xref: /dflybsd-src/sys/netinet/tcp_usrreq.c (revision f0140465f072dacd8b485e76b254a999329a1d3c)
1 /*
2  * Copyright (c) 2003, 2004 Jeffrey M. Hsu.  All rights reserved.
3  * Copyright (c) 2003, 2004 The DragonFly Project.  All rights reserved.
4  *
5  * This code is derived from software contributed to The DragonFly Project
6  * by Jeffrey M. Hsu.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. Neither the name of The DragonFly Project nor the names of its
17  *    contributors may be used to endorse or promote products derived
18  *    from this software without specific, prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
23  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
24  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
25  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
26  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
27  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
28  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
29  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
30  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  */
33 
34 /*
35  * Copyright (c) 1982, 1986, 1988, 1993
36  *	The Regents of the University of California.  All rights reserved.
37  *
38  * Redistribution and use in source and binary forms, with or without
39  * modification, are permitted provided that the following conditions
40  * are met:
41  * 1. Redistributions of source code must retain the above copyright
42  *    notice, this list of conditions and the following disclaimer.
43  * 2. Redistributions in binary form must reproduce the above copyright
44  *    notice, this list of conditions and the following disclaimer in the
45  *    documentation and/or other materials provided with the distribution.
46  * 3. All advertising materials mentioning features or use of this software
47  *    must display the following acknowledgement:
48  *	This product includes software developed by the University of
49  *	California, Berkeley and its contributors.
50  * 4. Neither the name of the University nor the names of its contributors
51  *    may be used to endorse or promote products derived from this software
52  *    without specific prior written permission.
53  *
54  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
55  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
56  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
57  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
58  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
59  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
60  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
61  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
62  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
63  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
64  * SUCH DAMAGE.
65  *
66  *	From: @(#)tcp_usrreq.c	8.2 (Berkeley) 1/3/94
67  * $FreeBSD: src/sys/netinet/tcp_usrreq.c,v 1.51.2.17 2002/10/11 11:46:44 ume Exp $
68  */
69 
70 #include "opt_ipsec.h"
71 #include "opt_inet.h"
72 #include "opt_inet6.h"
73 #include "opt_tcpdebug.h"
74 
75 #include <sys/param.h>
76 #include <sys/systm.h>
77 #include <sys/kernel.h>
78 #include <sys/malloc.h>
79 #include <sys/sysctl.h>
80 #include <sys/globaldata.h>
81 #include <sys/thread.h>
82 
83 #include <sys/mbuf.h>
84 #ifdef INET6
85 #include <sys/domain.h>
86 #endif /* INET6 */
87 #include <sys/socket.h>
88 #include <sys/socketvar.h>
89 #include <sys/socketops.h>
90 #include <sys/protosw.h>
91 
92 #include <sys/thread2.h>
93 #include <sys/msgport2.h>
94 #include <sys/socketvar2.h>
95 
96 #include <net/if.h>
97 #include <net/netisr.h>
98 #include <net/route.h>
99 
100 #include <net/netmsg2.h>
101 
102 #include <netinet/in.h>
103 #include <netinet/in_systm.h>
104 #ifdef INET6
105 #include <netinet/ip6.h>
106 #endif
107 #include <netinet/in_pcb.h>
108 #ifdef INET6
109 #include <netinet6/in6_pcb.h>
110 #endif
111 #include <netinet/in_var.h>
112 #include <netinet/ip_var.h>
113 #ifdef INET6
114 #include <netinet6/ip6_var.h>
115 #include <netinet6/tcp6_var.h>
116 #endif
117 #include <netinet/tcp.h>
118 #include <netinet/tcp_fsm.h>
119 #include <netinet/tcp_seq.h>
120 #include <netinet/tcp_timer.h>
121 #include <netinet/tcp_timer2.h>
122 #include <netinet/tcp_var.h>
123 #include <netinet/tcpip.h>
124 #ifdef TCPDEBUG
125 #include <netinet/tcp_debug.h>
126 #endif
127 
128 #ifdef IPSEC
129 #include <netinet6/ipsec.h>
130 #endif /*IPSEC*/
131 
132 /*
133  * TCP protocol interface to socket abstraction.
