 if (m->m_len < sizeof (struct tcpiphdr)) {
638 if ((m = m_pullup(m, sizeof (struct tcpiphdr)))
639 == NULL) {
640 TCPSTAT_INC(tcps_rcvshort);
641 return;
642 }
643 }
644 ip = mtod(m, struct ip *);
645 ipov = (struct ipovly *)ip;
646 th = (struct tcphdr *)((caddr_t)ip + off0);
647 tlen = ip->ip_len;
648
649 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
650 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
651 th->th_sum = m->m_pkthdr.csum_data;
652 else
653 th->th_sum = in_pseudo(ip->ip_src.s_addr,
654 ip->ip_dst.s_addr,
655 htonl(m->m_pkthdr.csum_data +
656 ip->ip_len +
657 IPPROTO_TCP));
658 th->th_sum ^= 0xffff;
659 #ifdef TCPDEBUG
660 ipov->ih_len = (u_short)tlen;
661 ipov->ih_len = htons(ipov->ih_len);
662 #endif
663 } else {
664 /*
665 * Checksum extended TCP header and data.
666 */
667 len = sizeof (struct ip) + tlen;
668 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
669 ipov->ih_len = (u_short)tlen;
670 ipov->ih_len = htons(ipov->ih_len);
671 th->th_sum = in_cksum(m, len);
672 }
673 if (th->th_sum) {
674 TCPSTAT_INC(tcps_rcvbadsum);
675 goto drop;
676 }
677 /* Re-initialization for later version check */
678 ip->ip_v = IPVERSION;
679 }
680
681 #ifdef INET6
682 if (isipv6)
683 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
684 else
685 #endif
686 iptos = ip->ip_tos;
687
688 /*
689 * Check that TCP offset makes sense,
690 * pull out TCP options and adjust length. XXX
691 */
692 off = th->th_off << 2;
693 if (off < sizeof (struct tcphdr) || off > tlen) {
694 TCPSTAT_INC(tcps_rcvbadoff);
695 goto drop;
696 }
697 tlen -= off; /* tlen is used instead of ti->ti_len */
698 if (off > sizeof (struct tcphdr)) {
699 if (isipv6) {
700 #ifdef INET6
701 IP6_EXTHDR_CHECK(m, off0, off, );
702 ip6 = mtod(m, struct ip6_hdr *);
703 th = (struct tcphdr *)((caddr_t)ip6 + off0);
704 #endif
705 } else {
706 if (m->m_len < sizeof(struct ip) + off) {
707 if ((m = m_pullup(m, sizeof (struct ip) + off))
708 == NULL) {
709 TCPSTAT_INC(tcps_rcvshort);
710 return;
711 }
712 ip = mtod(m, struct ip *);
713 ipov = (struct ipovly *)ip;
714 th = (struct tcphdr *)((caddr_t)ip + off0);
715 }
716 }
717 optlen = off - sizeof (struct tcphdr);
718 optp = (u_char *)(th + 1);
719 }
720 thflags = th->th_flags;
721
722 /*
723 * Convert TCP protocol specific fields to host format.
724 */
725 tcp_fields_to_host(th);
726
727 /*
728 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options.
729 */
730 drop_hdrlen = off0 + off;
731
732 /*
733 * Locate pcb for segment, which requires a lock on tcbinfo.
734 * Optimisticaly acquire a global read lock rather than a write lock
735 * unless header flags necessarily imply a state change. There are
736 * two cases where we might discover later we need a write lock
737 * despite the flags: ACKs moving a connection out of the syncache,
738 * and ACKs for a connection in TIMEWAIT.
739 */
740 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 ||
741 tcp_read_locking == 0) {
742 INP_INFO_WLOCK(&V_tcbinfo);
743 ti_locked = TI_WLOCKED;
744 } else {
745 INP_INFO_RLOCK(&V_tcbinfo);
746 ti_locked = TI_RLOCKED;
747 }
748
749 findpcb:
750 #ifdef INVARIANTS
751 if (ti_locked == TI_RLOCKED)
752 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
753 else if (ti_locked == TI_WLOCKED)
754 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
755 else
756 panic("%s: findpcb ti_locked %d\n", __func__, ti_locked);
757 #endif
758
759 #ifdef IPFIREWALL_FORWARD
760 /*
761 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
762 */
763 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
764
765 if (fwd_tag != NULL && isipv6 == 0) { /* IPv6 support is not yet */
766 struct sockaddr_in *next_hop;
767
768 next_hop = (struct sockaddr_in *)(fwd_tag+1);
769 /*
770 * Transparently forwarded. Pretend to be the destination.
771 * already got one like this?
772 */
773 inp = in_pcblookup_hash(&V_tcbinfo,
774 ip->ip_src, th->th_sport,
775 ip->ip_dst, th->th_dport,
776 0, m->m_pkthdr.rcvif);
777 if (!inp) {
778 /* It's new. Try to find the ambushing socket. */
779 inp = in_pcblookup_hash(&V_tcbinfo,
780 ip->ip_src, th->th_sport,
781 next_hop->sin_addr,
782 next_hop->sin_port ?
783 ntohs(next_hop->sin_port) :
784 th->th_dport,
785 INPLOOKUP_WILDCARD,
786 m->m_pkthdr.rcvif);
787 }
788 /* Remove the tag from the packet. We don't need it anymore. */
789 m_tag_delete(m, fwd_tag);
790 } else
791 #endif /* IPFIREWALL_FORWARD */
792 {
793 if (isipv6) {
794 #ifdef INET6
795 inp = in6_pcblookup_hash(&V_tcbinfo,
796 &ip6->ip6_src, th->th_sport,
797 &ip6->ip6_dst, th->th_dport,
798 INPLOOKUP_WILDCARD,
799 m->m_pkthdr.rcvif);
800 #endif
801 } else
802 inp = in_pcblookup_hash(&V_tcbinfo,
803 ip->ip_src, th->th_sport,
804 ip->ip_dst, th->th_dport,
805 INPLOOKUP_WILDCARD,
806 m->m_pkthdr.rcvif);
807 }
808
809 /*
810 * If the INPCB does not exist then all data in the incoming
811 * segment is discarded and an appropriate RST is sent back.
812 * XXX MRT Send RST using which routing table?
813 */
814 if (inp == NULL) {
815 /*
816 * Log communication attempts to ports that are not
817 * in use.
818 */
819 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) ||
820 tcp_log_in_vain == 2) {
821 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6)))
822 log(LOG_INFO, "%s; %s: Connection attempt "
823 "to closed port\n", s, __func__);
824 }
825 /*
826 * When blackholing do not respond with a RST but
827 * completely ignore the segment and drop it.
828 */
829 if ((V_blackhole == 1 && (thflags & TH_SYN)) ||
830 V_blackhole == 2)
831 goto dropunlock;
832
833 rstreason = BANDLIM_RST_CLOSEDPORT;
834 goto dropwithreset;
835 }
836 INP_WLOCK(inp);
837 if (!(inp->inp_flags & INP_HW_FLOWID)
838 && (m->m_flags & M_FLOWID)
839 && ((inp->inp_socket == NULL)
840 || !(inp->inp_socket->so_options & SO_ACCEPTCONN))) {
841 inp->inp_flags |= INP_HW_FLOWID;
842 inp->inp_flags &= ~INP_SW_FLOWID;
843 inp->inp_flowid = m->m_pkthdr.flowid;
844 }
845 #ifdef IPSEC
846 #ifdef INET6
847 if (isipv6 && ipsec6_in_reject(m, inp)) {
848 V_ipsec6stat.in_polvio++;
849 goto dropunlock;
850 } else
851 #endif /* INET6 */
852 if (ipsec4_in_reject(m, inp) != 0) {
853 V_ipsec4stat.in_polvio++;
854 goto dropunlock;
855 }
856 #endif /* IPSEC */
857
858 /*
859 * Check the minimum TTL for socket.
860 */
861 if (inp->inp_ip_minttl != 0) {
862 #ifdef INET6
863 if (isipv6 && inp->inp_ip_minttl > ip6->ip6_hlim)
864 goto dropunlock;
865 else
866 #endif
867 if (inp->inp_ip_minttl > ip->ip_ttl)
868 goto dropunlock;
869 }
870
871 /*
872 * A previous connection in TIMEWAIT state is supposed to catch stray
873 * or duplicate segments arriving late. If this segment was a
874 * legitimate new connection attempt the old INPCB gets removed and
875 * we can try again to find a listening socket.
876 *
877 * At this point, due to earlier optimism, we may hold a read lock on
878 * the inpcbinfo, rather than a write lock. If so, we need to
879 * upgrade, or if that fails, acquire a reference on the inpcb, drop
880 * all locks, acquire a global write lock, and then re-acquire the
881 * inpcb lock. We may at that point discover that another thread has
882 * tried to free the inpcb, in which case we need to loop back and
883 * try to find a new inpcb to deliver to.
884 */
885 relocked:
886 if (inp->inp_flags & INP_TIMEWAIT) {
887 KASSERT(ti_locked == TI_RLOCKED || ti_locked == TI_WLOCKED,
888 ("%s: INP_TIMEWAIT ti_locked %d", __func__, ti_locked));
889
890 if (ti_locked == TI_RLOCKED) {
891 if (INP_INFO_TRY_UPGRADE(&V_tcbinfo) == 0) {
892 in_pcbref(inp);
893 INP_WUNLOCK(inp);
894 INP_INFO_RUNLOCK(&V_tcbinfo);
895 INP_INFO_WLOCK(&V_tcbinfo);
896 ti_locked = TI_WLOCKED;
897 INP_WLOCK(inp);
898 if (in_pcbrele(inp)) {
899 inp = NULL;
900 goto findpcb;
901 }
902 } else
903 ti_locked = TI_WLOCKED;
904 }
905 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
906
907 if (thflags & TH_SYN)
908 tcp_dooptions(&to, optp, optlen, TO_SYN);
909 /*
910 * NB: tcp_twcheck unlocks the INP and frees the mbuf.
911 */
912 if (tcp_twcheck(inp, &to, th, m, tlen))
913 goto findpcb;
914 INP_INFO_WUNLOCK(&V_tcbinfo);
915 return;
916 }
917 /*
918 * The TCPCB may no longer exist if the connection is winding
919 * down or it is in the CLOSED state. Either way we drop the
920 * segment and send an appropriate response.
921 */
922 tp = intotcpcb(inp);
923 if (tp == NULL || tp->t_state == TCPS_CLOSED) {
924 rstreason = BANDLIM_RST_CLOSEDPORT;
925 goto dropwithreset;
926 }
927
928 /*
929 * We've identified a valid inpcb, but it could be that we need an
930 * inpcbinfo write lock and have only a read lock. In this case,
931 * attempt to upgrade/relock using the same strategy as the TIMEWAIT
932 * case above. If we relock, we have to jump back to 'relocked' as
933 * the connection might now be in TIMEWAIT.
934 */
935 if (tp->t_state != TCPS_ESTABLISHED ||
936 (thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 ||
937 tcp_read_locking == 0) {
938 KASSERT(ti_locked == TI_RLOCKED || ti_locked == TI_WLOCKED,
939 ("%s: upgrade check ti_locked %d", __func__, ti_locked));
940
941 if (ti_locked == TI_RLOCKED) {
942 if (INP_INFO_TRY_UPGRADE(&V_tcbinfo) == 0) {
943 in_pcbref(inp);
944 INP_WUNLOCK(inp);
945 INP_INFO_RUNLOCK(&V_tcbinfo);
946 INP_INFO_WLOCK(&V_tcbinfo);
947 ti_locked = TI_WLOCKED;
948 INP_WLOCK(inp);
949 if (in_pcbrele(inp)) {
950 inp = NULL;
951 goto findpcb;
952 }
953 goto relocked;
954 } else
955 ti_locked = TI_WLOCKED;
956 }
957 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
958 }
959
960 #ifdef MAC
961 INP_WLOCK_ASSERT(inp);
962 if (mac_inpcb_check_deliver(inp, m))
963 goto dropunlock;
964 #endif
965 so = inp->inp_socket;
966 KASSERT(so != NULL, ("%s: so == NULL", __func__));
967 #ifdef TCPDEBUG
968 if (so->so_options & SO_DEBUG) {
969 ostate = tp->t_state;
970 if (isipv6) {
971 #ifdef INET6
972 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
973 #endif
974 } else
975 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
976 tcp_savetcp = *th;
977 }
978 #endif
979 /*
980 * When the socket is accepting connections (the INPCB is in LISTEN
981 * state) we look into the SYN cache if this is a new connection
982 * attempt or the completion of a previous one.
983 */
984 if (so->so_options & SO_ACCEPTCONN) {
985 struct in_conninfo inc;
986
987 KASSERT(tp->t_state == TCPS_LISTEN, ("%s: so accepting but "
988 "tp not listening", __func__));
989
990 bzero(&inc, sizeof(inc));
991 #ifdef INET6
992 if (isipv6) {
993 inc.inc_flags |= INC_ISIPV6;
994 inc.inc6_faddr = ip6->ip6_src;
995 inc.inc6_laddr = ip6->ip6_dst;
996 } else
997 #endif
998 {
999 inc.inc_faddr = ip->ip_src;
1000 inc.inc_laddr = ip->ip_dst;
1001 }
1002 inc.inc_fport = th->th_sport;
1003 inc.inc_lport = th->th_dport;
1004 inc.inc_fibnum = so->so_fibnum;
1005
1006 /*
1007 * Check for an existing connection attempt in syncache if
1008 * the flag is only ACK. A successful lookup creates a new
1009 * socket appended to the listen queue in SYN_RECEIVED state.
1010 */
1011 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
1012 /*
1013 * Parse the TCP options here because
1014 * syncookies need access to the reflected
1015 * timestamp.
1016 */
1017 tcp_dooptions(&to, optp, optlen, 0);
1018 /*
1019 * NB: syncache_expand() doesn't unlock
1020 * inp and tcpinfo locks.
1021 */
1022 if (!syncache_expand(&inc, &to, th, &so, m)) {
1023 /*
1024 * No syncache entry or ACK was not
1025 * for our SYN/ACK. Send a RST.
1026 * NB: syncache did its own logging
1027 * of the failure cause.
1028 */
1029 rstreason = BANDLIM_RST_OPENPORT;
1030 goto dropwithreset;
1031 }
1032 if (so == NULL) {
1033 /*
1034 * We completed the 3-way handshake
1035 * but could not allocate a socket
1036 * either due to memory shortage,
1037 * listen queue length limits or
1038 * global socket limits. Send RST
1039 * or wait and have the remote end
1040 * retransmit the ACK for another
1041 * try.
1042 */
1043 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1044 log(LOG_DEBUG, "%s; %s: Listen socket: "
1045 "Socket allocation failed due to "
1046 "limits or memory shortage, %s\n",
1047 s, __func__,
1048 V_tcp_sc_rst_sock_fail ?
1049 "sending RST" : "try again");
1050 if (V_tcp_sc_rst_sock_fail) {
1051 rstreason = BANDLIM_UNLIMITED;
1052 goto dropwithreset;
1053 } else
1054 goto dropunlock;
1055 }
1056 /*
1057 * Socket is created in state SYN_RECEIVED.
1058 * Unlock the listen socket, lock the newly
1059 * created socket and update the tp variable.
1060 */
1061 INP_WUNLOCK(inp); /* listen socket */
1062 inp = sotoinpcb(so);
1063 INP_WLOCK(inp); /* new connection */
1064 tp = intotcpcb(inp);
1065 KASSERT(tp->t_state == TCPS_SYN_RECEIVED,
1066 ("%s: ", __func__));
1067 #ifdef TCP_SIGNATURE
1068 if (sig_checked == 0) {
1069 tcp_dooptions(&to, optp, optlen,
1070 (thflags & TH_SYN) ? TO_SYN : 0);
1071 if (!tcp_signature_verify_input(m, off0, tlen,
1072 optlen, &to, th, tp->t_flags)) {
1073
1074 /*
1075 * In SYN_SENT state if it receives an
1076 * RST, it is allowed for further
1077 * processing.
1078 */
1079 if ((thflags & TH_RST) == 0 ||
1080 (tp->t_state == TCPS_SYN_SENT) == 0)
1081 goto dropunlock;
1082 }
1083 sig_checked = 1;
1084 }
1085 #endif
1086
1087 /*
1088 * Process the segment and the data it
1089 * contains. tcp_do_segment() consumes
1090 * the mbuf chain and unlocks the inpcb.
1091 */
1092 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen,
1093 iptos, ti_locked);
1094 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1095 return;
1096 }
1097 /*
1098 * Segment flag validation for new connection attempts:
1099 *
1100 * Our (SYN|ACK) response was rejected.
1101 * Check with syncache and remove entry to prevent
1102 * retransmits.
1103 *
1104 * NB: syncache_chkrst does its own logging of failure
1105 * causes.
1106 */
1107 if (thflags & TH_RST) {
1108 syncache_chkrst(&inc, th);
1109 goto dropunlock;
1110 }
1111 /*
1112 * We can't do anything without SYN.
1113 */
1114 if ((thflags & TH_SYN) == 0) {
1115 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1116 log(LOG_DEBUG, "%s; %s: Listen socket: "
1117 "SYN is missing, segment ignored\n",
1118 s, __func__);
1119 TCPSTAT_INC(tcps_badsyn);
1120 goto dropunlock;
1121 }
1122 /*
1123 * (SYN|ACK) is bogus on a listen socket.
1124 */
1125 if (thflags & TH_ACK) {
1126 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1127 log(LOG_DEBUG, "%s; %s: Listen socket: "
1128 "SYN|ACK invalid, segment rejected\n",
1129 s, __func__);
1130 syncache_badack(&inc); /* XXX: Not needed! */
1131 TCPSTAT_INC(tcps_badsyn);
1132 rstreason = BANDLIM_RST_OPENPORT;
1133 goto dropwithreset;
1134 }
1135 /*
1136 * If the drop_synfin option is enabled, drop all
1137 * segments with both the SYN and FIN bits set.
1138 * This prevents e.g. nmap from identifying the
1139 * TCP/IP stack.
1140 * XXX: Poor reasoning. nmap has other methods
1141 * and is constantly refining its stack detection
1142 * strategies.
1143 * XXX: This is a violation of the TCP specification
1144 * and was used by RFC1644.
1145 */
1146 if ((thflags & TH_FIN) && V_drop_synfin) {
1147 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1148 log(LOG_DEBUG, "%s; %s: Listen socket: "
1149 "SYN|FIN segment ignored (based on "
1150 "sysctl setting)\n", s, __func__);
1151 TCPSTAT_INC(tcps_badsyn);
1152 goto dropunlock;
1153 }
1154 /*
1155 * Segment's flags are (SYN) or (SYN|FIN).
1156 *
1157 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored
1158 * as they do not affect the state of the TCP FSM.
1159 * The data pointed to by TH_URG and th_urp is ignored.
1160 */
1161 KASSERT((thflags & (TH_RST|TH_ACK)) == 0,
1162 ("%s: Listen socket: TH_RST or TH_ACK set", __func__));
1163 KASSERT(thflags & (TH_SYN),
1164 ("%s: Listen socket: TH_SYN not set", __func__));
1165 #ifdef INET6
1166 /*
1167 * If deprecated address is forbidden,
1168 * we do not accept SYN to deprecated interface
1169 * address to prevent any new inbound connection from
1170 * getting established.
1171 * When we do not accept SYN, we send a TCP RST,
1172 * with deprecated source address (instead of dropping
1173 * it). We compromise it as it is much better for peer
1174 * to send a RST, and RST will be the final packet
1175 * for the exchange.
1176 *
1177 * If we do not forbid deprecated addresses, we accept
1178 * the SYN packet. RFC2462 does not suggest dropping
1179 * SYN in this case.
1180 * If we decipher RFC2462 5.5.4, it says like this:
1181 * 1. use of deprecated addr with existing
1182 * communication is okay - "SHOULD continue to be
1183 * used"
1184 * 2. use of it with new communication:
1185 * (2a) "SHOULD NOT be used if alternate address
1186 * with sufficient scope is available"
1187 * (2b) nothing mentioned otherwise.
1188 * Here we fall into (2b) case as we have no choice in
1189 * our source address selection - we must obey the peer.
1190 *
1191 * The wording in RFC2462 is confusing, and there are
1192 * multiple description text for deprecated address
1193 * handling - worse, they are not exactly the same.
1194 * I believe 5.5.4 is the best one, so we follow 5.5.4.
1195 */
1196 if (isipv6 && !V_ip6_use_deprecated) {
1197 struct in6_ifaddr *ia6;
1198
1199 ia6 = ip6_getdstifaddr(m);
1200 if (ia6 != NULL &&
1201 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1202 ifa_free(&ia6->ia_ifa);
1203 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1204 log(LOG_DEBUG, "%s; %s: Listen socket: "
1205 "Connection attempt to deprecated "
1206 "IPv6 address rejected\n",
1207 s, __func__);
1208 rstreason = BANDLIM_RST_OPENPORT;
1209 goto dropwithreset;
1210 }
1211 if (ia6)
1212 ifa_free(&ia6->ia_ifa);
1213 }
1214 #endif
1215 /*
1216 * Basic sanity checks on incoming SYN requests:
1217 * Don't respond if the destination is a link layer
1218 * broadcast according to RFC1122 4.2.3.10, p. 104.
1219 * If it is from this socket it must be forged.
1220 * Don't respond if the source or destination is a
1221 * global or subnet broad- or multicast address.
1222 * Note that it is quite possible to receive unicast
1223 * link-layer packets with a broadcast IP address. Use
1224 * in_broadcast() to find them.
1225 */
1226 if (m->m_flags & (M_BCAST|M_MCAST)) {
1227 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1228 log(LOG_DEBUG, "%s; %s: Listen socket: "
1229 "Connection attempt from broad- or multicast "
1230 "link layer address ignored\n", s, __func__);
1231 goto dropunlock;
1232 }
1233 if (isipv6) {
1234 #ifdef INET6
1235 if (th->th_dport == th->th_sport &&
1236 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) {
1237 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1238 log(LOG_DEBUG, "%s; %s: Listen socket: "
1239 "Connection attempt to/from self "
1240 "ignored\n", s, __func__);
1241 goto dropunlock;
1242 }
1243 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1244 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) {
1245 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1246 log(LOG_DEBUG, "%s; %s: Listen socket: "
1247 "Connection attempt from/to multicast "
1248 "address ignored\n", s, __func__);
1249 goto dropunlock;
1250 }
1251 #endif
1252 } else {
1253 if (th->th_dport == th->th_sport &&
1254 ip->ip_dst.s_addr == ip->ip_src.s_addr) {
1255 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1256 log(LOG_DEBUG, "%s; %s: Listen socket: "
1257 "Connection attempt from/to self "
1258 "ignored\n", s, __func__);
1259 goto dropunlock;
1260 }
1261 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1262 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1263 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1264 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
1265 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1266 log(LOG_DEBUG, "%s; %s: Listen socket: "
1267 "Connection attempt from/to broad- "
1268 "or multicast address ignored\n",
1269 s, __func__);
1270 goto dropunlock;
1271 }
1272 }
1273 /*
1274 * SYN appears to be valid. Create compressed TCP state
1275 * for syncache.
1276 */
1277 #ifdef TCPDEBUG
1278 if (so->so_options & SO_DEBUG)
1279 tcp_trace(TA_INPUT, ostate, tp,
1280 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1281 #endif
1282 tcp_dooptions(&to, optp, optlen, TO_SYN);
1283 syncache_add(&inc, &to, th, inp, &so, m);
1284 /*
1285 * Entry added to syncache and mbuf consumed.
1286 * Everything already unlocked by syncache_add().
1287 */
1288 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1289 return;
1290 }
1291
1292 #ifdef TCP_SIGNATURE
1293 if (sig_checked == 0) {
1294 tcp_dooptions(&to, optp, optlen,
1295 (thflags & TH_SYN) ? TO_SYN : 0);
1296 if (!tcp_signature_verify_input(m, off0, tlen, optlen, &to,
1297 th, tp->t_flags)) {
1298
1299 /*
1300 * In SYN_SENT state if it receives an RST, it is
1301 * allowed for further processing.
1302 */
1303 if ((thflags & TH_RST) == 0 ||
1304 (tp->t_state == TCPS_SYN_SENT) == 0)
1305 goto dropunlock;
1306 }
1307 sig_checked = 1;
1308 }
1309 #endif
1310
1311 /*
1312 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later
1313 * state. tcp_do_segment() always consumes the mbuf chain, unlocks
1314 * the inpcb, and unlocks pcbinfo.
1315 */
1316 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos, ti_locked);
1317 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1318 return;
1319
1320 dropwithreset:
1321 if (ti_locked == TI_RLOCKED)
1322 INP_INFO_RUNLOCK(&V_tcbinfo);
1323 else if (ti_locked == TI_WLOCKED)
1324 INP_INFO_WUNLOCK(&V_tcbinfo);
1325 else
1326 panic("%s: dropwithreset ti_locked %d", __func__, ti_locked);
1327 ti_locked = TI_UNLOCKED;
1328
1329 if (inp != NULL) {
1330 tcp_dropwithreset(m, th, tp, tlen, rstreason);
1331 INP_WUNLOCK(inp);
1332 } else
1333 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
1334 m = NULL; /* mbuf chain got consumed. */
1335 goto drop;
1336
1337 dropunlock:
1338 if (ti_locked == TI_RLOCKED)
1339 INP_INFO_RUNLOCK(&V_tcbinfo);
1340 else if (ti_locked == TI_WLOCKED)
1341 INP_INFO_WUNLOCK(&V_tcbinfo);
1342 else
1343 panic("%s: dropunlock ti_locked %d", __func__, ti_locked);
1344 ti_locked = TI_UNLOCKED;
1345
1346 if (inp != NULL)
1347 INP_WUNLOCK(inp);
1348
1349 drop:
1350 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1351 if (s != NULL)
1352 free(s, M_TCPLOG);
1353 if (m != NULL)
1354 m_freem(m);
1355 }
1356
1357 static void
1358 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so,
1359 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos,
1360 int ti_locked)
1361 {
1362 int thflags, acked, ourfinisacked, needoutput = 0;
1363 int rstreason, todrop, win;
1364 u_long tiwin;
1365 struct tcpopt to;
1366
1367 #ifdef TCPDEBUG
1368 /*
1369 * The size of tcp_saveipgen must be the size of the max ip header,
1370 * now IPv6.
1371 */
1372 u_char tcp_saveipgen[IP6_HDR_LEN];
1373 struct tcphdr tcp_savetcp;
1374 short ostate = 0;
1375 #endif
1376 thflags = th->th_flags;
1377 tp->sackhint.last_sack_ack = 0;
1378
1379 /*
1380 * If this is either a state-changing packet or current state isn't
1381 * established, we require a write lock on tcbinfo. Otherwise, we
1382 * allow either a read lock or a write lock, as we may have acquired
1383 * a write lock due to a race.
1384 *
1385 * Require a global write lock for SYN/FIN/RST segments or
1386 * non-established connections; otherwise accept either a read or
1387 * write lock, as we may have conservatively acquired a write lock in
1388 * certain cases in tcp_input() (is this still true?). Currently we
1389 * will never enter with no lock, so we try to drop it quickly in the
1390 * common pure ack/pure data cases.
1391 */
1392 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 ||
1393 tp->t_state != TCPS_ESTABLISHED) {
1394 KASSERT(ti_locked == TI_WLOCKED, ("%s ti_locked %d for "
1395 "SYN/FIN/RST/!EST", __func__, ti_locked));
1396 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1397 } else {
1398 #ifdef INVARIANTS
1399 if (ti_locked == TI_RLOCKED)
1400 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1401 else if (ti_locked == TI_WLOCKED)
1402 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1403 else
1404 panic("%s: ti_locked %d for EST", __func__,
1405 ti_locked);
1406 #endif
1407 }
1408 INP_WLOCK_ASSERT(tp->t_inpcb);
1409 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN",
1410 __func__));
1411 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT",
1412 __func__));
1413
1414 /*
1415 * Segment received on connection.
1416 * Reset idle time and keep-alive timer.
1417 * XXX: This should be done after segment
1418 * validation to ignore broken/spoofed segs.
1419 */
1420 tp->t_rcvtime = ticks;
1421 if (TCPS_HAVEESTABLISHED(tp->t_state))
1422 tcp_timer_activate(tp, TT_KEEP, tcp_keepidle);
1423
1424 /*
1425 * Unscale the window into a 32-bit value.
1426 * For the SYN_SENT state the scale is zero.
1427 */
1428 tiwin = th->th_win << tp->snd_scale;
1429
1430 /*
1431 * TCP ECN processing.
1432 */
1433 if (tp->t_flags & TF_ECN_PERMIT) {
1434 if (thflags & TH_CWR)
1435 tp->t_flags &= ~TF_ECN_SND_ECE;
1436 switch (iptos & IPTOS_ECN_MASK) {
1437 case IPTOS_ECN_CE:
1438 tp->t_flags |= TF_ECN_SND_ECE;
1439 TCPSTAT_INC(tcps_ecn_ce);
1440 break;
1441 case IPTOS_ECN_ECT0:
1442 TCPSTAT_INC(tcps_ecn_ect0);
1443 break;
1444 case IPTOS_ECN_ECT1:
1445 TCPSTAT_INC(tcps_ecn_ect1);
1446 break;
1447 }
1448 /* Congestion experienced. */
1449 if (thflags & TH_ECE) {
1450 cc_cong_signal(tp, th, CC_ECN);
1451 }
1452 }
1453
1454 /*
1455 * Parse options on any incoming segment.
1456 */
1457 tcp_dooptions(&to, (u_char *)(th + 1),
1458 (th->th_off << 2) - sizeof(struct tcphdr),
1459 (thflags & TH_SYN) ? TO_SYN : 0);
1460
1461 /*
1462 * If echoed timestamp is later than the current time,
1463 * fall back to non RFC1323 RTT calculation. Normalize
1464 * timestamp if syncookies were used when this connection
1465 * was established.
1466 */
1467 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1468 to.to_tsecr -= tp->ts_offset;
1469 if (TSTMP_GT(to.to_tsecr, ticks))
1470 to.to_tsecr = 0;
1471 }
1472
1473 /*
1474 * Process options only when we get SYN/ACK back. The SYN case
1475 * for incoming connections is handled in tcp_syncache.
1476 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
1477 * or <SYN,ACK>) segment itself is never scaled.
1478 * XXX this is traditional behavior, may need to be cleaned up.
1479 */
1480 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1481 if ((to.to_flags & TOF_SCALE) &&
1482 (tp->t_flags & TF_REQ_SCALE)) {
1483 tp->t_flags |= TF_RCVD_SCALE;
1484 tp->snd_scale = to.to_wscale;
1485 }
1486 /*
1487 * Initial send window. It will be updated with
1488 * the next incoming segment to the scaled value.
1489 */
1490 tp->snd_wnd = th->th_win;
1491 if (to.to_flags & TOF_TS) {
1492 tp->t_flags |= TF_RCVD_TSTMP;
1493 tp->ts_recent = to.to_tsval;
1494 tp->ts_recent_age = ticks;
1495 }
1496 if (to.to_flags & TOF_MSS)
1497 tcp_mss(tp, to.to_mss);
1498 if ((tp->t_flags & TF_SACK_PERMIT) &&
1499 (to.to_flags & TOF_SACKPERM) == 0)
1500 tp->t_flags &= ~TF_SACK_PERMIT;
1501 }
1502
1503 /*
1504 * Header prediction: check for the two common cases
1505 * of a uni-directional data xfer. If the packet has
1506 * no control flags, is in-sequence, the window didn't
1507 * change and we're not retransmitting, it's a
1508 * candidate. If the length is zero and the ack moved
1509 * forward, we're the sender side of the xfer. Just
1510 * free the data acked & wake any higher level process
1511 * that was blocked waiting for space. If the length
1512 * is non-zero and the ack didn't move, we're the
1513 * receiver side. If we're getting packets in-order
1514 * (the reassembly queue is empty), add the data to
1515 * the socket buffer and note that we need a delayed ack.
1516 * Make sure that the hidden state-flags are also off.
1517 * Since we check for TCPS_ESTABLISHED first, it can only
1518 * be TH_NEEDSYN.
1519 */
1520 if (tp->t_state == TCPS_ESTABLISHED &&
1521 th->th_seq == tp->rcv_nxt &&
1522 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1523 tp->snd_nxt == tp->snd_max &&
1524 tiwin && tiwin == tp->snd_wnd &&
1525 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1526 LIST_EMPTY(&tp->t_segq) &&
1527 ((to.to_flags & TOF_TS) == 0 ||
1528 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) {
1529
1530 /*
1531 * If last ACK falls within this segment's sequence numbers,
1532 * record the timestamp.
1533 * NOTE that the test is modified according to the latest
1534 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1535 */
1536 if ((to.to_flags & TOF_TS) != 0 &&
1537 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1538 tp->ts_recent_age = ticks;
1539 tp->ts_recent = to.to_tsval;
1540 }
1541
1542 if (tlen == 0) {
1543 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1544 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1545 !IN_RECOVERY(tp->t_flags) &&
1546 (to.to_flags & TOF_SACK) == 0 &&
1547 TAILQ_EMPTY(&tp->snd_holes)) {
1548 /*
1549 * This is a pure ack for outstanding data.
1550 */
1551 if (ti_locked == TI_RLOCKED)
1552 INP_INFO_RUNLOCK(&V_tcbinfo);
1553 else if (ti_locked == TI_WLOCKED)
1554 INP_INFO_WUNLOCK(&V_tcbinfo);
1555 else
1556 panic("%s: ti_locked %d on pure ACK",
1557 __func__, ti_locked);
1558 ti_locked = TI_UNLOCKED;
1559
1560 TCPSTAT_INC(tcps_predack);
1561
1562 /*
1563 * "bad retransmit" recovery.
1564 */
1565 if (tp->t_rxtshift == 1 &&
1566 tp->t_flags & TF_PREVVALID &&
1567 (int)(ticks - tp->t_badrxtwin) < 0) {
1568 cc_cong_signal(tp, th, CC_RTO_ERR);
1569 }
1570
1571 /*
1572 * Recalculate the transmit timer / rtt.
1573 *
1574 * Some boxes send broken timestamp replies
1575 * during the SYN+ACK phase, ignore
1576 * timestamps of 0 or we could calculate a
1577 * huge RTT and blow up the retransmit timer.
1578 */
1579 if ((to.to_flags & TOF_TS) != 0 &&
1580 to.to_tsecr) {
1581 if (!tp->t_rttlow ||
1582 tp->t_rttlow > ticks - to.to_tsecr)
1583 tp->t_rttlow = ticks - to.to_tsecr;
1584 tcp_xmit_timer(tp,
1585 ticks - to.to_tsecr + 1);
1586 } else if (tp->t_rtttime &&
1587 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1588 if (!tp->t_rttlow ||
1589 tp->t_rttlow > ticks - tp->t_rtttime)
1590 tp->t_rttlow = ticks - tp->t_rtttime;
1591 tcp_xmit_timer(tp,
1592 ticks - tp->t_rtttime);
1593 }
1594 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1595 acked = BYTES_THIS_ACK(tp, th);
1596
1597 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
1598 hhook_run_tcp_est_in(tp, th, &to);
1599
1600 TCPSTAT_INC(tcps_rcvackpack);
1601 TCPSTAT_ADD(tcps_rcvackbyte, acked);
1602 sbdrop(&so->so_snd, acked);
1603 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1604 SEQ_LEQ(th->th_ack, tp->snd_recover))
1605 tp->snd_recover = th->th_ack - 1;
1606
1607 /*
1608 * Let the congestion control algorithm update
1609 * congestion control related information. This
1610 * typically means increasing the congestion
1611 * window.
1612 */
1613 cc_ack_received(tp, th, CC_ACK);
1614
1615 tp->snd_una = th->th_ack;
1616 /*
1617 * Pull snd_wl2 up to prevent seq wrap relative
1618 * to th_ack.
1619 */
1620 tp->snd_wl2 = th->th_ack;
1621 tp->t_dupacks = 0;
1622 m_freem(m);
1623 ND6_HINT(tp); /* Some progress has been made. */
1624
1625 /*
1626 * If all outstanding data are acked, stop
1627 * retransmit timer, otherwise restart timer
1628 * using current (possibly backed-off) value.
1629 * If process is waiting for space,
1630 * wakeup/selwakeup/signal. If data
1631 * are ready to send, let tcp_output
1632 * decide between more output or persist.
1633 */
1634 #ifdef TCPDEBUG
1635 if (so->so_options & SO_DEBUG)
1636 tcp_trace(TA_INPUT, ostate, tp,
1637 (void *)tcp_saveipgen,
1638 &tcp_savetcp, 0);
1639 #endif
1640 if (tp->snd_una == tp->snd_max)
1641 tcp_timer_activate(tp, TT_REXMT, 0);
1642 else if (!tcp_timer_active(tp, TT_PERSIST))
1643 tcp_timer_activate(tp, TT_REXMT,
1644 tp->t_rxtcur);
1645 sowwakeup(so);
1646 if (so->so_snd.sb_cc)
1647 (void) tcp_output(tp);
1648 goto check_delack;
1649 }
1650 } else if (th->th_ack == tp->snd_una &&
1651 tlen <= sbspace(&so->so_rcv)) {
1652 int newsize = 0; /* automatic sockbuf scaling */
1653
1654 /*
1655 * This is a pure, in-sequence data packet with
1656 * nothing on the reassembly queue and we have enough
1657 * buffer space to take it.
1658 */
1659 if (ti_locked == TI_RLOCKED)
1660 INP_INFO_RUNLOCK(&V_tcbinfo);
1661 else if (ti_locked == TI_WLOCKED)
1662 INP_INFO_WUNLOCK(&V_tcbinfo);
1663 else
1664 panic("%s: ti_locked %d on pure data "
1665 "segment", __func__, ti_locked);
1666 ti_locked = TI_UNLOCKED;
1667
1668 /* Clean receiver SACK report if present */
1669 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks)
1670 tcp_clean_sackreport(tp);
1671 TCPSTAT_INC(tcps_preddat);
1672 tp->rcv_nxt += tlen;
1673 /*
1674 * Pull snd_wl1 up to prevent seq wrap relative to
1675 * th_seq.
1676 */
1677 tp->snd_wl1 = th->th_seq;
1678 /*
1679 * Pull rcv_up up to prevent seq wrap relative to
1680 * rcv_nxt.
1681 */
1682 tp->rcv_up = tp->rcv_nxt;
1683 TCPSTAT_INC(tcps_rcvpack);
1684 TCPSTAT_ADD(tcps_rcvbyte, tlen);
1685 ND6_HINT(tp); /* Some progress has been made */
1686 #ifdef TCPDEBUG
1687 if (so->so_options & SO_DEBUG)
1688 tcp_trace(TA_INPUT, ostate, tp,
1689 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1690 #endif
1691 /*
1692 * Automatic sizing of receive socket buffer. Often the send
1693 * buffer size is not optimally adjusted to the actual network
1694 * conditions at hand (delay bandwidth product). Setting the
1695 * buffer size too small limits throughput on links with high
1696 * bandwidth and high delay (eg. trans-continental/oceanic links).
1697 *
1698 * On the receive side the socket buffer memory is only rarely
1699 * used to any significant extent. This allows us to be much
1700 * more aggressive in scaling the receive socket buffer. For
1701 * the case that the buffer space is actually used to a large
1702 * extent and we run out of kernel memory we can simply drop
1703 * the new segments; TCP on the sender will just retransmit it
1704 * later. Setting the buffer size too big may only consume too
1705 * much kernel memory if the application doesn't read() from
1706 * the socket or packet loss or reordering makes use of the
1707 * reassembly queue.
1708 *
1709 * The criteria to step up the receive buffer one notch are:
1710 * 1. the number of bytes received during the time it takes
1711 * one timestamp to be reflected back to us (the RTT);
1712 * 2. received bytes per RTT is within seven eighth of the
1713 * current socket buffer size;
1714 * 3. receive buffer size has not hit maximal automatic size;
1715 *
1716 * This algorithm does one step per RTT at most and only if
1717 * we receive a bulk stream w/o packet losses or reorderings.
1718 * Shrinking the buffer during idle times is not necessary as
1719 * it doesn't consume any memory when idle.
1720 *
1721 * TODO: Only step up if the application is actually serving
1722 * the buffer to better manage the socket buffer resources.
1723 */
1724 if (V_tcp_do_autorcvbuf &&
1725 to.to_tsecr &&
1726 (so->so_rcv.sb_flags & SB_AUTOSIZE)) {
1727 if (TSTMP_GT(to.to_tsecr, tp->rfbuf_ts) &&
1728 to.to_tsecr - tp->rfbuf_ts < hz) {
1729 if (tp->rfbuf_cnt >
1730 (so->so_rcv.sb_hiwat / 8 * 7) &&
1731 so->so_rcv.sb_hiwat <
1732 V_tcp_autorcvbuf_max) {
1733 newsize =
1734 min(so->so_rcv.sb_hiwat +
1735 V_tcp_autorcvbuf_inc,
1736 V_tcp_autorcvbuf_max);
1737 }
1738 /* Start over with next RTT. */
1739 tp->rfbuf_ts = 0;
1740 tp->rfbuf_cnt = 0;
1741 } else
1742 tp->rfbuf_cnt += tlen; /* add up */
1743 }
1744
1745 /* Add data to socket buffer. */
1746 SOCKBUF_LOCK(&so->so_rcv);
1747 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1748 m_freem(m);
1749 } else {
1750 /*
1751 * Set new socket buffer size.
1752 * Give up when limit is reached.
1753 */
1754 if (newsize)
1755 if (!sbreserve_locked(&so->so_rcv,
1756 newsize, so, NULL))
1757 so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1758 m_adj(m, drop_hdrlen); /* delayed header drop */
1759 sbappendstream_locked(&so->so_rcv, m);
1760 }
1761 /* NB: sorwakeup_locked() does an implicit unlock. */
1762 sorwakeup_locked(so);
1763 if (DELAY_ACK(tp)) {
1764 tp->t_flags |= TF_DELACK;
1765 } else {
1766 tp->t_flags |= TF_ACKNOW;
1767 tcp_output(tp);
1768 }
1769 goto check_delack;
1770 }
1771 }
1772
1773 /*
1774 * Calculate amount of space in receive window,
1775 * and then do TCP input processing.
1776 * Receive window is amount of space in rcv queue,
1777 * but not less than advertised window.
1778 */
1779 win = sbspace(&so->so_rcv);
1780 if (win < 0)
1781 win = 0;
1782 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1783
1784 /* Reset receive buffer auto scaling when not in bulk receive mode. */
1785 tp->rfbuf_ts = 0;
1786 tp->rfbuf_cnt = 0;
1787
1788 switch (tp->t_state) {
1789
1790 /*
1791 * If the state is SYN_RECEIVED:
1792 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1793 */
1794 case TCPS_SYN_RECEIVED:
1795 if ((thflags & TH_ACK) &&
1796 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1797 SEQ_GT(th->th_ack, tp->snd_max))) {
1798 rstreason = BANDLIM_RST_OPENPORT;
1799 goto dropwithreset;
1800 }
1801 break;
1802
1803 /*
1804 * If the state is SYN_SENT:
1805 * if seg contains an ACK, but not for our SYN, drop the input.
1806 * if seg contains a RST, then drop the connection.
1807 * if seg does not contain SYN, then drop it.
1808 * Otherwise this is an acceptable SYN segment
1809 * initialize tp->rcv_nxt and tp->irs
1810 * if seg contains ack then advance tp->snd_una
1811 * if seg contains an ECE and ECN support is enabled, the stream
1812 * is ECN capable.
1813 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1814 * arrange for segment to be acked (eventually)
1815 * continue processing rest of data/controls, beginning with URG
1816 */
1817 case TCPS_SYN_SENT:
1818 if ((thflags & TH_ACK) &&
1819 (SEQ_LEQ(th->th_ack, tp->iss) ||
1820 SEQ_GT(th->th_ack, tp->snd_max))) {
1821 rstreason = BANDLIM_UNLIMITED;
1822 goto dropwithreset;
1823 }
1824 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST))
1825 tp = tcp_drop(tp, ECONNREFUSED);
1826 if (thflags & TH_RST)
1827 goto drop;
1828 if (!(thflags & TH_SYN))
1829 goto drop;
1830
1831 tp->irs = th->th_seq;
1832 tcp_rcvseqinit(tp);
1833 if (thflags & TH_ACK) {
1834 TCPSTAT_INC(tcps_connects);
1835 soisconnected(so);
1836 #ifdef MAC
1837 mac_socketpeer_set_from_mbuf(m, so);
1838 #endif
1839 /* Do window scaling on this connection? */
1840 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1841 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1842 tp->rcv_scale = tp->request_r_scale;
1843 }
1844 tp->rcv_adv += imin(tp->rcv_wnd,
1845 TCP_MAXWIN << tp->rcv_scale);
1846 tp->snd_una++; /* SYN is acked */
1847 /*
1848 * If there's data, delay ACK; if there's also a FIN
1849 * ACKNOW will be turned on later.
1850 */
1851 if (DELAY_ACK(tp) && tlen != 0)
1852 tcp_timer_activate(tp, TT_DELACK,
1853 tcp_delacktime);
1854 else
1855 tp->t_flags |= TF_ACKNOW;
1856
1857 if ((thflags & TH_ECE) && V_tcp_do_ecn) {
1858 tp->t_flags |= TF_ECN_PERMIT;
1859 TCPSTAT_INC(tcps_ecn_shs);
1860 }
1861
1862 /*
1863 * Received <SYN,ACK> in SYN_SENT[*] state.
1864 * Transitions:
1865 * SYN_SENT --> ESTABLISHED
1866 * SYN_SENT* --> FIN_WAIT_1
1867 */
1868 tp->t_starttime = ticks;
1869 if (tp->t_flags & TF_NEEDFIN) {
1870 tp->t_state = TCPS_FIN_WAIT_1;
1871 tp->t_flags &= ~TF_NEEDFIN;
1872 thflags &= ~TH_SYN;
1873 } else {
1874 tp->t_state = TCPS_ESTABLISHED;
1875 cc_conn_init(tp);
1876 tcp_timer_activate(tp, TT_KEEP, tcp_keepidle);
1877 }
1878 } else {
1879 /*
1880 * Received initial SYN in SYN-SENT[*] state =>
1881 * simultaneous open. If segment contains CC option
1882 * and there is a cached CC, apply TAO test.
1883 * If it succeeds, connection is * half-synchronized.
1884 * Otherwise, do 3-way handshake:
1885 * SYN-SENT -> SYN-RECEIVED
1886 * SYN-SENT* -> SYN-RECEIVED*
1887 * If there was no CC option, clear cached CC value.
1888 */
1889 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
1890 tcp_timer_activate(tp, TT_REXMT, 0);
1891 tp->t_state = TCPS_SYN_RECEIVED;
1892 }
1893
1894 KASSERT(ti_locked == TI_WLOCKED, ("%s: trimthenstep6: "
1895 "ti_locked %d", __func__, ti_locked));
1896 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1897 INP_WLOCK_ASSERT(tp->t_inpcb);
1898
1899 /*
1900 * Advance th->th_seq to correspond to first data byte.
1901 * If data, trim to stay within window,
1902 * dropping FIN if necessary.
1903 */
1904 th->th_seq++;
1905 if (tlen > tp->rcv_wnd) {
1906 todrop = tlen - tp->rcv_wnd;
1907 m_adj(m, -todrop);
1908 tlen = tp->rcv_wnd;
1909 thflags &= ~TH_FIN;
1910 TCPSTAT_INC(tcps_rcvpackafterwin);
1911 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
1912 }
1913 tp->snd_wl1 = th->th_seq - 1;
1914 tp->rcv_up = th->th_seq;
1915 /*
1916 * Client side of transaction: already sent SYN and data.
1917 * If the remote host used T/TCP to validate the SYN,
1918 * our data will be ACK'd; if so, enter normal data segment
1919 * processing in the middle of step 5, ack processing.
1920 * Otherwise, goto step 6.
1921 */
1922 if (thflags & TH_ACK)
1923 goto process_ACK;
1924
1925 goto step6;
1926
1927 /*
1928 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
1929 * do normal processing.
1930 *
1931 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later.
1932 */
1933 case TCPS_LAST_ACK:
1934 case TCPS_CLOSING:
1935 break; /* continue normal processing */
1936 }
1937
1938 /*
1939 * States other than LISTEN or SYN_SENT.
1940 * First check the RST flag and sequence number since reset segments
1941 * are exempt from the timestamp and connection count tests. This
1942 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
1943 * below which allowed reset segments in half the sequence space
1944 * to fall though and be processed (which gives forged reset
1945 * segments with a random sequence number a 50 percent chance of
1946 * killing a connection).
1947 * Then check timestamp, if present.
1948 * Then check the connection count, if present.
1949 * Then check that at least some bytes of segment are within
1950 * receive window. If segment begins before rcv_nxt,
1951 * drop leading data (and SYN); if nothing left, just ack.
1952 *
1953 *
1954 * If the RST bit is set, check the sequence number to see
1955 * if this is a valid reset segment.
1956 * RFC 793 page 37:
1957 * In all states except SYN-SENT, all reset (RST) segments
1958 * are validated by checking their SEQ-fields. A reset is
1959 * valid if its sequence number is in the window.
1960 * Note: this does not take into account delayed ACKs, so
1961 * we should test against last_ack_sent instead of rcv_nxt.
1962 * The sequence number in the reset segment is normally an
1963 * echo of our outgoing acknowlegement numbers, but some hosts
1964 * send a reset with the sequence number at the rightmost edge
1965 * of our receive window, and we have to handle this case.
1966 * Note 2: Paul Watson's paper "Slipping in the Window" has shown
1967 * that brute force RST attacks are possible. To combat this,
1968 * we use a much stricter check while in the ESTABLISHED state,
1969 * only accepting RSTs where the sequence number is equal to
1970 * last_ack_sent. In all other states (the states in which a
1971 * RST is more likely), the more permissive check is used.
1972 * If we have multiple segments in flight, the initial reset
1973 * segment sequence numbers will be to the left of last_ack_sent,
1974 * but they will eventually catch up.
1975 * In any case, it never made sense to trim reset segments to
1976 * fit the receive window since RFC 1122 says:
1977 * 4.2.2.12 RST Segment: RFC-793 Section 3.4
1978 *
1979 * A TCP SHOULD allow a received RST segment to include data.
1980 *
1981 * DISCUSSION
1982 * It has been suggested that a RST segment could contain
1983 * ASCII text that encoded and explained the cause of the
1984 * RST. No standard has yet been established for such
1985 * data.
1986 *
1987 * If the reset segment passes the sequence number test examine
1988 * the state:
1989 * SYN_RECEIVED STATE:
1990 * If passive open, return to LISTEN state.
1991 * If active open, inform user that connection was refused.
1992 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
1993 * Inform user that connection was reset, and close tcb.
1994 * CLOSING, LAST_ACK STATES:
1995 * Close the tcb.
1996 * TIME_WAIT STATE:
1997 * Drop the segment - see Stevens, vol. 2, p. 964 and
1998 * RFC 1337.
1999 */
2000 if (thflags & TH_RST) {
2001 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) &&
2002 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
2003 switch (tp->t_state) {
2004
2005 case TCPS_SYN_RECEIVED:
2006 so->so_error = ECONNREFUSED;
2007 goto close;
2008
2009 case TCPS_ESTABLISHED:
2010 if (V_tcp_insecure_rst == 0 &&
2011 !(SEQ_GEQ(th->th_seq, tp->rcv_nxt - 1) &&
2012 SEQ_LEQ(th->th_seq, tp->rcv_nxt + 1)) &&
2013 !(SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) &&
2014 SEQ_LEQ(th->th_seq, tp->last_ack_sent + 1))) {
2015 TCPSTAT_INC(tcps_badrst);
2016 goto drop;
2017 }
2018 /* FALLTHROUGH */
2019 case TCPS_FIN_WAIT_1:
2020 case TCPS_FIN_WAIT_2:
2021 case TCPS_CLOSE_WAIT:
2022 so->so_error = ECONNRESET;
2023 close:
2024 KASSERT(ti_locked == TI_WLOCKED,
2025 ("tcp_do_segment: TH_RST 1 ti_locked %d",
2026 ti_locked));
2027 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2028
2029 tp->t_state = TCPS_CLOSED;
2030 TCPSTAT_INC(tcps_drops);
2031 tp = tcp_close(tp);
2032 break;
2033
2034 case TCPS_CLOSING:
2035 case TCPS_LAST_ACK:
2036 KASSERT(ti_locked == TI_WLOCKED,
2037 ("tcp_do_segment: TH_RST 2 ti_locked %d",
2038 ti_locked));
2039 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2040
2041 tp = tcp_close(tp);
2042 break;
2043 }
2044 }
2045 goto drop;
2046 }
2047
2048 /*
2049 * RFC 1323 PAWS: If we have a timestamp reply on this segment
2050 * and it's less than ts_recent, drop it.
2051 */
2052 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
2053 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
2054
2055 /* Check to see if ts_recent is over 24 days old. */
2056 if (ticks - tp->ts_recent_age > TCP_PAWS_IDLE) {
2057 /*
2058 * Invalidate ts_recent. If this segment updates
2059 * ts_recent, the age will be reset later and ts_recent
2060 * will get a valid value. If it does not, setting
2061 * ts_recent to zero will at least satisfy the
2062 * requirement that zero be placed in the timestamp
2063 * echo reply when ts_recent isn't valid. The
2064 * age isn't reset until we get a valid ts_recent
2065 * because we don't want out-of-order segments to be
2066 * dropped when ts_recent is old.
2067 */
2068 tp->ts_recent = 0;
2069 } else {
2070 TCPSTAT_INC(tcps_rcvduppack);
2071 TCPSTAT_ADD(tcps_rcvdupbyte, tlen);
2072 TCPSTAT_INC(tcps_pawsdrop);
2073 if (tlen)
2074 goto dropafterack;
2075 goto drop;
2076 }
2077 }
2078
2079 /*
2080 * In the SYN-RECEIVED state, validate that the packet belongs to
2081 * this connection before trimming the data to fit the receive
2082 * window. Check the sequence number versus IRS since we know
2083 * the sequence numbers haven't wrapped. This is a partial fix
2084 * for the "LAND" DoS attack.
2085 */
2086 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
2087 rstreason = BANDLIM_RST_OPENPORT;
2088 goto dropwithreset;
2089 }
2090
2091 todrop = tp->rcv_nxt - th->th_seq;
2092 if (todrop > 0) {
2093 /*
2094 * If this is a duplicate SYN for our current connection,
2095 * advance over it and pretend and it's not a SYN.
2096 */
2097 if (thflags & TH_SYN && th->th_seq == tp->irs) {
2098 thflags &= ~TH_SYN;
2099 th->th_seq++;
2100 if (th->th_urp > 1)
2101 th->th_urp--;
2102 else
2103 thflags &= ~TH_URG;
2104 todrop--;
2105 }
2106 /*
2107 * Following if statement from Stevens, vol. 2, p. 960.
2108 */
2109 if (todrop > tlen
2110 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
2111 /*
2112 * Any valid FIN must be to the left of the window.
2113 * At this point the FIN must be a duplicate or out
2114 * of sequence; drop it.
2115 */
2116 thflags &= ~TH_FIN;
2117
2118 /*
2119 * Send an ACK to resynchronize and drop any data.
2120 * But keep on processing for RST or ACK.
2121 */
2122 tp->t_flags |= TF_ACKNOW;
2123 todrop = tlen;
2124 TCPSTAT_INC(tcps_rcvduppack);
2125 TCPSTAT_ADD(tcps_rcvdupbyte, todrop);
2126 } else {
2127 TCPSTAT_INC(tcps_rcvpartduppack);
2128 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop);
2129 }
2130 drop_hdrlen += todrop; /* drop from the top afterwards */
2131 th->th_seq += todrop;
2132 tlen -= todrop;
2133 if (th->th_urp > todrop)
2134 th->th_urp -= todrop;
2135 else {
2136 thflags &= ~TH_URG;
2137 th->th_urp = 0;
2138 }
2139 }
2140
2141 /*
2142 * If new data are received on a connection after the
2143 * user processes are gone, then RST the other end.
2144 */
2145 if ((so->so_state & SS_NOFDREF) &&
2146 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
2147 char *s;
2148
2149 KASSERT(ti_locked == TI_WLOCKED, ("%s: SS_NOFDEREF && "
2150 "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked));
2151 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2152
2153 if ((s = tcp_log_addrs(&tp->t_inpcb->inp_inc, th, NULL, NULL))) {
2154 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data after socket "
2155 "was closed, sending RST and removing tcpcb\n",
2156 s, __func__, tcpstates[tp->t_state], tlen);
2157 free(s, M_TCPLOG);
2158 }
2159 tp = tcp_close(tp);
2160 TCPSTAT_INC(tcps_rcvafterclose);
2161 rstreason = BANDLIM_UNLIMITED;
2162 goto dropwithreset;
2163 }
2164
2165 /*
2166 * If segment ends after window, drop trailing data
2167 * (and PUSH and FIN); if nothing left, just ACK.
2168 */
2169 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
2170 if (todrop > 0) {
2171 TCPSTAT_INC(tcps_rcvpackafterwin);
2172 if (todrop >= tlen) {
2173 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen);
2174 /*
2175 * If window is closed can only take segments at
2176 * window edge, and have to drop data and PUSH from
2177 * incoming segments. Continue processing, but
2178 * remember to ack. Otherwise, drop segment
2179 * and ack.
2180 */
2181 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2182 tp->t_flags |= TF_ACKNOW;
2183 TCPSTAT_INC(tcps_rcvwinprobe);
2184 } else
2185 goto dropafterack;
2186 } else
2187 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2188 m_adj(m, -todrop);
2189 tlen -= todrop;
2190 thflags &= ~(TH_PUSH|TH_FIN);
2191 }
2192
2193 /*
2194 * If last ACK falls within this segment's sequence numbers,
2195 * record its timestamp.
2196 * NOTE:
2197 * 1) That the test incorporates suggestions from the latest
2198 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
2199 * 2) That updating only on newer timestamps interferes with
2200 * our earlier PAWS tests, so this check should be solely
2201 * predicated on the sequence space of this segment.
2202 * 3) That we modify the segment boundary check to be
2203 * Last.ACK.Sent <= SEG.SEQ + SEG.Len
2204 * instead of RFC1323's
2205 * Last.ACK.Sent < SEG.SEQ + SEG.Len,
2206 * This modified check allows us to overcome RFC1323's
2207 * limitations as described in Stevens TCP/IP Illustrated
2208 * Vol. 2 p.869. In such cases, we can still calculate the
2209 * RTT correctly when RCV.NXT == Last.ACK.Sent.
2210 */
2211 if ((to.to_flags & TOF_TS) != 0 &&
2212 SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2213 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
2214 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
2215 tp->ts_recent_age = ticks;
2216 tp->ts_recent = to.to_tsval;
2217 }
2218
2219 /*
2220 * If a SYN is in the window, then this is an
2221 * error and we send an RST and drop the connection.
2222 */
2223 if (thflags & TH_SYN) {
2224 KASSERT(ti_locked == TI_WLOCKED,
2225 ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked));
2226 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2227
2228 tp = tcp_drop(tp, ECONNRESET);
2229 rstreason = BANDLIM_UNLIMITED;
2230 goto drop;
2231 }
2232
2233 /*
2234 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
2235 * flag is on (half-synchronized state), then queue data for
2236 * later processing; else drop segment and return.
2237 */
2238 if ((thflags & TH_ACK) == 0) {
2239 if (tp->t_state == TCPS_SYN_RECEIVED ||
2240 (tp->t_flags & TF_NEEDSYN))
2241 goto step6;
2242 else if (tp->t_flags & TF_ACKNOW)
2243 goto dropafterack;
2244 else
2245 goto drop;
2246 }
2247
2248 /*
2249 * Ack processing.
2250 */
2251 switch (tp->t_state) {
2252
2253 /*
2254 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
2255 * ESTABLISHED state and continue processing.
2256 * The ACK was checked above.
2257 */
2258 case TCPS_SYN_RECEIVED:
2259
2260 TCPSTAT_INC(tcps_connects);
2261 soisconnected(so);
2262 /* Do window scaling? */
2263 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2264 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2265 tp->rcv_scale = tp->request_r_scale;
2266 tp->snd_wnd = tiwin;
2267 }
2268 /*
2269 * Make transitions:
2270 * SYN-RECEIVED -> ESTABLISHED
2271 * SYN-RECEIVED* -> FIN-WAIT-1
2272 */
2273 tp->t_starttime = ticks;
2274 if (tp->t_flags & TF_NEEDFIN) {
2275 tp->t_state = TCPS_FIN_WAIT_1;
2276 tp->t_flags &= ~TF_NEEDFIN;
2277 } else {
2278 tp->t_state = TCPS_ESTABLISHED;
2279 cc_conn_init(tp);
2280 tcp_timer_activate(tp, TT_KEEP, tcp_keepidle);
2281 }
2282 /*
2283 * If segment contains data or ACK, will call tcp_reass()
2284 * later; if not, do so now to pass queued data to user.
2285 */
2286 if (tlen == 0 && (thflags & TH_FIN) == 0)
2287 (void) tcp_reass(tp, (struct tcphdr *)0, 0,
2288 (struct mbuf *)0);
2289 tp->snd_wl1 = th->th_seq - 1;
2290 /* FALLTHROUGH */
2291
2292 /*
2293 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2294 * ACKs. If the ack is in the range
2295 * tp->snd_una < th->th_ack <= tp->snd_max
2296 * then advance tp->snd_una to th->th_ack and drop
2297 * data from the retransmission queue. If this ACK reflects
2298 * more up to date window information we update our window information.
2299 */
2300 case TCPS_ESTABLISHED:
2301 case TCPS_FIN_WAIT_1:
2302 case TCPS_FIN_WAIT_2:
2303 case TCPS_CLOSE_WAIT:
2304 case TCPS_CLOSING:
2305 case TCPS_LAST_ACK:
2306 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2307 TCPSTAT_INC(tcps_rcvacktoomuch);
2308 goto dropafterack;
2309 }
2310 if ((tp->t_flags & TF_SACK_PERMIT) &&
2311 ((to.to_flags & TOF_SACK) ||
2312 !TAILQ_EMPTY(&tp->snd_holes)))
2313 tcp_sack_doack(tp, &to, th->th_ack);
2314
2315 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
2316 hhook_run_tcp_est_in(tp, th, &to);
2317
2318 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2319 if (tlen == 0 && tiwin == tp->snd_wnd) {
2320 TCPSTAT_INC(tcps_rcvdupack);
2321 /*
2322 * If we have outstanding data (other than
2323 * a window probe), this is a completely
2324 * duplicate ack (ie, window info didn't
2325 * change), the ack is the biggest we've
2326 * seen and we've seen exactly our rexmt
2327 * threshhold of them, assume a packet
2328 * has been dropped and retransmit it.
2329 * Kludge snd_nxt & the congestion
2330 * window so we send only this one
2331 * packet.
2332 *
2333 * We know we're losing at the current
2334 * window size so do congestion avoidance
2335 * (set ssthresh to half the current window
2336 * and pull our congestion window back to
2337 * the new ssthresh).
2338 *
2339 * Dup acks mean that packets have left the
2340 * network (they're now cached at the receiver)
2341 * so bump cwnd by the amount in the receiver
2342 * to keep a constant cwnd packets in the
2343 * network.
2344 *
2345 * When using TCP ECN, notify the peer that
2346 * we reduced the cwnd.
2347 */
2348 if (!tcp_timer_active(tp, TT_REXMT) ||
2349 th->th_ack != tp->snd_una)
2350 tp->t_dupacks = 0;
2351 else if (++tp->t_dupacks > tcprexmtthresh ||
2352 IN_FASTRECOVERY(tp->t_flags)) {
2353 cc_ack_received(tp, th, CC_DUPACK);
2354 if ((tp->t_flags & TF_SACK_PERMIT) &&
2355 IN_FASTRECOVERY(tp->t_flags)) {
2356 int awnd;
2357
2358 /*
2359 * Compute the amount of data in flight first.
2360 * We can inject new data into the pipe iff
2361 * we have less than 1/2 the original window's
2362 * worth of data in flight.
2363 */
2364 awnd = (tp->snd_nxt - tp->snd_fack) +
2365 tp->sackhint.sack_bytes_rexmit;
2366 if (awnd < tp->snd_ssthresh) {
2367 tp->snd_cwnd += tp->t_maxseg;
2368 if (tp->snd_cwnd > tp->snd_ssthresh)
2369 tp->snd_cwnd = tp->snd_ssthresh;
2370 }
2371 } else
2372 tp->snd_cwnd += tp->t_maxseg;
2373 (void) tcp_output(tp);
2374 goto drop;
2375 } else if (tp->t_dupacks == tcprexmtthresh) {
2376 tcp_seq onxt = tp->snd_nxt;
2377
2378 /*
2379 * If we're doing sack, check to
2380 * see if we're already in sack
2381 * recovery. If we're not doing sack,
2382 * check to see if we're in newreno
2383 * recovery.
2384 */
2385 if (tp->t_flags & TF_SACK_PERMIT) {
2386 if (IN_FASTRECOVERY(tp->t_flags)) {
2387 tp->t_dupacks = 0;
2388 break;
2389 }
2390 } else {
2391 if (SEQ_LEQ(th->th_ack,
2392 tp->snd_recover)) {
2393 tp->t_dupacks = 0;
2394 break;
2395 }
2396 }
2397 /* Congestion signal before ack. */
2398 cc_cong_signal(tp, th, CC_NDUPACK);
2399 cc_ack_received(tp, th, CC_DUPACK);
2400 tcp_timer_activate(tp, TT_REXMT, 0);
2401 tp->t_rtttime = 0;
2402 if (tp->t_flags & TF_SACK_PERMIT) {
2403 TCPSTAT_INC(
2404 tcps_sack_recovery_episode);
2405 tp->sack_newdata = tp->snd_nxt;
2406 tp->snd_cwnd = tp->t_maxseg;
2407 (void) tcp_output(tp);
2408 goto drop;
2409 }
2410 tp->snd_nxt = th->th_ack;
2411 tp->snd_cwnd = tp->t_maxseg;
2412 (void) tcp_output(tp);
2413 KASSERT(tp->snd_limited <= 2,
2414 ("%s: tp->snd_limited too big",
2415 __func__));
2416 tp->snd_cwnd = tp->snd_ssthresh +
2417 tp->t_maxseg *
2418 (tp->t_dupacks - tp->snd_limited);
2419 if (SEQ_GT(onxt, tp->snd_nxt))
2420 tp->snd_nxt = onxt;
2421 goto drop;
2422 } else if (V_tcp_do_rfc3042) {
2423 cc_ack_received(tp, th, CC_DUPACK);
2424 u_long oldcwnd = tp->snd_cwnd;
2425 tcp_seq oldsndmax = tp->snd_max;
2426 u_int sent;
2427
2428 KASSERT(tp->t_dupacks == 1 ||
2429 tp->t_dupacks == 2,
2430 ("%s: dupacks not 1 or 2",
2431 __func__));
2432 if (tp->t_dupacks == 1)
2433 tp->snd_limited = 0;
2434 tp->snd_cwnd =
2435 (tp->snd_nxt - tp->snd_una) +
2436 (tp->t_dupacks - tp->snd_limited) *
2437 tp->t_maxseg;
2438 (void) tcp_output(tp);
2439 sent = tp->snd_max - oldsndmax;
2440 if (sent > tp->t_maxseg) {
2441 KASSERT((tp->t_dupacks == 2 &&
2442 tp->snd_limited == 0) ||
2443 (sent == tp->t_maxseg + 1 &&
2444 tp->t_flags & TF_SENTFIN),
2445 ("%s: sent too much",
2446 __func__));
2447 tp->snd_limited = 2;
2448 } else if (sent > 0)
2449 ++tp->snd_limited;
2450 tp->snd_cwnd = oldcwnd;
2451 goto drop;
2452 }
2453 } else
2454 tp->t_dupacks = 0;
2455 break;
2456 }
2457
2458 KASSERT(SEQ_GT(th->th_ack, tp->snd_una),
2459 ("%s: th_ack <= snd_una", __func__));
2460
2461 /*
2462 * If the congestion window was inflated to account
2463 * for the other side's cached packets, retract it.
2464 */
2465 if (IN_FASTRECOVERY(tp->t_flags)) {
2466 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2467 if (tp->t_flags & TF_SACK_PERMIT)
2468 tcp_sack_partialack(tp, th);
2469 else
2470 tcp_newreno_partial_ack(tp, th);
2471 } else
2472 cc_post_recovery(tp, th);
2473 }
2474 tp->t_dupacks = 0;
2475 /*
2476 * If we reach this point, ACK is not a duplicate,
2477 * i.e., it ACKs something we sent.
2478 */
2479 if (tp->t_flags & TF_NEEDSYN) {
2480 /*
2481 * T/TCP: Connection was half-synchronized, and our
2482 * SYN has been ACK'd (so connection is now fully
2483 * synchronized). Go to non-starred state,
2484 * increment snd_una for ACK of SYN, and check if
2485 * we can do window scaling.
2486 */
2487 tp->t_flags &= ~TF_NEEDSYN;
2488 tp->snd_una++;
2489 /* Do window scaling? */
2490 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2491 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2492 tp->rcv_scale = tp->request_r_scale;
2493 /* Send window already scaled. */
2494 }
2495 }
2496
2497 process_ACK:
2498 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
2499 KASSERT(ti_locked == TI_RLOCKED || ti_locked == TI_WLOCKED,
2500 ("tcp_input: process_ACK ti_locked %d", ti_locked));
2501 INP_WLOCK_ASSERT(tp->t_inpcb);
2502
2503 acked = BYTES_THIS_ACK(tp, th);
2504 TCPSTAT_INC(tcps_rcvackpack);
2505 TCPSTAT_ADD(tcps_rcvackbyte, acked);
2506
2507 /*
2508 * If we just performed our first retransmit, and the ACK
2509 * arrives within our recovery window, then it was a mistake
2510 * to do the retransmit in the first place. Recover our
2511 * original cwnd and ssthresh, and proceed to transmit where
2512 * we left off.
2513 */
2514 if (tp->t_rxtshift == 1 && tp->t_flags & TF_PREVVALID &&
2515 (int)(ticks - tp->t_badrxtwin) < 0)
2516 cc_cong_signal(tp, th, CC_RTO_ERR);
2517
2518 /*
2519 * If we have a timestamp reply, update smoothed
2520 * round trip time. If no timestamp is present but
2521 * transmit timer is running and timed sequence
2522 * number was acked, update smoothed round trip time.
2523 * Since we now have an rtt measurement, cancel the
2524 * timer backoff (cf., Phil Karn's retransmit alg.).
2525 * Recompute the initial retransmit timer.
2526 *
2527 * Some boxes send broken timestamp replies
2528 * during the SYN+ACK phase, ignore
2529 * timestamps of 0 or we could calculate a
2530 * huge RTT and blow up the retransmit timer.
2531 */
2532 if ((to.to_flags & TOF_TS) != 0 &&
2533 to.to_tsecr) {
2534 if (!tp->t_rttlow || tp->t_rttlow > ticks - to.to_tsecr)
2535 tp->t_rttlow = ticks - to.to_tsecr;
2536 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1);
2537 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2538 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2539 tp->t_rttlow = ticks - tp->t_rtttime;
2540 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2541 }
2542 tcp_xmit_bandwidth_limit(tp, th->th_ack);
2543
2544 /*
2545 * If all outstanding data is acked, stop retransmit
2546 * timer and remember to restart (more output or persist).
2547 * If there is more data to be acked, restart retransmit
2548 * timer, using current (possibly backed-off) value.
2549 */
2550 if (th->th_ack == tp->snd_max) {
2551 tcp_timer_activate(tp, TT_REXMT, 0);
2552 needoutput = 1;
2553 } else if (!tcp_timer_active(tp, TT_PERSIST))
2554 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
2555
2556 /*
2557 * If no data (only SYN) was ACK'd,
2558 * skip rest of ACK processing.
2559 */
2560 if (acked == 0)
2561 goto step6;
2562
2563 /*
2564 * Let the congestion control algorithm update congestion
2565 * control related information. This typically means increasing
2566 * the congestion window.
2567 */
2568 cc_ack_received(tp, th, CC_ACK);
2569
2570 SOCKBUF_LOCK(&so->so_snd);
2571 if (acked > so->so_snd.sb_cc) {
2572 tp->snd_wnd -= so->so_snd.sb_cc;
2573 sbdrop_locked(&so->so_snd, (int)so->so_snd.sb_cc);
2574 ourfinisacked = 1;
2575 } else {
2576 sbdrop_locked(&so->so_snd, acked);
2577 tp->snd_wnd -= acked;
2578 ourfinisacked = 0;
2579 }
2580 /* NB: sowwakeup_locked() does an implicit unlock. */
2581 sowwakeup_locked(so);
2582 /* Detect una wraparound. */
2583 if (!IN_RECOVERY(tp->t_flags) &&
2584 SEQ_GT(tp->snd_una, tp->snd_recover) &&
2585 SEQ_LEQ(th->th_ack, tp->snd_recover))
2586 tp->snd_recover = th->th_ack - 1;
2587 /* XXXLAS: Can this be moved up into cc_post_recovery? */
2588 if (IN_RECOVERY(tp->t_flags) &&
2589 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2590 EXIT_RECOVERY(tp->t_flags);
2591 }
2592 tp->snd_una = th->th_ack;
2593 if (tp->t_flags & TF_SACK_PERMIT) {
2594 if (SEQ_GT(tp->snd_una, tp->snd_recover))
2595 tp->snd_recover = tp->snd_una;
2596 }
2597 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2598 tp->snd_nxt = tp->snd_una;
2599
2600 switch (tp->t_state) {
2601
2602 /*
2603 * In FIN_WAIT_1 STATE in addition to the processing
2604 * for the ESTABLISHED state if our FIN is now acknowledged
2605 * then enter FIN_WAIT_2.
2606 */
2607 case TCPS_FIN_WAIT_1:
2608 if (ourfinisacked) {
2609 /*
2610 * If we can't receive any more
2611 * data, then closing user can proceed.
2612 * Starting the timer is contrary to the
2613 * specification, but if we don't get a FIN
2614 * we'll hang forever.
2615 *
2616 * XXXjl:
2617 * we should release the tp also, and use a
2618 * compressed state.
2619 */
2620 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2621 int timeout;
2622
2623 soisdisconnected(so);
2624 timeout = (tcp_fast_finwait2_recycle) ?
2625 tcp_finwait2_timeout : tcp_maxidle;
2626 tcp_timer_activate(tp, TT_2MSL, timeout);
2627 }
2628 tp->t_state = TCPS_FIN_WAIT_2;
2629 }
2630 break;
2631
2632 /*
2633 * In CLOSING STATE in addition to the processing for
2634 * the ESTABLISHED state if the ACK acknowledges our FIN
2635 * then enter the TIME-WAIT state, otherwise ignore
2636 * the segment.
2637 */
2638 case TCPS_CLOSING:
2639 if (ourfinisacked) {
2640 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2641 tcp_twstart(tp);
2642 INP_INFO_WUNLOCK(&V_tcbinfo);
2643 m_freem(m);
2644 return;
2645 }
2646 break;
2647
2648 /*
2649 * In LAST_ACK, we may still be waiting for data to drain
2650 * and/or to be acked, as well as for the ack of our FIN.
2651 * If our FIN is now acknowledged, delete the TCB,
2652 * enter the closed state and return.
2653 */
2654 case TCPS_LAST_ACK:
2655 if (ourfinisacked) {
2656 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2657 tp = tcp_close(tp);
2658 goto drop;
2659 }
2660 break;
2661 }
2662 }
2663
2664 step6:
2665 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
2666 KASSERT(ti_locked == TI_RLOCKED || ti_locked == TI_WLOCKED,
2667 ("tcp_do_segment: step6 ti_locked %d", ti_locked));
2668 INP_WLOCK_ASSERT(tp->t_inpcb);
2669
2670 /*
2671 * Update window information.
2672 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2673 */
2674 if ((thflags & TH_ACK) &&
2675 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2676 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2677 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2678 /* keep track of pure window updates */
2679 if (tlen == 0 &&
2680 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2681 TCPSTAT_INC(tcps_rcvwinupd);
2682 tp->snd_wnd = tiwin;
2683 tp->snd_wl1 = th->th_seq;
2684 tp->snd_wl2 = th->th_ack;
2685 if (tp->snd_wnd > tp->max_sndwnd)
2686 tp->max_sndwnd = tp->snd_wnd;
2687 needoutput = 1;
2688 }
2689
2690 /*
2691 * Process segments with URG.
2692 */
2693 if ((thflags & TH_URG) && th->th_urp &&
2694 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2695 /*
2696 * This is a kludge, but if we receive and accept
2697 * random urgent pointers, we'll crash in
2698 * soreceive. It's hard to imagine someone
2699 * actually wanting to send this much urgent data.
2700 */
2701 SOCKBUF_LOCK(&so->so_rcv);
2702 if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2703 th->th_urp = 0; /* XXX */
2704 thflags &= ~TH_URG; /* XXX */
2705 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */
2706 goto dodata; /* XXX */
2707 }
2708 /*
2709 * If this segment advances the known urgent pointer,
2710 * then mark the data stream. This should not happen
2711 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2712 * a FIN has been received from the remote side.
2713 * In these states we ignore the URG.
2714 *
2715 * According to RFC961 (Assigned Protocols),
2716 * the urgent pointer points to the last octet
2717 * of urgent data. We continue, however,
2718 * to consider it to indicate the first octet
2719 * of data past the urgent section as the original
2720 * spec states (in one of two places).
2721 */
2722 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2723 tp->rcv_up = th->th_seq + th->th_urp;
2724 so->so_oobmark = so->so_rcv.sb_cc +
2725 (tp->rcv_up - tp->rcv_nxt) - 1;
2726 if (so->so_oobmark == 0)
2727 so->so_rcv.sb_state |= SBS_RCVATMARK;
2728 sohasoutofband(so);
2729 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2730 }
2731 SOCKBUF_UNLOCK(&so->so_rcv);
2732 /*
2733 * Remove out of band data so doesn't get presented to user.
2734 * This can happen independent of advancing the URG pointer,
2735 * but if two URG's are pending at once, some out-of-band
2736 * data may creep in... ick.
2737 */
2738 if (th->th_urp <= (u_long)tlen &&
2739 !(so->so_options & SO_OOBINLINE)) {
2740 /* hdr drop is delayed */
2741 tcp_pulloutofband(so, th, m, drop_hdrlen);
2742 }
2743 } else {
2744 /*
2745 * If no out of band data is expected,
2746 * pull receive urgent pointer along
2747 * with the receive window.
2748 */
2749 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2750 tp->rcv_up = tp->rcv_nxt;
2751 }
2752 dodata: /* XXX */
2753 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
2754 KASSERT(ti_locked == TI_RLOCKED || ti_locked == TI_WLOCKED,
2755 ("tcp_do_segment: dodata ti_locked %d", ti_locked));
2756 INP_WLOCK_ASSERT(tp->t_inpcb);
2757
2758 /*
2759 * Process the segment text, merging it into the TCP sequencing queue,
2760 * and arranging for acknowledgment of receipt if necessary.
2761 * This process logically involves adjusting tp->rcv_wnd as data
2762 * is presented to the user (this happens in tcp_usrreq.c,
2763 * case PRU_RCVD). If a FIN has already been received on this
2764 * connection then we just ignore the text.
2765 */
2766 if ((tlen || (thflags & TH_FIN)) &&
2767 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2768 tcp_seq save_start = th->th_seq;
2769 m_adj(m, drop_hdrlen); /* delayed header drop */
2770 /*
2771 * Insert segment which includes th into TCP reassembly queue
2772 * with control block tp. Set thflags to whether reassembly now
2773 * includes a segment with FIN. This handles the common case
2774 * inline (segment is the next to be received on an established
2775 * connection, and the queue is empty), avoiding linkage into
2776 * and removal from the queue and repetition of various
2777 * conversions.
2778 * Set DELACK for segments received in order, but ack
2779 * immediately when segments are out of order (so
2780 * fast retransmit can work).
2781 */
2782 if (th->th_seq == tp->rcv_nxt &&
2783 LIST_EMPTY(&tp->t_segq) &&
2784 TCPS_HAVEESTABLISHED(tp->t_state)) {
2785 if (DELAY_ACK(tp))
2786 tp->t_flags |= TF_DELACK;
2787 else
2788 tp->t_flags |= TF_ACKNOW;
2789 tp->rcv_nxt += tlen;
2790 thflags = th->th_flags & TH_FIN;
2791 TCPSTAT_INC(tcps_rcvpack);
2792 TCPSTAT_ADD(tcps_rcvbyte, tlen);
2793 ND6_HINT(tp);
2794 SOCKBUF_LOCK(&so->so_rcv);
2795 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
2796 m_freem(m);
2797 else
2798 sbappendstream_locked(&so->so_rcv, m);
2799 /* NB: sorwakeup_locked() does an implicit unlock. */
2800 sorwakeup_locked(so);
2801 } else {
2802 /*
2803 * XXX: Due to the header drop above "th" is
2804 * theoretically invalid by now. Fortunately
2805 * m_adj() doesn't actually frees any mbufs
2806 * when trimming from the head.
2807 */
2808 thflags = tcp_reass(tp, th, &tlen, m);
2809 tp->t_flags |= TF_ACKNOW;
2810 }
2811 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT))
2812 tcp_update_sack_list(tp, save_start, save_start + tlen);
2813 #if 0
2814 /*
2815 * Note the amount of data that peer has sent into
2816 * our window, in order to estimate the sender's
2817 * buffer size.
2818 * XXX: Unused.
2819 */
2820 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
2821 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2822 else
2823 len = so->so_rcv.sb_hiwat;
2824 #endif
2825 } else {
2826 m_freem(m);
2827 thflags &= ~TH_FIN;
2828 }
2829
2830 /*
2831 * If FIN is received ACK the FIN and let the user know
2832 * that the connection is closing.
2833 */
2834 if (thflags & TH_FIN) {
2835 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2836 socantrcvmore(so);
2837 /*
2838 * If connection is half-synchronized
2839 * (ie NEEDSYN flag on) then delay ACK,
2840 * so it may be piggybacked when SYN is sent.
2841 * Otherwise, since we received a FIN then no
2842 * more input can be expected, send ACK now.
2843 */
2844 if (tp->t_flags & TF_NEEDSYN)
2845 tp->t_flags |= TF_DELACK;
2846 else
2847 tp->t_flags |= TF_ACKNOW;
2848 tp->rcv_nxt++;
2849 }
2850 switch (tp->t_state) {
2851
2852 /*
2853 * In SYN_RECEIVED and ESTABLISHED STATES
2854 * enter the CLOSE_WAIT state.
2855 */
2856 case TCPS_SYN_RECEIVED:
2857 tp->t_starttime = ticks;
2858 /* FALLTHROUGH */
2859 case TCPS_ESTABLISHED:
2860 tp->t_state = TCPS_CLOSE_WAIT;
2861 break;
2862
2863 /*
2864 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2865 * enter the CLOSING state.
2866 */
2867 case TCPS_FIN_WAIT_1:
2868 tp->t_state = TCPS_CLOSING;
2869 break;
2870
2871 /*
2872 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2873 * starting the time-wait timer, turning off the other
2874 * standard timers.
2875 */
2876 case TCPS_FIN_WAIT_2:
2877 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2878 KASSERT(ti_locked == TI_WLOCKED, ("%s: dodata "
2879 "TCP_FIN_WAIT_2 ti_locked: %d", __func__,
2880 ti_locked));
2881
2882 tcp_twstart(tp);
2883 INP_INFO_WUNLOCK(&V_tcbinfo);
2884 return;
2885 }
2886 }
2887 if (ti_locked == TI_RLOCKED)
2888 INP_INFO_RUNLOCK(&V_tcbinfo);
2889 else if (ti_locked == TI_WLOCKED)
2890 INP_INFO_WUNLOCK(&V_tcbinfo);
2891 else
2892 panic("%s: dodata epilogue ti_locked %d", __func__,
2893 ti_locked);
2894 ti_locked = TI_UNLOCKED;
2895
2896 #ifdef TCPDEBUG
2897 if (so->so_options & SO_DEBUG)
2898 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
2899 &tcp_savetcp, 0);
2900 #endif
2901
2902 /*
2903 * Return any desired output.
2904 */
2905 if (needoutput || (tp->t_flags & TF_ACKNOW))
2906 (void) tcp_output(tp);
2907
2908 check_delack:
2909 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d",
2910 __func__, ti_locked));
2911 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
2912 INP_WLOCK_ASSERT(tp->t_inpcb);
2913
2914 if (tp->t_flags & TF_DELACK) {
2915 tp->t_flags &= ~TF_DELACK;
2916 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
2917 }
2918 INP_WUNLOCK(tp->t_inpcb);
2919 return;
2920
2921 dropafterack:
2922 KASSERT(ti_locked == TI_RLOCKED || ti_locked == TI_WLOCKED,
2923 ("tcp_do_segment: dropafterack ti_locked %d", ti_locked));
2924
2925 /*
2926 * Generate an ACK dropping incoming segment if it occupies
2927 * sequence space, where the ACK reflects our state.
2928 *
2929 * We can now skip the test for the RST flag since all
2930 * paths to this code happen after packets containing
2931 * RST have been dropped.
2932 *
2933 * In the SYN-RECEIVED state, don't send an ACK unless the
2934 * segment we received passes the SYN-RECEIVED ACK test.
2935 * If it fails send a RST. This breaks the loop in the
2936 * "LAND" DoS attack, and also prevents an ACK storm
2937 * between two listening ports that have been sent forged
2938 * SYN segments, each with the source address of the other.
2939 */
2940 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
2941 (SEQ_GT(tp->snd_una, th->th_ack) ||
2942 SEQ_GT(th->th_ack, tp->snd_max)) ) {
2943 rstreason = BANDLIM_RST_OPENPORT;
2944 goto dropwithreset;
2945 }
2946 #ifdef TCPDEBUG
2947 if (so->so_options & SO_DEBUG)
2948 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2949 &tcp_savetcp, 0);
2950 #endif
2951 if (ti_locked == TI_RLOCKED)
2952 INP_INFO_RUNLOCK(&V_tcbinfo);
2953 else if (ti_locked == TI_WLOCKED)
2954 INP_INFO_WUNLOCK(&V_tcbinfo);
2955 else
2956 panic("%s: dropafterack epilogue ti_locked %d", __func__,
2957 ti_locked);
2958 ti_locked = TI_UNLOCKED;
2959
2960 tp->t_flags |= TF_ACKNOW;
2961 (void) tcp_output(tp);
2962 INP_WUNLOCK(tp->t_inpcb);
2963 m_freem(m);
2964 return;
2965
2966 dropwithreset:
2967 if (ti_locked == TI_RLOCKED)
2968 INP_INFO_RUNLOCK(&V_tcbinfo);
2969 else if (ti_locked == TI_WLOCKED)
2970 INP_INFO_WUNLOCK(&V_tcbinfo);
2971 else
2972 panic("%s: dropwithreset ti_locked %d", __func__, ti_locked);
2973 ti_locked = TI_UNLOCKED;
2974
2975 if (tp != NULL) {
2976 tcp_dropwithreset(m, th, tp, tlen, rstreason);
2977 INP_WUNLOCK(tp->t_inpcb);
2978 } else
2979 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
2980 return;
2981
2982 drop:
2983 if (ti_locked == TI_RLOCKED)
2984 INP_INFO_RUNLOCK(&V_tcbinfo);
2985 else if (ti_locked == TI_WLOCKED)
2986 INP_INFO_WUNLOCK(&V_tcbinfo);
2987 #ifdef INVARIANTS
2988 else
2989 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
2990 #endif
2991 ti_locked = TI_UNLOCKED;
2992
2993 /*
2994 * Drop space held by incoming segment and return.
2995 */
2996 #ifdef TCPDEBUG
2997 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2998 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2999 &tcp_savetcp, 0);
3000 #endif
3001 if (tp != NULL)
3002 INP_WUNLOCK(tp->t_inpcb);
3003 m_freem(m);
3004 }
3005
3006 /*
3007 * Issue RST and make ACK acceptable to originator of segment.
3008 * The mbuf must still include the original packet header.
3009 * tp may be NULL.
3010 */
3011 static void
3012 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
3013 int tlen, int rstreason)
3014 {
3015 struct ip *ip;
3016 #ifdef INET6
3017 struct ip6_hdr *ip6;
3018 #endif
3019
3020 if (tp != NULL) {
3021 INP_WLOCK_ASSERT(tp->t_inpcb);
3022 }
3023
3024 /* Don't bother if destination was broadcast/multicast. */
3025 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
3026 goto drop;
3027 #ifdef INET6
3028 if (mtod(m, struct ip *)->ip_v == 6) {
3029 ip6 = mtod(m, struct ip6_hdr *);
3030 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3031 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3032 goto drop;
3033 /* IPv6 anycast check is done at tcp6_input() */
3034 } else
3035 #endif
3036 {
3037 ip = mtod(m, struct ip *);
3038 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3039 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3040 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3041 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3042 goto drop;
3043 }
3044
3045 /* Perform bandwidth limiting. */
3046 if (badport_bandlim(rstreason) < 0)
3047 goto drop;
3048
3049 /* tcp_respond consumes the mbuf chain. */
3050 if (th->th_flags & TH_ACK) {
3051 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0,
3052 th->th_ack, TH_RST);
3053 } else {
3054 if (th->th_flags & TH_SYN)
3055 tlen++;
3056 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
3057 (tcp_seq)0, TH_RST|TH_ACK);
3058 }
3059 return;
3060 drop:
3061 m_freem(m);
3062 }
3063
3064 /*
3065 * Parse TCP options and place in tcpopt.
3066 */
3067 static void
3068 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags)
3069 {
3070 int opt, optlen;
3071
3072 to->to_flags = 0;
3073 for (; cnt > 0; cnt -= optlen, cp += optlen) {
3074 opt = cp[0];
3075 if (opt == TCPOPT_EOL)
3076 break;
3077 if (opt == TCPOPT_NOP)
3078 optlen = 1;
3079 else {
3080 if (cnt < 2)
3081 break;
3082 optlen = cp[1];
3083 if (optlen < 2 || optlen > cnt)
3084 break;
3085 }
3086 switch (opt) {
3087 case TCPOPT_MAXSEG:
3088 if (optlen != TCPOLEN_MAXSEG)
3089 continue;
3090 if (!(flags & TO_SYN))
3091 continue;
3092 to->to_flags |= TOF_MSS;
3093 bcopy((char *)cp + 2,
3094 (char *)&to->to_mss, sizeof(to->to_mss));
3095 to->to_mss = ntohs(to->to_mss);
3096 break;
3097 case TCPOPT_WINDOW:
3098 if (optlen != TCPOLEN_WINDOW)
3099 continue;
3100 if (!(flags & TO_SYN))
3101 continue;
3102 to->to_flags |= TOF_SCALE;
3103 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT);
3104 break;
3105 case TCPOPT_TIMESTAMP:
3106 if (optlen != TCPOLEN_TIMESTAMP)
3107 continue;
3108 to->to_flags |= TOF_TS;
3109 bcopy((char *)cp + 2,
3110 (char *)&to->to_tsval, sizeof(to->to_tsval));
3111 to->to_tsval = ntohl(to->to_tsval);
3112 bcopy((char *)cp + 6,
3113 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
3114 to->to_tsecr = ntohl(to->to_tsecr);
3115 break;
3116 #ifdef TCP_SIGNATURE
3117 /*
3118 * XXX In order to reply to a host which has set the
3119 * TCP_SIGNATURE option in its initial SYN, we have to
3120 * record the fact that the option was observed here
3121 * for the syncache code to perform the correct response.
3122 */
3123 case TCPOPT_SIGNATURE:
3124 if (optlen != TCPOLEN_SIGNATURE)
3125 continue;
3126 to->to_flags |= TOF_SIGNATURE;
3127 to->to_signature = cp + 2;
3128 break;
3129 #endif
3130 case TCPOPT_SACK_PERMITTED:
3131 if (optlen != TCPOLEN_SACK_PERMITTED)
3132 continue;
3133 if (!(flags & TO_SYN))
3134 continue;
3135 if (!V_tcp_do_sack)
3136 continue;
3137 to->to_flags |= TOF_SACKPERM;
3138 break;
3139 case TCPOPT_SACK:
3140 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
3141 continue;
3142 if (flags & TO_SYN)
3143 continue;
3144 to->to_flags |= TOF_SACK;
3145 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
3146 to->to_sacks = cp + 2;
3147 TCPSTAT_INC(tcps_sack_rcv_blocks);
3148 break;
3149 default:
3150 continue;
3151 }
3152 }
3153 }
3154
3155 /*
3156 * Pull out of band byte out of a segment so
3157 * it doesn't appear in the user's data queue.
3158 * It is still reflected in the segment length for
3159 * sequencing purposes.
3160 */
3161 static void
3162 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m,
3163 int off)
3164 {
3165 int cnt = off + th->th_urp - 1;
3166
3167 while (cnt >= 0) {
3168 if (m->m_len > cnt) {
3169 char *cp = mtod(m, caddr_t) + cnt;
3170 struct tcpcb *tp = sototcpcb(so);
3171
3172 INP_WLOCK_ASSERT(tp->t_inpcb);
3173
3174 tp->t_iobc = *cp;
3175 tp->t_oobflags |= TCPOOB_HAVEDATA;
3176 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3177 m->m_len--;
3178 if (m->m_flags & M_PKTHDR)
3179 m->m_pkthdr.len--;
3180 return;
3181 }
3182 cnt -= m->m_len;
3183 m = m->m_next;
3184 if (m == NULL)
3185 break;
3186 }
3187 panic("tcp_pulloutofband");
3188 }
3189
3190 /*
3191 * Collect new round-trip time estimate
3192 * and update averages and current timeout.
3193 */
3194 static void
3195 tcp_xmit_timer(struct tcpcb *tp, int rtt)
3196 {
3197 int delta;
3198
3199 INP_WLOCK_ASSERT(tp->t_inpcb);
3200
3201 TCPSTAT_INC(tcps_rttupdated);
3202 tp->t_rttupdated++;
3203 if (tp->t_srtt != 0) {
3204 /*
3205 * srtt is stored as fixed point with 5 bits after the
3206 * binary point (i.e., scaled by 8). The following magic
3207 * is equivalent to the smoothing algorithm in rfc793 with
3208 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3209 * point). Adjust rtt to origin 0.
