1 /*-
2 * Copyright (c) 1982, 1986, 1988, 1993
3 * The Regents of the University of California.
4 * Copyright (c) 2006-2007 Robert N. M. Watson
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 * 4. Neither the name of the University 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 REGENTS 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 REGENTS 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 * From: @(#)tcp_usrreq.c 8.2 (Berkeley) 1/3/94
32 */
33
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD: releng/8.2/sys/netinet/tcp_usrreq.c 218575 2011-02-11 15:14:51Z jhb $");
36
37 #include "opt_ddb.h"
38 #include "opt_inet.h"
39 #include "opt_inet6.h"
40 #include "opt_tcpdebug.h"
41
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/malloc.h>
45 #include <sys/kernel.h>
46 #include <sys/sysctl.h>
47 #include <sys/mbuf.h>
48 #ifdef INET6
49 #include <sys/domain.h>
50 #endif /* INET6 */
51 #include <sys/socket.h>
52 #include <sys/socketvar.h>
53 #include <sys/protosw.h>
54 #include <sys/proc.h>
55 #include <sys/jail.h>
56
57 #ifdef DDB
58 #include <ddb/ddb.h>
59 #endif
60
61 #include <net/if.h>
62 #include <net/route.h>
63 #include <net/vnet.h>
64
65 #include <netinet/in.h>
66 #include <netinet/in_systm.h>
67 #ifdef INET6
68 #include <netinet/ip6.h>
69 #endif
70 #include <netinet/in_pcb.h>
71 #ifdef INET6
72 #include <netinet6/in6_pcb.h>
73 #endif
74 #include <netinet/in_var.h>
75 #include <netinet/ip_var.h>
76 #ifdef INET6
77 #include <netinet6/ip6_var.h>
78 #include <netinet6/scope6_var.h>
79 #endif
80 #include <netinet/tcp.h>
81 #include <netinet/tcp_fsm.h>
82 #include <netinet/tcp_seq.h>
83 #include <netinet/tcp_timer.h>
84 #include <netinet/tcp_var.h>
85 #include <netinet/tcpip.h>
86 #ifdef TCPDEBUG
87 #include <netinet/tcp_debug.h>
88 #endif
89 #include <netinet/tcp_offload.h>
90
91 /*
92 * TCP protocol interface to socket abstraction.
93 */
94 static int tcp_attach(struct socket *);
95 static int tcp_connect(struct tcpcb *, struct sockaddr *,
96 struct thread *td);
97 #ifdef INET6
98 static int tcp6_connect(struct tcpcb *, struct sockaddr *,
99 struct thread *td);
100 #endif /* INET6 */
101 static void tcp_disconnect(struct tcpcb *);
102 static void tcp_usrclosed(struct tcpcb *);
103 static void tcp_fill_info(struct tcpcb *, struct tcp_info *);
104
105 #ifdef TCPDEBUG
106 #define TCPDEBUG0 int ostate = 0
107 #define TCPDEBUG1() ostate = tp ? tp->t_state : 0
108 #define TCPDEBUG2(req) if (tp && (so->so_options & SO_DEBUG)) \
109 tcp_trace(TA_USER, ostate, tp, 0, 0, req)
110 #else
111 #define TCPDEBUG0
112 #define TCPDEBUG1()
113 #define TCPDEBUG2(req)
114 #endif
115
116 /*
117 * TCP attaches to socket via pru_attach(), reserving space,
118 * and an internet control block.
119 */
120 static int
121 tcp_usr_attach(struct socket *so, int proto, struct thread *td)
122 {
123 struct inpcb *inp;
124 struct tcpcb *tp = NULL;
125 int error;
126 TCPDEBUG0;
127
128 inp = sotoinpcb(so);
129 KASSERT(inp == NULL, ("tcp_usr_attach: inp != NULL"));
130 TCPDEBUG1();
131
132 error = tcp_attach(so);
133 if (error)
134 goto out;
135
136 if ((so->so_options & SO_LINGER) && so->so_linger == 0)
137 so->so_linger = TCP_LINGERTIME;
138
139 inp = sotoinpcb(so);
140 tp = intotcpcb(inp);
141 out:
142 TCPDEBUG2(PRU_ATTACH);
143 return error;
144 }
145
146 /*
147 * tcp_detach is called when the socket layer loses its final reference
148 * to the socket, be it a file descriptor reference, a reference from TCP,
149 * etc. At this point, there is only one case in which we will keep around
150 * inpcb state: time wait.
151 *
152 * This function can probably be re-absorbed back into tcp_usr_detach() now
153 * that there is a single detach path.
154 */
155 static void
156 tcp_detach(struct socket *so, struct inpcb *inp)
157 {
158 struct tcpcb *tp;
159
160 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
161 INP_WLOCK_ASSERT(inp);
162
163 KASSERT(so->so_pcb == inp, ("tcp_detach: so_pcb != inp"));
164 KASSERT(inp->inp_socket == so, ("tcp_detach: inp_socket != so"));
165
166 tp = intotcpcb(inp);
167
168 if (inp->inp_flags & INP_TIMEWAIT) {
169 /*
170 * There are two cases to handle: one in which the time wait
171 * state is being discarded (INP_DROPPED), and one in which
172 * this connection will remain in timewait. In the former,
173 * it is time to discard all state (except tcptw, which has
174 * already been discarded by the timewait close code, which
175 * should be further up the call stack somewhere). In the
176 * latter case, we detach from the socket, but leave the pcb
177 * present until timewait ends.
178 *
179 * XXXRW: Would it be cleaner to free the tcptw here?
180 */
181 if (inp->inp_flags & INP_DROPPED) {
182 KASSERT(tp == NULL, ("tcp_detach: INP_TIMEWAIT && "
183 "INP_DROPPED && tp != NULL"));
184 in_pcbdetach(inp);
185 in_pcbfree(inp);
186 } else {
187 in_pcbdetach(inp);
188 INP_WUNLOCK(inp);
189 }
190 } else {
191 /*
192 * If the connection is not in timewait, we consider two
193 * two conditions: one in which no further processing is
194 * necessary (dropped || embryonic), and one in which TCP is
195 * not yet done, but no longer requires the socket, so the
196 * pcb will persist for the time being.
197 *
198 * XXXRW: Does the second case still occur?
199 */
200 if (inp->inp_flags & INP_DROPPED ||
201 tp->t_state < TCPS_SYN_SENT) {
202 tcp_discardcb(tp);
203 in_pcbdetach(inp);
204 in_pcbfree(inp);
205 } else
206 in_pcbdetach(inp);
207 }
208 }
209
210 /*
211 * pru_detach() detaches the TCP protocol from the socket.
212 * If the protocol state is non-embryonic, then can't
213 * do this directly: have to initiate a pru_disconnect(),
214 * which may finish later; embryonic TCB's can just
215 * be discarded here.
216 */
217 static void
218 tcp_usr_detach(struct socket *so)
219 {
220 struct inpcb *inp;
221
222 inp = sotoinpcb(so);
223 KASSERT(inp != NULL, ("tcp_usr_detach: inp == NULL"));
224 INP_INFO_WLOCK(&V_tcbinfo);
225 INP_WLOCK(inp);
226 KASSERT(inp->inp_socket != NULL,
227 ("tcp_usr_detach: inp_socket == NULL"));
228 tcp_detach(so, inp);
229 INP_INFO_WUNLOCK(&V_tcbinfo);
230 }
231
232 /*
233 * Give the socket an address.
234 */
235 static int
236 tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
237 {
238 int error = 0;
239 struct inpcb *inp;
240 struct tcpcb *tp = NULL;
241 struct sockaddr_in *sinp;
242
243 sinp = (struct sockaddr_in *)nam;
244 if (nam->sa_len != sizeof (*sinp))
245 return (EINVAL);
246 /*
247 * Must check for multicast addresses and disallow binding
248 * to them.
249 */
250 if (sinp->sin_family == AF_INET &&
251 IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
252 return (EAFNOSUPPORT);
253
254 TCPDEBUG0;
255 INP_INFO_WLOCK(&V_tcbinfo);
256 inp = sotoinpcb(so);
257 KASSERT(inp != NULL, ("tcp_usr_bind: inp == NULL"));
258 INP_WLOCK(inp);
259 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
260 error = EINVAL;
261 goto out;
262 }
263 tp = intotcpcb(inp);
264 TCPDEBUG1();
265 error = in_pcbbind(inp, nam, td->td_ucred);
266 out:
267 TCPDEBUG2(PRU_BIND);
268 INP_WUNLOCK(inp);
269 INP_INFO_WUNLOCK(&V_tcbinfo);
270
271 return (error);
272 }
273
274 #ifdef INET6
275 static int
276 tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
277 {
278 int error = 0;
279 struct inpcb *inp;
280 struct tcpcb *tp = NULL;
281 struct sockaddr_in6 *sin6p;
282
283 sin6p = (struct sockaddr_in6 *)nam;
284 if (nam->sa_len != sizeof (*sin6p))
285 return (EINVAL);
286 /*
287 * Must check for multicast addresses and disallow binding
288 * to them.
289 */
290 if (sin6p->sin6_family == AF_INET6 &&
291 IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr))
292 return (EAFNOSUPPORT);
293
294 TCPDEBUG0;
295 INP_INFO_WLOCK(&V_tcbinfo);
296 inp = sotoinpcb(so);
297 KASSERT(inp != NULL, ("tcp6_usr_bind: inp == NULL"));
298 INP_WLOCK(inp);
299 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
300 error = EINVAL;
301 goto out;
302 }
303 tp = intotcpcb(inp);
304 TCPDEBUG1();
305 inp->inp_vflag &= ~INP_IPV4;
306 inp->inp_vflag |= INP_IPV6;
307 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
308 if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr))
309 inp->inp_vflag |= INP_IPV4;
310 else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
311 struct sockaddr_in sin;
312
313 in6_sin6_2_sin(&sin, sin6p);
314 inp->inp_vflag |= INP_IPV4;
315 inp->inp_vflag &= ~INP_IPV6;
316 error = in_pcbbind(inp, (struct sockaddr *)&sin,
317 td->td_ucred);
318 goto out;
319 }
320 }
321 error = in6_pcbbind(inp, nam, td->td_ucred);
322 out:
323 TCPDEBUG2(PRU_BIND);
324 INP_WUNLOCK(inp);
325 INP_INFO_WUNLOCK(&V_tcbinfo);
326 return (error);
327 }
328 #endif /* INET6 */
329
330 /*
331 * Prepare to accept connections.
332 */
333 static int
334 tcp_usr_listen(struct socket *so, int backlog, struct thread *td)
335 {
336 int error = 0;
337 struct inpcb *inp;
338 struct tcpcb *tp = NULL;
339
340 TCPDEBUG0;
341 INP_INFO_WLOCK(&V_tcbinfo);
342 inp = sotoinpcb(so);
343 KASSERT(inp != NULL, ("tcp_usr_listen: inp == NULL"));
344 INP_WLOCK(inp);
345 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
346 error = EINVAL;
347 goto out;
348 }
349 tp = intotcpcb(inp);
350 TCPDEBUG1();
351 SOCK_LOCK(so);
352 error = solisten_proto_check(so);
353 if (error == 0 && inp->inp_lport == 0)
354 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
355 if (error == 0) {
356 tp->t_state = TCPS_LISTEN;
357 solisten_proto(so, backlog);
358 tcp_offload_listen_open(tp);
359 }
360 SOCK_UNLOCK(so);
361
362 out:
363 TCPDEBUG2(PRU_LISTEN);
364 INP_WUNLOCK(inp);
365 INP_INFO_WUNLOCK(&V_tcbinfo);
366 return (error);
367 }
368
369 #ifdef INET6
370 static int
371 tcp6_usr_listen(struct socket *so, int backlog, struct thread *td)
372 {
373 int error = 0;
374 struct inpcb *inp;
375 struct tcpcb *tp = NULL;
376
377 TCPDEBUG0;
378 INP_INFO_WLOCK(&V_tcbinfo);
379 inp = sotoinpcb(so);
380 KASSERT(inp != NULL, ("tcp6_usr_listen: inp == NULL"));
381 INP_WLOCK(inp);
382 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
383 error = EINVAL;
384 goto out;
385 }
386 tp = intotcpcb(inp);
387 TCPDEBUG1();
388 SOCK_LOCK(so);
389 error = solisten_proto_check(so);
390 if (error == 0 && inp->inp_lport == 0) {
391 inp->inp_vflag &= ~INP_IPV4;
392 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0)
393 inp->inp_vflag |= INP_IPV4;
394 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
395 }
396 if (error == 0) {
397 tp->t_state = TCPS_LISTEN;
398 solisten_proto(so, backlog);
399 }
400 SOCK_UNLOCK(so);
401
402 out:
403 TCPDEBUG2(PRU_LISTEN);
404 INP_WUNLOCK(inp);
405 INP_INFO_WUNLOCK(&V_tcbinfo);
406 return (error);
407 }
408 #endif /* INET6 */
409
410 /*
411 * Initiate connection to peer.
412 * Create a template for use in transmissions on this connection.
413 * Enter SYN_SENT state, and mark socket as connecting.
414 * Start keep-alive timer, and seed output sequence space.
415 * Send initial segment on connection.
416 */
417 static int
418 tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
419 {
420 int error = 0;
421 struct inpcb *inp;
422 struct tcpcb *tp = NULL;
423 struct sockaddr_in *sinp;
424
425 sinp = (struct sockaddr_in *)nam;
426 if (nam->sa_len != sizeof (*sinp))
427 return (EINVAL);
428 /*
429 * Must disallow TCP ``connections'' to multicast addresses.
430 */
431 if (sinp->sin_family == AF_INET
432 && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
433 return (EAFNOSUPPORT);
434 if ((error = prison_remote_ip4(td->td_ucred, &sinp->sin_addr)) != 0)
435 return (error);
436
437 TCPDEBUG0;
438 INP_INFO_WLOCK(&V_tcbinfo);
439 inp = sotoinpcb(so);
440 KASSERT(inp != NULL, ("tcp_usr_connect: inp == NULL"));
441 INP_WLOCK(inp);
442 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
443 error = EINVAL;
444 goto out;
445 }
446 tp = intotcpcb(inp);
447 TCPDEBUG1();
448 if ((error = tcp_connect(tp, nam, td)) != 0)
449 goto out;
450 error = tcp_output_connect(so, nam);
451 out:
452 TCPDEBUG2(PRU_CONNECT);
453 INP_WUNLOCK(inp);
454 INP_INFO_WUNLOCK(&V_tcbinfo);
455 return (error);
456 }
457
458 #ifdef INET6
459 static int
460 tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
461 {
462 int error = 0;
463 struct inpcb *inp;
464 struct tcpcb *tp = NULL;
465 struct sockaddr_in6 *sin6p;
466
467 TCPDEBUG0;
468
469 sin6p = (struct sockaddr_in6 *)nam;
470 if (nam->sa_len != sizeof (*sin6p))
471 return (EINVAL);
472 /*
473 * Must disallow TCP ``connections'' to multicast addresses.
474 */
475 if (sin6p->sin6_family == AF_INET6
476 && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr))
477 return (EAFNOSUPPORT);
478
479 INP_INFO_WLOCK(&V_tcbinfo);
480 inp = sotoinpcb(so);
481 KASSERT(inp != NULL, ("tcp6_usr_connect: inp == NULL"));
482 INP_WLOCK(inp);
483 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
484 error = EINVAL;
485 goto out;
486 }
487 tp = intotcpcb(inp);
488 TCPDEBUG1();
489 if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
490 struct sockaddr_in sin;
491
492 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
493 error = EINVAL;
494 goto out;
495 }
496
497 in6_sin6_2_sin(&sin, sin6p);
498 inp->inp_vflag |= INP_IPV4;
499 inp->inp_vflag &= ~INP_IPV6;
500 if ((error = prison_remote_ip4(td->td_ucred,
501 &sin.sin_addr)) != 0)
502 goto out;
503 if ((error = tcp_connect(tp, (struct sockaddr *)&sin, td)) != 0)
504 goto out;
505 error = tcp_output_connect(so, nam);
506 goto out;
507 }
508 inp->inp_vflag &= ~INP_IPV4;
509 inp->inp_vflag |= INP_IPV6;
510 inp->inp_inc.inc_flags |= INC_ISIPV6;
511 if ((error = prison_remote_ip6(td->td_ucred, &sin6p->sin6_addr)) != 0)
512 goto out;
513 if ((error = tcp6_connect(tp, nam, td)) != 0)
514 goto out;
515 error = tcp_output_connect(so, nam);
516
517 out:
518 TCPDEBUG2(PRU_CONNECT);
519 INP_WUNLOCK(inp);
520 INP_INFO_WUNLOCK(&V_tcbinfo);
521 return (error);
522 }
523 #endif /* INET6 */
524
525 /*
526 * Initiate disconnect from peer.
527 * If connection never passed embryonic stage, just drop;
528 * else if don't need to let data drain, then can just drop anyways,
529 * else have to begin TCP shutdown process: mark socket disconnecting,
530 * drain unread data, state switch to reflect user close, and
531 * send segment (e.g. FIN) to peer. Socket will be really disconnected
532 * when peer sends FIN and acks ours.
533 *
534 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
535 */
536 static int
537 tcp_usr_disconnect(struct socket *so)
538 {
539 struct inpcb *inp;
540 struct tcpcb *tp = NULL;
541 int error = 0;
542
543 TCPDEBUG0;
544 INP_INFO_WLOCK(&V_tcbinfo);
545 inp = sotoinpcb(so);
546 KASSERT(inp != NULL, ("tcp_usr_disconnect: inp == NULL"));
547 INP_WLOCK(inp);
548 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
549 error = ECONNRESET;
550 goto out;
551 }
552 tp = intotcpcb(inp);
553 TCPDEBUG1();
554 tcp_disconnect(tp);
555 out:
556 TCPDEBUG2(PRU_DISCONNECT);
557 INP_WUNLOCK(inp);
558 INP_INFO_WUNLOCK(&V_tcbinfo);
559 return (error);
560 }
561
562 /*
563 * Accept a connection. Essentially all the work is done at higher levels;
564 * just return the address of the peer, storing through addr.
565 *
566 * The rationale for acquiring the tcbinfo lock here is somewhat complicated,
567 * and is described in detail in the commit log entry for r175612. Acquiring
568 * it delays an accept(2) racing with sonewconn(), which inserts the socket
569 * before the inpcb address/port fields are initialized. A better fix would
570 * prevent the socket from being placed in the listen queue until all fields
571 * are fully initialized.
572 */
573 static int
574 tcp_usr_accept(struct socket *so, struct sockaddr **nam)
575 {
576 int error = 0;
577 struct inpcb *inp = NULL;
578 struct tcpcb *tp = NULL;
579 struct in_addr addr;
580 in_port_t port = 0;
581 TCPDEBUG0;
582
583 if (so->so_state & SS_ISDISCONNECTED)
584 return (ECONNABORTED);
585
586 inp = sotoinpcb(so);
587 KASSERT(inp != NULL, ("tcp_usr_accept: inp == NULL"));
588 INP_INFO_RLOCK(&V_tcbinfo);
589 INP_WLOCK(inp);
590 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
591 error = ECONNABORTED;
592 goto out;
593 }
594 tp = intotcpcb(inp);
595 TCPDEBUG1();
596
597 /*
598 * We inline in_getpeeraddr and COMMON_END here, so that we can
599 * copy the data of interest and defer the malloc until after we
600 * release the lock.
601 */
602 port = inp->inp_fport;
603 addr = inp->inp_faddr;
604
605 out:
606 TCPDEBUG2(PRU_ACCEPT);
607 INP_WUNLOCK(inp);
608 INP_INFO_RUNLOCK(&V_tcbinfo);
609 if (error == 0)
610 *nam = in_sockaddr(port, &addr);
611 return error;
612 }
613
614 #ifdef INET6
615 static int
616 tcp6_usr_accept(struct socket *so, struct sockaddr **nam)
617 {
618 struct inpcb *inp = NULL;
619 int error = 0;
620 struct tcpcb *tp = NULL;
621 struct in_addr addr;
622 struct in6_addr addr6;
623 in_port_t port = 0;
624 int v4 = 0;
625 TCPDEBUG0;
626
627 if (so->so_state & SS_ISDISCONNECTED)
628 return (ECONNABORTED);
629
630 inp = sotoinpcb(so);
631 KASSERT(inp != NULL, ("tcp6_usr_accept: inp == NULL"));
632 INP_WLOCK(inp);
633 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
634 error = ECONNABORTED;
635 goto out;
636 }
637 tp = intotcpcb(inp);
638 TCPDEBUG1();
639
640 /*
641 * We inline in6_mapped_peeraddr and COMMON_END here, so that we can
642 * copy the data of interest and defer the malloc until after we
643 * release the lock.
644 */
645 if (inp->inp_vflag & INP_IPV4) {
646 v4 = 1;
647 port = inp->inp_fport;
648 addr = inp->inp_faddr;
649 } else {
650 port = inp->inp_fport;
651 addr6 = inp->in6p_faddr;
652 }
653
654 out:
655 TCPDEBUG2(PRU_ACCEPT);
656 INP_WUNLOCK(inp);
657 if (error == 0) {
658 if (v4)
659 *nam = in6_v4mapsin6_sockaddr(port, &addr);
660 else
661 *nam = in6_sockaddr(port, &addr6);
662 }
663 return error;
664 }
665 #endif /* INET6 */
666
667 /*
668 * Mark the connection as being incapable of further output.
669 */
670 static int
671 tcp_usr_shutdown(struct socket *so)
672 {
673 int error = 0;
674 struct inpcb *inp;
675 struct tcpcb *tp = NULL;
676
677 TCPDEBUG0;
678 INP_INFO_WLOCK(&V_tcbinfo);
679 inp = sotoinpcb(so);
680 KASSERT(inp != NULL, ("inp == NULL"));
681 INP_WLOCK(inp);
682 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
683 error = ECONNRESET;
684 goto out;
685 }
686 tp = intotcpcb(inp);
687 TCPDEBUG1();
688 socantsendmore(so);
689 tcp_usrclosed(tp);
690 if (!(inp->inp_flags & INP_DROPPED))
691 error = tcp_output_disconnect(tp);
692
693 out:
694 TCPDEBUG2(PRU_SHUTDOWN);
695 INP_WUNLOCK(inp);
696 INP_INFO_WUNLOCK(&V_tcbinfo);
697
698 return (error);
699 }
700
701 /*
702 * After a receive, possibly send window update to peer.
703 */
704 static int
705 tcp_usr_rcvd(struct socket *so, int flags)
706 {
707 struct inpcb *inp;
708 struct tcpcb *tp = NULL;
709 int error = 0;
710
711 TCPDEBUG0;
712 inp = sotoinpcb(so);
713 KASSERT(inp != NULL, ("tcp_usr_rcvd: inp == NULL"));
714 INP_WLOCK(inp);
715 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
716 error = ECONNRESET;
717 goto out;
718 }
719 tp = intotcpcb(inp);
720 TCPDEBUG1();
721 tcp_output_rcvd(tp);
722
723 out:
724 TCPDEBUG2(PRU_RCVD);
725 INP_WUNLOCK(inp);
726 return (error);
727 }
728
729 /*
730 * Do a send by putting data in output queue and updating urgent
731 * marker if URG set. Possibly send more data. Unlike the other
732 * pru_*() routines, the mbuf chains are our responsibility. We
733 * must either enqueue them or free them. The other pru_* routines
734 * generally are caller-frees.
735 */
736 static int
737 tcp_usr_send(struct socket *so, int flags, struct mbuf *m,
738 struct sockaddr *nam, struct mbuf *control, struct thread *td)
739 {
740 int error = 0;
741 struct inpcb *inp;
742 struct tcpcb *tp = NULL;
743 int headlocked = 0;
744 #ifdef INET6
745 int isipv6;
746 #endif
747 TCPDEBUG0;
748
749 /*
750 * We require the pcbinfo lock in two cases:
751 *
752 * (1) An implied connect is taking place, which can result in
753 * binding IPs and ports and hence modification of the pcb hash
754 * chains.
755 *
756 * (2) PRUS_EOF is set, resulting in explicit close on the send.
757 */
758 if ((nam != NULL) || (flags & PRUS_EOF)) {
759 INP_INFO_WLOCK(&V_tcbinfo);
760 headlocked = 1;
761 }
762 inp = sotoinpcb(so);
763 KASSERT(inp != NULL, ("tcp_usr_send: inp == NULL"));
764 INP_WLOCK(inp);
765 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
766 if (control)
767 m_freem(control);
768 if (m)
769 m_freem(m);
770 error = ECONNRESET;
771 goto out;
772 }
773 #ifdef INET6
774 isipv6 = nam && nam->sa_family == AF_INET6;
775 #endif /* INET6 */
776 tp = intotcpcb(inp);
777 TCPDEBUG1();
778 if (control) {
779 /* TCP doesn't do control messages (rights, creds, etc) */
780 if (control->m_len) {
781 m_freem(control);
782 if (m)
783 m_freem(m);
784 error = EINVAL;
785 goto out;
786 }
787 m_freem(control); /* empty control, just free it */
788 }
789 if (!(flags & PRUS_OOB)) {
790 sbappendstream(&so->so_snd, m);
791 if (nam && tp->t_state < TCPS_SYN_SENT) {
792 /*
793 * Do implied connect if not yet connected,
794 * initialize window to default value, and
795 * initialize maxseg/maxopd using peer's cached
796 * MSS.
797 */
798 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
799 #ifdef INET6
800 if (isipv6)
801 error = tcp6_connect(tp, nam, td);
802 else
803 #endif /* INET6 */
804 error = tcp_connect(tp, nam, td);
805 if (error)
806 goto out;
807 tp->snd_wnd = TTCP_CLIENT_SND_WND;
808 tcp_mss(tp, -1);
809 }
810 if (flags & PRUS_EOF) {
811 /*
812 * Close the send side of the connection after
813 * the data is sent.
814 */
815 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
816 socantsendmore(so);
817 tcp_usrclosed(tp);
818 }
819 if (headlocked) {
820 INP_INFO_WUNLOCK(&V_tcbinfo);
821 headlocked = 0;
822 }
823 if (!(inp->inp_flags & INP_DROPPED)) {
824 if (flags & PRUS_MORETOCOME)
825 tp->t_flags |= TF_MORETOCOME;
826 error = tcp_output_send(tp);
827 if (flags & PRUS_MORETOCOME)
828 tp->t_flags &= ~TF_MORETOCOME;
829 }
830 } else {
831 /*
832 * XXXRW: PRUS_EOF not implemented with PRUS_OOB?
833 */
834 SOCKBUF_LOCK(&so->so_snd);
835 if (sbspace(&so->so_snd) < -512) {
836 SOCKBUF_UNLOCK(&so->so_snd);
837 m_freem(m);
838 error = ENOBUFS;
839 goto out;
840 }
841 /*
842 * According to RFC961 (Assigned Protocols),
843 * the urgent pointer points to the last octet
844 * of urgent data. We continue, however,
845 * to consider it to indicate the first octet
846 * of data past the urgent section.
847 * Otherwise, snd_up should be one lower.
848 */
849 sbappendstream_locked(&so->so_snd, m);
850 SOCKBUF_UNLOCK(&so->so_snd);
851 if (nam && tp->t_state < TCPS_SYN_SENT) {
852 /*
853 * Do implied connect if not yet connected,
854 * initialize window to default value, and
855 * initialize maxseg/maxopd using peer's cached
856 * MSS.
857 */
858 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
859 #ifdef INET6
860 if (isipv6)
861 error = tcp6_connect(tp, nam, td);
862 else
863 #endif /* INET6 */
864 error = tcp_connect(tp, nam, td);
865 if (error)
866 goto out;
867 tp->snd_wnd = TTCP_CLIENT_SND_WND;
868 tcp_mss(tp, -1);
869 INP_INFO_WUNLOCK(&V_tcbinfo);
870 headlocked = 0;
871 } else if (nam) {
872 INP_INFO_WUNLOCK(&V_tcbinfo);
873 headlocked = 0;
874 }
875 tp->snd_up = tp->snd_una + so->so_snd.sb_cc;
876 tp->t_flags |= TF_FORCEDATA;
877 error = tcp_output_send(tp);
878 tp->t_flags &= ~TF_FORCEDATA;
879 }
880 out:
881 TCPDEBUG2((flags & PRUS_OOB) ? PRU_SENDOOB :
882 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
883 INP_WUNLOCK(inp);
884 if (headlocked)
885 INP_INFO_WUNLOCK(&V_tcbinfo);
886 return (error);
887 }
888
889 /*
890 * Abort the TCP. Drop the connection abruptly.
891 */
892 static void
893 tcp_usr_abort(struct socket *so)
894 {
895 struct inpcb *inp;
896 struct tcpcb *tp = NULL;
897 TCPDEBUG0;
898
899 inp = sotoinpcb(so);
900 KASSERT(inp != NULL, ("tcp_usr_abort: inp == NULL"));
901
902 INP_INFO_WLOCK(&V_tcbinfo);
903 INP_WLOCK(inp);
904 KASSERT(inp->inp_socket != NULL,
905 ("tcp_usr_abort: inp_socket == NULL"));
906
907 /*
908 * If we still have full TCP state, and we're not dropped, drop.
909 */
910 if (!(inp->inp_flags & INP_TIMEWAIT) &&
911 !(inp->inp_flags & INP_DROPPED)) {
912 tp = intotcpcb(inp);
913 TCPDEBUG1();
914 tcp_drop(tp, ECONNABORTED);
915 TCPDEBUG2(PRU_ABORT);
916 }
917 if (!(inp->inp_flags & INP_DROPPED)) {
918 SOCK_LOCK(so);
919 so->so_state |= SS_PROTOREF;
920 SOCK_UNLOCK(so);
921 inp->inp_flags |= INP_SOCKREF;
922 }
923 INP_WUNLOCK(inp);
924 INP_INFO_WUNLOCK(&V_tcbinfo);
925 }
926
927 /*
928 * TCP socket is closed. Start friendly disconnect.
929 */
930 static void
931 tcp_usr_close(struct socket *so)
932 {
933 struct inpcb *inp;
934 struct tcpcb *tp = NULL;
935 TCPDEBUG0;
936
937 inp = sotoinpcb(so);
938 KASSERT(inp != NULL, ("tcp_usr_close: inp == NULL"));
939
940 INP_INFO_WLOCK(&V_tcbinfo);
941 INP_WLOCK(inp);
942 KASSERT(inp->inp_socket != NULL,
943 ("tcp_usr_close: inp_socket == NULL"));
944
945 /*
946 * If we still have full TCP state, and we're not dropped, initiate
947 * a disconnect.
948 */
949 if (!(inp->inp_flags & INP_TIMEWAIT) &&
950 !(inp->inp_flags & INP_DROPPED)) {
951 tp = intotcpcb(inp);
952 TCPDEBUG1();
953 tcp_disconnect(tp);
954 TCPDEBUG2(PRU_CLOSE);
955 }
956 if (!(inp->inp_flags & INP_DROPPED)) {
957 SOCK_LOCK(so);
958 so->so_state |= SS_PROTOREF;
959 SOCK_UNLOCK(so);
960 inp->inp_flags |= INP_SOCKREF;
961 }
962 INP_WUNLOCK(inp);
963 INP_INFO_WUNLOCK(&V_tcbinfo);
964 }
965
966 /*
967 * Receive out-of-band data.
968 */
969 static int
970 tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags)
971 {
972 int error = 0;
973 struct inpcb *inp;
974 struct tcpcb *tp = NULL;
975
976 TCPDEBUG0;
977 inp = sotoinpcb(so);
978 KASSERT(inp != NULL, ("tcp_usr_rcvoob: inp == NULL"));
979 INP_WLOCK(inp);
980 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
981 error = ECONNRESET;
982 goto out;
983 }
984 tp = intotcpcb(inp);
985 TCPDEBUG1();
986 if ((so->so_oobmark == 0 &&
987 (so->so_rcv.sb_state & SBS_RCVATMARK) == 0) ||
988 so->so_options & SO_OOBINLINE ||
989 tp->t_oobflags & TCPOOB_HADDATA) {
990 error = EINVAL;
991 goto out;
992 }
993 if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
994 error = EWOULDBLOCK;
995 goto out;
996 }
997 m->m_len = 1;
998 *mtod(m, caddr_t) = tp->t_iobc;
999 if ((flags & MSG_PEEK) == 0)
1000 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
1001
1002 out:
1003 TCPDEBUG2(PRU_RCVOOB);
1004 INP_WUNLOCK(inp);
1005 return (error);
1006 }
1007
1008 struct pr_usrreqs tcp_usrreqs = {
1009 .pru_abort = tcp_usr_abort,
1010 .pru_accept = tcp_usr_accept,
1011 .pru_attach = tcp_usr_attach,
1012 .pru_bind = tcp_usr_bind,
1013 .pru_connect = tcp_usr_connect,
1014 .pru_control = in_control,
1015 .pru_detach = tcp_usr_detach,
1016 .pru_disconnect = tcp_usr_disconnect,
1017 .pru_listen = tcp_usr_listen,
1018 .pru_peeraddr = in_getpeeraddr,
1019 .pru_rcvd = tcp_usr_rcvd,
1020 .pru_rcvoob = tcp_usr_rcvoob,
1021 .pru_send = tcp_usr_send,
1022 .pru_shutdown = tcp_usr_shutdown,
1023 .pru_sockaddr = in_getsockaddr,
1024 #if 0
1025 .pru_soreceive = soreceive_stream,
1026 #endif
1027 .pru_sosetlabel = in_pcbsosetlabel,
1028 .pru_close = tcp_usr_close,
1029 };
1030
1031 #ifdef INET6
1032 struct pr_usrreqs tcp6_usrreqs = {
1033 .pru_abort = tcp_usr_abort,
1034 .pru_accept = tcp6_usr_accept,
1035 .pru_attach = tcp_usr_attach,
1036 .pru_bind = tcp6_usr_bind,
1037 .pru_connect = tcp6_usr_connect,
1038 .pru_control = in6_control,
1039 .pru_detach = tcp_usr_detach,
1040 .pru_disconnect = tcp_usr_disconnect,
1041 .pru_listen = tcp6_usr_listen,
1042 .pru_peeraddr = in6_mapped_peeraddr,
1043 .pru_rcvd = tcp_usr_rcvd,
1044 .pru_rcvoob = tcp_usr_rcvoob,
1045 .pru_send = tcp_usr_send,
1046 .pru_shutdown = tcp_usr_shutdown,
1047 .pru_sockaddr = in6_mapped_sockaddr,
1048 #if 0
1049 .pru_soreceive = soreceive_stream,
1050 #endif
1051 .pru_sosetlabel = in_pcbsosetlabel,
1052 .pru_close = tcp_usr_close,
1053 };
1054 #endif /* INET6 */
1055
1056 /*
1057 * Common subroutine to open a TCP connection to remote host specified
1058 * by struct sockaddr_in in mbuf *nam. Call in_pcbbind to assign a local
1059 * port number if needed. Call in_pcbconnect_setup to do the routing and
1060 * to choose a local host address (interface). If there is an existing
1061 * incarnation of the same connection in TIME-WAIT state and if the remote
1062 * host was sending CC options and if the connection duration was < MSL, then
1063 * truncate the previous TIME-WAIT state and proceed.
1064 * Initialize connection parameters and enter SYN-SENT state.
1065 */
1066 static int
1067 tcp_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
1068 {
1069 struct inpcb *inp = tp->t_inpcb, *oinp;
1070 struct socket *so = inp->inp_socket;
1071 struct in_addr laddr;
1072 u_short lport;
1073 int error;
1074
1075 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1076 INP_WLOCK_ASSERT(inp);
1077
1078 if (inp->inp_lport == 0) {
1079 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1080 if (error)
1081 return error;
1082 }
1083
1084 /*
1085 * Cannot simply call in_pcbconnect, because there might be an
1086 * earlier incarnation of this same connection still in
1087 * TIME_WAIT state, creating an ADDRINUSE error.
1088 */
1089 laddr = inp->inp_laddr;
1090 lport = inp->inp_lport;
1091 error = in_pcbconnect_setup(inp, nam, &laddr.s_addr, &lport,
1092 &inp->inp_faddr.s_addr, &inp->inp_fport, &oinp, td->td_ucred);
1093 if (error && oinp == NULL)
1094 return error;
1095 if (oinp)
1096 return EADDRINUSE;
1097 inp->inp_laddr = laddr;
1098 in_pcbrehash(inp);
1099
1100 /*
1101 * Compute window scaling to request:
1102 * Scale to fit into sweet spot. See tcp_syncache.c.
1103 * XXX: This should move to tcp_output().
1104 */
1105 while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1106 (TCP_MAXWIN << tp->request_r_scale) < sb_max)
1107 tp->request_r_scale++;
1108
1109 soisconnecting(so);
1110 TCPSTAT_INC(tcps_connattempt);
1111 tp->t_state = TCPS_SYN_SENT;
1112 tcp_timer_activate(tp, TT_KEEP, tcp_keepinit);
1113 tp->iss = tcp_new_isn(tp);
1114 tp->t_bw_rtseq = tp->iss;
1115 tcp_sendseqinit(tp);
1116
1117 return 0;
1118 }
1119
1120 #ifdef INET6
1121 static int
1122 tcp6_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
1123 {
1124 struct inpcb *inp = tp->t_inpcb, *oinp;
1125 struct socket *so = inp->inp_socket;
1126 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam;
1127 struct in6_addr addr6;
1128 int error;
1129
1130 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1131 INP_WLOCK_ASSERT(inp);
1132
1133 if (inp->inp_lport == 0) {
1134 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1135 if (error)
1136 return error;
1137 }
1138
1139 /*
1140 * Cannot simply call in_pcbconnect, because there might be an
1141 * earlier incarnation of this same connection still in
1142 * TIME_WAIT state, creating an ADDRINUSE error.
1143 * in6_pcbladdr() also handles scope zone IDs.
1144 */
1145 error = in6_pcbladdr(inp, nam, &addr6);
1146 if (error)
1147 return error;
1148 oinp = in6_pcblookup_hash(inp->inp_pcbinfo,
1149 &sin6->sin6_addr, sin6->sin6_port,
1150 IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr)
1151 ? &addr6
1152 : &inp->in6p_laddr,
1153 inp->inp_lport, 0, NULL);
1154 if (oinp)
1155 return EADDRINUSE;
1156 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
1157 inp->in6p_laddr = addr6;
1158 inp->in6p_faddr = sin6->sin6_addr;
1159 inp->inp_fport = sin6->sin6_port;
1160 /* update flowinfo - draft-itojun-ipv6-flowlabel-api-00 */
1161 inp->inp_flow &= ~IPV6_FLOWLABEL_MASK;
1162 if (inp->inp_flags & IN6P_AUTOFLOWLABEL)
1163 inp->inp_flow |=
1164 (htonl(ip6_randomflowlabel()) & IPV6_FLOWLABEL_MASK);
1165 in_pcbrehash(inp);
1166
1167 /* Compute window scaling to request. */
1168 while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1169 (TCP_MAXWIN << tp->request_r_scale) < sb_max)
1170 tp->request_r_scale++;
1171
1172 soisconnecting(so);
1173 TCPSTAT_INC(tcps_connattempt);
1174 tp->t_state = TCPS_SYN_SENT;
1175 tcp_timer_activate(tp, TT_KEEP, tcp_keepinit);
1176 tp->iss = tcp_new_isn(tp);
1177 tp->t_bw_rtseq = tp->iss;
1178 tcp_sendseqinit(tp);
1179
1180 return 0;
1181 }
1182 #endif /* INET6 */
1183
1184 /*
1185 * Export TCP internal state information via a struct tcp_info, based on the
1186 * Linux 2.6 API. Not ABI compatible as our constants are mapped differently
1187 * (TCP state machine, etc). We export all information using FreeBSD-native
1188 * constants -- for example, the numeric values for tcpi_state will differ
1189 * from Linux.
1190 */
1191 static void
1192 tcp_fill_info(struct tcpcb *tp, struct tcp_info *ti)
1193 {
1194
1195 INP_WLOCK_ASSERT(tp->t_inpcb);
1196 bzero(ti, sizeof(*ti));
1197
1198 ti->tcpi_state = tp->t_state;
1199 if ((tp->t_flags & TF_REQ_TSTMP) && (tp->t_flags & TF_RCVD_TSTMP))
1200 ti->tcpi_options |= TCPI_OPT_TIMESTAMPS;
1201 if (tp->t_flags & TF_SACK_PERMIT)
1202 ti->tcpi_options |= TCPI_OPT_SACK;
1203 if ((tp->t_flags & TF_REQ_SCALE) && (tp->t_flags & TF_RCVD_SCALE)) {
1204 ti->tcpi_options |= TCPI_OPT_WSCALE;
1205 ti->tcpi_snd_wscale = tp->snd_scale;
1206 ti->tcpi_rcv_wscale = tp->rcv_scale;
1207 }
1208
1209 ti->tcpi_rto = tp->t_rxtcur * tick;
1210 ti->tcpi_last_data_recv = (long)(ticks - (int)tp->t_rcvtime) * tick;
1211 ti->tcpi_rtt = ((u_int64_t)tp->t_srtt * tick) >> TCP_RTT_SHIFT;
1212 ti->tcpi_rttvar = ((u_int64_t)tp->t_rttvar * tick) >> TCP_RTTVAR_SHIFT;
1213
1214 ti->tcpi_snd_ssthresh = tp->snd_ssthresh;
1215 ti->tcpi_snd_cwnd = tp->snd_cwnd;
1216
1217 /*
1218 * FreeBSD-specific extension fields for tcp_info.
1219 */
1220 ti->tcpi_rcv_space = tp->rcv_wnd;
1221 ti->tcpi_rcv_nxt = tp->rcv_nxt;
1222 ti->tcpi_snd_wnd = tp->snd_wnd;
1223 ti->tcpi_snd_bwnd = tp->snd_bwnd;
1224 ti->tcpi_snd_nxt = tp->snd_nxt;
1225 ti->tcpi_snd_mss = tp->t_maxseg;
1226 ti->tcpi_rcv_mss = tp->t_maxseg;
1227 if (tp->t_flags & TF_TOE)
1228 ti->tcpi_options |= TCPI_OPT_TOE;
1229 }
1230
1231 /*
1232 * tcp_ctloutput() must drop the inpcb lock before performing copyin on
1233 * socket option arguments. When it re-acquires the lock after the copy, it
1234 * has to revalidate that the connection is still valid for the socket
1235 * option.
1236 */
1237 #define INP_WLOCK_RECHECK(inp) do { \
1238 INP_WLOCK(inp); \
1239 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { \
1240 INP_WUNLOCK(inp); \
1241 return (ECONNRESET); \
1242 } \
1243 tp = intotcpcb(inp); \
1244 } while(0)
1245
1246 int
1247 tcp_ctloutput(struct socket *so, struct sockopt *sopt)
1248 {
1249 int error, opt, optval;
1250 struct inpcb *inp;
1251 struct tcpcb *tp;
1252 struct tcp_info ti;
1253
1254 error = 0;
1255 inp = sotoinpcb(so);
1256 KASSERT(inp != NULL, ("tcp_ctloutput: inp == NULL"));
1257 INP_WLOCK(inp);
1258 if (sopt->sopt_level != IPPROTO_TCP) {
1259 #ifdef INET6
1260 if (inp->inp_vflag & INP_IPV6PROTO) {
1261 INP_WUNLOCK(inp);
1262 error = ip6_ctloutput(so, sopt);
1263 } else {
1264 #endif /* INET6 */
1265 INP_WUNLOCK(inp);
1266 error = ip_ctloutput(so, sopt);
1267 #ifdef INET6
1268 }
1269 #endif
1270 return (error);
1271 }
1272 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1273 INP_WUNLOCK(inp);
1274 return (ECONNRESET);
1275 }
1276
1277 switch (sopt->sopt_dir) {
1278 case SOPT_SET:
1279 switch (sopt->sopt_name) {
1280 #ifdef TCP_SIGNATURE
1281 case TCP_MD5SIG:
1282 INP_WUNLOCK(inp);
1283 error = sooptcopyin(sopt, &optval, sizeof optval,
1284 sizeof optval);
1285 if (error)
1286 return (error);
1287
1288 INP_WLOCK_RECHECK(inp);
1289 if (optval > 0)
1290 tp->t_flags |= TF_SIGNATURE;
1291 else
1292 tp->t_flags &= ~TF_SIGNATURE;
1293 INP_WUNLOCK(inp);
1294 break;
1295 #endif /* TCP_SIGNATURE */
1296 case TCP_NODELAY:
1297 case TCP_NOOPT:
1298 INP_WUNLOCK(inp);
1299 error = sooptcopyin(sopt, &optval, sizeof optval,
1300 sizeof optval);
1301 if (error)
1302 return (error);
1303
1304 INP_WLOCK_RECHECK(inp);
1305 switch (sopt->sopt_name) {
1306 case TCP_NODELAY:
1307 opt = TF_NODELAY;
1308 break;
1309 case TCP_NOOPT:
1310 opt = TF_NOOPT;
1311 break;
1312 default:
1313 opt = 0; /* dead code to fool gcc */
1314 break;
1315 }
1316
1317 if (optval)
1318 tp->t_flags |= opt;
1319 else
1320 tp->t_flags &= ~opt;
1321 INP_WUNLOCK(inp);
1322 break;
1323
1324 case TCP_NOPUSH:
1325 INP_WUNLOCK(inp);
1326 error = sooptcopyin(sopt, &optval, sizeof optval,
1327 sizeof optval);
1328 if (error)
1329 return (error);
1330
1331 INP_WLOCK_RECHECK(inp);
1332 if (optval)
1333 tp->t_flags |= TF_NOPUSH;
1334 else if (tp->t_flags & TF_NOPUSH) {
1335 tp->t_flags &= ~TF_NOPUSH;
1336 if (TCPS_HAVEESTABLISHED(tp->t_state))
1337 error = tcp_output(tp);
1338 }
1339 INP_WUNLOCK(inp);
1340 break;
1341
1342 case TCP_MAXSEG:
1343 INP_WUNLOCK(inp);
1344 error = sooptcopyin(sopt, &optval, sizeof optval,
1345 sizeof optval);
1346 if (error)
1347 return (error);
1348
1349 INP_WLOCK_RECHECK(inp);
1350 if (optval > 0 && optval <= tp->t_maxseg &&
1351 optval + 40 >= V_tcp_minmss)
1352 tp->t_maxseg = optval;
1353 else
1354 error = EINVAL;
1355 INP_WUNLOCK(inp);
1356 break;
1357
1358 case TCP_INFO:
1359 INP_WUNLOCK(inp);
1360 error = EINVAL;
1361 break;
1362
1363 default:
1364 INP_WUNLOCK(inp);
1365 error = ENOPROTOOPT;
1366 break;
1367 }
1368 break;
1369
1370 case SOPT_GET:
1371 tp = intotcpcb(inp);
1372 switch (sopt->sopt_name) {
1373 #ifdef TCP_SIGNATURE
1374 case TCP_MD5SIG:
1375 optval = (tp->t_flags & TF_SIGNATURE) ? 1 : 0;
1376 INP_WUNLOCK(inp);
1377 error = sooptcopyout(sopt, &optval, sizeof optval);
1378 break;
1379 #endif
1380
1381 case TCP_NODELAY:
1382 optval = tp->t_flags & TF_NODELAY;
1383 INP_WUNLOCK(inp);
1384 error = sooptcopyout(sopt, &optval, sizeof optval);
1385 break;
1386 case TCP_MAXSEG:
1387 optval = tp->t_maxseg;
1388 INP_WUNLOCK(inp);
1389 error = sooptcopyout(sopt, &optval, sizeof optval);
1390 break;
1391 case TCP_NOOPT:
1392 optval = tp->t_flags & TF_NOOPT;
1393 INP_WUNLOCK(inp);
1394 error = sooptcopyout(sopt, &optval, sizeof optval);
1395 break;
1396 case TCP_NOPUSH:
1397 optval = tp->t_flags & TF_NOPUSH;
1398 INP_WUNLOCK(inp);
1399 error = sooptcopyout(sopt, &optval, sizeof optval);
1400 break;
1401 case TCP_INFO:
1402 tcp_fill_info(tp, &ti);
1403 INP_WUNLOCK(inp);
1404 error = sooptcopyout(sopt, &ti, sizeof ti);
1405 break;
1406 default:
1407 INP_WUNLOCK(inp);
1408 error = ENOPROTOOPT;
1409 break;
1410 }
1411 break;
1412 }
1413 return (error);
1414 }
1415 #undef INP_WLOCK_RECHECK
1416
1417 /*
1418 * tcp_sendspace and tcp_recvspace are the default send and receive window
1419 * sizes, respectively. These are obsolescent (this information should
1420 * be set by the route).
1421 */
1422 u_long tcp_sendspace = 1024*32;
1423 SYSCTL_ULONG(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW,
1424 &tcp_sendspace , 0, "Maximum outgoing TCP datagram size");
1425 u_long tcp_recvspace = 1024*64;
1426 SYSCTL_ULONG(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
1427 &tcp_recvspace , 0, "Maximum incoming TCP datagram size");
1428
1429 /*
1430 * Attach TCP protocol to socket, allocating
1431 * internet protocol control block, tcp control block,
1432 * bufer space, and entering LISTEN state if to accept connections.
1433 */
1434 static int
1435 tcp_attach(struct socket *so)
1436 {
1437 struct tcpcb *tp;
1438 struct inpcb *inp;
1439 int error;
1440
1441 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
1442 error = soreserve(so, tcp_sendspace, tcp_recvspace);
1443 if (error)
1444 return (error);
1445 }
1446 so->so_rcv.sb_flags |= SB_AUTOSIZE;
1447 so->so_snd.sb_flags |= SB_AUTOSIZE;
1448 INP_INFO_WLOCK(&V_tcbinfo);
1449 error = in_pcballoc(so, &V_tcbinfo);
1450 if (error) {
1451 INP_INFO_WUNLOCK(&V_tcbinfo);
1452 return (error);
1453 }
1454 inp = sotoinpcb(so);
1455 #ifdef INET6
1456 if (inp->inp_vflag & INP_IPV6PROTO) {
1457 inp->inp_vflag |= INP_IPV6;
1458 inp->in6p_hops = -1; /* use kernel default */
1459 }
1460 else
1461 #endif
1462 inp->inp_vflag |= INP_IPV4;
1463 tp = tcp_newtcpcb(inp);
1464 if (tp == NULL) {
1465 in_pcbdetach(inp);
1466 in_pcbfree(inp);
1467 INP_INFO_WUNLOCK(&V_tcbinfo);
1468 return (ENOBUFS);
1469 }
1470 tp->t_state = TCPS_CLOSED;
1471 INP_WUNLOCK(inp);
1472 INP_INFO_WUNLOCK(&V_tcbinfo);
1473 return (0);
1474 }
1475
1476 /*
1477 * Initiate (or continue) disconnect.
1478 * If embryonic state, just send reset (once).
1479 * If in ``let data drain'' option and linger null, just drop.
1480 * Otherwise (hard), mark socket disconnecting and drop
1481 * current input data; switch states based on user close, and
1482 * send segment to peer (with FIN).
1483 */
1484 static void
1485 tcp_disconnect(struct tcpcb *tp)
1486 {
1487 struct inpcb *inp = tp->t_inpcb;
1488 struct socket *so = inp->inp_socket;
1489
1490 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1491 INP_WLOCK_ASSERT(inp);
1492
1493 /*
1494 * Neither tcp_close() nor tcp_drop() should return NULL, as the
1495 * socket is still open.
1496 */
1497 if (tp->t_state < TCPS_ESTABLISHED) {
1498 tp = tcp_close(tp);
1499 KASSERT(tp != NULL,
1500 ("tcp_disconnect: tcp_close() returned NULL"));
1501 } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
1502 tp = tcp_drop(tp, 0);
1503 KASSERT(tp != NULL,
1504 ("tcp_disconnect: tcp_drop() returned NULL"));
1505 } else {
1506 soisdisconnecting(so);
1507 sbflush(&so->so_rcv);
1508 tcp_usrclosed(tp);
1509 if (!(inp->inp_flags & INP_DROPPED))
1510 tcp_output_disconnect(tp);
1511 }
1512 }
1513
1514 /*
1515 * User issued close, and wish to trail through shutdown states:
1516 * if never received SYN, just forget it. If got a SYN from peer,
1517 * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
1518 * If already got a FIN from peer, then almost done; go to LAST_ACK
1519 * state. In all other cases, have already sent FIN to peer (e.g.
1520 * after PRU_SHUTDOWN), and just have to play tedious game waiting
1521 * for peer to send FIN or not respond to keep-alives, etc.
1522 * We can let the user exit from the close as soon as the FIN is acked.
1523 */
1524 static void
1525 tcp_usrclosed(struct tcpcb *tp)
1526 {
1527
1528 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1529 INP_WLOCK_ASSERT(tp->t_inpcb);
1530
1531 switch (tp->t_state) {
1532 case TCPS_LISTEN:
1533 tcp_offload_listen_close(tp);
1534 /* FALLTHROUGH */
1535 case TCPS_CLOSED:
1536 tp->t_state = TCPS_CLOSED;
1537 tp = tcp_close(tp);
1538 /*
1539 * tcp_close() should never return NULL here as the socket is
1540 * still open.
1541 */
1542 KASSERT(tp != NULL,
1543 ("tcp_usrclosed: tcp_close() returned NULL"));
1544 break;
1545
1546 case TCPS_SYN_SENT:
1547 case TCPS_SYN_RECEIVED:
1548 tp->t_flags |= TF_NEEDFIN;
1549 break;
1550
1551 case TCPS_ESTABLISHED:
1552 tp->t_state = TCPS_FIN_WAIT_1;
1553 break;
1554
1555 case TCPS_CLOSE_WAIT:
1556 tp->t_state = TCPS_LAST_ACK;
1557 break;
1558 }
1559 if (tp->t_state >= TCPS_FIN_WAIT_2) {
1560 soisdisconnected(tp->t_inpcb->inp_socket);
1561 /* Prevent the connection hanging in FIN_WAIT_2 forever. */
1562 if (tp->t_state == TCPS_FIN_WAIT_2) {
1563 int timeout;
1564
1565 timeout = (tcp_fast_finwait2_recycle) ?
1566 tcp_finwait2_timeout : tcp_maxidle;
1567 tcp_timer_activate(tp, TT_2MSL, timeout);
1568 }
1569 }
1570 }
1571
1572 #ifdef DDB
1573 static void
1574 db_print_indent(int indent)
1575 {
1576 int i;
1577
1578 for (i = 0; i < indent; i++)
1579 db_printf(" ");
1580 }
1581
1582 static void
1583 db_print_tstate(int t_state)
1584 {
1585
1586 switch (t_state) {
1587 case TCPS_CLOSED:
1588 db_printf("TCPS_CLOSED");
1589 return;
1590
1591 case TCPS_LISTEN:
1592 db_printf("TCPS_LISTEN");
1593 return;
1594
1595 case TCPS_SYN_SENT:
1596 db_printf("TCPS_SYN_SENT");
1597 return;
1598
1599 case TCPS_SYN_RECEIVED:
1600 db_printf("TCPS_SYN_RECEIVED");
1601 return;
1602
1603 case TCPS_ESTABLISHED:
1604 db_printf("TCPS_ESTABLISHED");
1605 return;
1606
1607 case TCPS_CLOSE_WAIT:
1608 db_printf("TCPS_CLOSE_WAIT");
1609 return;
1610
1611 case TCPS_FIN_WAIT_1:
1612 db_printf("TCPS_FIN_WAIT_1");
1613 return;
1614
1615 case TCPS_CLOSING:
1616 db_printf("TCPS_CLOSING");
1617 return;
1618
1619 case TCPS_LAST_ACK:
1620 db_printf("TCPS_LAST_ACK");
1621 return;
1622
1623 case TCPS_FIN_WAIT_2:
1624 db_printf("TCPS_FIN_WAIT_2");
1625 return;
1626
1627 case TCPS_TIME_WAIT:
1628 db_printf("TCPS_TIME_WAIT");
1629 return;
1630
1631 default:
1632 db_printf("unknown");
1633 return;
1634 }
1635 }
1636
1637 static void
1638 db_print_tflags(u_int t_flags)
1639 {
1640 int comma;
1641
1642 comma = 0;
1643 if (t_flags & TF_ACKNOW) {
1644 db_printf("%sTF_ACKNOW", comma ? ", " : "");
1645 comma = 1;
1646 }
1647 if (t_flags & TF_DELACK) {
1648 db_printf("%sTF_DELACK", comma ? ", " : "");
1649 comma = 1;
1650 }
1651 if (t_flags & TF_NODELAY) {
1652 db_printf("%sTF_NODELAY", comma ? ", " : "");
1653 comma = 1;
1654 }
1655 if (t_flags & TF_NOOPT) {
1656 db_printf("%sTF_NOOPT", comma ? ", " : "");
1657 comma = 1;
1658 }
1659 if (t_flags & TF_SENTFIN) {
1660 db_printf("%sTF_SENTFIN", comma ? ", " : "");
1661 comma = 1;
1662 }
1663 if (t_flags & TF_REQ_SCALE) {
1664 db_printf("%sTF_REQ_SCALE", comma ? ", " : "");
1665 comma = 1;
1666 }
1667 if (t_flags & TF_RCVD_SCALE) {
1668 db_printf("%sTF_RECVD_SCALE", comma ? ", " : "");
1669 comma = 1;
1670 }
1671 if (t_flags & TF_REQ_TSTMP) {
1672 db_printf("%sTF_REQ_TSTMP", comma ? ", " : "");
1673 comma = 1;
1674 }
1675 if (t_flags & TF_RCVD_TSTMP) {
1676 db_printf("%sTF_RCVD_TSTMP", comma ? ", " : "");
1677 comma = 1;
1678 }
1679 if (t_flags & TF_SACK_PERMIT) {
1680 db_printf("%sTF_SACK_PERMIT", comma ? ", " : "");
1681 comma = 1;
1682 }
1683 if (t_flags & TF_NEEDSYN) {
1684 db_printf("%sTF_NEEDSYN", comma ? ", " : "");
1685 comma = 1;
1686 }
1687 if (t_flags & TF_NEEDFIN) {
1688 db_printf("%sTF_NEEDFIN", comma ? ", " : "");
1689 comma = 1;
1690 }
1691 if (t_flags & TF_NOPUSH) {
1692 db_printf("%sTF_NOPUSH", comma ? ", " : "");
1693 comma = 1;
1694 }
1695 if (t_flags & TF_NOPUSH) {
1696 db_printf("%sTF_NOPUSH", comma ? ", " : "");
1697 comma = 1;
1698 }
1699 if (t_flags & TF_MORETOCOME) {
1700 db_printf("%sTF_MORETOCOME", comma ? ", " : "");
1701 comma = 1;
1702 }
1703 if (t_flags & TF_LQ_OVERFLOW) {
1704 db_printf("%sTF_LQ_OVERFLOW", comma ? ", " : "");
1705 comma = 1;
1706 }
1707 if (t_flags & TF_LASTIDLE) {
1708 db_printf("%sTF_LASTIDLE", comma ? ", " : "");
1709 comma = 1;
1710 }
1711 if (t_flags & TF_RXWIN0SENT) {
1712 db_printf("%sTF_RXWIN0SENT", comma ? ", " : "");
1713 comma = 1;
1714 }
1715 if (t_flags & TF_FASTRECOVERY) {
1716 db_printf("%sTF_FASTRECOVERY", comma ? ", " : "");
1717 comma = 1;
1718 }
1719 if (t_flags & TF_WASFRECOVERY) {
1720 db_printf("%sTF_WASFRECOVERY", comma ? ", " : "");
1721 comma = 1;
1722 }
1723 if (t_flags & TF_SIGNATURE) {
1724 db_printf("%sTF_SIGNATURE", comma ? ", " : "");
1725 comma = 1;
1726 }
1727 if (t_flags & TF_FORCEDATA) {
1728 db_printf("%sTF_FORCEDATA", comma ? ", " : "");
1729 comma = 1;
1730 }
1731 if (t_flags & TF_TSO) {
1732 db_printf("%sTF_TSO", comma ? ", " : "");
1733 comma = 1;
1734 }
1735 if (t_flags & TF_ECN_PERMIT) {
1736 db_printf("%sTF_ECN_PERMIT", comma ? ", " : "");
1737 comma = 1;
1738 }
1739 }
1740
1741 static void
1742 db_print_toobflags(char t_oobflags)
1743 {
1744 int comma;
1745
1746 comma = 0;
1747 if (t_oobflags & TCPOOB_HAVEDATA) {
1748 db_printf("%sTCPOOB_HAVEDATA", comma ? ", " : "");
1749 comma = 1;
1750 }
1751 if (t_oobflags & TCPOOB_HADDATA) {
1752 db_printf("%sTCPOOB_HADDATA", comma ? ", " : "");
1753 comma = 1;
1754 }
1755 }
1756
1757 static void
1758 db_print_tcpcb(struct tcpcb *tp, const char *name, int indent)
1759 {
1760
1761 db_print_indent(indent);
1762 db_printf("%s at %p\n", name, tp);
1763
1764 indent += 2;
1765
1766 db_print_indent(indent);
1767 db_printf("t_segq first: %p t_segqlen: %d t_dupacks: %d\n",
1768 LIST_FIRST(&tp->t_segq), tp->t_segqlen, tp->t_dupacks);
1769
1770 db_print_indent(indent);
1771 db_printf("tt_rexmt: %p tt_persist: %p tt_keep: %p\n",
1772 &tp->t_timers->tt_rexmt, &tp->t_timers->tt_persist, &tp->t_timers->tt_keep);
1773
1774 db_print_indent(indent);
1775 db_printf("tt_2msl: %p tt_delack: %p t_inpcb: %p\n", &tp->t_timers->tt_2msl,
1776 &tp->t_timers->tt_delack, tp->t_inpcb);
1777
1778 db_print_indent(indent);
1779 db_printf("t_state: %d (", tp->t_state);
1780 db_print_tstate(tp->t_state);
1781 db_printf(")\n");
1782
1783 db_print_indent(indent);
1784 db_printf("t_flags: 0x%x (", tp->t_flags);
1785 db_print_tflags(tp->t_flags);
1786 db_printf(")\n");
1787
1788 db_print_indent(indent);
1789 db_printf("snd_una: 0x%08x snd_max: 0x%08x snd_nxt: x0%08x\n",
1790 tp->snd_una, tp->snd_max, tp->snd_nxt);
1791
1792 db_print_indent(indent);
1793 db_printf("snd_up: 0x%08x snd_wl1: 0x%08x snd_wl2: 0x%08x\n",
1794 tp->snd_up, tp->snd_wl1, tp->snd_wl2);
1795
1796 db_print_indent(indent);
1797 db_printf("iss: 0x%08x irs: 0x%08x rcv_nxt: 0x%08x\n",
1798 tp->iss, tp->irs, tp->rcv_nxt);
1799
1800 db_print_indent(indent);
1801 db_printf("rcv_adv: 0x%08x rcv_wnd: %lu rcv_up: 0x%08x\n",
1802 tp->rcv_adv, tp->rcv_wnd, tp->rcv_up);
1803
1804 db_print_indent(indent);
1805 db_printf("snd_wnd: %lu snd_cwnd: %lu snd_bwnd: %lu\n",
1806 tp->snd_wnd, tp->snd_cwnd, tp->snd_bwnd);
1807
1808 db_print_indent(indent);
1809 db_printf("snd_ssthresh: %lu snd_bandwidth: %lu snd_recover: "
1810 "0x%08x\n", tp->snd_ssthresh, tp->snd_bandwidth,
1811 tp->snd_recover);
1812
1813 db_print_indent(indent);
1814 db_printf("t_maxopd: %u t_rcvtime: %u t_startime: %u\n",
1815 tp->t_maxopd, tp->t_rcvtime, tp->t_starttime);
1816
1817 db_print_indent(indent);
1818 db_printf("t_rttime: %u t_rtsq: 0x%08x t_bw_rtttime: %u\n",
1819 tp->t_rtttime, tp->t_rtseq, tp->t_bw_rtttime);
1820
1821 db_print_indent(indent);
1822 db_printf("t_bw_rtseq: 0x%08x t_rxtcur: %d t_maxseg: %u "
1823 "t_srtt: %d\n", tp->t_bw_rtseq, tp->t_rxtcur, tp->t_maxseg,
1824 tp->t_srtt);
1825
1826 db_print_indent(indent);
1827 db_printf("t_rttvar: %d t_rxtshift: %d t_rttmin: %u "
1828 "t_rttbest: %u\n", tp->t_rttvar, tp->t_rxtshift, tp->t_rttmin,
1829 tp->t_rttbest);
1830
1831 db_print_indent(indent);
1832 db_printf("t_rttupdated: %lu max_sndwnd: %lu t_softerror: %d\n",
1833 tp->t_rttupdated, tp->max_sndwnd, tp->t_softerror);
1834
1835 db_print_indent(indent);
1836 db_printf("t_oobflags: 0x%x (", tp->t_oobflags);
1837 db_print_toobflags(tp->t_oobflags);
1838 db_printf(") t_iobc: 0x%02x\n", tp->t_iobc);
1839
1840 db_print_indent(indent);
1841 db_printf("snd_scale: %u rcv_scale: %u request_r_scale: %u\n",
1842 tp->snd_scale, tp->rcv_scale, tp->request_r_scale);
1843
1844 db_print_indent(indent);
1845 db_printf("ts_recent: %u ts_recent_age: %u\n",
1846 tp->ts_recent, tp->ts_recent_age);
1847
1848 db_print_indent(indent);
1849 db_printf("ts_offset: %u last_ack_sent: 0x%08x snd_cwnd_prev: "
1850 "%lu\n", tp->ts_offset, tp->last_ack_sent, tp->snd_cwnd_prev);
1851
1852 db_print_indent(indent);
1853 db_printf("snd_ssthresh_prev: %lu snd_recover_prev: 0x%08x "
1854 "t_badrxtwin: %u\n", tp->snd_ssthresh_prev,
1855 tp->snd_recover_prev, tp->t_badrxtwin);
1856
1857 db_print_indent(indent);
1858 db_printf("snd_numholes: %d snd_holes first: %p\n",
1859 tp->snd_numholes, TAILQ_FIRST(&tp->snd_holes));
1860
1861 db_print_indent(indent);
1862 db_printf("snd_fack: 0x%08x rcv_numsacks: %d sack_newdata: "
1863 "0x%08x\n", tp->snd_fack, tp->rcv_numsacks, tp->sack_newdata);
1864
1865 /* Skip sackblks, sackhint. */
1866
1867 db_print_indent(indent);
1868 db_printf("t_rttlow: %d rfbuf_ts: %u rfbuf_cnt: %d\n",
1869 tp->t_rttlow, tp->rfbuf_ts, tp->rfbuf_cnt);
1870 }
1871
1872 DB_SHOW_COMMAND(tcpcb, db_show_tcpcb)
1873 {
1874 struct tcpcb *tp;
1875
1876 if (!have_addr) {
1877 db_printf("usage: show tcpcb <addr>\n");
1878 return;
1879 }
1880 tp = (struct tcpcb *)addr;
1881
1882 db_print_tcpcb(tp, "tcpcb", 0);
1883 }
1884 #endif
Cache object: de88aa3d33c566d0870f1e8de64481bb
|