134  */
135 extern	char *tcpstates[];	/* XXX ??? */
136 
137 static int	tcp_attach (struct socket *, struct pru_attach_info *);
138 static void	tcp_connect (netmsg_t msg);
139 #ifdef INET6
140 static void	tcp6_connect (netmsg_t msg);
141 static int	tcp6_connect_oncpu(struct tcpcb *tp, int flags,
142 				struct mbuf **mp,
143 				struct sockaddr_in6 *sin6,
144 				struct in6_addr *addr6);
145 #endif /* INET6 */
146 static struct tcpcb *
147 		tcp_disconnect (struct tcpcb *);
148 static struct tcpcb *
149 		tcp_usrclosed (struct tcpcb *);
150 
151 #ifdef TCPDEBUG
152 #define	TCPDEBUG0	int ostate = 0
153 #define	TCPDEBUG1()	ostate = tp ? tp->t_state : 0
154 #define	TCPDEBUG2(req)	if (tp && (so->so_options & SO_DEBUG)) \
155 				tcp_trace(TA_USER, ostate, tp, 0, 0, req)
156 #else
157 #define	TCPDEBUG0
158 #define	TCPDEBUG1()
159 #define	TCPDEBUG2(req)
160 #endif
161 
162 static int	tcp_lport_extension = 1;
163 SYSCTL_INT(_net_inet_tcp, OID_AUTO, lportext, CTLFLAG_RW,
164     &tcp_lport_extension, 0, "");
165 
166 /*
167  * For some ill optimized programs, which try to use TCP_NOPUSH
168  * to improve performance, will have small amount of data sits
169  * in the sending buffer.  These small amount of data will _not_
170  * be pushed into the network until more data are written into
171  * the socket or the socket write side is shutdown.
172  */
173 static int	tcp_disable_nopush = 1;
174 SYSCTL_INT(_net_inet_tcp, OID_AUTO, disable_nopush, CTLFLAG_RW,
175     &tcp_disable_nopush, 0, "TCP_NOPUSH socket option will have no effect");
176 
177 /*
178  * TCP attaches to socket via pru_attach(), reserving space,
179  * and an internet control block.  This is likely occuring on
180  * cpu0 and may have to move later when we bind/connect.
181  */
182 static void
183 tcp_usr_attach(netmsg_t msg)
184 {
185 	struct socket *so = msg->base.nm_so;
186 	struct pru_attach_info *ai = msg->attach.nm_ai;
187 	int error;
188 	struct inpcb *inp;
189 	struct tcpcb *tp = NULL;
190 	TCPDEBUG0;
191 
192 	soreference(so);
193 	inp = so->so_pcb;
194 	TCPDEBUG1();
195 	if (inp) {
196 		error = EISCONN;
197 		goto out;
198 	}
199 
200 	error = tcp_attach(so, ai);
201 	if (error)
202 		goto out;
203 
204 	if ((so->so_options & SO_LINGER) && so->so_linger == 0)
205 		so->so_linger = TCP_LINGERTIME;
206 	tp = sototcpcb(so);
207 out:
208 	sofree(so);		/* from ref above */
209 	TCPDEBUG2(PRU_ATTACH);
210 	lwkt_replymsg(&msg->lmsg, error);
211 }
212 
213 /*
214  * pru_detach() detaches the TCP protocol from the socket.
215  * If the protocol state is non-embryonic, then can't
216  * do this directly: have to initiate a pru_disconnect(),
217  * which may finish later; embryonic TCB's can just
218  * be discarded here.
219  */
220 static void
221 tcp_usr_detach(netmsg_t msg)
222 {
223 	struct socket *so = msg->base.nm_so;
224 	int error = 0;
225 	struct inpcb *inp;
226 	struct tcpcb *tp;
227 	TCPDEBUG0;
228 
229 	inp = so->so_pcb;
230 
231 	/*
232 	 * If the inp is already detached it may have been due to an async
233 	 * close.  Just return as if no error occured.
234 	 *
235 	 * It's possible for the tcpcb (tp) to disconnect from the inp due
236 	 * to tcp_drop()->tcp_close() being called.  This may occur *after*
237 	 * the detach message has been queued so we may find a NULL tp here.
238 	 */
239 	if (inp) {
240 		if ((tp = intotcpcb(inp)) != NULL) {
241 			TCPDEBUG1();
242 			tp = tcp_disconnect(tp);
243 			TCPDEBUG2(PRU_DETACH);
244 		}
245 	}
246 	lwkt_replymsg(&msg->lmsg, error);
247 }
248 
249 /*
250  * NOTE: ignore_error is non-zero for certain disconnection races
251  * which we want to silently allow, otherwise close() may return
252  * an unexpected error.
253  *
254  * NOTE: The variables (msg) and (tp) are assumed.
255  */
256 #define	COMMON_START(so, inp, ignore_error)			\
257 	TCPDEBUG0; 						\
258 								\
259 	inp = so->so_pcb; 					\
260 	do {							\
261 		 if (inp == NULL) {				\
262 			error = ignore_error ? 0 : EINVAL;	\
263 			tp = NULL;				\
264 			goto out;				\
265 		 }						\
266 		 tp = intotcpcb(inp);				\
267 		 TCPDEBUG1();					\
268 	} while(0)
269 
270 #define COMMON_END1(req, noreply)				\
271 	out: do {						\
272 		TCPDEBUG2(req);					\
273 		if (!(noreply))					\
274 			lwkt_replymsg(&msg->lmsg, error);	\
275 		return;						\
276 	} while(0)
277 
278 #define COMMON_END(req)		COMMON_END1((req), 0)
279 
280 /*
281  * Give the socket an address.
282  */
283 static void
284 tcp_usr_bind(netmsg_t msg)
285 {
286 	struct socket *so = msg->bind.base.nm_so;
287 	struct sockaddr *nam = msg->bind.nm_nam;
288 	struct thread *td = msg->bind.nm_td;
289 	int error = 0;
290 	struct inpcb *inp;
291 	struct tcpcb *tp;
292 	struct sockaddr_in *sinp;
293 
294 	COMMON_START(so, inp, 0);
295 
296 	/*
297 	 * Must check for multicast addresses and disallow binding
298 	 * to them.
299 	 */
300 	sinp = (struct sockaddr_in *)nam;
301 	if (sinp->sin_family == AF_INET &&
302 	    IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
303 		error = EAFNOSUPPORT;
304 		goto out;
305 	}
306 	error = in_pcbbind(inp, nam, td);
307 	if (error)
308 		goto out;
309 	COMMON_END(PRU_BIND);
310 
311 }
312 
313 #ifdef INET6
314 
315 static void
316 tcp6_usr_bind(netmsg_t msg)
317 {
318 	struct socket *so = msg->bind.base.nm_so;
319 	struct sockaddr *nam = msg->bind.nm_nam;
320 	struct thread *td = msg->bind.nm_td;
321 	int error = 0;
322 	struct inpcb *inp;
323 	struct tcpcb *tp;
324 	struct sockaddr_in6 *sin6p;
325 
326 	COMMON_START(so, inp, 0);
327 
328 	/*
329 	 * Must check for multicast addresses and disallow binding
330 	 * to them.
331 	 */
332 	sin6p = (struct sockaddr_in6 *)nam;
333 	if (sin6p->sin6_family == AF_INET6 &&
334 	    IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
335 		error = EAFNOSUPPORT;
336 		goto out;
337 	}
338 	inp->inp_vflag &= ~INP_IPV4;
339 	inp->inp_vflag |= INP_IPV6;
340 	if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
341 		if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr))
342 			inp->inp_vflag |= INP_IPV4;
343 		else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
344 			struct sockaddr_in sin;
345 
346 			in6_sin6_2_sin(&sin, sin6p);
347 			inp->inp_vflag |= INP_IPV4;
348 			inp->inp_vflag &= ~INP_IPV6;
349 			error = in_pcbbind(inp, (struct sockaddr *)&sin, td);
350 			goto out;
351 		}
352 	}
353 	error = in6_pcbbind(inp, nam, td);
354 	if (error)
355 		goto out;
356 	COMMON_END(PRU_BIND);
357 }
358 #endif /* INET6 */
359 
360 #ifdef SMP
361 
362 struct netmsg_inswildcard {
363 	struct netmsg_base	base;
364 	struct inpcb		*nm_inp;
365 };
366 
367 static void
368 in_pcbinswildcardhash_handler(netmsg_t msg)
369 {
370 	struct netmsg_inswildcard *nm = (struct netmsg_inswildcard *)msg;
371 	int cpu = mycpuid, nextcpu;
372 
373 	in_pcbinswildcardhash_oncpu(nm->nm_inp, &tcbinfo[cpu]);
374 
375 	nextcpu = cpu + 1;
376 	if (nextcpu < ncpus2)
377 		lwkt_forwardmsg(netisr_portfn(nextcpu), &nm->base.lmsg);
378 	else
379 		lwkt_replymsg(&nm->base.lmsg, 0);
380 }
381 
382 #endif
383 
384 /*
385  * Prepare to accept connections.
386  */
387 static void
388 tcp_usr_listen(netmsg_t msg)
389 {
390 	struct socket *so = msg->listen.base.nm_so;
391 	struct thread *td = msg->listen.nm_td;
392 	int error = 0;
393 	struct inpcb *inp;
394 	struct tcpcb *tp;
395 #ifdef SMP
396 	struct netmsg_inswildcard nm;
397 #endif
398 
399 	COMMON_START(so, inp, 0);
400 
401 	if (tp->t_flags & TF_LISTEN)
402 		goto out;
403 
404 	if (inp->inp_lport == 0) {
405 		error = in_pcbbind(inp, NULL, td);
406 		if (error)
407 			goto out;
408 	}
409 
410 	tp->t_state = TCPS_LISTEN;
411 	tp->t_flags |= TF_LISTEN;
412 	tp->tt_msg = NULL; /* Catch any invalid timer usage */
413 
414 #ifdef SMP
415 	if (ncpus > 1) {
416 		/*
417 		 * We have to set the flag because we can't have other cpus
418 		 * messing with our inp's flags.
419 		 */
420 		KASSERT(!(inp->inp_flags & INP_CONNECTED),
421 			("already on connhash"));
422 		KASSERT(!(inp->inp_flags & INP_WILDCARD),
423 			("already on wildcardhash"));
424 		KASSERT(!(inp->inp_flags & INP_WILDCARD_MP),
425 			("already on MP wildcardhash"));
426 		inp->inp_flags |= INP_WILDCARD_MP;
427 
428 		KKASSERT(so->so_port == netisr_portfn(0));
429 		KKASSERT(&curthread->td_msgport == netisr_portfn(0));
430 		KKASSERT(inp->inp_pcbinfo == &tcbinfo[0]);
431 
432 		netmsg_init(&nm.base, NULL, &curthread->td_msgport,
433 			    MSGF_PRIORITY, in_pcbinswildcardhash_handler);
434 		nm.nm_inp = inp;
435 		lwkt_domsg(netisr_portfn(1), &nm.base.lmsg, 0);
436 	}
437 #endif
438 	in_pcbinswildcardhash(inp);
439 	COMMON_END(PRU_LISTEN);
440 }
441 
442 #ifdef INET6
443 
444 static void
445 tcp6_usr_listen(netmsg_t msg)
446 {
447 	struct socket *so = msg->listen.base.nm_so;
448 	struct thread *td = msg->listen.nm_td;
449 	int error = 0;
450 	struct inpcb *inp;
451 	struct tcpcb *tp;
452 #ifdef SMP
453 	struct netmsg_inswildcard nm;
454 #endif
455 
456 	COMMON_START(so, inp, 0);
457 
458 	if (tp->t_flags & TF_LISTEN)
459 		goto out;
460 
461 	if (inp->inp_lport == 0) {
462 		if (!(inp->inp_flags & IN6P_IPV6_V6ONLY))
463 			inp->inp_vflag |= INP_IPV4;
464 		else
465 			inp->inp_vflag &= ~INP_IPV4;
466 		error = in6_pcbbind(inp, NULL, td);
467 		if (error)
468 			goto out;
469 	}
470 
471 	tp->t_state = TCPS_LISTEN;
472 	tp->t_flags |= TF_LISTEN;
473 	tp->tt_msg = NULL; /* Catch any invalid timer usage */
474 
475 #ifdef SMP
476 	if (ncpus > 1) {
477 		/*
478 		 * We have to set the flag because we can't have other cpus
479 		 * messing with our inp's flags.
480 		 */
481 		KASSERT(!(inp->inp_flags & INP_CONNECTED),
482 			("already on connhash"));
483 		KASSERT(!(inp->inp_flags & INP_WILDCARD),
484 			("already on wildcardhash"));
485 		KASSERT(!(inp->inp_flags & INP_WILDCARD_MP),
486 			("already on MP wildcardhash"));
487 		inp->inp_flags |= INP_WILDCARD_MP;
488 
489 		KKASSERT(so->so_port == netisr_portfn(0));
490 		KKASSERT(&curthread->td_msgport == netisr_portfn(0));
491 		KKASSERT(inp->inp_pcbinfo == &tcbinfo[0]);
492 
493 		netmsg_init(&nm.base, NULL, &curthread->td_msgport,
494 			    MSGF_PRIORITY, in_pcbinswildcardhash_handler);
495 		nm.nm_inp = inp;
496 		lwkt_domsg(netisr_portfn(1), &nm.base.lmsg, 0);
497 	}
498 #endif
499 	in_pcbinswildcardhash(inp);
500 	COMMON_END(PRU_LISTEN);
501 }
502 #endif /* INET6 */
503 
504 /*
505  * Initiate connection to peer.
506  * