Fix a client tail -f vs server-appended file data corruption case by
[dragonfly.git] / sys / netinet / tcp_usrreq.c
CommitLineData
984263bc
MD
1/*
2 * Copyright (c) 1982, 1986, 1988, 1993
3 * The Regents of the University of California. All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 3. All advertising materials mentioning features or use of this software
14 * must display the following acknowledgement:
15 * This product includes software developed by the University of
16 * California, Berkeley and its contributors.
17 * 4. Neither the name of the University nor the names of its contributors
18 * may be used to endorse or promote products derived from this software
19 * without specific prior written permission.
20 *
21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * SUCH DAMAGE.
32 *
33 * From: @(#)tcp_usrreq.c 8.2 (Berkeley) 1/3/94
34 * $FreeBSD: src/sys/netinet/tcp_usrreq.c,v 1.51.2.17 2002/10/11 11:46:44 ume Exp $
21a9b6a9 35 * $DragonFly: src/sys/netinet/tcp_usrreq.c,v 1.18 2004/04/22 04:31:27 dillon Exp $
984263bc
MD
36 */
37
38#include "opt_ipsec.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/kernel.h>
7fe56515 45#include <sys/malloc.h>
984263bc 46#include <sys/sysctl.h>
d371a63a
JH
47#include <sys/globaldata.h>
48#include <sys/thread.h>
49
984263bc
MD
50#include <sys/mbuf.h>
51#ifdef INET6
52#include <sys/domain.h>
53#endif /* INET6 */
54#include <sys/socket.h>
55#include <sys/socketvar.h>
56#include <sys/protosw.h>
57
7fe56515
JH
58#include <sys/msgport2.h>
59
984263bc 60#include <net/if.h>
7fe56515 61#include <net/netisr.h>
984263bc
MD
62#include <net/route.h>
63
64#include <netinet/in.h>
65#include <netinet/in_systm.h>
66#ifdef INET6
67#include <netinet/ip6.h>
68#endif
69#include <netinet/in_pcb.h>
70#ifdef INET6
71#include <netinet6/in6_pcb.h>
72#endif
73#include <netinet/in_var.h>
74#include <netinet/ip_var.h>
75#ifdef INET6
76#include <netinet6/ip6_var.h>
77#endif
78#include <netinet/tcp.h>
79#include <netinet/tcp_fsm.h>
80#include <netinet/tcp_seq.h>
81#include <netinet/tcp_timer.h>
82#include <netinet/tcp_var.h>
83#include <netinet/tcpip.h>
84#ifdef TCPDEBUG
85#include <netinet/tcp_debug.h>
86#endif
87
88#ifdef IPSEC
89#include <netinet6/ipsec.h>
90#endif /*IPSEC*/
91
92/*
93 * TCP protocol interface to socket abstraction.
94 */
95extern char *tcpstates[]; /* XXX ??? */
96
e4700d00 97static int tcp_attach (struct socket *, struct pru_attach_info *);
42a7fc75
RG
98static int tcp_connect (struct tcpcb *, struct sockaddr *,
99 struct thread *);
984263bc 100#ifdef INET6
42a7fc75
RG
101static int tcp6_connect (struct tcpcb *, struct sockaddr *,
102 struct thread *);
984263bc
MD
103#endif /* INET6 */
104static struct tcpcb *
42a7fc75 105 tcp_disconnect (struct tcpcb *);
984263bc 106static struct tcpcb *
42a7fc75 107 tcp_usrclosed (struct tcpcb *);
984263bc
MD
108
109#ifdef TCPDEBUG
110#define TCPDEBUG0 int ostate = 0
111#define TCPDEBUG1() ostate = tp ? tp->t_state : 0
112#define TCPDEBUG2(req) if (tp && (so->so_options & SO_DEBUG)) \
113 tcp_trace(TA_USER, ostate, tp, 0, 0, req)
114#else
115#define TCPDEBUG0
116#define TCPDEBUG1()
117#define TCPDEBUG2(req)
118#endif
119
120/*
121 * TCP attaches to socket via pru_attach(), reserving space,
122 * and an internet control block.
123 */
124static int
e4700d00 125tcp_usr_attach(struct socket *so, int proto, struct pru_attach_info *ai)
984263bc
MD
126{
127 int s = splnet();
128 int error;
129 struct inpcb *inp = sotoinpcb(so);
130 struct tcpcb *tp = 0;
131 TCPDEBUG0;
132
133 TCPDEBUG1();
134 if (inp) {
135 error = EISCONN;
136 goto out;
137 }
138
e4700d00 139 error = tcp_attach(so, ai);
984263bc
MD
140 if (error)
141 goto out;
142
143 if ((so->so_options & SO_LINGER) && so->so_linger == 0)
144 so->so_linger = TCP_LINGERTIME;
145 tp = sototcpcb(so);
146out:
147 TCPDEBUG2(PRU_ATTACH);
148 splx(s);
149 return error;
150}
151
152/*
153 * pru_detach() detaches the TCP protocol from the socket.
154 * If the protocol state is non-embryonic, then can't
155 * do this directly: have to initiate a pru_disconnect(),
156 * which may finish later; embryonic TCB's can just
157 * be discarded here.
158 */
159static int
160tcp_usr_detach(struct socket *so)
161{
162 int s = splnet();
163 int error = 0;
164 struct inpcb *inp = sotoinpcb(so);
165 struct tcpcb *tp;
166 TCPDEBUG0;
167
168 if (inp == 0) {
169 splx(s);
170 return EINVAL; /* XXX */
171 }
172 tp = intotcpcb(inp);
173 TCPDEBUG1();
174 tp = tcp_disconnect(tp);
175
176 TCPDEBUG2(PRU_DETACH);
177 splx(s);
178 return error;
179}
180
181#define COMMON_START() TCPDEBUG0; \
182 do { \
183 if (inp == 0) { \
184 splx(s); \
185 return EINVAL; \
186 } \
187 tp = intotcpcb(inp); \
188 TCPDEBUG1(); \
189 } while(0)
190
191#define COMMON_END(req) out: TCPDEBUG2(req); splx(s); return error; goto out
192
193
194/*
195 * Give the socket an address.
196 */
197static int
dadab5e9 198tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
984263bc
MD
199{
200 int s = splnet();
201 int error = 0;
202 struct inpcb *inp = sotoinpcb(so);
203 struct tcpcb *tp;
204 struct sockaddr_in *sinp;
205
206 COMMON_START();
207
208 /*
209 * Must check for multicast addresses and disallow binding
210 * to them.
211 */
212 sinp = (struct sockaddr_in *)nam;
213 if (sinp->sin_family == AF_INET &&
214 IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
215 error = EAFNOSUPPORT;
216 goto out;
217 }
dadab5e9 218 error = in_pcbbind(inp, nam, td);
984263bc
MD
219 if (error)
220 goto out;
221 COMMON_END(PRU_BIND);
222
223}
224
225#ifdef INET6
226static int
dadab5e9 227tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
984263bc
MD
228{
229 int s = splnet();
230 int error = 0;
231 struct inpcb *inp = sotoinpcb(so);
232 struct tcpcb *tp;
233 struct sockaddr_in6 *sin6p;
234
235 COMMON_START();
236
237 /*
238 * Must check for multicast addresses and disallow binding
239 * to them.
240 */
241 sin6p = (struct sockaddr_in6 *)nam;
242 if (sin6p->sin6_family == AF_INET6 &&
243 IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
244 error = EAFNOSUPPORT;
245 goto out;
246 }
247 inp->inp_vflag &= ~INP_IPV4;
248 inp->inp_vflag |= INP_IPV6;
249 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
250 if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr))
251 inp->inp_vflag |= INP_IPV4;
252 else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
253 struct sockaddr_in sin;
254
255 in6_sin6_2_sin(&sin, sin6p);
256 inp->inp_vflag |= INP_IPV4;
257 inp->inp_vflag &= ~INP_IPV6;
dadab5e9 258 error = in_pcbbind(inp, (struct sockaddr *)&sin, td);
984263bc
MD
259 goto out;
260 }
261 }
dadab5e9 262 error = in6_pcbbind(inp, nam, td);
984263bc
MD
263 if (error)
264 goto out;
265 COMMON_END(PRU_BIND);
266}
267#endif /* INET6 */
268
269/*
270 * Prepare to accept connections.
271 */
272static int
dadab5e9 273tcp_usr_listen(struct socket *so, struct thread *td)
984263bc
MD
274{
275 int s = splnet();
276 int error = 0;
277 struct inpcb *inp = sotoinpcb(so);
278 struct tcpcb *tp;
279
280 COMMON_START();
281 if (inp->inp_lport == 0)
dadab5e9 282 error = in_pcbbind(inp, (struct sockaddr *)0, td);
e61cbecc 283 if (error == 0) {
984263bc 284 tp->t_state = TCPS_LISTEN;
e61cbecc
JH
285 in_pcbinswildcardhash(inp);
286 }
984263bc
MD
287 COMMON_END(PRU_LISTEN);
288}
289
290#ifdef INET6
291static int
dadab5e9 292tcp6_usr_listen(struct socket *so, struct thread *td)
984263bc
MD
293{
294 int s = splnet();
295 int error = 0;
296 struct inpcb *inp = sotoinpcb(so);
297 struct tcpcb *tp;
298
299 COMMON_START();
300 if (inp->inp_lport == 0) {
301 inp->inp_vflag &= ~INP_IPV4;
302 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0)
303 inp->inp_vflag |= INP_IPV4;
dadab5e9 304 error = in6_pcbbind(inp, (struct sockaddr *)0, td);
984263bc
MD
305 }
306 if (error == 0)
307 tp->t_state = TCPS_LISTEN;
308 COMMON_END(PRU_LISTEN);
309}
310#endif /* INET6 */
311
312/*
313 * Initiate connection to peer.
314 * Create a template for use in transmissions on this connection.
315 * Enter SYN_SENT state, and mark socket as connecting.
316 * Start keep-alive timer, and seed output sequence space.
317 * Send initial segment on connection.
318 */
319static int
dadab5e9 320tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
984263bc
MD
321{
322 int s = splnet();
323 int error = 0;
324 struct inpcb *inp = sotoinpcb(so);
325 struct tcpcb *tp;
326 struct sockaddr_in *sinp;
327
328 COMMON_START();
329
330 /*
331 * Must disallow TCP ``connections'' to multicast addresses.
332 */
333 sinp = (struct sockaddr_in *)nam;
334 if (sinp->sin_family == AF_INET
335 && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
336 error = EAFNOSUPPORT;
337 goto out;
338 }
339
dadab5e9 340 prison_remote_ip(td, 0, &sinp->sin_addr.s_addr);
984263bc 341
dadab5e9 342 if ((error = tcp_connect(tp, nam, td)) != 0)
984263bc
MD
343 goto out;
344 error = tcp_output(tp);
345 COMMON_END(PRU_CONNECT);
346}
347
348#ifdef INET6
349static int
dadab5e9 350tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
984263bc
MD
351{
352 int s = splnet();
353 int error = 0;
354 struct inpcb *inp = sotoinpcb(so);
355 struct tcpcb *tp;
356 struct sockaddr_in6 *sin6p;
357
358 COMMON_START();
359
360 /*
361 * Must disallow TCP ``connections'' to multicast addresses.
362 */
363 sin6p = (struct sockaddr_in6 *)nam;
364 if (sin6p->sin6_family == AF_INET6
365 && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
366 error = EAFNOSUPPORT;
367 goto out;
368 }
369
370 if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
371 struct sockaddr_in sin;
372
373 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
374 error = EINVAL;
375 goto out;
376 }
377
378 in6_sin6_2_sin(&sin, sin6p);
379 inp->inp_vflag |= INP_IPV4;
380 inp->inp_vflag &= ~INP_IPV6;
dadab5e9 381 if ((error = tcp_connect(tp, (struct sockaddr *)&sin, td)) != 0)
984263bc
MD
382 goto out;
383 error = tcp_output(tp);
384 goto out;
385 }
386 inp->inp_vflag &= ~INP_IPV4;
387 inp->inp_vflag |= INP_IPV6;
388 inp->inp_inc.inc_isipv6 = 1;
dadab5e9 389 if ((error = tcp6_connect(tp, nam, td)) != 0)
984263bc
MD
390 goto out;
391 error = tcp_output(tp);
392 COMMON_END(PRU_CONNECT);
393}
394#endif /* INET6 */
395
396/*
397 * Initiate disconnect from peer.
398 * If connection never passed embryonic stage, just drop;
399 * else if don't need to let data drain, then can just drop anyways,
400 * else have to begin TCP shutdown process: mark socket disconnecting,
401 * drain unread data, state switch to reflect user close, and
402 * send segment (e.g. FIN) to peer. Socket will be really disconnected
403 * when peer sends FIN and acks ours.
404 *
405 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
406 */
407static int
408tcp_usr_disconnect(struct socket *so)
409{
410 int s = splnet();
411 int error = 0;
412 struct inpcb *inp = sotoinpcb(so);
413 struct tcpcb *tp;
414
415 COMMON_START();
416 tp = tcp_disconnect(tp);
417 COMMON_END(PRU_DISCONNECT);
418}
419
420/*
421 * Accept a connection. Essentially all the work is
422 * done at higher levels; just return the address
423 * of the peer, storing through addr.
424 */
425static int
426tcp_usr_accept(struct socket *so, struct sockaddr **nam)
427{
428 int s = splnet();
429 int error = 0;
430 struct inpcb *inp = sotoinpcb(so);
431 struct tcpcb *tp = NULL;
432 TCPDEBUG0;
433
434 if (so->so_state & SS_ISDISCONNECTED) {
435 error = ECONNABORTED;
436 goto out;
437 }
438 if (inp == 0) {
439 splx(s);
440 return (EINVAL);
441 }
442 tp = intotcpcb(inp);
443 TCPDEBUG1();
444 in_setpeeraddr(so, nam);
445 COMMON_END(PRU_ACCEPT);
446}
447
448#ifdef INET6
449static int
450tcp6_usr_accept(struct socket *so, struct sockaddr **nam)
451{
452 int s = splnet();
453 int error = 0;
454 struct inpcb *inp = sotoinpcb(so);
455 struct tcpcb *tp = NULL;
456 TCPDEBUG0;
457
458 if (so->so_state & SS_ISDISCONNECTED) {
459 error = ECONNABORTED;
460 goto out;
461 }
462 if (inp == 0) {
463 splx(s);
464 return (EINVAL);
465 }
466 tp = intotcpcb(inp);
467 TCPDEBUG1();
468 in6_mapped_peeraddr(so, nam);
469 COMMON_END(PRU_ACCEPT);
470}
471#endif /* INET6 */
472/*
473 * Mark the connection as being incapable of further output.
474 */
475static int
476tcp_usr_shutdown(struct socket *so)
477{
478 int s = splnet();
479 int error = 0;
480 struct inpcb *inp = sotoinpcb(so);
481 struct tcpcb *tp;
482
483 COMMON_START();
484 socantsendmore(so);
485 tp = tcp_usrclosed(tp);
486 if (tp)
487 error = tcp_output(tp);
488 COMMON_END(PRU_SHUTDOWN);
489}
490
491/*
492 * After a receive, possibly send window update to peer.
493 */
494static int
495tcp_usr_rcvd(struct socket *so, int flags)
496{
497 int s = splnet();
498 int error = 0;
499 struct inpcb *inp = sotoinpcb(so);
500 struct tcpcb *tp;
501
502 COMMON_START();
503 tcp_output(tp);
504 COMMON_END(PRU_RCVD);
505}
506
507/*
508 * Do a send by putting data in output queue and updating urgent
509 * marker if URG set. Possibly send more data. Unlike the other
510 * pru_*() routines, the mbuf chains are our responsibility. We
511 * must either enqueue them or free them. The other pru_* routines
512 * generally are caller-frees.
513 */
514static int
515tcp_usr_send(struct socket *so, int flags, struct mbuf *m,
dadab5e9 516 struct sockaddr *nam, struct mbuf *control, struct thread *td)
984263bc
MD
517{
518 int s = splnet();
519 int error = 0;
520 struct inpcb *inp = sotoinpcb(so);
521 struct tcpcb *tp;
522#ifdef INET6
523 int isipv6;
524#endif
525 TCPDEBUG0;
526
527 if (inp == NULL) {
528 /*
529 * OOPS! we lost a race, the TCP session got reset after
530 * we checked SS_CANTSENDMORE, eg: while doing uiomove or a
531 * network interrupt in the non-splnet() section of sosend().
532 */
533 if (m)
534 m_freem(m);
535 if (control)
536 m_freem(control);
537 error = ECONNRESET; /* XXX EPIPE? */
538 tp = NULL;
539 TCPDEBUG1();
540 goto out;
541 }
542#ifdef INET6
543 isipv6 = nam && nam->sa_family == AF_INET6;
544#endif /* INET6 */
545 tp = intotcpcb(inp);
546 TCPDEBUG1();
547 if (control) {
548 /* TCP doesn't do control messages (rights, creds, etc) */
549 if (control->m_len) {
550 m_freem(control);
551 if (m)
552 m_freem(m);
553 error = EINVAL;
554 goto out;
555 }
556 m_freem(control); /* empty control, just free it */
557 }
558 if(!(flags & PRUS_OOB)) {
559 sbappend(&so->so_snd, m);
560 if (nam && tp->t_state < TCPS_SYN_SENT) {
561 /*
562 * Do implied connect if not yet connected,
563 * initialize window to default value, and
564 * initialize maxseg/maxopd using peer's cached
565 * MSS.
566 */
567#ifdef INET6
568 if (isipv6)
dadab5e9 569 error = tcp6_connect(tp, nam, td);
984263bc
MD
570 else
571#endif /* INET6 */
dadab5e9 572 error = tcp_connect(tp, nam, td);
984263bc
MD
573 if (error)
574 goto out;
575 tp->snd_wnd = TTCP_CLIENT_SND_WND;
576 tcp_mss(tp, -1);
577 }
578
579 if (flags & PRUS_EOF) {
580 /*
581 * Close the send side of the connection after
582 * the data is sent.
583 */
584 socantsendmore(so);
585 tp = tcp_usrclosed(tp);
586 }
587 if (tp != NULL) {
588 if (flags & PRUS_MORETOCOME)
589 tp->t_flags |= TF_MORETOCOME;
590 error = tcp_output(tp);
591 if (flags & PRUS_MORETOCOME)
592 tp->t_flags &= ~TF_MORETOCOME;
593 }
594 } else {
595 if (sbspace(&so->so_snd) < -512) {
596 m_freem(m);
597 error = ENOBUFS;
598 goto out;
599 }
600 /*
601 * According to RFC961 (Assigned Protocols),
602 * the urgent pointer points to the last octet
603 * of urgent data. We continue, however,
604 * to consider it to indicate the first octet
605 * of data past the urgent section.
606 * Otherwise, snd_up should be one lower.
607 */
608 sbappend(&so->so_snd, m);
609 if (nam && tp->t_state < TCPS_SYN_SENT) {
610 /*
611 * Do implied connect if not yet connected,
612 * initialize window to default value, and
613 * initialize maxseg/maxopd using peer's cached
614 * MSS.
615 */
616#ifdef INET6
617 if (isipv6)
dadab5e9 618 error = tcp6_connect(tp, nam, td);
984263bc
MD
619 else
620#endif /* INET6 */
dadab5e9 621 error = tcp_connect(tp, nam, td);
984263bc
MD
622 if (error)
623 goto out;
624 tp->snd_wnd = TTCP_CLIENT_SND_WND;
625 tcp_mss(tp, -1);
626 }
627 tp->snd_up = tp->snd_una + so->so_snd.sb_cc;
628 tp->t_force = 1;
629 error = tcp_output(tp);
630 tp->t_force = 0;
631 }
632 COMMON_END((flags & PRUS_OOB) ? PRU_SENDOOB :
633 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
634}
635
636/*
637 * Abort the TCP.
638 */
639static int
640tcp_usr_abort(struct socket *so)
641{
642 int s = splnet();
643 int error = 0;
644 struct inpcb *inp = sotoinpcb(so);
645 struct tcpcb *tp;
646
647 COMMON_START();
648 tp = tcp_drop(tp, ECONNABORTED);
649 COMMON_END(PRU_ABORT);
650}
651
652/*
653 * Receive out-of-band data.
654 */
655static int
656tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags)
657{
658 int s = splnet();
659 int error = 0;
660 struct inpcb *inp = sotoinpcb(so);
661 struct tcpcb *tp;
662
663 COMMON_START();
664 if ((so->so_oobmark == 0 &&
665 (so->so_state & SS_RCVATMARK) == 0) ||
666 so->so_options & SO_OOBINLINE ||
667 tp->t_oobflags & TCPOOB_HADDATA) {
668 error = EINVAL;
669 goto out;
670 }
671 if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
672 error = EWOULDBLOCK;
673 goto out;
674 }
675 m->m_len = 1;
676 *mtod(m, caddr_t) = tp->t_iobc;
677 if ((flags & MSG_PEEK) == 0)
678 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
679 COMMON_END(PRU_RCVOOB);
680}
681
682/* xxx - should be const */
683struct pr_usrreqs tcp_usrreqs = {
684 tcp_usr_abort, tcp_usr_accept, tcp_usr_attach, tcp_usr_bind,
685 tcp_usr_connect, pru_connect2_notsupp, in_control, tcp_usr_detach,
686 tcp_usr_disconnect, tcp_usr_listen, in_setpeeraddr, tcp_usr_rcvd,
687 tcp_usr_rcvoob, tcp_usr_send, pru_sense_null, tcp_usr_shutdown,
688 in_setsockaddr, sosend, soreceive, sopoll
689};
690
691#ifdef INET6
692struct pr_usrreqs tcp6_usrreqs = {
693 tcp_usr_abort, tcp6_usr_accept, tcp_usr_attach, tcp6_usr_bind,
694 tcp6_usr_connect, pru_connect2_notsupp, in6_control, tcp_usr_detach,
695 tcp_usr_disconnect, tcp6_usr_listen, in6_mapped_peeraddr, tcp_usr_rcvd,
696 tcp_usr_rcvoob, tcp_usr_send, pru_sense_null, tcp_usr_shutdown,
697 in6_mapped_sockaddr, sosend, soreceive, sopoll
698};
699#endif /* INET6 */
700
7fe56515
JH
701struct netmsg_tcp_connect {
702 struct lwkt_msg nm_lmsg;
7fe56515
JH
703 struct tcpcb *nm_tp;
704 struct sockaddr_in *nm_sin;
705 struct sockaddr_in *nm_ifsin;
706};
707
984263bc 708static int
7fe56515
JH
709tcp_connect_oncpu(struct tcpcb *tp, struct sockaddr_in *sin,
710 struct sockaddr_in *if_sin)
984263bc
MD
711{
712 struct inpcb *inp = tp->t_inpcb, *oinp;
713 struct socket *so = inp->inp_socket;
714 struct tcpcb *otp;
984263bc
MD
715 struct rmxp_tao *taop;
716 struct rmxp_tao tao_noncached;
984263bc 717
7fe56515 718 oinp = in_pcblookup_hash(&tcbinfo[mycpu->gd_cpuid],
984263bc 719 sin->sin_addr, sin->sin_port,
13d8907a
JH
720 inp->inp_laddr.s_addr != INADDR_ANY ?
721 inp->inp_laddr : if_sin->sin_addr,
7fe56515 722 inp->inp_lport, 0, NULL);
13d8907a 723 if (oinp != NULL) {
984263bc 724 if (oinp != inp && (otp = intotcpcb(oinp)) != NULL &&
13d8907a 725 otp->t_state == TCPS_TIME_WAIT &&
984263bc
MD
726 (ticks - otp->t_starttime) < tcp_msl &&
727 (otp->t_flags & TF_RCVD_CC))
13d8907a 728 (void) tcp_close(otp);
984263bc 729 else
13d8907a 730 return (EADDRINUSE);
984263bc
MD
731 }
732 if (inp->inp_laddr.s_addr == INADDR_ANY)
13d8907a 733 inp->inp_laddr = if_sin->sin_addr;
984263bc
MD
734 inp->inp_faddr = sin->sin_addr;
735 inp->inp_fport = sin->sin_port;
7fe56515 736 inp->inp_cpcbinfo = &tcbinfo[mycpu->gd_cpuid];
13d8907a 737 in_pcbinsconnhash(inp);
984263bc
MD
738
739 /* Compute window scaling to request. */
740 while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
741 (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.sb_hiwat)
742 tp->request_r_scale++;
743
744 soisconnecting(so);
745 tcpstat.tcps_connattempt++;
746 tp->t_state = TCPS_SYN_SENT;
747 callout_reset(tp->tt_keep, tcp_keepinit, tcp_timer_keep, tp);
748 tp->iss = tcp_new_isn(tp);
749 tp->t_bw_rtseq = tp->iss;
750 tcp_sendseqinit(tp);
751
752 /*
753 * Generate a CC value for this connection and
754 * check whether CC or CCnew should be used.
755 */
756 if ((taop = tcp_gettaocache(&tp->t_inpcb->inp_inc)) == NULL) {
757 taop = &tao_noncached;
758 bzero(taop, sizeof(*taop));
759 }
760
761 tp->cc_send = CC_INC(tcp_ccgen);
762 if (taop->tao_ccsent != 0 &&
763 CC_GEQ(tp->cc_send, taop->tao_ccsent)) {
764 taop->tao_ccsent = tp->cc_send;
765 } else {
766 taop->tao_ccsent = 0;
767 tp->t_flags |= TF_SENDCCNEW;
768 }
769
13d8907a 770 return (0);
984263bc
MD
771}
772
b44419cb
MD
773#if defined(SMP)
774
775static int
776tcp_connect_handler(lwkt_msg_t lmsg)
7fe56515 777{
b44419cb 778 struct netmsg_tcp_connect *msg = (void *)lmsg;
7fe56515
JH
779 int error;
780
781 error = tcp_connect_oncpu(msg->nm_tp, msg->nm_sin, msg->nm_ifsin);
b44419cb
MD
782 lwkt_replymsg(lmsg, error);
783 return(EASYNC);
7fe56515
JH
784}
785
b44419cb
MD
786#endif
787
7fe56515
JH
788/*
789 * Common subroutine to open a TCP connection to remote host specified
790 * by struct sockaddr_in in mbuf *nam. Call in_pcbbind to assign a local
791 * port number if needed. Call in_pcbladdr to do the routing and to choose
792 * a local host address (interface). If there is an existing incarnation
793 * of the same connection in TIME-WAIT state and if the remote host was
794 * sending CC options and if the connection duration was < MSL, then
795 * truncate the previous TIME-WAIT state and proceed.
796 * Initialize connection parameters and enter SYN-SENT state.
797 */
798static int
799tcp_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
800{
801 struct inpcb *inp = tp->t_inpcb;
802 struct sockaddr_in *sin = (struct sockaddr_in *)nam;
803 struct sockaddr_in *if_sin;
804 int error;
805 boolean_t didbind = FALSE;
b44419cb 806#if defined(SMP)
7fe56515
JH
807 lwkt_port_t port;
808#endif
809
810 if (inp->inp_lport == 0) {
811 error = in_pcbbind(inp, (struct sockaddr *)NULL, td);
812 if (error)
813 return (error);
814 didbind = TRUE;
815 }
816
817 /*
818 * Cannot simply call in_pcbconnect, because there might be an
819 * earlier incarnation of this same connection still in
820 * TIME_WAIT state, creating an ADDRINUSE error.
821 */
822 error = in_pcbladdr(inp, nam, &if_sin);
823 if (error)
824 return (error);
825
b44419cb 826#if defined(SMP)
7fe56515
JH
827 port = tcp_addrport(sin->sin_addr.s_addr, sin->sin_port,
828 inp->inp_laddr.s_addr ?
829 inp->inp_laddr.s_addr : if_sin->sin_addr.s_addr,
830 inp->inp_lport);
831
832 if (port->mp_td != curthread) {
833 struct netmsg_tcp_connect *msg;
834
835 msg = malloc(sizeof(struct netmsg_tcp_connect), M_LWKTMSG,
21a9b6a9 836 M_INTWAIT);
b44419cb
MD
837 lwkt_initmsg(&msg->nm_lmsg, &curthread->td_msgport, 0,
838 lwkt_cmd_func(tcp_connect_handler),
839 lwkt_cmd_op_none);
7fe56515
JH
840 msg->nm_tp = tp;
841 msg->nm_sin = sin;
842 msg->nm_ifsin = if_sin;
843 error = lwkt_domsg(port, &msg->nm_lmsg);
b76bed62 844 free(msg, M_LWKTMSG);
7fe56515
JH
845 } else
846#endif
847 error = tcp_connect_oncpu(tp, sin, if_sin);
848
849 return (error);
850}
851
984263bc
MD
852#ifdef INET6
853static int
dadab5e9 854tcp6_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
984263bc
MD
855{
856 struct inpcb *inp = tp->t_inpcb, *oinp;
857 struct socket *so = inp->inp_socket;
858 struct tcpcb *otp;
859 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam;
860 struct in6_addr *addr6;
861 struct rmxp_tao *taop;
862 struct rmxp_tao tao_noncached;
863 int error;
864
865 if (inp->inp_lport == 0) {
dadab5e9 866 error = in6_pcbbind(inp, (struct sockaddr *)0, td);
984263bc
MD
867 if (error)
868 return error;
869 }
870
871 /*
872 * Cannot simply call in_pcbconnect, because there might be an
873 * earlier incarnation of this same connection still in
874 * TIME_WAIT state, creating an ADDRINUSE error.
875 */
876 error = in6_pcbladdr(inp, nam, &addr6);
877 if (error)
878 return error;
7fe56515 879 oinp = in6_pcblookup_hash(inp->inp_cpcbinfo,
984263bc 880 &sin6->sin6_addr, sin6->sin6_port,
13d8907a
JH
881 IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) ?
882 addr6 : &inp->in6p_laddr,
984263bc
MD
883 inp->inp_lport, 0, NULL);
884 if (oinp) {
885 if (oinp != inp && (otp = intotcpcb(oinp)) != NULL &&
886 otp->t_state == TCPS_TIME_WAIT &&
887 (ticks - otp->t_starttime) < tcp_msl &&
888 (otp->t_flags & TF_RCVD_CC))
889 otp = tcp_close(otp);
890 else
13d8907a 891 return (EADDRINUSE);
984263bc
MD
892 }
893 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
894 inp->in6p_laddr = *addr6;
895 inp->in6p_faddr = sin6->sin6_addr;
896 inp->inp_fport = sin6->sin6_port;
897 if ((sin6->sin6_flowinfo & IPV6_FLOWINFO_MASK) != NULL)
898 inp->in6p_flowinfo = sin6->sin6_flowinfo;
13d8907a 899 in_pcbinsconnhash(inp);
984263bc
MD
900
901 /* Compute window scaling to request. */
902 while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
903 (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.sb_hiwat)
904 tp->request_r_scale++;
905
906 soisconnecting(so);
907 tcpstat.tcps_connattempt++;
908 tp->t_state = TCPS_SYN_SENT;
909 callout_reset(tp->tt_keep, tcp_keepinit, tcp_timer_keep, tp);
910 tp->iss = tcp_new_isn(tp);
911 tp->t_bw_rtseq = tp->iss;
912 tcp_sendseqinit(tp);
913
914 /*
915 * Generate a CC value for this connection and
916 * check whether CC or CCnew should be used.
917 */
918 if ((taop = tcp_gettaocache(&tp->t_inpcb->inp_inc)) == NULL) {
919 taop = &tao_noncached;
920 bzero(taop, sizeof(*taop));
921 }
922
923 tp->cc_send = CC_INC(tcp_ccgen);
924 if (taop->tao_ccsent != 0 &&
925 CC_GEQ(tp->cc_send, taop->tao_ccsent)) {
926 taop->tao_ccsent = tp->cc_send;
927 } else {
928 taop->tao_ccsent = 0;
929 tp->t_flags |= TF_SENDCCNEW;
930 }
931
13d8907a 932 return (0);
984263bc
MD
933}
934#endif /* INET6 */
935
936/*
937 * The new sockopt interface makes it possible for us to block in the
938 * copyin/out step (if we take a page fault). Taking a page fault at
939 * splnet() is probably a Bad Thing. (Since sockets and pcbs both now
940 * use TSM, there probably isn't any need for this function to run at
941 * splnet() any more. This needs more examination.)
942 */
943int
944tcp_ctloutput(so, sopt)
945 struct socket *so;
946 struct sockopt *sopt;
947{
948 int error, opt, optval, s;
949 struct inpcb *inp;
950 struct tcpcb *tp;
951
952 error = 0;
953 s = splnet(); /* XXX */
954 inp = sotoinpcb(so);
955 if (inp == NULL) {
956 splx(s);
957 return (ECONNRESET);
958 }
959 if (sopt->sopt_level != IPPROTO_TCP) {
960#ifdef INET6
961 if (INP_CHECK_SOCKAF(so, AF_INET6))
962 error = ip6_ctloutput(so, sopt);
963 else
964#endif /* INET6 */
965 error = ip_ctloutput(so, sopt);
966 splx(s);
967 return (error);
968 }
969 tp = intotcpcb(inp);
970
971 switch (sopt->sopt_dir) {
972 case SOPT_SET:
973 switch (sopt->sopt_name) {
974 case TCP_NODELAY:
975 case TCP_NOOPT:
976 error = sooptcopyin(sopt, &optval, sizeof optval,
977 sizeof optval);
978 if (error)
979 break;
980
981 switch (sopt->sopt_name) {
982 case TCP_NODELAY:
983 opt = TF_NODELAY;
984 break;
985 case TCP_NOOPT:
986 opt = TF_NOOPT;
987 break;
988 default:
989 opt = 0; /* dead code to fool gcc */
990 break;
991 }
992
993 if (optval)
994 tp->t_flags |= opt;
995 else
996 tp->t_flags &= ~opt;
997 break;
998
999 case TCP_NOPUSH:
1000 error = sooptcopyin(sopt, &optval, sizeof optval,
1001 sizeof optval);
1002 if (error)
1003 break;
1004
1005 if (optval)
1006 tp->t_flags |= TF_NOPUSH;
1007 else {
1008 tp->t_flags &= ~TF_NOPUSH;
1009 error = tcp_output(tp);
1010 }
1011 break;
1012
1013 case TCP_MAXSEG:
1014 error = sooptcopyin(sopt, &optval, sizeof optval,
1015 sizeof optval);
1016 if (error)
1017 break;
1018
1019 if (optval > 0 && optval <= tp->t_maxseg)
1020 tp->t_maxseg = optval;
1021 else
1022 error = EINVAL;
1023 break;
1024
1025 default:
1026 error = ENOPROTOOPT;
1027 break;
1028 }
1029 break;
1030
1031 case SOPT_GET:
1032 switch (sopt->sopt_name) {
1033 case TCP_NODELAY:
1034 optval = tp->t_flags & TF_NODELAY;
1035 break;
1036 case TCP_MAXSEG:
1037 optval = tp->t_maxseg;
1038 break;
1039 case TCP_NOOPT:
1040 optval = tp->t_flags & TF_NOOPT;
1041 break;
1042 case TCP_NOPUSH:
1043 optval = tp->t_flags & TF_NOPUSH;
1044 break;
1045 default:
1046 error = ENOPROTOOPT;
1047 break;
1048 }
1049 if (error == 0)
1050 error = sooptcopyout(sopt, &optval, sizeof optval);
1051 break;
1052 }
1053 splx(s);
1054 return (error);
1055}
1056
1057/*
1058 * tcp_sendspace and tcp_recvspace are the default send and receive window
1059 * sizes, respectively. These are obsolescent (this information should
1060 * be set by the route).
1061 */
1062u_long tcp_sendspace = 1024*32;
1063SYSCTL_INT(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW,
1064 &tcp_sendspace , 0, "Maximum outgoing TCP datagram size");
1065u_long tcp_recvspace = 57344; /* largest multiple of PAGE_SIZE < 64k */
1066SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
1067 &tcp_recvspace , 0, "Maximum incoming TCP datagram size");
1068
1069/*
1070 * Attach TCP protocol to socket, allocating
1071 * internet protocol control block, tcp control block,
1072 * bufer space, and entering LISTEN state if to accept connections.
1073 */
1074static int
e4700d00 1075tcp_attach(struct socket *so, struct pru_attach_info *ai)
984263bc 1076{
2256ba69 1077 struct tcpcb *tp;
984263bc
MD
1078 struct inpcb *inp;
1079 int error;
6ca1a1cd 1080 int cpu;
984263bc
MD
1081#ifdef INET6
1082 int isipv6 = INP_CHECK_SOCKAF(so, AF_INET6) != NULL;
1083#endif
1084
1085 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
e4700d00
JH
1086 error = soreserve(so, tcp_sendspace, tcp_recvspace,
1087 ai->sb_rlimit);
984263bc
MD
1088 if (error)
1089 return (error);
1090 }
6ca1a1cd 1091 cpu = mycpu->gd_cpuid;
6ca1a1cd 1092 error = in_pcballoc(so, &tcbinfo[cpu]);
984263bc
MD
1093 if (error)
1094 return (error);
1095 inp = sotoinpcb(so);
1096#ifdef INET6
1097 if (isipv6) {
1098 inp->inp_vflag |= INP_IPV6;
1099 inp->in6p_hops = -1; /* use kernel default */
1100 }
1101 else
1102#endif
1103 inp->inp_vflag |= INP_IPV4;
1104 tp = tcp_newtcpcb(inp);
1105 if (tp == 0) {
1106 int nofd = so->so_state & SS_NOFDREF; /* XXX */
1107
1108 so->so_state &= ~SS_NOFDREF; /* don't free the socket yet */
1109#ifdef INET6
1110 if (isipv6)
1111 in6_pcbdetach(inp);
1112 else
1113#endif
1114 in_pcbdetach(inp);
1115 so->so_state |= nofd;
1116 return (ENOBUFS);
1117 }
1118 tp->t_state = TCPS_CLOSED;
1119 return (0);
1120}
1121
1122/*
1123 * Initiate (or continue) disconnect.
1124 * If embryonic state, just send reset (once).
1125 * If in ``let data drain'' option and linger null, just drop.
1126 * Otherwise (hard), mark socket disconnecting and drop
1127 * current input data; switch states based on user close, and
1128 * send segment to peer (with FIN).
1129 */
1130static struct tcpcb *
1131tcp_disconnect(tp)
2256ba69 1132 struct tcpcb *tp;
984263bc
MD
1133{
1134 struct socket *so = tp->t_inpcb->inp_socket;
1135
1136 if (tp->t_state < TCPS_ESTABLISHED)
1137 tp = tcp_close(tp);
1138 else if ((so->so_options & SO_LINGER) && so->so_linger == 0)
1139 tp = tcp_drop(tp, 0);
1140 else {
1141 soisdisconnecting(so);
1142 sbflush(&so->so_rcv);
1143 tp = tcp_usrclosed(tp);
1144 if (tp)
1145 (void) tcp_output(tp);
1146 }
1147 return (tp);
1148}
1149
1150/*
1151 * User issued close, and wish to trail through shutdown states:
1152 * if never received SYN, just forget it. If got a SYN from peer,
1153 * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
1154 * If already got a FIN from peer, then almost done; go to LAST_ACK
1155 * state. In all other cases, have already sent FIN to peer (e.g.
1156 * after PRU_SHUTDOWN), and just have to play tedious game waiting
1157 * for peer to send FIN or not respond to keep-alives, etc.
1158 * We can let the user exit from the close as soon as the FIN is acked.
1159 */
1160static struct tcpcb *
1161tcp_usrclosed(tp)
2256ba69 1162 struct tcpcb *tp;
984263bc
MD
1163{
1164
1165 switch (tp->t_state) {
1166
1167 case TCPS_CLOSED:
1168 case TCPS_LISTEN:
1169 tp->t_state = TCPS_CLOSED;
1170 tp = tcp_close(tp);
1171 break;
1172
1173 case TCPS_SYN_SENT:
1174 case TCPS_SYN_RECEIVED:
1175 tp->t_flags |= TF_NEEDFIN;
1176 break;
1177
1178 case TCPS_ESTABLISHED:
1179 tp->t_state = TCPS_FIN_WAIT_1;
1180 break;
1181
1182 case TCPS_CLOSE_WAIT:
1183 tp->t_state = TCPS_LAST_ACK;
1184 break;
1185 }
1186 if (tp && tp->t_state >= TCPS_FIN_WAIT_2) {
1187 soisdisconnected(tp->t_inpcb->inp_socket);
1188 /* To prevent the connection hanging in FIN_WAIT_2 forever. */
1189 if (tp->t_state == TCPS_FIN_WAIT_2)
1190 callout_reset(tp->tt_2msl, tcp_maxidle,
1191 tcp_timer_2msl, tp);
1192 }
1193 return (tp);
1194}
1195