1 /* $NetBSD: tcp_subr.c,v 1.296 2022/11/04 09:01:53 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 1.296 2022/11/04 09:01:53 ozaki-r Exp $");
95
96 #ifdef _KERNEL_OPT
97 #include "opt_inet.h"
98 #include "opt_ipsec.h"
99 #include "opt_inet_csum.h"
100 #include "opt_mbuftrace.h"
101 #endif
102
103 #include <sys/param.h>
104 #include <sys/atomic.h>
105 #include <sys/proc.h>
106 #include <sys/systm.h>
107 #include <sys/mbuf.h>
108 #include <sys/once.h>
109 #include <sys/socket.h>
110 #include <sys/socketvar.h>
111 #include <sys/protosw.h>
112 #include <sys/errno.h>
113 #include <sys/kernel.h>
114 #include <sys/pool.h>
115 #include <sys/md5.h>
116 #include <sys/cprng.h>
117
118 #include <net/route.h>
119 #include <net/if.h>
120
121 #include <netinet/in.h>
122 #include <netinet/in_systm.h>
123 #include <netinet/ip.h>
124 #include <netinet/in_pcb.h>
125 #include <netinet/ip_var.h>
126 #include <netinet/ip_icmp.h>
127
128 #ifdef INET6
129 #include <netinet/ip6.h>
130 #include <netinet6/in6_pcb.h>
131 #include <netinet6/ip6_var.h>
132 #include <netinet6/in6_var.h>
133 #include <netinet6/ip6protosw.h>
134 #include <netinet/icmp6.h>
135 #include <netinet6/nd6.h>
136 #endif
137
138 #include <netinet/tcp.h>
139 #include <netinet/tcp_fsm.h>
140 #include <netinet/tcp_seq.h>
141 #include <netinet/tcp_timer.h>
142 #include <netinet/tcp_var.h>
143 #include <netinet/tcp_vtw.h>
144 #include <netinet/tcp_private.h>
145 #include <netinet/tcp_congctl.h>
146 #include <netinet/tcp_syncache.h>
147
148 #ifdef IPSEC
149 #include <netipsec/ipsec.h>
150 #ifdef INET6
151 #include <netipsec/ipsec6.h>
152 #endif
153 #include <netipsec/key.h>
154 #endif
155
156
157 struct inpcbtable tcbtable; /* head of queue of active tcpcb's */
158 u_int32_t tcp_now; /* slow ticks, for RFC 1323 timestamps */
159
160 percpu_t *tcpstat_percpu;
161
162 /* patchable/settable parameters for tcp */
163 int tcp_mssdflt = TCP_MSS;
164 int tcp_minmss = TCP_MINMSS;
165 int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
166 int tcp_do_rfc1323 = 1; /* window scaling / timestamps (obsolete) */
167 int tcp_do_rfc1948 = 0; /* ISS by cryptographic hash */
168 int tcp_do_sack = 1; /* selective acknowledgement */
169 int tcp_do_win_scale = 1; /* RFC1323 window scaling */
170 int tcp_do_timestamps = 1; /* RFC1323 timestamps */
171 int tcp_ack_on_push = 0; /* set to enable immediate ACK-on-PUSH */
172 int tcp_do_ecn = 0; /* Explicit Congestion Notification */
173 #ifndef TCP_INIT_WIN
174 #define TCP_INIT_WIN 4 /* initial slow start window */
175 #endif
176 #ifndef TCP_INIT_WIN_LOCAL
177 #define TCP_INIT_WIN_LOCAL 4 /* initial slow start window for local nets */
178 #endif
179 /*
180 * Up to 5 we scale linearly, to reach 3 * 1460; then (iw) * 1460.
181 * This is to simulate current behavior for iw == 4
182 */
183 int tcp_init_win_max[] = {
184 1 * 1460,
185 1 * 1460,
186 2 * 1460,
187 2 * 1460,
188 3 * 1460,
189 5 * 1460,
190 6 * 1460,
191 7 * 1460,
192 8 * 1460,
193 9 * 1460,
194 10 * 1460
195 };
196 int tcp_init_win = TCP_INIT_WIN;
197 int tcp_init_win_local = TCP_INIT_WIN_LOCAL;
198 int tcp_mss_ifmtu = 0;
199 int tcp_rst_ppslim = 100; /* 100pps */
200 int tcp_ackdrop_ppslim = 100; /* 100pps */
201 int tcp_do_loopback_cksum = 0;
202 int tcp_do_abc = 1; /* RFC3465 Appropriate byte counting. */
203 int tcp_abc_aggressive = 1; /* 1: L=2*SMSS 0: L=1*SMSS */
204 int tcp_sack_tp_maxholes = 32;
205 int tcp_sack_globalmaxholes = 1024;
206 int tcp_sack_globalholes = 0;
207 int tcp_ecn_maxretries = 1;
208 int tcp_msl_enable = 1; /* enable TIME_WAIT truncation */
209 int tcp_msl_loop = PR_SLOWHZ; /* MSL for loopback */
210 int tcp_msl_local = 5 * PR_SLOWHZ; /* MSL for 'local' */
211 int tcp_msl_remote = TCPTV_MSL; /* MSL otherwise */
212 int tcp_msl_remote_threshold = TCPTV_SRTTDFLT; /* RTT threshold */
213 int tcp_rttlocal = 0; /* Use RTT to decide who's 'local' */
214
215 int tcp4_vtw_enable = 0; /* 1 to enable */
216 int tcp6_vtw_enable = 0; /* 1 to enable */
217 int tcp_vtw_was_enabled = 0;
218 int tcp_vtw_entries = 1 << 4; /* 16 vestigial TIME_WAIT entries */
219
220 /* tcb hash */
221 #ifndef TCBHASHSIZE
222 #define TCBHASHSIZE 128
223 #endif
224 int tcbhashsize = TCBHASHSIZE;
225
226 int tcp_freeq(struct tcpcb *);
227 static int tcp_iss_secret_init(void);
228
229 static void tcp_mtudisc_callback(struct in_addr);
230
231 #ifdef INET6
232 static void tcp6_mtudisc(struct inpcb *, int);
233 #endif
234
235 static struct pool tcpcb_pool;
236
237 static int tcp_drainwanted;
238
239 #ifdef TCP_CSUM_COUNTERS
240 #include <sys/device.h>
241
242 struct evcnt tcp_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
243 NULL, "tcp", "hwcsum bad");
244 struct evcnt tcp_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
245 NULL, "tcp", "hwcsum ok");
246 struct evcnt tcp_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
247 NULL, "tcp", "hwcsum data");
248 struct evcnt tcp_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
249 NULL, "tcp", "swcsum");
250
251 EVCNT_ATTACH_STATIC(tcp_hwcsum_bad);
252 EVCNT_ATTACH_STATIC(tcp_hwcsum_ok);
253 EVCNT_ATTACH_STATIC(tcp_hwcsum_data);
254 EVCNT_ATTACH_STATIC(tcp_swcsum);
255
256 #if defined(INET6)
257 struct evcnt tcp6_hwcsum_bad = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
258 NULL, "tcp6", "hwcsum bad");
259 struct evcnt tcp6_hwcsum_ok = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
260 NULL, "tcp6", "hwcsum ok");
261 struct evcnt tcp6_hwcsum_data = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
262 NULL, "tcp6", "hwcsum data");
263 struct evcnt tcp6_swcsum = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
264 NULL, "tcp6", "swcsum");
265
266 EVCNT_ATTACH_STATIC(tcp6_hwcsum_bad);
267 EVCNT_ATTACH_STATIC(tcp6_hwcsum_ok);
268 EVCNT_ATTACH_STATIC(tcp6_hwcsum_data);
269 EVCNT_ATTACH_STATIC(tcp6_swcsum);
270 #endif /* defined(INET6) */
271 #endif /* TCP_CSUM_COUNTERS */
272
273
274 #ifdef TCP_OUTPUT_COUNTERS
275 #include <sys/device.h>
276
277 struct evcnt tcp_output_bigheader = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
278 NULL, "tcp", "output big header");
279 struct evcnt tcp_output_predict_hit = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
280 NULL, "tcp", "output predict hit");
281 struct evcnt tcp_output_predict_miss = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
282 NULL, "tcp", "output predict miss");
283 struct evcnt tcp_output_copysmall = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
284 NULL, "tcp", "output copy small");
285 struct evcnt tcp_output_copybig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
286 NULL, "tcp", "output copy big");
287 struct evcnt tcp_output_refbig = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
288 NULL, "tcp", "output reference big");
289
290 EVCNT_ATTACH_STATIC(tcp_output_bigheader);
291 EVCNT_ATTACH_STATIC(tcp_output_predict_hit);
292 EVCNT_ATTACH_STATIC(tcp_output_predict_miss);
293 EVCNT_ATTACH_STATIC(tcp_output_copysmall);
294 EVCNT_ATTACH_STATIC(tcp_output_copybig);
295 EVCNT_ATTACH_STATIC(tcp_output_refbig);
296
297 #endif /* TCP_OUTPUT_COUNTERS */
298
299 #ifdef TCP_REASS_COUNTERS
300 #include <sys/device.h>
301
302 struct evcnt tcp_reass_ = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
303 NULL, "tcp_reass", "calls");
304 struct evcnt tcp_reass_empty = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
305 &tcp_reass_, "tcp_reass", "insert into empty queue");
306 struct evcnt tcp_reass_iteration[8] = {
307 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", ">7 iterations"),
308 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "1 iteration"),
309 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "2 iterations"),
310 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "3 iterations"),
311 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "4 iterations"),
312 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "5 iterations"),
313 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "6 iterations"),
314 EVCNT_INITIALIZER(EVCNT_TYPE_MISC, &tcp_reass_, "tcp_reass", "7 iterations"),
315 };
316 struct evcnt tcp_reass_prependfirst = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
317 &tcp_reass_, "tcp_reass", "prepend to first");
318 struct evcnt tcp_reass_prepend = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
319 &tcp_reass_, "tcp_reass", "prepend");