3210 */
3211 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
3212 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
3213
3214 if ((tp->t_srtt += delta) <= 0)
3215 tp->t_srtt = 1;
3216
3217 /*
3218 * We accumulate a smoothed rtt variance (actually, a
3219 * smoothed mean difference), then set the retransmit
3220 * timer to smoothed rtt + 4 times the smoothed variance.
3221 * rttvar is stored as fixed point with 4 bits after the
3222 * binary point (scaled by 16). The following is
3223 * equivalent to rfc793 smoothing with an alpha of .75
3224 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
3225 * rfc793's wired-in beta.
3226 */
3227 if (delta < 0)
3228 delta = -delta;
3229 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
3230 if ((tp->t_rttvar += delta) <= 0)
3231 tp->t_rttvar = 1;
3232 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
3233 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3234 } else {
3235 /*
3236 * No rtt measurement yet - use the unsmoothed rtt.
3237 * Set the variance to half the rtt (so our first
3238 * retransmit happens at 3*rtt).
3239 */
3240 tp->t_srtt = rtt << TCP_RTT_SHIFT;
3241 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
3242 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3243 }
3244 tp->t_rtttime = 0;
3245 tp->t_rxtshift = 0;
3246
3247 /*
3248 * the retransmit should happen at rtt + 4 * rttvar.
3249 * Because of the way we do the smoothing, srtt and rttvar
3250 * will each average +1/2 tick of bias. When we compute
3251 * the retransmit timer, we want 1/2 tick of rounding and
3252 * 1 extra tick because of +-1/2 tick uncertainty in the
3253 * firing of the timer. The bias will give us exactly the
3254 * 1.5 tick we need. But, because the bias is
3255 * statistical, we have to test that we don't drop below
3256 * the minimum feasible timer (which is 2 ticks).
3257 */
3258 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3259 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3260
3261 /*
3262 * We received an ack for a packet that wasn't retransmitted;
3263 * it is probably safe to discard any error indications we've
3264 * received recently. This isn't quite right, but close enough
3265 * for now (a route might have failed after we sent a segment,
3266 * and the return path might not be symmetrical).
3267 */
3268 tp->t_softerror = 0;
3269 }
3270
3271 /*
3272 * Determine a reasonable value for maxseg size.
3273 * If the route is known, check route for mtu.
3274 * If none, use an mss that can be handled on the outgoing
3275 * interface without forcing IP to fragment; if bigger than
3276 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
3277 * to utilize large mbufs. If no route is found, route has no mtu,
3278 * or the destination isn't local, use a default, hopefully conservative
3279 * size (usually 512 or the default IP max size, but no more than the mtu
3280 * of the interface), as we can't discover anything about intervening
3281 * gateways or networks. We also initialize the congestion/slow start
3282 * window to be a single segment if the destination isn't local.
3283 * While looking at the routing entry, we also initialize other path-dependent
3284 * parameters from pre-set or cached values in the routing entry.
3285 *
3286 * Also take into account the space needed for options that we
3287 * send regularly. Make maxseg shorter by that amount to assure
3288 * that we can send maxseg amount of data even when the options
3289 * are present. Store the upper limit of the length of options plus
3290 * data in maxopd.
3291 *
3292 * In case of T/TCP, we call this routine during implicit connection
3293 * setup as well (offer = -1), to initialize maxseg from the cached
3294 * MSS of our peer.
3295 *
3296 * NOTE that this routine is only called when we process an incoming
3297 * segment. Outgoing SYN/ACK MSS settings are handled in tcp_mssopt().
3298 */
3299 void
3300 tcp_mss_update(struct tcpcb *tp, int offer,
3301 struct hc_metrics_lite *metricptr, int *mtuflags)
3302 {
3303 int mss;
3304 u_long maxmtu;
3305 struct inpcb *inp = tp->t_inpcb;
3306 struct hc_metrics_lite metrics;
3307 int origoffer = offer;
3308 #ifdef INET6
3309 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
3310 size_t min_protoh = isipv6 ?
3311 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
3312 sizeof (struct tcpiphdr);
3313 #else
3314 const size_t min_protoh = sizeof(struct tcpiphdr);
3315 #endif
3316
3317 INP_WLOCK_ASSERT(tp->t_inpcb);
3318
3319 /* Initialize. */
3320 #ifdef INET6
3321 if (isipv6) {
3322 maxmtu = tcp_maxmtu6(&inp->inp_inc, mtuflags);
3323 tp->t_maxopd = tp->t_maxseg = V_tcp_v6mssdflt;
3324 } else
3325 #endif
3326 {
3327 maxmtu = tcp_maxmtu(&inp->inp_inc, mtuflags);
3328 tp->t_maxopd = tp->t_maxseg = V_tcp_mssdflt;
3329 }
3330
3331 /*
3332 * No route to sender, stay with default mss and return.
3333 */
3334 if (maxmtu == 0) {
3335 /*
3336 * In case we return early we need to initialize metrics
3337 * to a defined state as tcp_hc_get() would do for us
3338 * if there was no cache hit.
3339 */
3340 if (metricptr != NULL)
3341 bzero(metricptr, sizeof(struct hc_metrics_lite));
3342 return;
3343 }
3344
3345 /* What have we got? */
3346 switch (offer) {
3347 case 0:
3348 /*
3349 * Offer == 0 means that there was no MSS on the SYN
3350 * segment, in this case we use tcp_mssdflt as
3351 * already assigned to t_maxopd above.
3352 */
3353 offer = tp->t_maxopd;
3354 break;
3355
3356 case -1:
3357 /*
3358 * Offer == -1 means that we didn't receive SYN yet.
3359 */
3360 /* FALLTHROUGH */
3361
3362 default:
3363 /*
3364 * Prevent DoS attack with too small MSS. Round up
3365 * to at least minmss.
3366 */
3367 offer = max(offer, V_tcp_minmss);
3368 }
3369
3370 /*
3371 * rmx information is now retrieved from tcp_hostcache.
3372 */
3373 tcp_hc_get(&inp->inp_inc, &metrics);
3374 if (metricptr != NULL)
3375 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite));
3376
3377 /*
3378 * If there's a discovered mtu int tcp hostcache, use it
3379 * else, use the link mtu.
3380 */
3381 if (metrics.rmx_mtu)
3382 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
3383 else {
3384 #ifdef INET6
3385 if (isipv6) {
3386 mss = maxmtu - min_protoh;
3387 if (!V_path_mtu_discovery &&
3388 !in6_localaddr(&inp->in6p_faddr))
3389 mss = min(mss, V_tcp_v6mssdflt);
3390 } else
3391 #endif
3392 {
3393 mss = maxmtu - min_protoh;
3394 if (!V_path_mtu_discovery &&
3395 !in_localaddr(inp->inp_faddr))
3396 mss = min(mss, V_tcp_mssdflt);
3397 }
3398 /*
3399 * XXX - The above conditional (mss = maxmtu - min_protoh)
3400 * probably violates the TCP spec.
3401 * The problem is that, since we don't know the
3402 * other end's MSS, we are supposed to use a conservative
3403 * default. But, if we do that, then MTU discovery will
3404 * never actually take place, because the conservative
3405 * default is much less than the MTUs typically seen
3406 * on the Internet today. For the moment, we'll sweep
3407 * this under the carpet.
3408 *
3409 * The conservative default might not actually be a problem
3410 * if the only case this occurs is when sending an initial
3411 * SYN with options and data to a host we've never talked
3412 * to before. Then, they will reply with an MSS value which
3413 * will get recorded and the new parameters should get
3414 * recomputed. For Further Study.
3415 */
3416 }
3417 mss = min(mss, offer);
3418
3419 /*
3420 * Sanity check: make sure that maxopd will be large
3421 * enough to allow some data on segments even if the
3422 * all the option space is used (40bytes). Otherwise
3423 * funny things may happen in tcp_output.
3424 */
3425 mss = max(mss, 64);
3426
3427 /*
3428 * maxopd stores the maximum length of data AND options
3429 * in a segment; maxseg is the amount of data in a normal
3430 * segment. We need to store this value (maxopd) apart
3431 * from maxseg, because now every segment carries options
3432 * and thus we normally have somewhat less data in segments.
3433 */
3434 tp->t_maxopd = mss;
3435
3436 /*
3437 * origoffer==-1 indicates that no segments were received yet.
3438 * In this case we just guess.
3439 */
3440 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
3441 (origoffer == -1 ||
3442 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
3443 mss -= TCPOLEN_TSTAMP_APPA;
3444
3445 #if (MCLBYTES & (MCLBYTES - 1)) == 0
3446 if (mss > MCLBYTES)
3447 mss &= ~(MCLBYTES-1);
3448 #else
3449 if (mss > MCLBYTES)
3450 mss = mss / MCLBYTES * MCLBYTES;
3451 #endif
3452 tp->t_maxseg = mss;
3453 }
3454
3455 void
3456 tcp_mss(struct tcpcb *tp, int offer)
3457 {
3458 int mss;
3459 u_long bufsize;
3460 struct inpcb *inp;
3461 struct socket *so;
3462 struct hc_metrics_lite metrics;
3463 int mtuflags = 0;
3464
3465 KASSERT(tp != NULL, ("%s: tp == NULL", __func__));
3466
3467 tcp_mss_update(tp, offer, &metrics, &mtuflags);
3468
3469 mss = tp->t_maxseg;
3470 inp = tp->t_inpcb;
3471
3472 /*
3473 * If there's a pipesize, change the socket buffer to that size,
3474 * don't change if sb_hiwat is different than default (then it
3475 * has been changed on purpose with setsockopt).
3476 * Make the socket buffers an integral number of mss units;
3477 * if the mss is larger than the socket buffer, decrease the mss.
3478 */
3479 so = inp->inp_socket;
3480 SOCKBUF_LOCK(&so->so_snd);
3481 if ((so->so_snd.sb_hiwat == tcp_sendspace) && metrics.rmx_sendpipe)
3482 bufsize = metrics.rmx_sendpipe;
3483 else
3484 bufsize = so->so_snd.sb_hiwat;
3485 if (bufsize < mss)
3486 mss = bufsize;
3487 else {
3488 bufsize = roundup(bufsize, mss);
3489 if (bufsize > sb_max)
3490 bufsize = sb_max;
3491 if (bufsize > so->so_snd.sb_hiwat)
3492 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
3493 }
3494 SOCKBUF_UNLOCK(&so->so_snd);
3495 tp->t_maxseg = mss;
3496
3497 SOCKBUF_LOCK(&so->so_rcv);
3498 if ((so->so_rcv.sb_hiwat == tcp_recvspace) && metrics.rmx_recvpipe)
3499 bufsize = metrics.rmx_recvpipe;
3500 else
3501 bufsize = so->so_rcv.sb_hiwat;
3502 if (bufsize > mss) {
3503 bufsize = roundup(bufsize, mss);
3504 if (bufsize > sb_max)
3505 bufsize = sb_max;
3506 if (bufsize > so->so_rcv.sb_hiwat)
3507 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
3508 }
3509 SOCKBUF_UNLOCK(&so->so_rcv);
3510
3511 /* Check the interface for TSO capabilities. */
3512 if (mtuflags & CSUM_TSO)
3513 tp->t_flags |= TF_TSO;
3514 }
3515
3516 /*
3517 * Determine the MSS option to send on an outgoing SYN.
3518 */
3519 int
3520 tcp_mssopt(struct in_conninfo *inc)
3521 {
3522 int mss = 0;
3523 u_long maxmtu = 0;
3524 u_long thcmtu = 0;
3525 size_t min_protoh;
3526
3527 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3528
3529 #ifdef INET6
3530 if (inc->inc_flags & INC_ISIPV6) {
3531 mss = V_tcp_v6mssdflt;
3532 maxmtu = tcp_maxmtu6(inc, NULL);
3533 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3534 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3535 } else
3536 #endif
3537 {
3538 mss = V_tcp_mssdflt;
3539 maxmtu = tcp_maxmtu(inc, NULL);
3540 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3541 min_protoh = sizeof(struct tcpiphdr);
3542 }
3543 if (maxmtu && thcmtu)
3544 mss = min(maxmtu, thcmtu) - min_protoh;
3545 else if (maxmtu || thcmtu)
3546 mss = max(maxmtu, thcmtu) - min_protoh;
3547
3548 return (mss);
3549 }
3550
3551
3552 /*
3553 * On a partial ack arrives, force the retransmission of the
3554 * next unacknowledged segment. Do not clear tp->t_dupacks.
3555 * By setting snd_nxt to ti_ack, this forces retransmission timer to
3556 * be started again.
3557 */
3558 static void
3559 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th)
3560 {
3561 tcp_seq onxt = tp->snd_nxt;
3562 u_long ocwnd = tp->snd_cwnd;
3563
3564 INP_WLOCK_ASSERT(tp->t_inpcb);
3565
3566 tcp_timer_activate(tp, TT_REXMT, 0);
3567 tp->t_rtttime = 0;
3568 tp->snd_nxt = th->th_ack;
3569 /*
3570 * Set snd_cwnd to one segment beyond acknowledged offset.
3571 * (tp->snd_una has not yet been updated when this function is called.)
3572 */
3573 tp->snd_cwnd = tp->t_maxseg + BYTES_THIS_ACK(tp, th);
3574 tp->t_flags |= TF_ACKNOW;
3575 (void) tcp_output(tp);
3576 tp->snd_cwnd = ocwnd;
3577 if (SEQ_GT(onxt, tp->snd_nxt))
3578 tp->snd_nxt = onxt;
3579 /*
3580 * Partial window deflation. Relies on fact that tp->snd_una
3581 * not updated yet.
3582 */
3583 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th))
3584 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th);
3585 else
3586 tp->snd_cwnd = 0;
3587 tp->snd_cwnd += tp->t_maxseg;
3588 }
Cache object: 0ae51a58042034daae2a3a501fbd25fa
|