1/* $NetBSD: tcp_subr.c,v 1.267 2016/11/09 03:33:30 ozaki-r Exp $ */
2
3/*
4 * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 3. Neither the name of the project nor the names of its contributors
16 * may be used to endorse or promote products derived from this software
17 * without specific prior written permission.
18 *
19 * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29 * SUCH DAMAGE.
30 */
31
32/*-
33 * Copyright (c) 1997, 1998, 2000, 2001, 2008 The NetBSD Foundation, Inc.
34 * All rights reserved.
35 *
36 * This code is derived from software contributed to The NetBSD Foundation
37 * by Jason R. Thorpe and Kevin M. Lahey of the Numerical Aerospace Simulation
38 * Facility, NASA Ames Research Center.
39 *
40 * Redistribution and use in source and binary forms, with or without
41 * modification, are permitted provided that the following conditions
42 * are met:
43 * 1. Redistributions of source code must retain the above copyright
44 * notice, this list of conditions and the following disclaimer.
45 * 2. Redistributions in binary form must reproduce the above copyright
46 * notice, this list of conditions and the following disclaimer in the
47 * documentation and/or other materials provided with the distribution.
48 *
49 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
50 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
51 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
52 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
53 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
54 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
55 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
56 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
57 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
58 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
59 * POSSIBILITY OF SUCH DAMAGE.
60 */
61
62/*
63 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
64 * The Regents of the University of California. All rights reserved.
65 *
66 * Redistribution and use in source and binary forms, with or without
67 * modification, are permitted provided that the following conditions
68 * are met:
69 * 1. Redistributions of source code must retain the above copyright
70 * notice, this list of conditions and the following disclaimer.
71 * 2. Redistributions in binary form must reproduce the above copyright
72 * notice, this list of conditions and the following disclaimer in the
73 * documentation and/or other materials provided with the distribution.
74 * 3. Neither the name of the University nor the names of its contributors
75 * may be used to endorse or promote products derived from this software
76 * without specific prior written permission.
77 *
78 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
79 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
80 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
81 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
82 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
83 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
84 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
85 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
86 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
87 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
88 * SUCH DAMAGE.
89 *
90 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95
91 */
92
93#include <sys/cdefs.h>
94__KERNEL_RCSID(0, "$NetBSD: tcp_subr.c,v 1.267 2016/11/09 03:33:30 ozaki-r Exp $");
95
96#ifdef _KERNEL_OPT
97#include "opt_inet.h"
98#include "opt_ipsec.h"
99#include "opt_tcp_compat_42.h"
100#include "opt_inet_csum.h"
101#include "opt_mbuftrace.h"
102#endif
103
104#include <sys/param.h>
105#include <sys/atomic.h>
106#include <sys/proc.h>
107#include <sys/systm.h>
108#include <sys/mbuf.h>
109#include <sys/once.h>
110#include <sys/socket.h>
111#include <sys/socketvar.h>
112#include <sys/protosw.h>
113#include <sys/errno.h>
114#include <sys/kernel.h>
115#include <sys/pool.h>
116#include <sys/md5.h>
117#include <sys/cprng.h>
118
119#include <net/route.h>
120#include <net/if.h>
121
122#include <netinet/in.h>
123#include <netinet/in_systm.h>
124#include <netinet/ip.h>
125#include <netinet/in_pcb.h>
126#include <netinet/ip_var.h>
127#include <netinet/ip_icmp.h>
128
129#ifdef INET6
130#ifndef INET
131#include <netinet/in.h>
132#endif
133#include <netinet/ip6.h>
134#include <netinet6/in6_pcb.h>
135#include <netinet6/ip6_var.h>
136#include <netinet6/in6_var.h>
137#include <netinet6/ip6protosw.h>
138#include <netinet/icmp6.h>
139#include <netinet6/nd6.h>
140#endif
141
142#include <netinet/tcp.h>
143#include <netinet/tcp_fsm.h>
144#include <netinet/tcp_seq.h>
145#include <netinet/tcp_timer.h>
146#include <netinet/tcp_var.h>
147#include <netinet/tcp_vtw.h>
148#include <netinet/tcp_private.h>
149#include <netinet/tcp_congctl.h>
150#include <netinet/tcpip.h>
151
152#ifdef IPSEC
153#include <netipsec/ipsec.h>
154#include <netipsec/xform.h>
155#ifdef INET6
156#include <netipsec/ipsec6.h>
157#endif
158 #include <netipsec/key.h>
159#endif /* IPSEC*/
160
161
162struct inpcbtable tcbtable; /* head of queue of active tcpcb's */
163u_int32_t tcp_now; /* slow ticks, for RFC 1323 timestamps */
164
165percpu_t *tcpstat_percpu;
166
167/* patchable/settable parameters for tcp */
168int tcp_mssdflt = TCP_MSS;
169int tcp_minmss = TCP_MINMSS;
170int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
171int tcp_do_rfc1323 = 1; /* window scaling / timestamps (obsolete) */
172int tcp_do_rfc1948 = 0; /* ISS by cryptographic hash */
173int tcp_do_sack = 1; /* selective acknowledgement */
174int tcp_do_win_scale = 1; /* RFC1323 window scaling */
175int tcp_do_timestamps = 1; /* RFC1323 timestamps */
176int tcp_ack_on_push = 0; /* set to enable immediate ACK-on-PUSH */
177int tcp_do_ecn = 0; /* Explicit Congestion Notification */
178#ifndef TCP_INIT_WIN
179#define TCP_INIT_WIN 4 /* initial slow start window */
180#endif
181#ifndef TCP_INIT_WIN_LOCAL
182#define TCP_INIT_WIN_LOCAL 4 /* initial slow start window for local nets */
183#endif
184/*
185 * Up to 5 we scale linearly, to reach 3 * 1460; then (iw) * 1460.
186 * This is to simulate current behavior for iw == 4
187 */
188int tcp_init_win_max[] = {
189 1 * 1460,
190 1 * 1460,
191 2 * 1460,
192 2 * 1460,
193 3 * 1460,
194 5 * 1460,
195 6 * 1460,
196 7 * 1460,
197 8 * 1460,
198 9 * 1460,
199 10 * 1460
200};
201int tcp_init_win = TCP_INIT_WIN;
202int tcp_init_win_local = TCP_INIT_WIN_LOCAL;
203int tcp_mss_ifmtu = 0;
204#ifdef TCP_COMPAT_42
205int tcp_compat_42 = 1;
206#else
207int tcp_compat_42 = 0;
208#endif
209int tcp_rst_ppslim = 100; /* 100pps */
210int tcp_ackdrop_ppslim = 100; /* 100pps */
211int tcp_do_loopback_cksum = 0;
212int tcp_do_abc = 1; /* RFC3465 Appropriate byte counting. */
213int tcp_abc_aggressive = 1; /* 1: L=2*SMSS 0: L=1*SMSS */
214int tcp_sack_tp_maxholes = 32;
215int tcp_sack_globalmaxholes = 1024;
216int tcp_sack_globalholes = 0;
217int tcp_ecn_maxretries = 1;
218int tcp_msl_enable = 1; /* enable TIME_WAIT truncation */
219int tcp_msl_loop = PR_SLOWHZ; /* MSL for loopback */
220int tcp_msl_local = 5 * PR_SLOWHZ; /* MSL for 'local' */
221int tcp_msl_remote = TCPTV_MSL; /* MSL otherwise */
222int tcp_msl_remote_threshold = TCPTV_SRTTDFLT; /* RTT threshold */
223int tcp_rttlocal = 0; /* Use RTT to decide who's 'local' */
224
225int tcp4_vtw_enable = 0; /* 1 to enable */
226int tcp6_vtw_enable = 0; /* 1 to enable */
227int tcp_vtw_was_enabled = 0;
228int tcp_vtw_entries = 1 << 4; /* 16 vestigial TIME_WAIT entries */
229
230/* tcb hash */
231#ifndef TCBHASHSIZE
232#define TCBHASHSIZE 128
233#endif
234int tcbhashsize = TCBHASHSIZE;
235
236/* syn hash parameters */
237#define TCP_SYN_HASH_SIZE 293
238#define TCP_SYN_BUCKET_SIZE 35
239int tcp_syn_cache_size = TCP_SYN_HASH_SIZE;
240int tcp_syn_cache_limit = TCP_SYN_HASH_SIZE*TCP_SYN_BUCKET_SIZE;
241int tcp_syn_bucket_limit = 3*TCP_SYN_BUCKET_SIZE;
242struct syn_cache_head tcp_syn_cache[TCP_SYN_HASH_SIZE];
243
244int tcp_freeq(struct tcpcb *);
245static int tcp_iss_secret_init(void);
246
247#ifdef INET
248static void tcp_mtudisc_callback(struct in_addr);
249#endif
250
251#ifdef INET6
252void tcp6_mtudisc(struct in6pcb *, int);
253#endif
254
255static struct pool tcpcb_pool;
256
257static int tcp_drainwanted;
258
259#ifdef TCP_CSUM_COUNTERS
260#include <sys/device.h>
261
262#if defined(INET)
263struct evcnt tcp_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
264 NULL, "tcp", "hwcsum bad");
265struct evcnt tcp_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
266 NULL, "tcp", "hwcsum ok");
267struct evcnt tcp_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
268 NULL, "tcp", "hwcsum data");
269struct evcnt tcp_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
270 NULL, "tcp", "swcsum");
271
272EVCNT_ATTACH_STATIC(tcp_hwcsum_bad);
273EVCNT_ATTACH_STATIC(tcp_hwcsum_ok);
274EVCNT_ATTACH_STATIC(tcp_hwcsum_data);
275EVCNT_ATTACH_STATIC(tcp_swcsum);
276#endif /* defined(INET) */
277
278#if defined(INET6)
279struct evcnt tcp6_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
280 NULL, "tcp6", "hwcsum bad");
281struct evcnt tcp6_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
282 NULL, "tcp6", "hwcsum ok");
283struct evcnt tcp6_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
284 NULL, "tcp6", "hwcsum data");
285struct evcnt tcp6_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
286 NULL, "tcp6", "swcsum");
287
288EVCNT_ATTACH_STATIC(tcp6_hwcsum_bad);
289EVCNT_ATTACH_STATIC(tcp6_hwcsum_ok);
290EVCNT_ATTACH_STATIC(tcp6_hwcsum_data);
291EVCNT_ATTACH_STATIC(tcp6_swcsum);
292#endif /* defined(INET6) */
293#endif /* TCP_CSUM_COUNTERS */
294
295
296#ifdef TCP_OUTPUT_COUNTERS
297#include <sys/device.h>
298
299struct evcnt tcp_output_bigheader = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
300 NULL, "tcp", "output big header");
301struct evcnt tcp_output_predict_hit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
302 NULL, "tcp", "output predict hit");
303struct evcnt tcp_output_predict_miss = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
304 NULL, "tcp", "output predict miss");
305struct evcnt tcp_output_copysmall = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
306 NULL, "tcp", "output copy small");
307struct evcnt tcp_output_copybig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
308 NULL, "tcp", "output copy big");
309struct evcnt tcp_output_refbig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
310 NULL, "tcp", "output reference big");
311
312EVCNT_ATTACH_STATIC(tcp_output_bigheader);
313EVCNT_ATTACH_STATIC(tcp_output_predict_hit);
314EVCNT_ATTACH_STATIC(tcp_output_predict_miss);
315EVCNT_ATTACH_STATIC(tcp_output_copysmall);
316EVCNT_ATTACH_STATIC(tcp_output_copybig);
317EVCNT_ATTACH_STATIC(tcp_output_refbig);
318
319#endif /* TCP_OUTPUT_COUNTERS */
320
321#ifdef TCP_REASS_COUNTERS
322#include <sys/device.h>
323
324struct evcnt tcp_reass_ = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
325 NULL, "tcp_reass", "calls");
326struct evcnt tcp_reass_empty = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
327 &tcp_reass_, "tcp_reass", "insert into empty queue");
328struct evcnt tcp_reass_iteration[8] = {
329 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", ">7 iterations"),
330 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "1 iteration"),
331 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "2 iterations"),
332 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "3 iterations"),
333 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "4 iterations"),
334 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "5 iterations"),
335 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "6 iterations"),
336 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "7 iterations"),
337};
338struct evcnt tcp_reass_prependfirst = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
339 &tcp_reass_, "tcp_reass", "prepend to first");
340struct evcnt tcp_reass_prepend = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
341 &tcp_reass_, "tcp_reass", "prepend");
342struct evcnt tcp_reass_insert = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
343 &tcp_reass_, "tcp_reass", "insert");
344struct evcnt tcp_reass_inserttail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
345 &tcp_reass_, "tcp_reass", "insert at tail");
346struct evcnt tcp_reass_append = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
347 &tcp_reass_, "tcp_reass", "append");
348struct evcnt tcp_reass_appendtail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
349 &tcp_reass_, "tcp_reass", "append to tail fragment");
350struct evcnt tcp_reass_overlaptail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
351 &tcp_reass_, "tcp_reass", "overlap at end");
352struct evcnt tcp_reass_overlapfront = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
353 &tcp_reass_, "tcp_reass", "overlap at start");
354struct evcnt tcp_reass_segdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
355 &tcp_reass_, "tcp_reass", "duplicate segment");
356struct evcnt tcp_reass_fragdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
357 &tcp_reass_, "tcp_reass", "duplicate fragment");
358
359EVCNT_ATTACH_STATIC(tcp_reass_);
360EVCNT_ATTACH_STATIC(tcp_reass_empty);
361EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 0);
362EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 1);
363EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 2);
364EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 3);
365EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 4);
366EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 5);
367EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 6);
368EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 7);
369EVCNT_ATTACH_STATIC(tcp_reass_prependfirst);
370EVCNT_ATTACH_STATIC(tcp_reass_prepend);
371EVCNT_ATTACH_STATIC(tcp_reass_insert);
372EVCNT_ATTACH_STATIC(tcp_reass_inserttail);
373EVCNT_ATTACH_STATIC(tcp_reass_append);
374EVCNT_ATTACH_STATIC(tcp_reass_appendtail);
375EVCNT_ATTACH_STATIC(tcp_reass_overlaptail);
376EVCNT_ATTACH_STATIC(tcp_reass_overlapfront);
377EVCNT_ATTACH_STATIC(tcp_reass_segdup);
378EVCNT_ATTACH_STATIC(tcp_reass_fragdup);
379
380#endif /* TCP_REASS_COUNTERS */
381
382#ifdef MBUFTRACE
383struct mowner tcp_mowner = MOWNER_INIT("tcp", "");
384struct mowner tcp_rx_mowner = MOWNER_INIT("tcp", "rx");
385struct mowner tcp_tx_mowner = MOWNER_INIT("tcp", "tx");
386struct mowner tcp_sock_mowner = MOWNER_INIT("tcp", "sock");
387struct mowner tcp_sock_rx_mowner = MOWNER_INIT("tcp", "sock rx");
388struct mowner tcp_sock_tx_mowner = MOWNER_INIT("tcp", "sock tx");
389#endif
390
391callout_t tcp_slowtimo_ch;
392
393static int
394do_tcpinit(void)
395{
396
397 in_pcbinit(&tcbtable, tcbhashsize, tcbhashsize);
398 pool_init(&tcpcb_pool, sizeof(struct tcpcb), 0, 0, 0, "tcpcbpl",
399 NULL, IPL_SOFTNET);
400
401 tcp_usrreq_init();
402
403 /* Initialize timer state. */
404 tcp_timer_init();
405
406 /* Initialize the compressed state engine. */
407 syn_cache_init();
408
409 /* Initialize the congestion control algorithms. */
410 tcp_congctl_init();
411
412 /* Initialize the TCPCB template. */
413 tcp_tcpcb_template();
414
415 /* Initialize reassembly queue */
416 tcpipqent_init();
417
418 /* SACK */
419 tcp_sack_init();
420
421 MOWNER_ATTACH(&tcp_tx_mowner);
422 MOWNER_ATTACH(&tcp_rx_mowner);
423 MOWNER_ATTACH(&tcp_reass_mowner);
424 MOWNER_ATTACH(&tcp_sock_mowner);
425 MOWNER_ATTACH(&tcp_sock_tx_mowner);
426 MOWNER_ATTACH(&tcp_sock_rx_mowner);
427 MOWNER_ATTACH(&tcp_mowner);
428
429 tcpstat_percpu = percpu_alloc(sizeof(uint64_t) * TCP_NSTATS);
430
431 vtw_earlyinit();
432
433 callout_init(&tcp_slowtimo_ch, CALLOUT_MPSAFE);
434 callout_reset(&tcp_slowtimo_ch, 1, tcp_slowtimo, NULL);
435
436 return 0;
437}
438
439void
440tcp_init_common(unsigned basehlen)
441{
442 static ONCE_DECL(dotcpinit);
443 unsigned hlen = basehlen + sizeof(struct tcphdr);
444 unsigned oldhlen;
445
446 if (max_linkhdr + hlen > MHLEN)
447 panic("tcp_init");
448 while ((oldhlen = max_protohdr) < hlen)
449 atomic_cas_uint(&max_protohdr, oldhlen, hlen);
450
451 RUN_ONCE(&dotcpinit, do_tcpinit);
452}
453
454/*
455 * Tcp initialization
456 */
457void
458tcp_init(void)
459{
460
461 icmp_mtudisc_callback_register(tcp_mtudisc_callback);
462
463 tcp_init_common(sizeof(struct ip));
464}
465
466/*
467 * Create template to be used to send tcp packets on a connection.
468 * Call after host entry created, allocates an mbuf and fills
469 * in a skeletal tcp/ip header, minimizing the amount of work
470 * necessary when the connection is used.
471 */
472struct mbuf *
473tcp_template(struct tcpcb *tp)
474{
475 struct inpcb *inp = tp->t_inpcb;
476#ifdef INET6
477 struct in6pcb *in6p = tp->t_in6pcb;
478#endif
479 struct tcphdr *n;
480 struct mbuf *m;
481 int hlen;
482
483 switch (tp->t_family) {
484 case AF_INET:
485 hlen = sizeof(struct ip);
486 if (inp)
487 break;
488#ifdef INET6
489 if (in6p) {
490 /* mapped addr case */
491 if (IN6_IS_ADDR_V4MAPPED(&in6p->in6p_laddr)
492 && IN6_IS_ADDR_V4MAPPED(&in6p->in6p_faddr))
493 break;
494 }
495#endif
496 return NULL; /*EINVAL*/
497#ifdef INET6
498 case AF_INET6:
499 hlen = sizeof(struct ip6_hdr);
500 if (in6p) {
501 /* more sainty check? */
502 break;
503 }
504 return NULL; /*EINVAL*/
505#endif
506 default:
507 hlen = 0; /*pacify gcc*/
508 return NULL; /*EAFNOSUPPORT*/
509 }
510#ifdef DIAGNOSTIC
511 if (hlen + sizeof(struct tcphdr) > MCLBYTES)
512 panic("mclbytes too small for t_template");
513#endif
514 m = tp->t_template;
515 if (m && m->m_len == hlen + sizeof(struct tcphdr))
516 ;
517 else {
518 if (m)
519 m_freem(m);
520 m = tp->t_template = NULL;
521 MGETHDR(m, M_DONTWAIT, MT_HEADER);
522 if (m && hlen + sizeof(struct tcphdr) > MHLEN) {
523 MCLGET(m, M_DONTWAIT);
524 if ((m->m_flags & M_EXT) == 0) {
525 m_free(m);
526 m = NULL;
527 }
528 }
529 if (m == NULL)
530 return NULL;
531 MCLAIM(m, &tcp_mowner);
532 m->m_pkthdr.len = m->m_len = hlen + sizeof(struct tcphdr);
533 }
534
535 memset(mtod(m, void *), 0, m->m_len);
536
537 n = (struct tcphdr *)(mtod(m, char *) + hlen);
538
539 switch (tp->t_family) {
540 case AF_INET:
541 {
542 struct ipovly *ipov;
543 mtod(m, struct ip *)->ip_v = 4;
544 mtod(m, struct ip *)->ip_hl = hlen >> 2;
545 ipov = mtod(m, struct ipovly *);
546 ipov->ih_pr = IPPROTO_TCP;
547 ipov->ih_len = htons(sizeof(struct tcphdr));
548 if (inp) {
549 ipov->ih_src = inp->inp_laddr;
550 ipov->ih_dst = inp->inp_faddr;
551 }
552#ifdef INET6
553 else if (in6p) {
554 /* mapped addr case */
555 bcopy(&in6p->in6p_laddr.s6_addr32[3], &ipov->ih_src,
556 sizeof(ipov->ih_src));
557 bcopy(&in6p->in6p_faddr.s6_addr32[3], &ipov->ih_dst,
558 sizeof(ipov->ih_dst));
559 }
560#endif
561 /*
562 * Compute the pseudo-header portion of the checksum
563 * now. We incrementally add in the TCP option and
564 * payload lengths later, and then compute the TCP
565 * checksum right before the packet is sent off onto
566 * the wire.
567 */
568 n->th_sum = in_cksum_phdr(ipov->ih_src.s_addr,
569 ipov->ih_dst.s_addr,
570 htons(sizeof(struct tcphdr) + IPPROTO_TCP));
571 break;
572 }
573#ifdef INET6
574 case AF_INET6:
575 {
576 struct ip6_hdr *ip6;
577 mtod(m, struct ip *)->ip_v = 6;
578 ip6 = mtod(m, struct ip6_hdr *);
579 ip6->ip6_nxt = IPPROTO_TCP;
580 ip6->ip6_plen = htons(sizeof(struct tcphdr));
581 ip6->ip6_src = in6p->in6p_laddr;
582 ip6->ip6_dst = in6p->in6p_faddr;
583 ip6->ip6_flow = in6p->in6p_flowinfo & IPV6_FLOWINFO_MASK;
584 if (ip6_auto_flowlabel) {
585 ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK;
586 ip6->ip6_flow |=
587 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK);
588 }
589 ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
590 ip6->ip6_vfc |= IPV6_VERSION;
591
592 /*
593 * Compute the pseudo-header portion of the checksum
594 * now. We incrementally add in the TCP option and
595 * payload lengths later, and then compute the TCP
596 * checksum right before the packet is sent off onto
597 * the wire.
598 */
599 n->th_sum = in6_cksum_phdr(&in6p->in6p_laddr,
600 &in6p->in6p_faddr, htonl(sizeof(struct tcphdr)),
601 htonl(IPPROTO_TCP));
602 break;
603 }
604#endif
605 }
606 if (inp) {
607 n->th_sport = inp->inp_lport;
608 n->th_dport = inp->inp_fport;
609 }
610#ifdef INET6
611 else if (in6p) {
612 n->th_sport = in6p->in6p_lport;
613 n->th_dport = in6p->in6p_fport;
614 }
615#endif
616 n->th_seq = 0;
617 n->th_ack = 0;
618 n->th_x2 = 0;
619 n->th_off = 5;
620 n->th_flags = 0;
621 n->th_win = 0;
622 n->th_urp = 0;
623 return (m);
624}
625
626/*
627 * Send a single message to the TCP at address specified by
628 * the given TCP/IP header. If m == 0, then we make a copy
629 * of the tcpiphdr at ti and send directly to the addressed host.
630 * This is used to force keep alive messages out using the TCP
631 * template for a connection tp->t_template. If flags are given
632 * then we send a message back to the TCP which originated the
633 * segment ti, and discard the mbuf containing it and any other
634 * attached mbufs.
635 *
636 * In any case the ack and sequence number of the transmitted
637 * segment are as specified by the parameters.
638 */
639int
640tcp_respond(struct tcpcb *tp, struct mbuf *mtemplate, struct mbuf *m,
641 struct tcphdr *th0, tcp_seq ack, tcp_seq seq, int flags)
642{
643 struct route *ro;
644 int error, tlen, win = 0;
645 int hlen;
646 struct ip *ip;
647#ifdef INET6
648 struct ip6_hdr *ip6;
649#endif
650 int family; /* family on packet, not inpcb/in6pcb! */
651 struct tcphdr *th;
652 struct socket *so;
653
654 if (tp != NULL && (flags & TH_RST) == 0) {
655#ifdef DIAGNOSTIC
656 if (tp->t_inpcb && tp->t_in6pcb)
657 panic("tcp_respond: both t_inpcb and t_in6pcb are set");
658#endif
659#ifdef INET
660 if (tp->t_inpcb)
661 win = sbspace(&tp->t_inpcb->inp_socket->so_rcv);
662#endif
663#ifdef INET6
664 if (tp->t_in6pcb)
665 win = sbspace(&tp->t_in6pcb->in6p_socket->so_rcv);
666#endif
667 }
668
669 th = NULL; /* Quell uninitialized warning */
670 ip = NULL;
671#ifdef INET6
672 ip6 = NULL;
673#endif
674 if (m == 0) {
675 if (!mtemplate)
676 return EINVAL;
677
678 /* get family information from template */
679 switch (mtod(mtemplate, struct ip *)->ip_v) {
680 case 4:
681 family = AF_INET;
682 hlen = sizeof(struct ip);
683 break;
684#ifdef INET6
685 case 6:
686 family = AF_INET6;
687 hlen = sizeof(struct ip6_hdr);
688 break;
689#endif
690 default:
691 return EAFNOSUPPORT;
692 }
693
694 MGETHDR(m, M_DONTWAIT, MT_HEADER);
695 if (m) {
696 MCLAIM(m, &tcp_tx_mowner);
697 MCLGET(m, M_DONTWAIT);
698 if ((m->m_flags & M_EXT) == 0) {
699 m_free(m);
700 m = NULL;
701 }
702 }
703 if (m == NULL)
704 return (ENOBUFS);
705
706 if (tcp_compat_42)
707 tlen = 1;
708 else
709 tlen = 0;
710
711 m->m_data += max_linkhdr;
712 bcopy(mtod(mtemplate, void *), mtod(m, void *),
713 mtemplate->m_len);
714 switch (family) {
715 case AF_INET:
716 ip = mtod(m, struct ip *);
717 th = (struct tcphdr *)(ip + 1);
718 break;
719#ifdef INET6
720 case AF_INET6:
721 ip6 = mtod(m, struct ip6_hdr *);
722 th = (struct tcphdr *)(ip6 + 1);
723 break;
724#endif
725#if 0
726 default:
727 /* noone will visit here */
728 m_freem(m);
729 return EAFNOSUPPORT;
730#endif
731 }
732 flags = TH_ACK;
733 } else {
734
735 if ((m->m_flags & M_PKTHDR) == 0) {
736#if 0
737 printf("non PKTHDR to tcp_respond\n");
738#endif
739 m_freem(m);
740 return EINVAL;
741 }
742#ifdef DIAGNOSTIC
743 if (!th0)
744 panic("th0 == NULL in tcp_respond");
745#endif
746
747 /* get family information from m */
748 switch (mtod(m, struct ip *)->ip_v) {
749 case 4:
750 family = AF_INET;
751 hlen = sizeof(struct ip);
752 ip = mtod(m, struct ip *);
753 break;
754#ifdef INET6
755 case 6:
756 family = AF_INET6;
757 hlen = sizeof(struct ip6_hdr);
758 ip6 = mtod(m, struct ip6_hdr *);
759 break;
760#endif
761 default:
762 m_freem(m);
763 return EAFNOSUPPORT;
764 }
765 /* clear h/w csum flags inherited from rx packet */
766 m->m_pkthdr.csum_flags = 0;
767
768 if ((flags & TH_SYN) == 0 || sizeof(*th0) > (th0->th_off << 2))
769 tlen = sizeof(*th0);
770 else
771 tlen = th0->th_off << 2;
772
773 if (m->m_len > hlen + tlen && (m->m_flags & M_EXT) == 0 &&
774 mtod(m, char *) + hlen == (char *)th0) {
775 m->m_len = hlen + tlen;
776 m_freem(m->m_next);
777 m->m_next = NULL;
778 } else {
779 struct mbuf *n;
780
781#ifdef DIAGNOSTIC
782 if (max_linkhdr + hlen + tlen > MCLBYTES) {
783 m_freem(m);
784 return EMSGSIZE;
785 }
786#endif
787 MGETHDR(n, M_DONTWAIT, MT_HEADER);
788 if (n && max_linkhdr + hlen + tlen > MHLEN) {
789 MCLGET(n, M_DONTWAIT);
790 if ((n->m_flags & M_EXT) == 0) {
791 m_freem(n);
792 n = NULL;
793 }
794 }
795 if (!n) {
796 m_freem(m);
797 return ENOBUFS;
798 }
799
800 MCLAIM(n, &tcp_tx_mowner);
801 n->m_data += max_linkhdr;
802 n->m_len = hlen + tlen;
803 m_copyback(n, 0, hlen, mtod(m, void *));
804 m_copyback(n, hlen, tlen, (void *)th0);
805
806 m_freem(m);
807 m = n;
808 n = NULL;
809 }
810
811#define xchg(a,b,type) { type t; t=a; a=b; b=t; }
812 switch (family) {
813 case AF_INET:
814 ip = mtod(m, struct ip *);
815 th = (struct tcphdr *)(ip + 1);
816 ip->ip_p = IPPROTO_TCP;
817 xchg(ip->ip_dst, ip->ip_src, struct in_addr);
818 ip->ip_p = IPPROTO_TCP;
819 break;
820#ifdef INET6
821 case AF_INET6:
822 ip6 = mtod(m, struct ip6_hdr *);
823 th = (struct tcphdr *)(ip6 + 1);
824 ip6->ip6_nxt = IPPROTO_TCP;
825 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
826 ip6->ip6_nxt = IPPROTO_TCP;
827 break;
828#endif
829#if 0
830 default:
831 /* noone will visit here */
832 m_freem(m);
833 return EAFNOSUPPORT;
834#endif
835 }
836 xchg(th->th_dport, th->th_sport, u_int16_t);
837#undef xchg
838 tlen = 0; /*be friendly with the following code*/
839 }
840 th->th_seq = htonl(seq);
841 th->th_ack = htonl(ack);
842 th->th_x2 = 0;
843 if ((flags & TH_SYN) == 0) {
844 if (tp)
845 win >>= tp->rcv_scale;
846 if (win > TCP_MAXWIN)
847 win = TCP_MAXWIN;
848 th->th_win = htons((u_int16_t)win);
849 th->th_off = sizeof (struct tcphdr) >> 2;
850 tlen += sizeof(*th);
851 } else
852 tlen += th->th_off << 2;
853 m->m_len = hlen + tlen;
854 m->m_pkthdr.len = hlen + tlen;
855 m_reset_rcvif(m);
856 th->th_flags = flags;
857 th->th_urp = 0;
858
859 switch (family) {
860#ifdef INET
861 case AF_INET:
862 {
863 struct ipovly *ipov = (struct ipovly *)ip;
864 memset(ipov->ih_x1, 0, sizeof ipov->ih_x1);
865 ipov->ih_len = htons((u_int16_t)tlen);
866
867 th->th_sum = 0;
868 th->th_sum = in_cksum(m, hlen + tlen);
869 ip->ip_len = htons(hlen + tlen);
870 ip->ip_ttl = ip_defttl;
871 break;
872 }
873#endif
874#ifdef INET6
875 case AF_INET6:
876 {
877 th->th_sum = 0;
878 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr),
879 tlen);
880 ip6->ip6_plen = htons(tlen);
881 if (tp && tp->t_in6pcb)
882 ip6->ip6_hlim = in6_selecthlim_rt(tp->t_in6pcb);
883 else
884 ip6->ip6_hlim = ip6_defhlim;
885 ip6->ip6_flow &= ~IPV6_FLOWINFO_MASK;
886 if (ip6_auto_flowlabel) {
887 ip6->ip6_flow |=
888 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK);
889 }
890 break;
891 }
892#endif
893 }
894
895 if (tp && tp->t_inpcb)
896 so = tp->t_inpcb->inp_socket;
897#ifdef INET6
898 else if (tp && tp->t_in6pcb)
899 so = tp->t_in6pcb->in6p_socket;
900#endif
901 else
902 so = NULL;
903
904 if (tp != NULL && tp->t_inpcb != NULL) {
905 ro = &tp->t_inpcb->inp_route;
906#ifdef DIAGNOSTIC
907 if (family != AF_INET)
908 panic("tcp_respond: address family mismatch");
909 if (!in_hosteq(ip->ip_dst, tp->t_inpcb->inp_faddr)) {
910 panic("tcp_respond: ip_dst %x != inp_faddr %x",
911 ntohl(ip->ip_dst.s_addr),
912 ntohl(tp->t_inpcb->inp_faddr.s_addr));
913 }
914#endif
915 }
916#ifdef INET6
917 else if (tp != NULL && tp->t_in6pcb != NULL) {
918 ro = (struct route *)&tp->t_in6pcb->in6p_route;
919#ifdef DIAGNOSTIC
920 if (family == AF_INET) {
921 if (!IN6_IS_ADDR_V4MAPPED(&tp->t_in6pcb->in6p_faddr))
922 panic("tcp_respond: not mapped addr");
923 if (memcmp(&ip->ip_dst,
924 &tp->t_in6pcb->in6p_faddr.s6_addr32[3],
925 sizeof(ip->ip_dst)) != 0) {
926 panic("tcp_respond: ip_dst != in6p_faddr");
927 }
928 } else if (family == AF_INET6) {
929 if (!IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
930 &tp->t_in6pcb->in6p_faddr))
931 panic("tcp_respond: ip6_dst != in6p_faddr");
932 } else
933 panic("tcp_respond: address family mismatch");
934#endif
935 }
936#endif
937 else
938 ro = NULL;
939
940 switch (family) {
941#ifdef INET
942 case AF_INET:
943 error = ip_output(m, NULL, ro,
944 (tp && tp->t_mtudisc ? IP_MTUDISC : 0), NULL, so);
945 break;
946#endif
947#ifdef INET6
948 case AF_INET6:
949 error = ip6_output(m, NULL, ro, 0, NULL, so, NULL);
950 break;
951#endif
952 default:
953 error = EAFNOSUPPORT;
954 break;
955 }
956
957 return (error);
958}
959
960/*
961 * Template TCPCB. Rather than zeroing a new TCPCB and initializing
962 * a bunch of members individually, we maintain this template for the
963 * static and mostly-static components of the TCPCB, and copy it into
964 * the new TCPCB instead.
965 */
966static struct tcpcb tcpcb_template = {
967 .t_srtt = TCPTV_SRTTBASE,
968 .t_rttmin = TCPTV_MIN,
969
970 .snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT,
971 .snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT,
972 .snd_numholes = 0,
973 .snd_cubic_wmax = 0,
974 .snd_cubic_wmax_last = 0,
975 .snd_cubic_ctime = 0,
976
977 .t_partialacks = -1,
978 .t_bytes_acked = 0,
979 .t_sndrexmitpack = 0,
980 .t_rcvoopack = 0,
981 .t_sndzerowin = 0,
982};
983
984/*
985 * Updates the TCPCB template whenever a parameter that would affect
986 * the template is changed.
987 */
988void
989tcp_tcpcb_template(void)
990{
991 struct tcpcb *tp = &tcpcb_template;
992 int flags;
993
994 tp->t_peermss = tcp_mssdflt;
995 tp->t_ourmss = tcp_mssdflt;
996 tp->t_segsz = tcp_mssdflt;
997
998 flags = 0;
999 if (tcp_do_rfc1323 && tcp_do_win_scale)
1000 flags |= TF_REQ_SCALE;
1001 if (tcp_do_rfc1323 && tcp_do_timestamps)
1002 flags |= TF_REQ_TSTMP;
1003 tp->t_flags = flags;
1004
1005 /*
1006 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
1007 * rtt estimate. Set rttvar so that srtt + 2 * rttvar gives
1008 * reasonable initial retransmit time.
1009 */
1010 tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1);
1011 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
1012 TCPTV_MIN, TCPTV_REXMTMAX);
1013
1014 /* Keep Alive */
1015 tp->t_keepinit = tcp_keepinit;
1016 tp->t_keepidle = tcp_keepidle;
1017 tp->t_keepintvl = tcp_keepintvl;
1018 tp->t_keepcnt = tcp_keepcnt;
1019 tp->t_maxidle = tp->t_keepcnt * tp->t_keepintvl;
1020
1021 /* MSL */
1022 tp->t_msl = TCPTV_MSL;
1023}
1024
1025/*
1026 * Create a new TCP control block, making an
1027 * empty reassembly queue and hooking it to the argument
1028 * protocol control block.
1029 */
1030/* family selects inpcb, or in6pcb */
1031struct tcpcb *
1032tcp_newtcpcb(int family, void *aux)
1033{
1034 struct tcpcb *tp;
1035 int i;
1036
1037 /* XXX Consider using a pool_cache for speed. */
1038 tp = pool_get(&tcpcb_pool, PR_NOWAIT); /* splsoftnet via tcp_usrreq */
1039 if (tp == NULL)
1040 return (NULL);
1041 memcpy(tp, &tcpcb_template, sizeof(*tp));
1042 TAILQ_INIT(&tp->segq);
1043 TAILQ_INIT(&tp->timeq);
1044 tp->t_family = family; /* may be overridden later on */
1045 TAILQ_INIT(&tp->snd_holes);
1046 LIST_INIT(&tp->t_sc); /* XXX can template this */
1047
1048 /* Don't sweat this loop; hopefully the compiler will unroll it. */
1049 for (i = 0; i < TCPT_NTIMERS; i++) {
1050 callout_init(&tp->t_timer[i], CALLOUT_MPSAFE);
1051 TCP_TIMER_INIT(tp, i);
1052 }
1053 callout_init(&tp->t_delack_ch, CALLOUT_MPSAFE);
1054
1055 switch (family) {
1056 case AF_INET:
1057 {
1058 struct inpcb *inp = (struct inpcb *)aux;
1059
1060 inp->inp_ip.ip_ttl = ip_defttl;
1061 inp->inp_ppcb = (void *)tp;
1062
1063 tp->t_inpcb = inp;
1064 tp->t_mtudisc = ip_mtudisc;
1065 break;
1066 }
1067#ifdef INET6
1068 case AF_INET6:
1069 {
1070 struct in6pcb *in6p = (struct in6pcb *)aux;
1071
1072 in6p->in6p_ip6.ip6_hlim = in6_selecthlim_rt(in6p);
1073 in6p->in6p_ppcb = (void *)tp;
1074
1075 tp->t_in6pcb = in6p;
1076 /* for IPv6, always try to run path MTU discovery */
1077 tp->t_mtudisc = 1;
1078 break;
1079 }
1080#endif /* INET6 */
1081 default:
1082 for (i = 0; i < TCPT_NTIMERS; i++)
1083 callout_destroy(&tp->t_timer[i]);
1084 callout_destroy(&tp->t_delack_ch);
1085 pool_put(&tcpcb_pool, tp); /* splsoftnet via tcp_usrreq */
1086 return (NULL);
1087 }
1088
1089 /*
1090 * Initialize our timebase. When we send timestamps, we take
1091 * the delta from tcp_now -- this means each connection always
1092 * gets a timebase of 1, which makes it, among other things,
1093 * more difficult to determine how long a system has been up,
1094 * and thus how many TCP sequence increments have occurred.
1095 *
1096 * We start with 1, because 0 doesn't work with linux, which
1097 * considers timestamp 0 in a SYN packet as a bug and disables
1098 * timestamps.
1099 */
1100 tp->ts_timebase = tcp_now - 1;
1101
1102 tcp_congctl_select(tp, tcp_congctl_global_name);
1103
1104 return (tp);
1105}
1106
1107/*
1108 * Drop a TCP connection, reporting
1109 * the specified error. If connection is synchronized,
1110 * then send a RST to peer.
1111 */
1112struct tcpcb *
1113tcp_drop(struct tcpcb *tp, int errno)
1114{
1115 struct socket *so = NULL;
1116
1117#ifdef DIAGNOSTIC
1118 if (tp->t_inpcb && tp->t_in6pcb)
1119 panic("tcp_drop: both t_inpcb and t_in6pcb are set");
1120#endif
1121#ifdef INET
1122 if (tp->t_inpcb)
1123 so = tp->t_inpcb->inp_socket;
1124#endif
1125#ifdef INET6
1126 if (tp->t_in6pcb)
1127 so = tp->t_in6pcb->in6p_socket;
1128#endif
1129 if (!so)
1130 return NULL;
1131
1132 if (TCPS_HAVERCVDSYN(tp->t_state)) {
1133 tp->t_state = TCPS_CLOSED;
1134 (void) tcp_output(tp);
1135 TCP_STATINC(TCP_STAT_DROPS);
1136 } else
1137 TCP_STATINC(TCP_STAT_CONNDROPS);
1138 if (errno == ETIMEDOUT && tp->t_softerror)
1139 errno = tp->t_softerror;
1140 so->so_error = errno;
1141 return (tcp_close(tp));
1142}
1143
1144/*
1145 * Close a TCP control block:
1146 * discard all space held by the tcp
1147 * discard internet protocol block
1148 * wake up any sleepers
1149 */
1150struct tcpcb *
1151tcp_close(struct tcpcb *tp)
1152{
1153 struct inpcb *inp;
1154#ifdef INET6
1155 struct in6pcb *in6p;
1156#endif
1157 struct socket *so;
1158#ifdef RTV_RTT
1159 struct rtentry *rt;
1160#endif
1161 struct route *ro;
1162 int j;
1163
1164 inp = tp->t_inpcb;
1165#ifdef INET6
1166 in6p = tp->t_in6pcb;
1167#endif
1168 so = NULL;
1169 ro = NULL;
1170 if (inp) {
1171 so = inp->inp_socket;
1172 ro = &inp->inp_route;
1173 }
1174#ifdef INET6
1175 else if (in6p) {
1176 so = in6p->in6p_socket;
1177 ro = (struct route *)&in6p->in6p_route;
1178 }
1179#endif
1180
1181#ifdef RTV_RTT
1182 /*
1183 * If we sent enough data to get some meaningful characteristics,
1184 * save them in the routing entry. 'Enough' is arbitrarily
1185 * defined as the sendpipesize (default 4K) * 16. This would
1186 * give us 16 rtt samples assuming we only get one sample per
1187 * window (the usual case on a long haul net). 16 samples is
1188 * enough for the srtt filter to converge to within 5% of the correct
1189 * value; fewer samples and we could save a very bogus rtt.
1190 *
1191 * Don't update the default route's characteristics and don't
1192 * update anything that the user "locked".
1193 */
1194 if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) &&
1195 ro && (rt = rtcache_validate(ro)) != NULL &&
1196 !in_nullhost(satocsin(rt_getkey(rt))->sin_addr)) {
1197 u_long i = 0;
1198
1199 if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
1200 i = tp->t_srtt *
1201 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
1202 if (rt->rt_rmx.rmx_rtt && i)
1203 /*
1204 * filter this update to half the old & half
1205 * the new values, converting scale.
1206 * See route.h and tcp_var.h for a
1207 * description of the scaling constants.
1208 */
1209 rt->rt_rmx.rmx_rtt =
1210 (rt->rt_rmx.rmx_rtt + i) / 2;
1211 else
1212 rt->rt_rmx.rmx_rtt = i;
1213 }
1214 if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
1215 i = tp->t_rttvar *
1216 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2));
1217 if (rt->rt_rmx.rmx_rttvar && i)
1218 rt->rt_rmx.rmx_rttvar =
1219 (rt->rt_rmx.rmx_rttvar + i) / 2;
1220 else
1221 rt->rt_rmx.rmx_rttvar = i;
1222 }
1223 /*
1224 * update the pipelimit (ssthresh) if it has been updated
1225 * already or if a pipesize was specified & the threshhold
1226 * got below half the pipesize. I.e., wait for bad news
1227 * before we start updating, then update on both good
1228 * and bad news.
1229 */
1230 if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
1231 (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) ||
1232 i < (rt->rt_rmx.rmx_sendpipe / 2)) {
1233 /*
1234 * convert the limit from user data bytes to
1235 * packets then to packet data bytes.
1236 */
1237 i = (i + tp->t_segsz / 2) / tp->t_segsz;
1238 if (i < 2)
1239 i = 2;
1240 i *= (u_long)(tp->t_segsz + sizeof (struct tcpiphdr));
1241 if (rt->rt_rmx.rmx_ssthresh)
1242 rt->rt_rmx.rmx_ssthresh =
1243 (rt->rt_rmx.rmx_ssthresh + i) / 2;
1244 else
1245 rt->rt_rmx.rmx_ssthresh = i;
1246 }
1247 }
1248#endif /* RTV_RTT */
1249 /* free the reassembly queue, if any */
1250 TCP_REASS_LOCK(tp);
1251 (void) tcp_freeq(tp);
1252 TCP_REASS_UNLOCK(tp);
1253
1254 /* free the SACK holes list. */
1255 tcp_free_sackholes(tp);
1256 tcp_congctl_release(tp);
1257 syn_cache_cleanup(tp);
1258
1259 if (tp->t_template) {
1260 m_free(tp->t_template);
1261 tp->t_template = NULL;
1262 }
1263
1264 /*
1265 * Detaching the pcb will unlock the socket/tcpcb, and stopping
1266 * the timers can also drop the lock. We need to prevent access
1267 * to the tcpcb as it's half torn down. Flag the pcb as dead
1268 * (prevents access by timers) and only then detach it.
1269 */
1270 tp->t_flags |= TF_DEAD;
1271 if (inp) {
1272 inp->inp_ppcb = 0;
1273 soisdisconnected(so);
1274 in_pcbdetach(inp);
1275 }
1276#ifdef INET6
1277 else if (in6p) {
1278 in6p->in6p_ppcb = 0;
1279 soisdisconnected(so);
1280 in6_pcbdetach(in6p);
1281 }
1282#endif
1283 /*
1284 * pcb is no longer visble elsewhere, so we can safely release
1285 * the lock in callout_halt() if needed.
1286 */
1287 TCP_STATINC(TCP_STAT_CLOSED);
1288 for (j = 0; j < TCPT_NTIMERS; j++) {
1289 callout_halt(&tp->t_timer[j], softnet_lock);
1290 callout_destroy(&tp->t_timer[j]);
1291 }
1292 callout_halt(&tp->t_delack_ch, softnet_lock);
1293 callout_destroy(&tp->t_delack_ch);
1294 pool_put(&tcpcb_pool, tp);
1295
1296 return NULL;
1297}
1298
1299int
1300tcp_freeq(struct tcpcb *tp)
1301{
1302 struct ipqent *qe;
1303 int rv = 0;
1304#ifdef TCPREASS_DEBUG
1305 int i = 0;
1306#endif
1307
1308 TCP_REASS_LOCK_CHECK(tp);
1309
1310 while ((qe = TAILQ_FIRST(&tp->segq)) != NULL) {
1311#ifdef TCPREASS_DEBUG
1312 printf("tcp_freeq[%p,%d]: %u:%u(%u) 0x%02x\n",
1313 tp, i++, qe->ipqe_seq, qe->ipqe_seq + qe->ipqe_len,
1314 qe->ipqe_len, qe->ipqe_flags & (TH_SYN|TH_FIN|TH_RST));
1315#endif
1316 TAILQ_REMOVE(&tp->segq, qe, ipqe_q);
1317 TAILQ_REMOVE(&tp->timeq, qe, ipqe_timeq);
1318 m_freem(qe->ipqe_m);
1319 tcpipqent_free(qe);
1320 rv = 1;
1321 }
1322 tp->t_segqlen = 0;
1323 KASSERT(TAILQ_EMPTY(&tp->timeq));
1324 return (rv);
1325}
1326
1327void
1328tcp_fasttimo(void)
1329{
1330 if (tcp_drainwanted) {
1331 tcp_drain();
1332 tcp_drainwanted = 0;
1333 }
1334}
1335
1336void
1337tcp_drainstub(void)
1338{
1339 tcp_drainwanted = 1;
1340}
1341
1342/*
1343 * Protocol drain routine. Called when memory is in short supply.
1344 * Called from pr_fasttimo thus a callout context.
1345 */
1346void
1347tcp_drain(void)
1348{
1349 struct inpcb_hdr *inph;
1350 struct tcpcb *tp;
1351
1352 mutex_enter(softnet_lock);
1353 KERNEL_LOCK(1, NULL);
1354
1355 /*
1356 * Free the sequence queue of all TCP connections.
1357 */
1358 TAILQ_FOREACH(inph, &tcbtable.inpt_queue, inph_queue) {
1359 switch (inph->inph_af) {
1360 case AF_INET:
1361 tp = intotcpcb((struct inpcb *)inph);
1362 break;
1363#ifdef INET6
1364 case AF_INET6:
1365 tp = in6totcpcb((struct in6pcb *)inph);
1366 break;
1367#endif
1368 default:
1369 tp = NULL;
1370 break;
1371 }
1372 if (tp != NULL) {
1373 /*
1374 * We may be called from a device's interrupt
1375 * context. If the tcpcb is already busy,
1376 * just bail out now.
1377 */
1378 if (tcp_reass_lock_try(tp) == 0)
1379 continue;
1380 if (tcp_freeq(tp))
1381 TCP_STATINC(TCP_STAT_CONNSDRAINED);
1382 TCP_REASS_UNLOCK(tp);
1383 }
1384 }
1385
1386 KERNEL_UNLOCK_ONE(NULL);
1387 mutex_exit(softnet_lock);
1388}
1389
1390/*
1391 * Notify a tcp user of an asynchronous error;
1392 * store error as soft error, but wake up user
1393 * (for now, won't do anything until can select for soft error).
1394 */
1395void
1396tcp_notify(struct inpcb *inp, int error)
1397{
1398 struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
1399 struct socket *so = inp->inp_socket;
1400
1401 /*
1402 * Ignore some errors if we are hooked up.
1403 * If connection hasn't completed, has retransmitted several times,
1404 * and receives a second error, give up now. This is better
1405 * than waiting a long time to establish a connection that
1406 * can never complete.
1407 */
1408 if (tp->t_state == TCPS_ESTABLISHED &&
1409 (error == EHOSTUNREACH || error == ENETUNREACH ||
1410 error == EHOSTDOWN)) {
1411 return;
1412 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
1413 tp->t_rxtshift > 3 && tp->t_softerror)
1414 so->so_error = error;
1415 else
1416 tp->t_softerror = error;
1417 cv_broadcast(&so->so_cv);
1418 sorwakeup(so);
1419 sowwakeup(so);
1420}
1421
1422#ifdef INET6
1423void
1424tcp6_notify(struct in6pcb *in6p, int error)
1425{
1426 struct tcpcb *tp = (struct tcpcb *)in6p->in6p_ppcb;
1427 struct socket *so = in6p->in6p_socket;
1428
1429 /*
1430 * Ignore some errors if we are hooked up.
1431 * If connection hasn't completed, has retransmitted several times,
1432 * and receives a second error, give up now. This is better
1433 * than waiting a long time to establish a connection that
1434 * can never complete.
1435 */
1436 if (tp->t_state == TCPS_ESTABLISHED &&
1437 (error == EHOSTUNREACH || error == ENETUNREACH ||
1438 error == EHOSTDOWN)) {
1439 return;
1440 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
1441 tp->t_rxtshift > 3 && tp->t_softerror)
1442 so->so_error = error;
1443 else
1444 tp->t_softerror = error;
1445 cv_broadcast(&so->so_cv);
1446 sorwakeup(so);
1447 sowwakeup(so);
1448}
1449#endif
1450
1451#ifdef INET6
1452void *
1453tcp6_ctlinput(int cmd, const struct sockaddr *sa, void *d)
1454{
1455 struct tcphdr th;
1456 void (*notify)(struct in6pcb *, int) = tcp6_notify;
1457 int nmatch;
1458 struct ip6_hdr *ip6;
1459 const struct sockaddr_in6 *sa6_src = NULL;
1460 const struct sockaddr_in6 *sa6 = (const struct sockaddr_in6 *)sa;
1461 struct mbuf *m;
1462 int off;
1463
1464 if (sa->sa_family != AF_INET6 ||
1465 sa->sa_len != sizeof(struct sockaddr_in6))
1466 return NULL;
1467 if ((unsigned)cmd >= PRC_NCMDS)
1468 return NULL;
1469 else if (cmd == PRC_QUENCH) {
1470 /*
1471 * Don't honor ICMP Source Quench messages meant for
1472 * TCP connections.
1473 */
1474 return NULL;
1475 } else if (PRC_IS_REDIRECT(cmd))
1476 notify = in6_rtchange, d = NULL;
1477 else if (cmd == PRC_MSGSIZE)
1478 ; /* special code is present, see below */
1479 else if (cmd == PRC_HOSTDEAD)
1480 d = NULL;
1481 else if (inet6ctlerrmap[cmd] == 0)
1482 return NULL;
1483
1484 /* if the parameter is from icmp6, decode it. */
1485 if (d != NULL) {
1486 struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d;
1487 m = ip6cp->ip6c_m;
1488 ip6 = ip6cp->ip6c_ip6;
1489 off = ip6cp->ip6c_off;
1490 sa6_src = ip6cp->ip6c_src;
1491 } else {
1492 m = NULL;
1493 ip6 = NULL;
1494 sa6_src = &sa6_any;
1495 off = 0;
1496 }
1497
1498 if (ip6) {
1499 /*
1500 * XXX: We assume that when ip6 is non NULL,
1501 * M and OFF are valid.
1502 */
1503
1504 /* check if we can safely examine src and dst ports */
1505 if (m->m_pkthdr.len < off + sizeof(th)) {
1506 if (cmd == PRC_MSGSIZE)
1507 icmp6_mtudisc_update((struct ip6ctlparam *)d, 0);
1508 return NULL;
1509 }
1510
1511 memset(&th, 0, sizeof(th));
1512 m_copydata(m, off, sizeof(th), (void *)&th);
1513
1514 if (cmd == PRC_MSGSIZE) {
1515 int valid = 0;
1516
1517 /*
1518 * Check to see if we have a valid TCP connection
1519 * corresponding to the address in the ICMPv6 message
1520 * payload.
1521 */
1522 if (in6_pcblookup_connect(&tcbtable, &sa6->sin6_addr,
1523 th.th_dport,
1524 (const struct in6_addr *)&sa6_src->sin6_addr,
1525 th.th_sport, 0, 0))
1526 valid++;
1527
1528 /*
1529 * Depending on the value of "valid" and routing table
1530 * size (mtudisc_{hi,lo}wat), we will:
1531 * - recalcurate the new MTU and create the
1532 * corresponding routing entry, or
1533 * - ignore the MTU change notification.
1534 */
1535 icmp6_mtudisc_update((struct ip6ctlparam *)d, valid);
1536
1537 /*
1538 * no need to call in6_pcbnotify, it should have been
1539 * called via callback if necessary
1540 */
1541 return NULL;
1542 }
1543
1544 nmatch = in6_pcbnotify(&tcbtable, sa, th.th_dport,
1545 (const struct sockaddr *)sa6_src, th.th_sport, cmd, NULL, notify);
1546 if (nmatch == 0 && syn_cache_count &&
1547 (inet6ctlerrmap[cmd] == EHOSTUNREACH ||
1548 inet6ctlerrmap[cmd] == ENETUNREACH ||
1549 inet6ctlerrmap[cmd] == EHOSTDOWN))
1550 syn_cache_unreach((const struct sockaddr *)sa6_src,
1551 sa, &th);
1552 } else {
1553 (void) in6_pcbnotify(&tcbtable, sa, 0,
1554 (const struct sockaddr *)sa6_src, 0, cmd, NULL, notify);
1555 }
1556
1557 return NULL;
1558}
1559#endif
1560
1561#ifdef INET
1562/* assumes that ip header and tcp header are contiguous on mbuf */
1563void *
1564tcp_ctlinput(int cmd, const struct sockaddr *sa, void *v)
1565{
1566 struct ip *ip = v;
1567 struct tcphdr *th;
1568 struct icmp *icp;
1569 extern const int inetctlerrmap[];
1570 void (*notify)(struct inpcb *, int) = tcp_notify;
1571 int errno;
1572 int nmatch;
1573 struct tcpcb *tp;
1574 u_int mtu;
1575 tcp_seq seq;
1576 struct inpcb *inp;
1577#ifdef INET6
1578 struct in6pcb *in6p;
1579 struct in6_addr src6, dst6;
1580#endif
1581
1582 if (sa->sa_family != AF_INET ||
1583 sa->sa_len != sizeof(struct sockaddr_in))
1584 return NULL;
1585 if ((unsigned)cmd >= PRC_NCMDS)
1586 return NULL;
1587 errno = inetctlerrmap[cmd];
1588 if (cmd == PRC_QUENCH)
1589 /*
1590 * Don't honor ICMP Source Quench messages meant for
1591 * TCP connections.
1592 */
1593 return NULL;
1594 else if (PRC_IS_REDIRECT(cmd))
1595 notify = in_rtchange, ip = 0;
1596 else if (cmd == PRC_MSGSIZE && ip && ip->ip_v == 4) {
1597 /*
1598 * Check to see if we have a valid TCP connection
1599 * corresponding to the address in the ICMP message
1600 * payload.
1601 *
1602 * Boundary check is made in icmp_input(), with ICMP_ADVLENMIN.
1603 */
1604 th = (struct tcphdr *)((char *)ip + (ip->ip_hl << 2));
1605#ifdef INET6
1606 in6_in_2_v4mapin6(&ip->ip_src, &src6);
1607 in6_in_2_v4mapin6(&ip->ip_dst, &dst6);
1608#endif
1609 if ((inp = in_pcblookup_connect(&tcbtable, ip->ip_dst,
1610 th->th_dport, ip->ip_src, th->th_sport, 0)) != NULL)
1611#ifdef INET6
1612 in6p = NULL;
1613#else
1614 ;
1615#endif
1616#ifdef INET6
1617 else if ((in6p = in6_pcblookup_connect(&tcbtable, &dst6,
1618 th->th_dport, &src6, th->th_sport, 0, 0)) != NULL)
1619 ;
1620#endif
1621 else
1622 return NULL;
1623
1624 /*
1625 * Now that we've validated that we are actually communicating
1626 * with the host indicated in the ICMP message, locate the
1627 * ICMP header, recalculate the new MTU, and create the
1628 * corresponding routing entry.
1629 */
1630 icp = (struct icmp *)((char *)ip -
1631 offsetof(struct icmp, icmp_ip));
1632 if (inp) {
1633 if ((tp = intotcpcb(inp)) == NULL)
1634 return NULL;
1635 }
1636#ifdef INET6
1637 else if (in6p) {
1638 if ((tp = in6totcpcb(in6p)) == NULL)
1639 return NULL;
1640 }
1641#endif
1642 else
1643 return NULL;
1644 seq = ntohl(th->th_seq);
1645 if (SEQ_LT(seq, tp->snd_una) || SEQ_GT(seq, tp->snd_max))
1646 return NULL;
1647 /*
1648 * If the ICMP message advertises a Next-Hop MTU
1649 * equal or larger than the maximum packet size we have
1650 * ever sent, drop the message.
1651 */
1652 mtu = (u_int)ntohs(icp->icmp_nextmtu);
1653 if (mtu >= tp->t_pmtud_mtu_sent)
1654 return NULL;
1655 if (mtu >= tcp_hdrsz(tp) + tp->t_pmtud_mss_acked) {
1656 /*
1657 * Calculate new MTU, and create corresponding
1658 * route (traditional PMTUD).
1659 */
1660 tp->t_flags &= ~TF_PMTUD_PEND;
1661 icmp_mtudisc(icp, ip->ip_dst);
1662 } else {
1663 /*
1664 * Record the information got in the ICMP
1665 * message; act on it later.
1666 * If we had already recorded an ICMP message,
1667 * replace the old one only if the new message
1668 * refers to an older TCP segment
1669 */
1670 if (tp->t_flags & TF_PMTUD_PEND) {
1671 if (SEQ_LT(tp->t_pmtud_th_seq, seq))
1672 return NULL;
1673 } else
1674 tp->t_flags |= TF_PMTUD_PEND;
1675 tp->t_pmtud_th_seq = seq;
1676 tp->t_pmtud_nextmtu = icp->icmp_nextmtu;
1677 tp->t_pmtud_ip_len = icp->icmp_ip.ip_len;
1678 tp->t_pmtud_ip_hl = icp->icmp_ip.ip_hl;
1679 }
1680 return NULL;
1681 } else if (cmd == PRC_HOSTDEAD)
1682 ip = 0;
1683 else if (errno == 0)
1684 return NULL;
1685 if (ip && ip->ip_v == 4 && sa->sa_family == AF_INET) {
1686 th = (struct tcphdr *)((char *)ip + (ip->ip_hl << 2));
1687 nmatch = in_pcbnotify(&tcbtable, satocsin(sa)->sin_addr,
1688 th->th_dport, ip->ip_src, th->th_sport, errno, notify);
1689 if (nmatch == 0 && syn_cache_count &&
1690 (inetctlerrmap[cmd] == EHOSTUNREACH ||
1691 inetctlerrmap[cmd] == ENETUNREACH ||
1692 inetctlerrmap[cmd] == EHOSTDOWN)) {
1693 struct sockaddr_in sin;
1694 memset(&sin, 0, sizeof(sin));
1695 sin.sin_len = sizeof(sin);
1696 sin.sin_family = AF_INET;
1697 sin.sin_port = th->th_sport;
1698 sin.sin_addr = ip->ip_src;
1699 syn_cache_unreach((struct sockaddr *)&sin, sa, th);
1700 }
1701
1702 /* XXX mapped address case */
1703 } else
1704 in_pcbnotifyall(&tcbtable, satocsin(sa)->sin_addr, errno,
1705 notify);
1706 return NULL;
1707}
1708
1709/*
1710 * When a source quench is received, we are being notified of congestion.
1711 * Close the congestion window down to the Loss Window (one segment).
1712 * We will gradually open it again as we proceed.
1713 */
1714void
1715tcp_quench(struct inpcb *inp, int errno)
1716{
1717 struct tcpcb *tp = intotcpcb(inp);
1718
1719 if (tp) {
1720 tp->snd_cwnd = tp->t_segsz;
1721 tp->t_bytes_acked = 0;
1722 }
1723}
1724#endif
1725
1726#ifdef INET6
1727void
1728tcp6_quench(struct in6pcb *in6p, int errno)
1729{
1730 struct tcpcb *tp = in6totcpcb(in6p);
1731
1732 if (tp) {
1733 tp->snd_cwnd = tp->t_segsz;
1734 tp->t_bytes_acked = 0;
1735 }
1736}
1737#endif
1738
1739#ifdef INET
1740/*
1741 * Path MTU Discovery handlers.
1742 */
1743void
1744tcp_mtudisc_callback(struct in_addr faddr)
1745{
1746#ifdef INET6
1747 struct in6_addr in6;
1748#endif
1749
1750 in_pcbnotifyall(&tcbtable, faddr, EMSGSIZE, tcp_mtudisc);
1751#ifdef INET6
1752 in6_in_2_v4mapin6(&faddr, &in6);
1753 tcp6_mtudisc_callback(&in6);
1754#endif
1755}
1756
1757/*
1758 * On receipt of path MTU corrections, flush old route and replace it
1759 * with the new one. Retransmit all unacknowledged packets, to ensure
1760 * that all packets will be received.
1761 */
1762void
1763tcp_mtudisc(struct inpcb *inp, int errno)
1764{
1765 struct tcpcb *tp = intotcpcb(inp);
1766 struct rtentry *rt;
1767
1768 if (tp == NULL)
1769 return;
1770
1771 rt = in_pcbrtentry(inp);
1772 if (rt != NULL) {
1773 /*
1774 * If this was not a host route, remove and realloc.
1775 */
1776 if ((rt->rt_flags & RTF_HOST) == 0) {
1777 in_rtchange(inp, errno);
1778 if ((rt = in_pcbrtentry(inp)) == NULL)
1779 return;
1780 }
1781
1782 /*
1783 * Slow start out of the error condition. We
1784 * use the MTU because we know it's smaller
1785 * than the previously transmitted segment.
1786 *
1787 * Note: This is more conservative than the
1788 * suggestion in draft-floyd-incr-init-win-03.
1789 */
1790 if (rt->rt_rmx.rmx_mtu != 0)
1791 tp->snd_cwnd =
1792 TCP_INITIAL_WINDOW(tcp_init_win,
1793 rt->rt_rmx.rmx_mtu);
1794 }
1795
1796 /*
1797 * Resend unacknowledged packets.
1798 */
1799 tp->snd_nxt = tp->sack_newdata = tp->snd_una;
1800 tcp_output(tp);
1801}
1802#endif /* INET */
1803
1804#ifdef INET6
1805/*
1806 * Path MTU Discovery handlers.
1807 */
1808void
1809tcp6_mtudisc_callback(struct in6_addr *faddr)
1810{
1811 struct sockaddr_in6 sin6;
1812
1813 memset(&sin6, 0, sizeof(sin6));
1814 sin6.sin6_family = AF_INET6;
1815 sin6.sin6_len = sizeof(struct sockaddr_in6);
1816 sin6.sin6_addr = *faddr;
1817 (void) in6_pcbnotify(&tcbtable, (struct sockaddr *)&sin6, 0,
1818 (const struct sockaddr *)&sa6_any, 0, PRC_MSGSIZE, NULL, tcp6_mtudisc);
1819}
1820
1821void
1822tcp6_mtudisc(struct in6pcb *in6p, int errno)
1823{
1824 struct tcpcb *tp = in6totcpcb(in6p);
1825 struct rtentry *rt;
1826
1827 if (tp == NULL)
1828 return;
1829
1830 rt = in6_pcbrtentry(in6p);
1831 if (rt != NULL) {
1832 /*
1833 * If this was not a host route, remove and realloc.
1834 */
1835 if ((rt->rt_flags & RTF_HOST) == 0) {
1836 in6_rtchange(in6p, errno);
1837 rt = in6_pcbrtentry(in6p);
1838 if (rt == NULL)
1839 return;
1840 }
1841
1842 /*
1843 * Slow start out of the error condition. We
1844 * use the MTU because we know it's smaller
1845 * than the previously transmitted segment.
1846 *
1847 * Note: This is more conservative than the
1848 * suggestion in draft-floyd-incr-init-win-03.
1849 */
1850 if (rt->rt_rmx.rmx_mtu != 0) {
1851 tp->snd_cwnd = TCP_INITIAL_WINDOW(tcp_init_win,
1852 rt->rt_rmx.rmx_mtu);
1853 }
1854 }
1855
1856 /*
1857 * Resend unacknowledged packets.
1858 */
1859 tp->snd_nxt = tp->sack_newdata = tp->snd_una;
1860 tcp_output(tp);
1861}
1862#endif /* INET6 */
1863
1864/*
1865 * Compute the MSS to advertise to the peer. Called only during
1866 * the 3-way handshake. If we are the server (peer initiated
1867 * connection), we are called with a pointer to the interface
1868 * on which the SYN packet arrived. If we are the client (we
1869 * initiated connection), we are called with a pointer to the
1870 * interface out which this connection should go.
1871 *
1872 * NOTE: Do not subtract IP option/extension header size nor IPsec
1873 * header size from MSS advertisement. MSS option must hold the maximum
1874 * segment size we can accept, so it must always be:
1875 * max(if mtu) - ip header - tcp header
1876 */
1877u_long
1878tcp_mss_to_advertise(const struct ifnet *ifp, int af)
1879{
1880 extern u_long in_maxmtu;
1881 u_long mss = 0;
1882 u_long hdrsiz;
1883
1884 /*
1885 * In order to avoid defeating path MTU discovery on the peer,
1886 * we advertise the max MTU of all attached networks as our MSS,
1887 * per RFC 1191, section 3.1.
1888 *
1889 * We provide the option to advertise just the MTU of
1890 * the interface on which we hope this connection will
1891 * be receiving. If we are responding to a SYN, we
1892 * will have a pretty good idea about this, but when
1893 * initiating a connection there is a bit more doubt.
1894 *
1895 * We also need to ensure that loopback has a large enough
1896 * MSS, as the loopback MTU is never included in in_maxmtu.
1897 */
1898
1899 if (ifp != NULL)
1900 switch (af) {
1901 case AF_INET:
1902 mss = ifp->if_mtu;
1903 break;
1904#ifdef INET6
1905 case AF_INET6:
1906 mss = IN6_LINKMTU(ifp);
1907 break;
1908#endif
1909 }
1910
1911 if (tcp_mss_ifmtu == 0)
1912 switch (af) {
1913 case AF_INET:
1914 mss = max(in_maxmtu, mss);
1915 break;
1916#ifdef INET6
1917 case AF_INET6:
1918 mss = max(in6_maxmtu, mss);
1919 break;
1920#endif
1921 }
1922
1923 switch (af) {
1924 case AF_INET:
1925 hdrsiz = sizeof(struct ip);
1926 break;
1927#ifdef INET6
1928 case AF_INET6:
1929 hdrsiz = sizeof(struct ip6_hdr);
1930 break;
1931#endif
1932 default:
1933 hdrsiz = 0;
1934 break;
1935 }
1936 hdrsiz += sizeof(struct tcphdr);
1937 if (mss > hdrsiz)
1938 mss -= hdrsiz;
1939
1940 mss = max(tcp_mssdflt, mss);
1941 return (mss);
1942}
1943
1944/*
1945 * Set connection variables based on the peer's advertised MSS.
1946 * We are passed the TCPCB for the actual connection. If we
1947 * are the server, we are called by the compressed state engine
1948 * when the 3-way handshake is complete. If we are the client,
1949 * we are called when we receive the SYN,ACK from the server.
1950 *
1951 * NOTE: Our advertised MSS value must be initialized in the TCPCB
1952 * before this routine is called!
1953 */
1954void
1955tcp_mss_from_peer(struct tcpcb *tp, int offer)
1956{
1957 struct socket *so;
1958#if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1959 struct rtentry *rt;
1960#endif
1961 u_long bufsize;
1962 int mss;
1963
1964#ifdef DIAGNOSTIC
1965 if (tp->t_inpcb && tp->t_in6pcb)
1966 panic("tcp_mss_from_peer: both t_inpcb and t_in6pcb are set");
1967#endif
1968 so = NULL;
1969 rt = NULL;
1970#ifdef INET
1971 if (tp->t_inpcb) {
1972 so = tp->t_inpcb->inp_socket;
1973#if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1974 rt = in_pcbrtentry(tp->t_inpcb);
1975#endif
1976 }
1977#endif
1978#ifdef INET6
1979 if (tp->t_in6pcb) {
1980 so = tp->t_in6pcb->in6p_socket;
1981#if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1982 rt = in6_pcbrtentry(tp->t_in6pcb);
1983#endif
1984 }
1985#endif
1986
1987 /*
1988 * As per RFC1122, use the default MSS value, unless they
1989 * sent us an offer. Do not accept offers less than 256 bytes.
1990 */
1991 mss = tcp_mssdflt;
1992 if (offer)
1993 mss = offer;
1994 mss = max(mss, 256); /* sanity */
1995 tp->t_peermss = mss;
1996 mss -= tcp_optlen(tp);
1997#ifdef INET
1998 if (tp->t_inpcb)
1999 mss -= ip_optlen(tp->t_inpcb);
2000#endif
2001#ifdef INET6
2002 if (tp->t_in6pcb)
2003 mss -= ip6_optlen(tp->t_in6pcb);
2004#endif
2005
2006 /*
2007 * If there's a pipesize, change the socket buffer to that size.
2008 * Make the socket buffer an integral number of MSS units. If
2009 * the MSS is larger than the socket buffer, artificially decrease
2010 * the MSS.
2011 */
2012#ifdef RTV_SPIPE
2013 if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0)
2014 bufsize = rt->rt_rmx.rmx_sendpipe;
2015 else
2016#endif
2017 {
2018 KASSERT(so != NULL);
2019 bufsize = so->so_snd.sb_hiwat;
2020 }
2021 if (bufsize < mss)
2022 mss = bufsize;
2023 else {
2024 bufsize = roundup(bufsize, mss);
2025 if (bufsize > sb_max)
2026 bufsize = sb_max;
2027 (void) sbreserve(&so->so_snd, bufsize, so);
2028 }
2029 tp->t_segsz = mss;
2030
2031#ifdef RTV_SSTHRESH
2032 if (rt != NULL && rt->rt_rmx.rmx_ssthresh) {
2033 /*
2034 * There's some sort of gateway or interface buffer
2035 * limit on the path. Use this to set the slow
2036 * start threshold, but set the threshold to no less
2037 * than 2 * MSS.
2038 */
2039 tp->snd_ssthresh = max(2 * mss, rt->rt_rmx.rmx_ssthresh);
2040 }
2041#endif
2042}
2043
2044/*
2045 * Processing necessary when a TCP connection is established.
2046 */
2047void
2048tcp_established(struct tcpcb *tp)
2049{
2050 struct socket *so;
2051#ifdef RTV_RPIPE
2052 struct rtentry *rt;
2053#endif
2054 u_long bufsize;
2055
2056#ifdef DIAGNOSTIC
2057 if (tp->t_inpcb && tp->t_in6pcb)
2058 panic("tcp_established: both t_inpcb and t_in6pcb are set");
2059#endif
2060 so = NULL;
2061 rt = NULL;
2062#ifdef INET
2063 /* This is a while() to reduce the dreadful stairstepping below */
2064 while (tp->t_inpcb) {
2065 so = tp->t_inpcb->inp_socket;
2066#if defined(RTV_RPIPE)
2067 rt = in_pcbrtentry(tp->t_inpcb);
2068#endif
2069 if (__predict_true(tcp_msl_enable)) {
2070 if (tp->t_inpcb->inp_laddr.s_addr == INADDR_LOOPBACK) {
2071 tp->t_msl = tcp_msl_loop ? tcp_msl_loop : (TCPTV_MSL >> 2);
2072 break;
2073 }
2074
2075 if (__predict_false(tcp_rttlocal)) {
2076 /* This may be adjusted by tcp_input */
2077 tp->t_msl = tcp_msl_local ? tcp_msl_local : (TCPTV_MSL >> 1);
2078 break;
2079 }
2080 if (in_localaddr(tp->t_inpcb->inp_faddr)) {
2081 tp->t_msl = tcp_msl_local ? tcp_msl_local : (TCPTV_MSL >> 1);
2082 break;
2083 }
2084 }
2085 tp->t_msl = tcp_msl_remote ? tcp_msl_remote : TCPTV_MSL;
2086 break;
2087 }
2088#endif
2089#ifdef INET6
2090 /* The !tp->t_inpcb lets the compiler know it can't be v4 *and* v6 */
2091 while (!tp->t_inpcb && tp->t_in6pcb) {
2092 so = tp->t_in6pcb->in6p_socket;
2093#if defined(RTV_RPIPE)
2094 rt = in6_pcbrtentry(tp->t_in6pcb);
2095#endif
2096 if (__predict_true(tcp_msl_enable)) {
2097 extern const struct in6_addr in6addr_loopback;
2098
2099 if (IN6_ARE_ADDR_EQUAL(&tp->t_in6pcb->in6p_laddr,
2100 &in6addr_loopback)) {
2101 tp->t_msl = tcp_msl_loop ? tcp_msl_loop : (TCPTV_MSL >> 2);
2102 break;
2103 }
2104
2105 if (__predict_false(tcp_rttlocal)) {
2106 /* This may be adjusted by tcp_input */
2107 tp->t_msl = tcp_msl_local ? tcp_msl_local : (TCPTV_MSL >> 1);
2108 break;
2109 }
2110 if (in6_localaddr(&tp->t_in6pcb->in6p_faddr)) {
2111 tp->t_msl = tcp_msl_local ? tcp_msl_local : (TCPTV_MSL >> 1);
2112 break;
2113 }
2114 }
2115 tp->t_msl = tcp_msl_remote ? tcp_msl_remote : TCPTV_MSL;
2116 break;
2117 }
2118#endif
2119
2120 tp->t_state = TCPS_ESTABLISHED;
2121 TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepidle);
2122
2123#ifdef RTV_RPIPE
2124 if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0)
2125 bufsize = rt->rt_rmx.rmx_recvpipe;
2126 else
2127#endif
2128 {
2129 KASSERT(so != NULL);
2130 bufsize = so->so_rcv.sb_hiwat;
2131 }
2132 if (bufsize > tp->t_ourmss) {
2133 bufsize = roundup(bufsize, tp->t_ourmss);
2134 if (bufsize > sb_max)
2135 bufsize = sb_max;
2136 (void) sbreserve(&so->so_rcv, bufsize, so);
2137 }
2138}
2139
2140/*
2141 * Check if there's an initial rtt or rttvar. Convert from the
2142 * route-table units to scaled multiples of the slow timeout timer.
2143 * Called only during the 3-way handshake.
2144 */
2145void
2146tcp_rmx_rtt(struct tcpcb *tp)
2147{
2148#ifdef RTV_RTT
2149 struct rtentry *rt = NULL;
2150 int rtt;
2151
2152#ifdef DIAGNOSTIC
2153 if (tp->t_inpcb && tp->t_in6pcb)
2154 panic("tcp_rmx_rtt: both t_inpcb and t_in6pcb are set");
2155#endif
2156#ifdef INET
2157 if (tp->t_inpcb)
2158 rt = in_pcbrtentry(tp->t_inpcb);
2159#endif
2160#ifdef INET6
2161 if (tp->t_in6pcb)
2162 rt = in6_pcbrtentry(tp->t_in6pcb);
2163#endif
2164 if (rt == NULL)
2165 return;
2166
2167 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
2168 /*
2169 * XXX The lock bit for MTU indicates that the value
2170 * is also a minimum value; this is subject to time.
2171 */
2172 if (rt->rt_rmx.rmx_locks & RTV_RTT)
2173 TCPT_RANGESET(tp->t_rttmin,
2174 rtt / (RTM_RTTUNIT / PR_SLOWHZ),
2175 TCPTV_MIN, TCPTV_REXMTMAX);
2176 tp->t_srtt = rtt /
2177 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
2178 if (rt->rt_rmx.rmx_rttvar) {
2179 tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
2180 ((RTM_RTTUNIT / PR_SLOWHZ) >>
2181 (TCP_RTTVAR_SHIFT + 2));
2182 } else {
2183 /* Default variation is +- 1 rtt */
2184 tp->t_rttvar =
2185 tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT);
2186 }
2187 TCPT_RANGESET(tp->t_rxtcur,
2188 ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2),
2189 tp->t_rttmin, TCPTV_REXMTMAX);
2190 }
2191#endif
2192}
2193
2194tcp_seq tcp_iss_seq = 0; /* tcp initial seq # */
2195
2196/*
2197 * Get a new sequence value given a tcp control block
2198 */
2199tcp_seq
2200tcp_new_iss(struct tcpcb *tp, tcp_seq addin)
2201{
2202
2203#ifdef INET
2204 if (tp->t_inpcb != NULL) {
2205 return (tcp_new_iss1(&tp->t_inpcb->inp_laddr,
2206 &tp->t_inpcb->inp_faddr, tp->t_inpcb->inp_lport,
2207 tp->t_inpcb->inp_fport, sizeof(tp->t_inpcb->inp_laddr),
2208 addin));
2209 }
2210#endif
2211#ifdef INET6
2212 if (tp->t_in6pcb != NULL) {
2213 return (tcp_new_iss1(&tp->t_in6pcb->in6p_laddr,
2214 &tp->t_in6pcb->in6p_faddr, tp->t_in6pcb->in6p_lport,
2215 tp->t_in6pcb->in6p_fport, sizeof(tp->t_in6pcb->in6p_laddr),
2216 addin));
2217 }
2218#endif
2219 /* Not possible. */
2220 panic("tcp_new_iss");
2221}
2222
2223static u_int8_t tcp_iss_secret[16]; /* 128 bits; should be plenty */
2224
2225/*
2226 * Initialize RFC 1948 ISS Secret
2227 */
2228static int
2229tcp_iss_secret_init(void)
2230{
2231 cprng_strong(kern_cprng,
2232 tcp_iss_secret, sizeof(tcp_iss_secret), 0);
2233
2234 return 0;
2235}
2236
2237/*
2238 * This routine actually generates a new TCP initial sequence number.
2239 */
2240tcp_seq
2241tcp_new_iss1(void *laddr, void *faddr, u_int16_t lport, u_int16_t fport,
2242 size_t addrsz, tcp_seq addin)
2243{
2244 tcp_seq tcp_iss;
2245
2246 if (tcp_do_rfc1948) {
2247 MD5_CTX ctx;
2248 u_int8_t hash[16]; /* XXX MD5 knowledge */
2249 static ONCE_DECL(tcp_iss_secret_control);
2250
2251 /*
2252 * If we haven't been here before, initialize our cryptographic
2253 * hash secret.
2254 */
2255 RUN_ONCE(&tcp_iss_secret_control, tcp_iss_secret_init);
2256
2257 /*
2258 * Compute the base value of the ISS. It is a hash
2259 * of (saddr, sport, daddr, dport, secret).
2260 */
2261 MD5Init(&ctx);
2262
2263 MD5Update(&ctx, (u_char *) laddr, addrsz);
2264 MD5Update(&ctx, (u_char *) &lport, sizeof(lport));
2265
2266 MD5Update(&ctx, (u_char *) faddr, addrsz);
2267 MD5Update(&ctx, (u_char *) &fport, sizeof(fport));
2268
2269 MD5Update(&ctx, tcp_iss_secret, sizeof(tcp_iss_secret));
2270
2271 MD5Final(hash, &ctx);
2272
2273 memcpy(&tcp_iss, hash, sizeof(tcp_iss));
2274
2275 /*
2276 * Now increment our "timer", and add it in to
2277 * the computed value.
2278 *
2279 * XXX Use `addin'?
2280 * XXX TCP_ISSINCR too large to use?
2281 */
2282 tcp_iss_seq += TCP_ISSINCR;
2283#ifdef TCPISS_DEBUG
2284 printf("ISS hash 0x%08x, ", tcp_iss);
2285#endif
2286 tcp_iss += tcp_iss_seq + addin;
2287#ifdef TCPISS_DEBUG
2288 printf("new ISS 0x%08x\n", tcp_iss);
2289#endif
2290 } else {
2291 /*
2292 * Randomize.
2293 */
2294 tcp_iss = cprng_fast32();
2295
2296 /*
2297 * If we were asked to add some amount to a known value,
2298 * we will take a random value obtained above, mask off
2299 * the upper bits, and add in the known value. We also
2300 * add in a constant to ensure that we are at least a
2301 * certain distance from the original value.
2302 *
2303 * This is used when an old connection is in timed wait
2304 * and we have a new one coming in, for instance.
2305 */
2306 if (addin != 0) {
2307#ifdef TCPISS_DEBUG
2308 printf("Random %08x, ", tcp_iss);
2309#endif
2310 tcp_iss &= TCP_ISS_RANDOM_MASK;
2311 tcp_iss += addin + TCP_ISSINCR;
2312#ifdef TCPISS_DEBUG
2313 printf("Old ISS %08x, ISS %08x\n", addin, tcp_iss);
2314#endif
2315 } else {
2316 tcp_iss &= TCP_ISS_RANDOM_MASK;
2317 tcp_iss += tcp_iss_seq;
2318 tcp_iss_seq += TCP_ISSINCR;
2319#ifdef TCPISS_DEBUG
2320 printf("ISS %08x\n", tcp_iss);
2321#endif
2322 }
2323 }
2324
2325 if (tcp_compat_42) {
2326 /*
2327 * Limit it to the positive range for really old TCP
2328 * implementations.
2329 * Just AND off the top bit instead of checking if
2330 * is set first - saves a branch 50% of the time.
2331 */
2332 tcp_iss &= 0x7fffffff; /* XXX */
2333 }
2334
2335 return (tcp_iss);
2336}
2337
2338#if defined(IPSEC)
2339/* compute ESP/AH header size for TCP, including outer IP header. */
2340size_t
2341ipsec4_hdrsiz_tcp(struct tcpcb *tp)
2342{
2343 struct inpcb *inp;
2344 size_t hdrsiz;
2345
2346 /* XXX mapped addr case (tp->t_in6pcb) */
2347 if (!tp || !tp->t_template || !(inp = tp->t_inpcb))
2348 return 0;
2349 switch (tp->t_family) {
2350 case AF_INET:
2351 /* XXX: should use currect direction. */
2352 hdrsiz = ipsec4_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, inp);
2353 break;
2354 default:
2355 hdrsiz = 0;
2356 break;
2357 }
2358
2359 return hdrsiz;
2360}
2361
2362#ifdef INET6
2363size_t
2364ipsec6_hdrsiz_tcp(struct tcpcb *tp)
2365{
2366 struct in6pcb *in6p;
2367 size_t hdrsiz;
2368
2369 if (!tp || !tp->t_template || !(in6p = tp->t_in6pcb))
2370 return 0;
2371 switch (tp->t_family) {
2372 case AF_INET6:
2373 /* XXX: should use currect direction. */
2374 hdrsiz = ipsec6_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, in6p);
2375 break;
2376 case AF_INET:
2377 /* mapped address case - tricky */
2378 default:
2379 hdrsiz = 0;
2380 break;
2381 }
2382
2383 return hdrsiz;
2384}
2385#endif
2386#endif /*IPSEC*/
2387
2388/*
2389 * Determine the length of the TCP options for this connection.
2390 *
2391 * XXX: What do we do for SACK, when we add that? Just reserve
2392 * all of the space? Otherwise we can't exactly be incrementing
2393 * cwnd by an amount that varies depending on the amount we last
2394 * had to SACK!
2395 */
2396
2397u_int
2398tcp_optlen(struct tcpcb *tp)
2399{
2400 u_int optlen;
2401
2402 optlen = 0;
2403 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) ==
2404 (TF_REQ_TSTMP | TF_RCVD_TSTMP))
2405 optlen += TCPOLEN_TSTAMP_APPA;
2406
2407#ifdef TCP_SIGNATURE
2408 if (tp->t_flags & TF_SIGNATURE)
2409 optlen += TCPOLEN_SIGNATURE + 2;
2410#endif /* TCP_SIGNATURE */
2411
2412 return optlen;
2413}
2414
2415u_int
2416tcp_hdrsz(struct tcpcb *tp)
2417{
2418 u_int hlen;
2419
2420 switch (tp->t_family) {
2421#ifdef INET6
2422 case AF_INET6:
2423 hlen = sizeof(struct ip6_hdr);
2424 break;
2425#endif
2426 case AF_INET:
2427 hlen = sizeof(struct ip);
2428 break;
2429 default:
2430 hlen = 0;
2431 break;
2432 }
2433 hlen += sizeof(struct tcphdr);
2434
2435 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
2436 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)
2437 hlen += TCPOLEN_TSTAMP_APPA;
2438#ifdef TCP_SIGNATURE
2439 if (tp->t_flags & TF_SIGNATURE)
2440 hlen += TCPOLEN_SIGLEN;
2441#endif
2442 return hlen;
2443}
2444
2445void
2446tcp_statinc(u_int stat)
2447{
2448
2449 KASSERT(stat < TCP_NSTATS);
2450 TCP_STATINC(stat);
2451}
2452
2453void
2454tcp_statadd(u_int stat, uint64_t val)
2455{
2456
2457 KASSERT(stat < TCP_NSTATS);
2458 TCP_STATADD(stat, val);
2459}
2460