320 struct evcnt tcp_reass_insert = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
321 &tcp_reass_, "tcp_reass", "insert");
322 struct evcnt tcp_reass_inserttail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
323 &tcp_reass_, "tcp_reass", "insert at tail");
324 struct evcnt tcp_reass_append = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
325 &tcp_reass_, "tcp_reass", "append");
326 struct evcnt tcp_reass_appendtail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
327 &tcp_reass_, "tcp_reass", "append to tail fragment");
328 struct evcnt tcp_reass_overlaptail = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
329 &tcp_reass_, "tcp_reass", "overlap at end");
330 struct evcnt tcp_reass_overlapfront = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
331 &tcp_reass_, "tcp_reass", "overlap at start");
332 struct evcnt tcp_reass_segdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
333 &tcp_reass_, "tcp_reass", "duplicate segment");
334 struct evcnt tcp_reass_fragdup = EVCNT_INITIALIZER(EVCNT_TYPE_MISC,
335 &tcp_reass_, "tcp_reass", "duplicate fragment");
336
337 EVCNT_ATTACH_STATIC(tcp_reass_);
338 EVCNT_ATTACH_STATIC(tcp_reass_empty);
339 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 0);
340 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 1);
341 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 2);
342 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 3);
343 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 4);
344 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 5);
345 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 6);
346 EVCNT_ATTACH_STATIC2(tcp_reass_iteration, 7);
347 EVCNT_ATTACH_STATIC(tcp_reass_prependfirst);
348 EVCNT_ATTACH_STATIC(tcp_reass_prepend);
349 EVCNT_ATTACH_STATIC(tcp_reass_insert);
350 EVCNT_ATTACH_STATIC(tcp_reass_inserttail);
351 EVCNT_ATTACH_STATIC(tcp_reass_append);
352 EVCNT_ATTACH_STATIC(tcp_reass_appendtail);
353 EVCNT_ATTACH_STATIC(tcp_reass_overlaptail);
354 EVCNT_ATTACH_STATIC(tcp_reass_overlapfront);
355 EVCNT_ATTACH_STATIC(tcp_reass_segdup);
356 EVCNT_ATTACH_STATIC(tcp_reass_fragdup);
357
358 #endif /* TCP_REASS_COUNTERS */
359
360 #ifdef MBUFTRACE
361 struct mowner tcp_mowner = MOWNER_INIT("tcp", "");
362 struct mowner tcp_rx_mowner = MOWNER_INIT("tcp", "rx");
363 struct mowner tcp_tx_mowner = MOWNER_INIT("tcp", "tx");
364 struct mowner tcp_sock_mowner = MOWNER_INIT("tcp", "sock");
365 struct mowner tcp_sock_rx_mowner = MOWNER_INIT("tcp", "sock rx");
366 struct mowner tcp_sock_tx_mowner = MOWNER_INIT("tcp", "sock tx");
367 #endif
368
369 static int
do_tcpinit(void)370 do_tcpinit(void)
371 {
372
373 inpcb_init(&tcbtable, tcbhashsize, tcbhashsize);
374 pool_init(&tcpcb_pool, sizeof(struct tcpcb), 0, 0, 0, "tcpcbpl",
375 NULL, IPL_SOFTNET);
376
377 tcp_usrreq_init();
378
379 /* Initialize timer state. */
380 tcp_timer_init();
381
382 /* Initialize the compressed state engine. */
383 syn_cache_init();
384
385 /* Initialize the congestion control algorithms. */
386 tcp_congctl_init();
387
388 /* Initialize the TCPCB template. */
389 tcp_tcpcb_template();
390
391 /* Initialize reassembly queue */
392 tcpipqent_init();
393
394 /* SACK */
395 tcp_sack_init();
396
397 MOWNER_ATTACH(&tcp_tx_mowner);
398 MOWNER_ATTACH(&tcp_rx_mowner);
399 MOWNER_ATTACH(&tcp_reass_mowner);
400 MOWNER_ATTACH(&tcp_sock_mowner);
401 MOWNER_ATTACH(&tcp_sock_tx_mowner);
402 MOWNER_ATTACH(&tcp_sock_rx_mowner);
403 MOWNER_ATTACH(&tcp_mowner);
404
405 tcpstat_percpu = percpu_alloc(sizeof(uint64_t) * TCP_NSTATS);
406
407 vtw_earlyinit();
408
409 tcp_slowtimo_init();
410
411 return 0;
412 }
413
414 void
tcp_init_common(unsigned basehlen)415 tcp_init_common(unsigned basehlen)
416 {
417 static ONCE_DECL(dotcpinit);
418 unsigned hlen = basehlen + sizeof(struct tcphdr);
419 unsigned oldhlen;
420
421 if (max_linkhdr + hlen > MHLEN)
422 panic("tcp_init");
423 while ((oldhlen = max_protohdr) < hlen)
424 atomic_cas_uint(&max_protohdr, oldhlen, hlen);
425
426 RUN_ONCE(&dotcpinit, do_tcpinit);
427 }
428
429 /*
430 * Tcp initialization
431 */
432 void
tcp_init(void)433 tcp_init(void)
434 {
435
436 icmp_mtudisc_callback_register(tcp_mtudisc_callback);
437
438 tcp_init_common(sizeof(struct ip));
439 }
440
441 /*
442 * Create template to be used to send tcp packets on a connection.
443 * Call after host entry created, allocates an mbuf and fills
444 * in a skeletal tcp/ip header, minimizing the amount of work
445 * necessary when the connection is used.
446 */
447 struct mbuf *
tcp_template(struct tcpcb * tp)448 tcp_template(struct tcpcb *tp)
449 {
450 struct inpcb *inp = tp->t_inpcb;
451 struct tcphdr *n;
452 struct mbuf *m;
453 int hlen;
454
455 switch (tp->t_family) {
456 case AF_INET:
457 hlen = sizeof(struct ip);
458 if (inp->inp_af == AF_INET)
459 break;
460 #ifdef INET6
461 if (inp->inp_af == AF_INET6) {
462 /* mapped addr case */
463 if (IN6_IS_ADDR_V4MAPPED(&in6p_laddr(inp))
464 && IN6_IS_ADDR_V4MAPPED(&in6p_faddr(inp)))
465 break;
466 }
467 #endif
468 return NULL; /*EINVAL*/
469 #ifdef INET6
470 case AF_INET6:
471 hlen = sizeof(struct ip6_hdr);
472 if (inp != NULL) {
473 /* more sainty check? */
474 break;
475 }
476 return NULL; /*EINVAL*/
477 #endif
478 default:
479 return NULL; /*EAFNOSUPPORT*/
480 }
481
482 KASSERT(hlen + sizeof(struct tcphdr) <= MCLBYTES);
483
484 m = tp->t_template;
485 if (m && m->m_len == hlen + sizeof(struct tcphdr)) {
486 ;
487 } else {
488 if (m)
489 m_freem(m);
490 m = tp->t_template = NULL;
491 MGETHDR(m, M_DONTWAIT, MT_HEADER);
492 if (m && hlen + sizeof(struct tcphdr) > MHLEN) {
493 MCLGET(m, M_DONTWAIT);
494 if ((m->m_flags & M_EXT) == 0) {
495 m_free(m);
496 m = NULL;
497 }
498 }
499 if (m == NULL)
500 return NULL;
501 MCLAIM(m, &tcp_mowner);
502 m->m_pkthdr.len = m->m_len = hlen + sizeof(struct tcphdr);
503 }
504
505 memset(mtod(m, void *), 0, m->m_len);
506
507 n = (struct tcphdr *)(mtod(m, char *) + hlen);
508
509 switch (tp->t_family) {
510 case AF_INET:
511 {
512 struct ipovly *ipov;
513 mtod(m, struct ip *)->ip_v = 4;
514 mtod(m, struct ip *)->ip_hl = hlen >> 2;
515 ipov = mtod(m, struct ipovly *);
516 ipov->ih_pr = IPPROTO_TCP;
517 ipov->ih_len = htons(sizeof(struct tcphdr));
518 if (inp->inp_af == AF_INET) {
519 ipov->ih_src = in4p_laddr(inp);
520 ipov->ih_dst = in4p_faddr(inp);
521 }
522 #ifdef INET6
523 else if (inp->inp_af == AF_INET6) {
524 /* mapped addr case */
525 bcopy(&in6p_laddr(inp).s6_addr32[3], &ipov->ih_src,
526 sizeof(ipov->ih_src));
527 bcopy(&in6p_faddr(inp).s6_addr32[3], &ipov->ih_dst,
528 sizeof(ipov->ih_dst));
529 }
530 #endif
531
532 /*
533 * Compute the pseudo-header portion of the checksum
534 * now. We incrementally add in the TCP option and
535 * payload lengths later, and then compute the TCP
536 * checksum right before the packet is sent off onto
537 * the wire.
538 */
539 n->th_sum = in_cksum_phdr(ipov->ih_src.s_addr,
540 ipov->ih_dst.s_addr,
541 htons(sizeof(struct tcphdr) + IPPROTO_TCP));
542 break;
543 }
544 #ifdef INET6
545 case AF_INET6:
546 {
547 struct ip6_hdr *ip6;
548 mtod(m, struct ip *)->ip_v = 6;
549 ip6 = mtod(m, struct ip6_hdr *);
550 ip6->ip6_nxt = IPPROTO_TCP;
551 ip6->ip6_plen = htons(sizeof(struct tcphdr));
552 ip6->ip6_src = in6p_laddr(inp);
553 ip6->ip6_dst = in6p_faddr(inp);
554 ip6->ip6_flow = in6p_flowinfo(inp) & IPV6_FLOWINFO_MASK;
555 if (ip6_auto_flowlabel) {
556 ip6->ip6_flow &= ~IPV6_FLOWLABEL_MASK;
557 ip6->ip6_flow |=
558 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK);
559 }
560 ip6->ip6_vfc &= ~IPV6_VERSION_MASK;
561 ip6->ip6_vfc |= IPV6_VERSION;
562
563 /*
564 * Compute the pseudo-header portion of the checksum
565 * now. We incrementally add in the TCP option and
566 * payload lengths later, and then compute the TCP
567 * checksum right before the packet is sent off onto
568 * the wire.
569 */
570 n->th_sum = in6_cksum_phdr(&in6p_laddr(inp),
571 &in6p_faddr(inp), htonl(sizeof(struct tcphdr)),
572 htonl(IPPROTO_TCP));
573 break;
574 }
575 #endif
576 }
577
578 n->th_sport = inp->inp_lport;
579 n->th_dport = inp->inp_fport;
580
581 n->th_seq = 0;
582 n->th_ack = 0;
583 n->th_x2 = 0;
584 n->th_off = 5;
585 n->th_flags = 0;
586 n->th_win = 0;
587 n->th_urp = 0;
588 return m;
589 }
590
591 /*
592 * Send a single message to the TCP at address specified by
593 * the given TCP/IP header. If m == 0, then we make a copy
594 * of the tcpiphdr at ti and send directly to the addressed host.
595 * This is used to force keep alive messages out using the TCP
596 * template for a connection tp->t_template. If flags are given
597 * then we send a message back to the TCP which originated the
598 * segment ti, and discard the mbuf containing it and any other
599 * attached mbufs.
600 *
601 * In any case the ack and sequence number of the transmitted
602 * segment are as specified by the parameters.
603 */
604 int
tcp_respond(struct tcpcb * tp,struct mbuf * mtemplate,struct mbuf * m,struct tcphdr * th0,tcp_seq ack,tcp_seq seq,int flags)605 tcp_respond(struct tcpcb *tp, struct mbuf *mtemplate, struct mbuf *m,
606 struct tcphdr *th0, tcp_seq ack, tcp_seq seq, int flags)
607 {
608 struct route *ro;
609 int error, tlen, win = 0;
610 int hlen;
611 struct ip *ip;
612 #ifdef INET6
613 struct ip6_hdr *ip6;
614 #endif
615 int family; /* family on packet, not inpcb! */
616 struct tcphdr *th;
617
618 if (tp != NULL && (flags & TH_RST) == 0) {
619 KASSERT(tp->t_inpcb != NULL);
620
621 win = sbspace(&tp->t_inpcb->inp_socket->so_rcv);
622 }
623
624 th = NULL; /* Quell uninitialized warning */
625 ip = NULL;
626 #ifdef INET6
627 ip6 = NULL;
628 #endif
629 if (m == NULL) {
630 if (!mtemplate)
631 return EINVAL;
632
633 /* get family information from template */
634 switch (mtod(mtemplate, struct ip *)->ip_v) {
635 case 4:
636 family = AF_INET;
637 hlen = sizeof(struct ip);
638 break;
639 #ifdef INET6
640 case 6:
641 family = AF_INET6;
642 hlen = sizeof(struct ip6_hdr);
643 break;
644 #endif
645 default:
646 return EAFNOSUPPORT;
647 }
648
649 MGETHDR(m, M_DONTWAIT, MT_HEADER);
650 if (m) {
651 MCLAIM(m, &tcp_tx_mowner);
652 MCLGET(m, M_DONTWAIT);
653 if ((m->m_flags & M_EXT) == 0) {
654 m_free(m);
655 m = NULL;
656 }
657 }
658 if (m == NULL)
659 return ENOBUFS;
660
661 tlen = 0;
662
663 m->m_data += max_linkhdr;
664 bcopy(mtod(mtemplate, void *), mtod(m, void *),
665 mtemplate->m_len);
666 switch (family) {
667 case AF_INET:
668 ip = mtod(m, struct ip *);
669 th = (struct tcphdr *)(ip + 1);
670 break;
671 #ifdef INET6
672 case AF_INET6:
673 ip6 = mtod(m, struct ip6_hdr *);
674 th = (struct tcphdr *)(ip6 + 1);
675 break;
676 #endif
677 }
678 flags = TH_ACK;
679 } else {
680 if ((m->m_flags & M_PKTHDR) == 0) {
681 m_freem(m);
682 return EINVAL;
683 }
684 KASSERT(th0 != NULL);
685
686 /* get family information from m */
687 switch (mtod(m, struct ip *)->ip_v) {
688 case 4:
689 family = AF_INET;
690 hlen = sizeof(struct ip);
691 ip = mtod(m, struct ip *);
692 break;
693 #ifdef INET6
694 case 6:
695 family = AF_INET6;
696 hlen = sizeof(struct ip6_hdr);
697 ip6 = mtod(m, struct ip6_hdr *);
698 break;
699 #endif
700 default:
701 m_freem(m);
702 return EAFNOSUPPORT;
703 }
704 /* clear h/w csum flags inherited from rx packet */
705 m->m_pkthdr.csum_flags = 0;
706
707 if ((flags & TH_SYN) == 0 || sizeof(*th0) > (th0->th_off << 2))
708 tlen = sizeof(*th0);
709 else
710 tlen = th0->th_off << 2;
711
712 if (m->m_len > hlen + tlen && (m->m_flags & M_EXT) == 0 &&
713 mtod(m, char *) + hlen == (char *)th0) {
714 m->m_len = hlen + tlen;
715 m_freem(m->m_next);
716 m->m_next = NULL;
717 } else {
718 struct mbuf *n;
719
720 KASSERT(max_linkhdr + hlen + tlen <= MCLBYTES);
721
722 MGETHDR(n, M_DONTWAIT, MT_HEADER);
723 if (n && max_linkhdr + hlen + tlen > MHLEN) {
724 MCLGET(n, M_DONTWAIT);
725 if ((n->m_flags & M_EXT) == 0) {
726 m_freem(n);
727 n = NULL;
728 }
729 }
730 if (!n) {
731 m_freem(m);
732 return ENOBUFS;
733 }
734
735 MCLAIM(n, &tcp_tx_mowner);
736 n->m_data += max_linkhdr;
737 n->m_len = hlen + tlen;
738 m_copyback(n, 0, hlen, mtod(m, void *));
739 m_copyback(n, hlen, tlen, (void *)th0);
740
741 m_freem(m);
742 m = n;
743 n = NULL;
744 }
745
746 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
747 switch (family) {
748 case AF_INET:
749 ip = mtod(m, struct ip *);
750 th = (struct tcphdr *)(ip + 1);
751 ip->ip_p = IPPROTO_TCP;
752 xchg(ip->ip_dst, ip->ip_src, struct in_addr);
753 ip->ip_p = IPPROTO_TCP;
754 break;
755 #ifdef INET6
756 case AF_INET6:
757 ip6 = mtod(m, struct ip6_hdr *);
758 th = (struct tcphdr *)(ip6 + 1);
759 ip6->ip6_nxt = IPPROTO_TCP;
760 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
761 ip6->ip6_nxt = IPPROTO_TCP;
762 break;
763 #endif
764 }
765 xchg(th->th_dport, th->th_sport, u_int16_t);
766 #undef xchg
767 tlen = 0; /*be friendly with the following code*/
768 }
769 th->th_seq = htonl(seq);
770 th->th_ack = htonl(ack);
771 th->th_x2 = 0;
772 if ((flags & TH_SYN) == 0) {
773 if (tp)
774 win >>= tp->rcv_scale;
775 if (win > TCP_MAXWIN)
776 win = TCP_MAXWIN;
777 th->th_win = htons((u_int16_t)win);
778 th->th_off = sizeof (struct tcphdr) >> 2;
779 tlen += sizeof(*th);
780 } else {
781 tlen += th->th_off << 2;
782 }
783 m->m_len = hlen + tlen;
784 m->m_pkthdr.len = hlen + tlen;
785 m_reset_rcvif(m);
786 th->th_flags = flags;
787 th->th_urp = 0;
788
789 switch (family) {
790 case AF_INET:
791 {
792 struct ipovly *ipov = (struct ipovly *)ip;
793 memset(ipov->ih_x1, 0, sizeof ipov->ih_x1);
794 ipov->ih_len = htons((u_int16_t)tlen);
795
796 th->th_sum = 0;
797 th->th_sum = in_cksum(m, hlen + tlen);
798 ip->ip_len = htons(hlen + tlen);
799 ip->ip_ttl = ip_defttl;
800 break;
801 }
802 #ifdef INET6
803 case AF_INET6:
804 {
805 th->th_sum = 0;
806 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr),
807 tlen);
808 ip6->ip6_plen = htons(tlen);
809 if (tp && tp->t_inpcb->inp_af == AF_INET6)
810 ip6->ip6_hlim = in6pcb_selecthlim_rt(tp->t_inpcb);
811 else
812 ip6->ip6_hlim = ip6_defhlim;
813 ip6->ip6_flow &= ~IPV6_FLOWINFO_MASK;
814 if (ip6_auto_flowlabel) {
815 ip6->ip6_flow |=
816 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK);
817 }
818 break;
819 }
820 #endif
821 }
822
823 if (tp != NULL && tp->t_inpcb->inp_af == AF_INET) {
824 ro = &tp->t_inpcb->inp_route;
825 KASSERT(family == AF_INET);
826 KASSERT(in_hosteq(ip->ip_dst, in4p_faddr(tp->t_inpcb)));
827 }
828 #ifdef INET6
829 else if (tp != NULL && tp->t_inpcb->inp_af == AF_INET6) {
830 ro = (struct route *)&tp->t_inpcb->inp_route;
831
832 #ifdef DIAGNOSTIC
833 if (family == AF_INET) {
834 if (!IN6_IS_ADDR_V4MAPPED(&in6p_faddr(tp->t_inpcb)))
835 panic("tcp_respond: not mapped addr");
836 if (memcmp(&ip->ip_dst,
837 &in6p_faddr(tp->t_inpcb).s6_addr32[3],
838 sizeof(ip->ip_dst)) != 0) {
839 panic("tcp_respond: ip_dst != in6p_faddr");
840 }
841 } else if (family == AF_INET6) {
842 if (!IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst,
843 &in6p_faddr(tp->t_inpcb)))
844 panic("tcp_respond: ip6_dst != in6p_faddr");
845 } else
846 panic("tcp_respond: address family mismatch");
847 #endif
848 }
849 #endif
850 else
851 ro = NULL;
852
853 switch (family) {
854 case AF_INET:
855 error = ip_output(m, NULL, ro,
856 (tp && tp->t_mtudisc ? IP_MTUDISC : 0), NULL,
857 tp ? tp->t_inpcb : NULL);
858 break;
859 #ifdef INET6
860 case AF_INET6:
861 error = ip6_output(m, NULL, ro, 0, NULL,
862 tp ? tp->t_inpcb : NULL, NULL);
863 break;
864 #endif
865 default:
866 error = EAFNOSUPPORT;
867 break;
868 }
869
870 return error;
871 }
872
873 /*
874 * Template TCPCB. Rather than zeroing a new TCPCB and initializing
875 * a bunch of members individually, we maintain this template for the
876 * static and mostly-static components of the TCPCB, and copy it into
877 * the new TCPCB instead.
878 */
879 static struct tcpcb tcpcb_template = {
880 .t_srtt = TCPTV_SRTTBASE,
881 .t_rttmin = TCPTV_MIN,
882
883 .snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT,
884 .snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT,
885 .snd_numholes = 0,
886 .snd_cubic_wmax = 0,
887 .snd_cubic_wmax_last = 0,
888 .snd_cubic_ctime = 0,
889
890 .t_partialacks = -1,
891 .t_bytes_acked = 0,
892 .t_sndrexmitpack = 0,
893 .t_rcvoopack = 0,
894 .t_sndzerowin = 0,
895 };
896
897 /*
898 * Updates the TCPCB template whenever a parameter that would affect
899 * the template is changed.
900 */
901 void
tcp_tcpcb_template(void)902 tcp_tcpcb_template(void)
903 {
904 struct tcpcb *tp = &tcpcb_template;
905 int flags;
906
907 tp->t_peermss = tcp_mssdflt;
908 tp->t_ourmss = tcp_mssdflt;
909 tp->t_segsz = tcp_mssdflt;
910
911 flags = 0;
912 if (tcp_do_rfc1323 && tcp_do_win_scale)
913 flags |= TF_REQ_SCALE;
914 if (tcp_do_rfc1323 && tcp_do_timestamps)
915 flags |= TF_REQ_TSTMP;
916 tp->t_flags = flags;
917
918 /*
919 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
920 * rtt estimate. Set rttvar so that srtt + 2 * rttvar gives
921 * reasonable initial retransmit time.
922 */
923 tp->t_rttvar = tcp_rttdflt * PR_SLOWHZ << (TCP_RTTVAR_SHIFT + 2 - 1);
924 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
925 TCPTV_MIN, TCPTV_REXMTMAX);
926
927 /* Keep Alive */
928 tp->t_keepinit = MIN(tcp_keepinit, TCP_TIMER_MAXTICKS);
929 tp->t_keepidle = MIN(tcp_keepidle, TCP_TIMER_MAXTICKS);
930 tp->t_keepintvl = MIN(tcp_keepintvl, TCP_TIMER_MAXTICKS);
931 tp->t_keepcnt = MAX(1, MIN(tcp_keepcnt, TCP_TIMER_MAXTICKS));
932 tp->t_maxidle = tp->t_keepcnt * MIN(tp->t_keepintvl,
933 TCP_TIMER_MAXTICKS/tp->t_keepcnt);
934
935 /* MSL */
936 tp->t_msl = TCPTV_MSL;
937 }
938
939 /*
940 * Create a new TCP control block, making an
941 * empty reassembly queue and hooking it to the argument
942 * protocol control block.
943 */
944 struct tcpcb *
tcp_newtcpcb(int family,struct inpcb * inp)945 tcp_newtcpcb(int family, struct inpcb *inp)
946 {
947 struct tcpcb *tp;
948 int i;
949
950 /* XXX Consider using a pool_cache for speed. */
951 tp = pool_get(&tcpcb_pool, PR_NOWAIT); /* splsoftnet via tcp_usrreq */
952 if (tp == NULL)
953 return NULL;
954 memcpy(tp, &tcpcb_template, sizeof(*tp));
955 TAILQ_INIT(&tp->segq);
956 TAILQ_INIT(&tp->timeq);
957 tp->t_family = family; /* may be overridden later on */
958 TAILQ_INIT(&tp->snd_holes);
959 LIST_INIT(&tp->t_sc); /* XXX can template this */
960
961 /* Don't sweat this loop; hopefully the compiler will unroll it. */
962 for (i = 0; i < TCPT_NTIMERS; i++) {
963 callout_init(&tp->t_timer[i], CALLOUT_MPSAFE);
964 TCP_TIMER_INIT(tp, i);
965 }
966 callout_init(&tp->t_delack_ch, CALLOUT_MPSAFE);
967
968 switch (family) {
969 case AF_INET:
970 in4p_ip(inp).ip_ttl = ip_defttl;
971 inp->inp_ppcb = (void *)tp;
972
973 tp->t_inpcb = inp;
974 tp->t_mtudisc = ip_mtudisc;
975 break;
976 #ifdef INET6
977 case AF_INET6:
978 in6p_ip6(inp).ip6_hlim = in6pcb_selecthlim_rt(inp);
979 inp->inp_ppcb = (void *)tp;
980
981 tp->t_inpcb = inp;
982 /* for IPv6, always try to run path MTU discovery */
983 tp->t_mtudisc = 1;
984 break;
985 #endif /* INET6 */
986 default:
987 for (i = 0; i < TCPT_NTIMERS; i++)
988 callout_destroy(&tp->t_timer[i]);
989 callout_destroy(&tp->t_delack_ch);
990 pool_put(&tcpcb_pool, tp); /* splsoftnet via tcp_usrreq */
991 return NULL;
992 }
993
994 /*
995 * Initialize our timebase. When we send timestamps, we take
996 * the delta from tcp_now -- this means each connection always
997 * gets a timebase of 1, which makes it, among other things,
998 * more difficult to determine how long a system has been up,
999 * and thus how many TCP sequence increments have occurred.
1000 *
1001 * We start with 1, because 0 doesn't work with linux, which
1002 * considers timestamp 0 in a SYN packet as a bug and disables
1003 * timestamps.
1004 */
1005 tp->ts_timebase = tcp_now - 1;
1006
1007 tcp_congctl_select(tp, tcp_congctl_global_name);
1008
1009 return tp;
1010 }
1011
1012 /*
1013 * Drop a TCP connection, reporting
1014 * the specified error. If connection is synchronized,
1015 * then send a RST to peer.
1016 */
1017 struct tcpcb *
tcp_drop(struct tcpcb * tp,int errno)1018 tcp_drop(struct tcpcb *tp, int errno)
1019 {
1020 struct socket *so;
1021
1022 KASSERT(tp->t_inpcb != NULL);
1023
1024 so = tp->t_inpcb->inp_socket;
1025 if (so == NULL)
1026 return NULL;
1027
1028 if (TCPS_HAVERCVDSYN(tp->t_state)) {
1029 tp->t_state = TCPS_CLOSED;
1030 (void) tcp_output(tp);
1031 TCP_STATINC(TCP_STAT_DROPS);
1032 } else
1033 TCP_STATINC(TCP_STAT_CONNDROPS);
1034 if (errno == ETIMEDOUT && tp->t_softerror)
1035 errno = tp->t_softerror;
1036 so->so_error = errno;
1037 return (tcp_close(tp));
1038 }
1039
1040 /*
1041 * Close a TCP control block:
1042 * discard all space held by the tcp
1043 * discard internet protocol block
1044 * wake up any sleepers
1045 */
1046 struct tcpcb *
tcp_close(struct tcpcb * tp)1047 tcp_close(struct tcpcb *tp)
1048 {
1049 struct inpcb *inp;
1050 struct socket *so;
1051 #ifdef RTV_RTT
1052 struct rtentry *rt = NULL;
1053 #endif
1054 struct route *ro;
1055 int j;
1056
1057 inp = tp->t_inpcb;
1058 so = inp->inp_socket;
1059 ro = &inp->inp_route;
1060
1061 #ifdef RTV_RTT
1062 /*
1063 * If we sent enough data to get some meaningful characteristics,
1064 * save them in the routing entry. 'Enough' is arbitrarily
1065 * defined as the sendpipesize (default 4K) * 16. This would
1066 * give us 16 rtt samples assuming we only get one sample per
1067 * window (the usual case on a long haul net). 16 samples is
1068 * enough for the srtt filter to converge to within 5% of the correct
1069 * value; fewer samples and we could save a very bogus rtt.
1070 *
1071 * Don't update the default route's characteristics and don't
1072 * update anything that the user "locked".
1073 */
1074 if (SEQ_LT(tp->iss + so->so_snd.sb_hiwat * 16, tp->snd_max) &&
1075 ro && (rt = rtcache_validate(ro)) != NULL &&
1076 !in_nullhost(satocsin(rt_getkey(rt))->sin_addr)) {
1077 u_long i = 0;
1078
1079 if ((rt->rt_rmx.rmx_locks & RTV_RTT) == 0) {
1080 i = tp->t_srtt *
1081 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
1082 if (rt->rt_rmx.rmx_rtt && i)
1083 /*
1084 * filter this update to half the old & half
1085 * the new values, converting scale.
1086 * See route.h and tcp_var.h for a
1087 * description of the scaling constants.
1088 */
1089 rt->rt_rmx.rmx_rtt =
1090 (rt->rt_rmx.rmx_rtt + i) / 2;
1091 else
1092 rt->rt_rmx.rmx_rtt = i;
1093 }
1094 if ((rt->rt_rmx.rmx_locks & RTV_RTTVAR) == 0) {
1095 i = tp->t_rttvar *
1096 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTTVAR_SHIFT + 2));
1097 if (rt->rt_rmx.rmx_rttvar && i)
1098 rt->rt_rmx.rmx_rttvar =
1099 (rt->rt_rmx.rmx_rttvar + i) / 2;
1100 else
1101 rt->rt_rmx.rmx_rttvar = i;
1102 }
1103 /*
1104 * update the pipelimit (ssthresh) if it has been updated
1105 * already or if a pipesize was specified & the threshold
1106 * got below half the pipesize. I.e., wait for bad news
1107 * before we start updating, then update on both good
1108 * and bad news.
1109 */
1110 if (((rt->rt_rmx.rmx_locks & RTV_SSTHRESH) == 0 &&
1111 (i = tp->snd_ssthresh) && rt->rt_rmx.rmx_ssthresh) ||
1112 i < (rt->rt_rmx.rmx_sendpipe / 2)) {
1113 /*
1114 * convert the limit from user data bytes to
1115 * packets then to packet data bytes.
1116 */
1117 i = (i + tp->t_segsz / 2) / tp->t_segsz;
1118 if (i < 2)
1119 i = 2;
1120 i *= (u_long)(tp->t_segsz + sizeof (struct tcpiphdr));
1121 if (rt->rt_rmx.rmx_ssthresh)
1122 rt->rt_rmx.rmx_ssthresh =
1123 (rt->rt_rmx.rmx_ssthresh + i) / 2;
1124 else
1125 rt->rt_rmx.rmx_ssthresh = i;
1126 }
1127 }
1128 rtcache_unref(rt, ro);
1129 #endif /* RTV_RTT */
1130 /* free the reassembly queue, if any */
1131 TCP_REASS_LOCK(tp);
1132 (void) tcp_freeq(tp);
1133 TCP_REASS_UNLOCK(tp);
1134
1135 /* free the SACK holes list. */
1136 tcp_free_sackholes(tp);
1137 tcp_congctl_release(tp);
1138 syn_cache_cleanup(tp);
1139
1140 if (tp->t_template) {
1141 m_free(tp->t_template);
1142 tp->t_template = NULL;
1143 }
1144
1145 /*
1146 * Detaching the pcb will unlock the socket/tcpcb, and stopping
1147 * the timers can also drop the lock. We need to prevent access
1148 * to the tcpcb as it's half torn down. Flag the pcb as dead
1149 * (prevents access by timers) and only then detach it.
1150 */
1151 tp->t_flags |= TF_DEAD;
1152 inp->inp_ppcb = NULL;
1153 soisdisconnected(so);
1154 inpcb_destroy(inp);
1155 /*
1156 * pcb is no longer visble elsewhere, so we can safely release
1157 * the lock in callout_halt() if needed.
1158 */
1159 TCP_STATINC(TCP_STAT_CLOSED);
1160 for (j = 0; j < TCPT_NTIMERS; j++) {
1161 callout_halt(&tp->t_timer[j], softnet_lock);
1162 callout_destroy(&tp->t_timer[j]);
1163 }
1164 callout_halt(&tp->t_delack_ch, softnet_lock);
1165 callout_destroy(&tp->t_delack_ch);
1166 pool_put(&tcpcb_pool, tp);
1167
1168 return NULL;
1169 }
1170
1171 int
tcp_freeq(struct tcpcb * tp)1172 tcp_freeq(struct tcpcb *tp)
1173 {
1174 struct ipqent *qe;
1175 int rv = 0;
1176
1177 TCP_REASS_LOCK_CHECK(tp);
1178
1179 while ((qe = TAILQ_FIRST(&tp->segq)) != NULL) {
1180 TAILQ_REMOVE(&tp->segq, qe, ipqe_q);
1181 TAILQ_REMOVE(&tp->timeq, qe, ipqe_timeq);
1182 m_freem(qe->ipqe_m);
1183 tcpipqent_free(qe);
1184 rv = 1;
1185 }
1186 tp->t_segqlen = 0;
1187 KASSERT(TAILQ_EMPTY(&tp->timeq));
1188 return (rv);
1189 }
1190
1191 void
tcp_fasttimo(void)1192 tcp_fasttimo(void)
1193 {
1194 if (tcp_drainwanted) {
1195 tcp_drain();
1196 tcp_drainwanted = 0;
1197 }
1198 }
1199
1200 void
tcp_drainstub(void)1201 tcp_drainstub(void)
1202 {
1203 tcp_drainwanted = 1;
1204 }
1205
1206 /*
1207 * Protocol drain routine. Called when memory is in short supply.
1208 * Called from pr_fasttimo thus a callout context.
1209 */
1210 void
tcp_drain(void)1211 tcp_drain(void)
1212 {
1213 struct inpcb *inp;
1214 struct tcpcb *tp;
1215
1216 mutex_enter(softnet_lock);
1217 KERNEL_LOCK(1, NULL);
1218
1219 /*
1220 * Free the sequence queue of all TCP connections.
1221 */
1222 TAILQ_FOREACH(inp, &tcbtable.inpt_queue, inp_queue) {
1223 tp = intotcpcb(inp);
1224 if (tp != NULL) {
1225 /*
1226 * If the tcpcb is already busy,
1227 * just bail out now.
1228 */
1229 if (tcp_reass_lock_try(tp) == 0)
1230 continue;
1231 if (tcp_freeq(tp))
1232 TCP_STATINC(TCP_STAT_CONNSDRAINED);
1233 TCP_REASS_UNLOCK(tp);
1234 }
1235 }
1236
1237 KERNEL_UNLOCK_ONE(NULL);
1238 mutex_exit(softnet_lock);
1239 }
1240
1241 /*
1242 * Notify a tcp user of an asynchronous error;
1243 * store error as soft error, but wake up user
1244 * (for now, won't do anything until can select for soft error).
1245 */
1246 void
tcp_notify(struct inpcb * inp,int error)1247 tcp_notify(struct inpcb *inp, int error)
1248 {
1249 struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
1250 struct socket *so = inp->inp_socket;
1251
1252 /*
1253 * Ignore some errors if we are hooked up.
1254 * If connection hasn't completed, has retransmitted several times,
1255 * and receives a second error, give up now. This is better
1256 * than waiting a long time to establish a connection that
1257 * can never complete.
1258 */
1259 if (tp->t_state == TCPS_ESTABLISHED &&
1260 (error == EHOSTUNREACH || error == ENETUNREACH ||
1261 error == EHOSTDOWN)) {
1262 return;
1263 } else if (TCPS_HAVEESTABLISHED(tp->t_state) == 0 &&
1264 tp->t_rxtshift > 3 && tp->t_softerror)
1265 so->so_error = error;
1266 else
1267 tp->t_softerror = error;
1268 cv_broadcast(&so->so_cv);
1269 sorwakeup(so);
1270 sowwakeup(so);
1271 }
1272
1273 #ifdef INET6
1274 void *
tcp6_ctlinput(int cmd,const struct sockaddr * sa,void * d)1275 tcp6_ctlinput(int cmd, const struct sockaddr *sa, void *d)
1276 {
1277 struct tcphdr th;
1278 void (*notify)(struct inpcb *, int) = tcp_notify;
1279 int nmatch;
1280 struct ip6_hdr *ip6;
1281 const struct sockaddr_in6 *sa6_src = NULL;
1282 const struct sockaddr_in6 *sa6 = (const struct sockaddr_in6 *)sa;
1283 struct mbuf *m;
1284 int off;
1285
1286 if (sa->sa_family != AF_INET6 ||
1287 sa->sa_len != sizeof(struct sockaddr_in6))
1288 return NULL;
1289 if ((unsigned)cmd >= PRC_NCMDS)
1290 return NULL;
1291 else if (cmd == PRC_QUENCH) {
1292 /*
1293 * Don't honor ICMP Source Quench messages meant for
1294 * TCP connections.
1295 */
1296 return NULL;
1297 } else if (PRC_IS_REDIRECT(cmd))
1298 notify = in6pcb_rtchange, d = NULL;
1299 else if (cmd == PRC_MSGSIZE)
1300 ; /* special code is present, see below */
1301 else if (cmd == PRC_HOSTDEAD)
1302 d = NULL;
1303 else if (inet6ctlerrmap[cmd] == 0)
1304 return NULL;
1305
1306 /* if the parameter is from icmp6, decode it. */
1307 if (d != NULL) {
1308 struct ip6ctlparam *ip6cp = (struct ip6ctlparam *)d;
1309 m = ip6cp->ip6c_m;
1310 ip6 = ip6cp->ip6c_ip6;
1311 off = ip6cp->ip6c_off;
1312 sa6_src = ip6cp->ip6c_src;
1313 } else {
1314 m = NULL;
1315 ip6 = NULL;
1316 sa6_src = &sa6_any;
1317 off = 0;
1318 }
1319
1320 if (ip6) {
1321 /* check if we can safely examine src and dst ports */
1322 if (m->m_pkthdr.len < off + sizeof(th)) {
1323 if (cmd == PRC_MSGSIZE)
1324 icmp6_mtudisc_update((struct ip6ctlparam *)d, 0);
1325 return NULL;
1326 }
1327
1328 memset(&th, 0, sizeof(th));
1329 m_copydata(m, off, sizeof(th), (void *)&th);
1330
1331 if (cmd == PRC_MSGSIZE) {
1332 int valid = 0;
1333
1334 /*
1335 * Check to see if we have a valid TCP connection
1336 * corresponding to the address in the ICMPv6 message
1337 * payload.
1338 */
1339 if (in6pcb_lookup(&tcbtable, &sa6->sin6_addr,
1340 th.th_dport,
1341 (const struct in6_addr *)&sa6_src->sin6_addr,
1342 th.th_sport, 0, 0))
1343 valid++;
1344
1345 /*
1346 * Depending on the value of "valid" and routing table
1347 * size (mtudisc_{hi,lo}wat), we will:
1348 * - recalcurate the new MTU and create the
1349 * corresponding routing entry, or
1350 * - ignore the MTU change notification.
1351 */
1352 icmp6_mtudisc_update((struct ip6ctlparam *)d, valid);
1353
1354 /*
1355 * no need to call in6pcb_notify, it should have been
1356 * called via callback if necessary
1357 */
1358 return NULL;
1359 }
1360
1361 nmatch = in6pcb_notify(&tcbtable, sa, th.th_dport,
1362 (const struct sockaddr *)sa6_src, th.th_sport, cmd, NULL, notify);
1363 if (nmatch == 0 && syn_cache_count &&
1364 (inet6ctlerrmap[cmd] == EHOSTUNREACH ||
1365 inet6ctlerrmap[cmd] == ENETUNREACH ||
1366 inet6ctlerrmap[cmd] == EHOSTDOWN))
1367 syn_cache_unreach((const struct sockaddr *)sa6_src,
1368 sa, &th);
1369 } else {
1370 (void) in6pcb_notify(&tcbtable, sa, 0,
1371 (const struct sockaddr *)sa6_src, 0, cmd, NULL, notify);
1372 }
1373
1374 return NULL;
1375 }
1376 #endif
1377
1378 /* assumes that ip header and tcp header are contiguous on mbuf */
1379 void *
tcp_ctlinput(int cmd,const struct sockaddr * sa,void * v)1380 tcp_ctlinput(int cmd, const struct sockaddr *sa, void *v)
1381 {
1382 struct ip *ip = v;
1383 struct tcphdr *th;
1384 struct icmp *icp;
1385 extern const int inetctlerrmap[];
1386 void (*notify)(struct inpcb *, int) = tcp_notify;
1387 int errno;
1388 int nmatch;
1389 struct tcpcb *tp;
1390 u_int mtu;
1391 tcp_seq seq;
1392 struct inpcb *inp;
1393 #ifdef INET6
1394 struct in6_addr src6, dst6;
1395 #endif
1396
1397 if (sa->sa_family != AF_INET ||
1398 sa->sa_len != sizeof(struct sockaddr_in))
1399 return NULL;
1400 if ((unsigned)cmd >= PRC_NCMDS)
1401 return NULL;
1402 errno = inetctlerrmap[cmd];
1403 if (cmd == PRC_QUENCH)
1404 /*
1405 * Don't honor ICMP Source Quench messages meant for
1406 * TCP connections.
1407 */
1408 return NULL;
1409 else if (PRC_IS_REDIRECT(cmd))
1410 notify = inpcb_rtchange, ip = 0;
1411 else if (cmd == PRC_MSGSIZE && ip && ip->ip_v == 4) {
1412 /*
1413 * Check to see if we have a valid TCP connection
1414 * corresponding to the address in the ICMP message
1415 * payload.
1416 *
1417 * Boundary check is made in icmp_input(), with ICMP_ADVLENMIN.
1418 */
1419 th = (struct tcphdr *)((char *)ip + (ip->ip_hl << 2));
1420 #ifdef INET6
1421 in6_in_2_v4mapin6(&ip->ip_src, &src6);
1422 in6_in_2_v4mapin6(&ip->ip_dst, &dst6);
1423 #endif
1424 if ((inp = inpcb_lookup(&tcbtable, ip->ip_dst,
1425 th->th_dport, ip->ip_src, th->th_sport, 0)) != NULL)
1426 ;
1427 #ifdef INET6
1428 else if ((inp = in6pcb_lookup(&tcbtable, &dst6,
1429 th->th_dport, &src6, th->th_sport, 0, 0)) != NULL)
1430 ;
1431 #endif
1432 else
1433 return NULL;
1434
1435 /*
1436 * Now that we've validated that we are actually communicating
1437 * with the host indicated in the ICMP message, locate the
1438 * ICMP header, recalculate the new MTU, and create the
1439 * corresponding routing entry.
1440 */
1441 icp = (struct icmp *)((char *)ip -
1442 offsetof(struct icmp, icmp_ip));
1443 tp = intotcpcb(inp);
1444 if (tp == NULL)
1445 return NULL;
1446 seq = ntohl(th->th_seq);
1447 if (SEQ_LT(seq, tp->snd_una) || SEQ_GT(seq, tp->snd_max))
1448 return NULL;
1449 /*
1450 * If the ICMP message advertises a Next-Hop MTU
1451 * equal or larger than the maximum packet size we have
1452 * ever sent, drop the message.
1453 */
1454 mtu = (u_int)ntohs(icp->icmp_nextmtu);
1455 if (mtu >= tp->t_pmtud_mtu_sent)
1456 return NULL;
1457 if (mtu >= tcp_hdrsz(tp) + tp->t_pmtud_mss_acked) {
1458 /*
1459 * Calculate new MTU, and create corresponding
1460 * route (traditional PMTUD).
1461 */
1462 tp->t_flags &= ~TF_PMTUD_PEND;
1463 icmp_mtudisc(icp, ip->ip_dst);
1464 } else {
1465 /*
1466 * Record the information got in the ICMP
1467 * message; act on it later.
1468 * If we had already recorded an ICMP message,
1469 * replace the old one only if the new message
1470 * refers to an older TCP segment
1471 */
1472 if (tp->t_flags & TF_PMTUD_PEND) {
1473 if (SEQ_LT(tp->t_pmtud_th_seq, seq))
1474 return NULL;
1475 } else
1476 tp->t_flags |= TF_PMTUD_PEND;
1477 tp->t_pmtud_th_seq = seq;
1478 tp->t_pmtud_nextmtu = icp->icmp_nextmtu;
1479 tp->t_pmtud_ip_len = icp->icmp_ip.ip_len;
1480 tp->t_pmtud_ip_hl = icp->icmp_ip.ip_hl;
1481 }
1482 return NULL;
1483 } else if (cmd == PRC_HOSTDEAD)
1484 ip = 0;
1485 else if (errno == 0)
1486 return NULL;
1487 if (ip && ip->ip_v == 4 && sa->sa_family == AF_INET) {
1488 th = (struct tcphdr *)((char *)ip + (ip->ip_hl << 2));
1489 nmatch = inpcb_notify(&tcbtable, satocsin(sa)->sin_addr,
1490 th->th_dport, ip->ip_src, th->th_sport, errno, notify);
1491 if (nmatch == 0 && syn_cache_count &&
1492 (inetctlerrmap[cmd] == EHOSTUNREACH ||
1493 inetctlerrmap[cmd] == ENETUNREACH ||
1494 inetctlerrmap[cmd] == EHOSTDOWN)) {
1495 struct sockaddr_in sin;
1496 memset(&sin, 0, sizeof(sin));
1497 sin.sin_len = sizeof(sin);
1498 sin.sin_family = AF_INET;
1499 sin.sin_port = th->th_sport;
1500 sin.sin_addr = ip->ip_src;
1501 syn_cache_unreach((struct sockaddr *)&sin, sa, th);
1502 }
1503
1504 /* XXX mapped address case */
1505 } else
1506 inpcb_notifyall(&tcbtable, satocsin(sa)->sin_addr, errno,
1507 notify);
1508 return NULL;
1509 }
1510
1511 /*
1512 * When a source quench is received, we are being notified of congestion.
1513 * Close the congestion window down to the Loss Window (one segment).
1514 * We will gradually open it again as we proceed.
1515 */
1516 void
tcp_quench(struct inpcb * inp)1517 tcp_quench(struct inpcb *inp)
1518 {
1519 struct tcpcb *tp = intotcpcb(inp);
1520
1521 if (tp) {
1522 tp->snd_cwnd = tp->t_segsz;
1523 tp->t_bytes_acked = 0;
1524 }
1525 }
1526
1527 /*
1528 * Path MTU Discovery handlers.
1529 */
1530 void
tcp_mtudisc_callback(struct in_addr faddr)1531 tcp_mtudisc_callback(struct in_addr faddr)
1532 {
1533 #ifdef INET6
1534 struct in6_addr in6;
1535 #endif
1536
1537 inpcb_notifyall(&tcbtable, faddr, EMSGSIZE, tcp_mtudisc);
1538 #ifdef INET6
1539 in6_in_2_v4mapin6(&faddr, &in6);
1540 tcp6_mtudisc_callback(&in6);
1541 #endif
1542 }
1543
1544 /*
1545 * On receipt of path MTU corrections, flush old route and replace it
1546 * with the new one. Retransmit all unacknowledged packets, to ensure
1547 * that all packets will be received.
1548 */
1549 void
tcp_mtudisc(struct inpcb * inp,int errno)1550 tcp_mtudisc(struct inpcb *inp, int errno)
1551 {
1552 struct tcpcb *tp = intotcpcb(inp);
1553 struct rtentry *rt;
1554
1555 if (tp == NULL)
1556 return;
1557
1558 rt = inpcb_rtentry(inp);
1559 if (rt != NULL) {
1560 /*
1561 * If this was not a host route, remove and realloc.
1562 */
1563 if ((rt->rt_flags & RTF_HOST) == 0) {
1564 inpcb_rtentry_unref(rt, inp);
1565 inpcb_rtchange(inp, errno);
1566 if ((rt = inpcb_rtentry(inp)) == NULL)
1567 return;
1568 }
1569
1570 /*
1571 * Slow start out of the error condition. We
1572 * use the MTU because we know it's smaller
1573 * than the previously transmitted segment.
1574 *
1575 * Note: This is more conservative than the
1576 * suggestion in draft-floyd-incr-init-win-03.
1577 */
1578 if (rt->rt_rmx.rmx_mtu != 0)
1579 tp->snd_cwnd =
1580 TCP_INITIAL_WINDOW(tcp_init_win,
1581 rt->rt_rmx.rmx_mtu);
1582 inpcb_rtentry_unref(rt, inp);
1583 }
1584
1585 /*
1586 * Resend unacknowledged packets.
1587 */
1588 tp->snd_nxt = tp->sack_newdata = tp->snd_una;
1589 tcp_output(tp);
1590 }
1591
1592 #ifdef INET6
1593 /*
1594 * Path MTU Discovery handlers.
1595 */
1596 void
tcp6_mtudisc_callback(struct in6_addr * faddr)1597 tcp6_mtudisc_callback(struct in6_addr *faddr)
1598 {
1599 struct sockaddr_in6 sin6;
1600
1601 memset(&sin6, 0, sizeof(sin6));
1602 sin6.sin6_family = AF_INET6;
1603 sin6.sin6_len = sizeof(struct sockaddr_in6);
1604 sin6.sin6_addr = *faddr;
1605 (void) in6pcb_notify(&tcbtable, (struct sockaddr *)&sin6, 0,
1606 (const struct sockaddr *)&sa6_any, 0, PRC_MSGSIZE, NULL, tcp6_mtudisc);
1607 }
1608
1609 void
tcp6_mtudisc(struct inpcb * inp,int errno)1610 tcp6_mtudisc(struct inpcb *inp, int errno)
1611 {
1612 struct tcpcb *tp = intotcpcb(inp);
1613 struct rtentry *rt;
1614
1615 if (tp == NULL)
1616 return;
1617
1618 rt = in6pcb_rtentry(inp);
1619 if (rt != NULL) {
1620 /*
1621 * If this was not a host route, remove and realloc.
1622 */
1623 if ((rt->rt_flags & RTF_HOST) == 0) {
1624 in6pcb_rtentry_unref(rt, inp);
1625 in6pcb_rtchange(inp, errno);
1626 rt = in6pcb_rtentry(inp);
1627 if (rt == NULL)
1628 return;
1629 }
1630
1631 /*
1632 * Slow start out of the error condition. We
1633 * use the MTU because we know it's smaller
1634 * than the previously transmitted segment.
1635 *
1636 * Note: This is more conservative than the
1637 * suggestion in draft-floyd-incr-init-win-03.
1638 */
1639 if (rt->rt_rmx.rmx_mtu != 0) {
1640 tp->snd_cwnd = TCP_INITIAL_WINDOW(tcp_init_win,
1641 rt->rt_rmx.rmx_mtu);
1642 }
1643 in6pcb_rtentry_unref(rt, inp);
1644 }
1645
1646 /*
1647 * Resend unacknowledged packets.
1648 */
1649 tp->snd_nxt = tp->sack_newdata = tp->snd_una;
1650 tcp_output(tp);
1651 }
1652 #endif /* INET6 */
1653
1654 /*
1655 * Compute the MSS to advertise to the peer. Called only during
1656 * the 3-way handshake. If we are the server (peer initiated
1657 * connection), we are called with a pointer to the interface
1658 * on which the SYN packet arrived. If we are the client (we
1659 * initiated connection), we are called with a pointer to the
1660 * interface out which this connection should go.
1661 *
1662 * NOTE: Do not subtract IP option/extension header size nor IPsec
1663 * header size from MSS advertisement. MSS option must hold the maximum
1664 * segment size we can accept, so it must always be:
1665 * max(if mtu) - ip header - tcp header
1666 */
1667 u_long
tcp_mss_to_advertise(const struct ifnet * ifp,int af)1668 tcp_mss_to_advertise(const struct ifnet *ifp, int af)
1669 {
1670 extern u_long in_maxmtu;
1671 u_long mss = 0;
1672 u_long hdrsiz;
1673
1674 /*
1675 * In order to avoid defeating path MTU discovery on the peer,
1676 * we advertise the max MTU of all attached networks as our MSS,
1677 * per RFC 1191, section 3.1.
1678 *
1679 * We provide the option to advertise just the MTU of
1680 * the interface on which we hope this connection will
1681 * be receiving. If we are responding to a SYN, we
1682 * will have a pretty good idea about this, but when
1683 * initiating a connection there is a bit more doubt.
1684 *
1685 * We also need to ensure that loopback has a large enough
1686 * MSS, as the loopback MTU is never included in in_maxmtu.
1687 */
1688
1689 if (ifp != NULL)
1690 switch (af) {
1691 #ifdef INET6
1692 case AF_INET6: /* FALLTHROUGH */
1693 #endif
1694 case AF_INET:
1695 mss = ifp->if_mtu;
1696 break;
1697 }
1698
1699 if (tcp_mss_ifmtu == 0)
1700 switch (af) {
1701 #ifdef INET6
1702 case AF_INET6: /* FALLTHROUGH */
1703 #endif
1704 case AF_INET:
1705 mss = uimax(in_maxmtu, mss);
1706 break;
1707 }
1708
1709 switch (af) {
1710 case AF_INET:
1711 hdrsiz = sizeof(struct ip);
1712 break;
1713 #ifdef INET6
1714 case AF_INET6:
1715 hdrsiz = sizeof(struct ip6_hdr);
1716 break;
1717 #endif
1718 default:
1719 hdrsiz = 0;
1720 break;
1721 }
1722 hdrsiz += sizeof(struct tcphdr);
1723 if (mss > hdrsiz)
1724 mss -= hdrsiz;
1725
1726 mss = uimax(tcp_mssdflt, mss);
1727 return (mss);
1728 }
1729
1730 /*
1731 * Set connection variables based on the peer's advertised MSS.
1732 * We are passed the TCPCB for the actual connection. If we
1733 * are the server, we are called by the compressed state engine
1734 * when the 3-way handshake is complete. If we are the client,
1735 * we are called when we receive the SYN,ACK from the server.
1736 *
1737 * NOTE: Our advertised MSS value must be initialized in the TCPCB
1738 * before this routine is called!
1739 */
1740 void
tcp_mss_from_peer(struct tcpcb * tp,int offer)1741 tcp_mss_from_peer(struct tcpcb *tp, int offer)
1742 {
1743 struct socket *so;
1744 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1745 struct rtentry *rt;
1746 #endif
1747 u_long bufsize;
1748 int mss;
1749
1750 KASSERT(tp->t_inpcb != NULL);
1751
1752 so = NULL;
1753 rt = NULL;
1754
1755 so = tp->t_inpcb->inp_socket;
1756 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1757 rt = inpcb_rtentry(tp->t_inpcb);
1758 #endif
1759
1760 /*
1761 * As per RFC1122, use the default MSS value, unless they
1762 * sent us an offer. Do not accept offers less than 256 bytes.
1763 */
1764 mss = tcp_mssdflt;
1765 if (offer)
1766 mss = offer;
1767 mss = uimax(mss, 256); /* sanity */
1768 tp->t_peermss = mss;
1769 mss -= tcp_optlen(tp);
1770 if (tp->t_inpcb->inp_af == AF_INET)
1771 mss -= ip_optlen(tp->t_inpcb);
1772 #ifdef INET6
1773 if (tp->t_inpcb->inp_af == AF_INET6)
1774 mss -= ip6_optlen(tp->t_inpcb);
1775 #endif
1776 /*
1777 * XXX XXX What if mss goes negative or zero? This can happen if a
1778 * socket has large IPv6 options. We crash below.
1779 */
1780
1781 /*
1782 * If there's a pipesize, change the socket buffer to that size.
1783 * Make the socket buffer an integral number of MSS units. If
1784 * the MSS is larger than the socket buffer, artificially decrease
1785 * the MSS.
1786 */
1787 #ifdef RTV_SPIPE
1788 if (rt != NULL && rt->rt_rmx.rmx_sendpipe != 0)
1789 bufsize = rt->rt_rmx.rmx_sendpipe;
1790 else
1791 #endif
1792 {
1793 KASSERT(so != NULL);
1794 bufsize = so->so_snd.sb_hiwat;
1795 }
1796 if (bufsize < mss)
1797 mss = bufsize;
1798 else {
1799 bufsize = roundup(bufsize, mss);
1800 if (bufsize > sb_max)
1801 bufsize = sb_max;
1802 (void) sbreserve(&so->so_snd, bufsize, so);
1803 }
1804 tp->t_segsz = mss;
1805
1806 #ifdef RTV_SSTHRESH
1807 if (rt != NULL && rt->rt_rmx.rmx_ssthresh) {
1808 /*
1809 * There's some sort of gateway or interface buffer
1810 * limit on the path. Use this to set the slow
1811 * start threshold, but set the threshold to no less
1812 * than 2 * MSS.
1813 */
1814 tp->snd_ssthresh = uimax(2 * mss, rt->rt_rmx.rmx_ssthresh);
1815 }
1816 #endif
1817 #if defined(RTV_SPIPE) || defined(RTV_SSTHRESH)
1818 inpcb_rtentry_unref(rt, tp->t_inpcb);
1819 #endif
1820 }
1821
1822 /*
1823 * Processing necessary when a TCP connection is established.
1824 */
1825 void
tcp_established(struct tcpcb * tp)1826 tcp_established(struct tcpcb *tp)
1827 {
1828 struct socket *so;
1829 #ifdef RTV_RPIPE
1830 struct rtentry *rt;
1831 #endif
1832 u_long bufsize;
1833
1834 KASSERT(tp->t_inpcb != NULL);
1835
1836 so = NULL;
1837 rt = NULL;
1838
1839 /* This is a while() to reduce the dreadful stairstepping below */
1840 while (tp->t_inpcb->inp_af == AF_INET) {
1841 so = tp->t_inpcb->inp_socket;
1842 #if defined(RTV_RPIPE)
1843 rt = inpcb_rtentry(tp->t_inpcb);
1844 #endif
1845 if (__predict_true(tcp_msl_enable)) {
1846 if (in4p_laddr(tp->t_inpcb).s_addr == INADDR_LOOPBACK) {
1847 tp->t_msl = tcp_msl_loop ? tcp_msl_loop : (TCPTV_MSL >> 2);
1848 break;
1849 }
1850
1851 if (__predict_false(tcp_rttlocal)) {
1852 /* This may be adjusted by tcp_input */
1853 tp->t_msl = tcp_msl_local ? tcp_msl_local : (TCPTV_MSL >> 1);
1854 break;
1855 }
1856 if (in_localaddr(in4p_faddr(tp->t_inpcb))) {
1857 tp->t_msl = tcp_msl_local ? tcp_msl_local : (TCPTV_MSL >> 1);
1858 break;
1859 }
1860 }
1861 tp->t_msl = tcp_msl_remote ? tcp_msl_remote : TCPTV_MSL;
1862 break;
1863 }
1864
1865 /* Clamp to a reasonable range. */
1866 tp->t_msl = MIN(tp->t_msl, TCP_MAXMSL);
1867
1868 #ifdef INET6
1869 while (tp->t_inpcb->inp_af == AF_INET6) {
1870 so = tp->t_inpcb->inp_socket;
1871 #if defined(RTV_RPIPE)
1872 rt = in6pcb_rtentry(tp->t_inpcb);
1873 #endif
1874 if (__predict_true(tcp_msl_enable)) {
1875 extern const struct in6_addr in6addr_loopback;
1876
1877 if (IN6_ARE_ADDR_EQUAL(&in6p_laddr(tp->t_inpcb),
1878 &in6addr_loopback)) {
1879 tp->t_msl = tcp_msl_loop ? tcp_msl_loop : (TCPTV_MSL >> 2);
1880 break;
1881 }
1882
1883 if (__predict_false(tcp_rttlocal)) {
1884 /* This may be adjusted by tcp_input */
1885 tp->t_msl = tcp_msl_local ? tcp_msl_local : (TCPTV_MSL >> 1);
1886 break;
1887 }
1888 if (in6_localaddr(&in6p_faddr(tp->t_inpcb))) {
1889 tp->t_msl = tcp_msl_local ? tcp_msl_local : (TCPTV_MSL >> 1);
1890 break;
1891 }
1892 }
1893 tp->t_msl = tcp_msl_remote ? tcp_msl_remote : TCPTV_MSL;
1894 break;
1895 }
1896
1897 /* Clamp to a reasonable range. */
1898 tp->t_msl = MIN(tp->t_msl, TCP_MAXMSL);
1899 #endif
1900
1901 tp->t_state = TCPS_ESTABLISHED;
1902 TCP_TIMER_ARM(tp, TCPT_KEEP, tp->t_keepidle);
1903
1904 #ifdef RTV_RPIPE
1905 if (rt != NULL && rt->rt_rmx.rmx_recvpipe != 0)
1906 bufsize = rt->rt_rmx.rmx_recvpipe;
1907 else
1908 #endif
1909 {
1910 KASSERT(so != NULL);
1911 bufsize = so->so_rcv.sb_hiwat;
1912 }
1913 if (bufsize > tp->t_ourmss) {
1914 bufsize = roundup(bufsize, tp->t_ourmss);
1915 if (bufsize > sb_max)
1916 bufsize = sb_max;
1917 (void) sbreserve(&so->so_rcv, bufsize, so);
1918 }
1919 #ifdef RTV_RPIPE
1920 inpcb_rtentry_unref(rt, tp->t_inpcb);
1921 #endif
1922 }
1923
1924 /*
1925 * Check if there's an initial rtt or rttvar. Convert from the
1926 * route-table units to scaled multiples of the slow timeout timer.
1927 * Called only during the 3-way handshake.
1928 */
1929 void
tcp_rmx_rtt(struct tcpcb * tp)1930 tcp_rmx_rtt(struct tcpcb *tp)
1931 {
1932 #ifdef RTV_RTT
1933 struct rtentry *rt = NULL;
1934 int rtt;
1935
1936 KASSERT(tp->t_inpcb != NULL);
1937
1938 rt = inpcb_rtentry(tp->t_inpcb);
1939 if (rt == NULL)
1940 return;
1941
1942 if (tp->t_srtt == 0 && (rtt = rt->rt_rmx.rmx_rtt)) {
1943 /*
1944 * XXX The lock bit for MTU indicates that the value
1945 * is also a minimum value; this is subject to time.
1946 */
1947 if (rt->rt_rmx.rmx_locks & RTV_RTT)
1948 TCPT_RANGESET(tp->t_rttmin,
1949 rtt / (RTM_RTTUNIT / PR_SLOWHZ),
1950 TCPTV_MIN, TCPTV_REXMTMAX);
1951 tp->t_srtt = rtt /
1952 ((RTM_RTTUNIT / PR_SLOWHZ) >> (TCP_RTT_SHIFT + 2));
1953 if (rt->rt_rmx.rmx_rttvar) {
1954 tp->t_rttvar = rt->rt_rmx.rmx_rttvar /
1955 ((RTM_RTTUNIT / PR_SLOWHZ) >>
1956 (TCP_RTTVAR_SHIFT + 2));
1957 } else {
1958 /* Default variation is +- 1 rtt */
1959 tp->t_rttvar =
1960 tp->t_srtt >> (TCP_RTT_SHIFT - TCP_RTTVAR_SHIFT);
1961 }
1962 TCPT_RANGESET(tp->t_rxtcur,
1963 ((tp->t_srtt >> 2) + tp->t_rttvar) >> (1 + 2),
1964 tp->t_rttmin, TCPTV_REXMTMAX);
1965 }
1966 inpcb_rtentry_unref(rt, tp->t_inpcb);
1967 #endif
1968 }
1969
1970 tcp_seq tcp_iss_seq = 0; /* tcp initial seq # */
1971
1972 /*
1973 * Get a new sequence value given a tcp control block
1974 */
1975 tcp_seq
tcp_new_iss(struct tcpcb * tp)1976 tcp_new_iss(struct tcpcb *tp)
1977 {
1978
1979 if (tp->t_inpcb->inp_af == AF_INET) {
1980 return tcp_new_iss1(&in4p_laddr(tp->t_inpcb),
1981 &in4p_faddr(tp->t_inpcb), tp->t_inpcb->inp_lport,
1982 tp->t_inpcb->inp_fport, sizeof(in4p_laddr(tp->t_inpcb)));
1983 }
1984 #ifdef INET6
1985 if (tp->t_inpcb->inp_af == AF_INET6) {
1986 return tcp_new_iss1(&in6p_laddr(tp->t_inpcb),
1987 &in6p_faddr(tp->t_inpcb), tp->t_inpcb->inp_lport,
1988 tp->t_inpcb->inp_fport, sizeof(in6p_laddr(tp->t_inpcb)));
1989 }
1990 #endif
1991
1992 panic("tcp_new_iss: unreachable");
1993 }
1994
1995 static u_int8_t tcp_iss_secret[16]; /* 128 bits; should be plenty */
1996
1997 /*
1998 * Initialize RFC 1948 ISS Secret
1999 */
2000 static int
tcp_iss_secret_init(void)2001 tcp_iss_secret_init(void)
2002 {
2003 cprng_strong(kern_cprng,
2004 tcp_iss_secret, sizeof(tcp_iss_secret), 0);
2005
2006 return 0;
2007 }
2008
2009 /*
2010 * This routine actually generates a new TCP initial sequence number.
2011 */
2012 tcp_seq
tcp_new_iss1(void * laddr,void * faddr,u_int16_t lport,u_int16_t fport,size_t addrsz)2013 tcp_new_iss1(void *laddr, void *faddr, u_int16_t lport, u_int16_t fport,
2014 size_t addrsz)
2015 {
2016 tcp_seq tcp_iss;
2017
2018 if (tcp_do_rfc1948) {
2019 MD5_CTX ctx;
2020 u_int8_t hash[16]; /* XXX MD5 knowledge */
2021 static ONCE_DECL(tcp_iss_secret_control);
2022
2023 /*
2024 * If we haven't been here before, initialize our cryptographic
2025 * hash secret.
2026 */
2027 RUN_ONCE(&tcp_iss_secret_control, tcp_iss_secret_init);
2028
2029 /*
2030 * Compute the base value of the ISS. It is a hash
2031 * of (saddr, sport, daddr, dport, secret).
2032 */
2033 MD5Init(&ctx);
2034
2035 MD5Update(&ctx, (u_char *) laddr, addrsz);
2036 MD5Update(&ctx, (u_char *) &lport, sizeof(lport));
2037
2038 MD5Update(&ctx, (u_char *) faddr, addrsz);
2039 MD5Update(&ctx, (u_char *) &fport, sizeof(fport));
2040
2041 MD5Update(&ctx, tcp_iss_secret, sizeof(tcp_iss_secret));
2042
2043 MD5Final(hash, &ctx);
2044
2045 memcpy(&tcp_iss, hash, sizeof(tcp_iss));
2046
2047 #ifdef TCPISS_DEBUG
2048 printf("ISS hash 0x%08x, ", tcp_iss);
2049 #endif
2050 } else {
2051 /*
2052 * Randomize.
2053 */
2054 tcp_iss = cprng_fast32() & TCP_ISS_RANDOM_MASK;
2055 #ifdef TCPISS_DEBUG
2056 printf("ISS random 0x%08x, ", tcp_iss);
2057 #endif
2058 }
2059
2060 /*
2061 * Add the offset in to the computed value.
2062 */
2063 tcp_iss += tcp_iss_seq;
2064 #ifdef TCPISS_DEBUG
2065 printf("ISS %08x\n", tcp_iss);
2066 #endif
2067 return tcp_iss;
2068 }
2069
2070 #if defined(IPSEC)
2071 /* compute ESP/AH header size for TCP, including outer IP header. */
2072 size_t
ipsec4_hdrsiz_tcp(struct tcpcb * tp)2073 ipsec4_hdrsiz_tcp(struct tcpcb *tp)
2074 {
2075 struct inpcb *inp;
2076 size_t hdrsiz;
2077
2078 /* XXX mapped addr case (tp->t_inpcb) */
2079 if (!tp || !tp->t_template || !(inp = tp->t_inpcb))
2080 return 0;
2081 switch (tp->t_family) {
2082 case AF_INET:
2083 /* XXX: should use correct direction. */
2084 hdrsiz = ipsec_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, inp);
2085 break;
2086 default:
2087 hdrsiz = 0;
2088 break;
2089 }
2090
2091 return hdrsiz;
2092 }
2093
2094 #ifdef INET6
2095 size_t
ipsec6_hdrsiz_tcp(struct tcpcb * tp)2096 ipsec6_hdrsiz_tcp(struct tcpcb *tp)
2097 {
2098 struct inpcb *inp;
2099 size_t hdrsiz;
2100
2101 if (!tp || !tp->t_template || !(inp = tp->t_inpcb))
2102 return 0;
2103 switch (tp->t_family) {
2104 case AF_INET6:
2105 /* XXX: should use correct direction. */
2106 hdrsiz = ipsec_hdrsiz(tp->t_template, IPSEC_DIR_OUTBOUND, inp);
2107 break;
2108 case AF_INET:
2109 /* mapped address case - tricky */
2110 default:
2111 hdrsiz = 0;
2112 break;
2113 }
2114
2115 return hdrsiz;
2116 }
2117 #endif
2118 #endif /*IPSEC*/
2119
2120 /*
2121 * Determine the length of the TCP options for this connection.
2122 *
2123 * XXX: What do we do for SACK, when we add that? Just reserve
2124 * all of the space? Otherwise we can't exactly be incrementing
2125 * cwnd by an amount that varies depending on the amount we last
2126 * had to SACK!
2127 */
2128
2129 u_int
tcp_optlen(struct tcpcb * tp)2130 tcp_optlen(struct tcpcb *tp)
2131 {
2132 u_int optlen;
2133
2134 optlen = 0;
2135 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) ==
2136 (TF_REQ_TSTMP | TF_RCVD_TSTMP))
2137 optlen += TCPOLEN_TSTAMP_APPA;
2138
2139 #ifdef TCP_SIGNATURE
2140 if (tp->t_flags & TF_SIGNATURE)
2141 optlen += TCPOLEN_SIGLEN;
2142 #endif
2143
2144 return optlen;
2145 }
2146
2147 u_int
tcp_hdrsz(struct tcpcb * tp)2148 tcp_hdrsz(struct tcpcb *tp)
2149 {
2150 u_int hlen;
2151
2152 switch (tp->t_family) {
2153 #ifdef INET6
2154 case AF_INET6:
2155 hlen = sizeof(struct ip6_hdr);
2156 break;
2157 #endif
2158 case AF_INET:
2159 hlen = sizeof(struct ip);
2160 break;
2161 default:
2162 hlen = 0;
2163 break;
2164 }
2165 hlen += sizeof(struct tcphdr);
2166
2167 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
2168 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)
2169 hlen += TCPOLEN_TSTAMP_APPA;
2170 #ifdef TCP_SIGNATURE
2171 if (tp->t_flags & TF_SIGNATURE)
2172 hlen += TCPOLEN_SIGLEN;
2173 #endif
2174 return hlen;
2175 }
2176
2177 void
tcp_statinc(u_int stat)2178 tcp_statinc(u_int stat)
2179 {
2180
2181 KASSERT(stat < TCP_NSTATS);
2182 TCP_STATINC(stat);
2183 }
2184
2185 void
tcp_statadd(u_int stat,uint64_t val)2186 tcp_statadd(u_int stat, uint64_t val)
2187 {
2188
2189 KASSERT(stat < TCP_NSTATS);
2190 TCP_STATADD(stat, val);
2191 }
2192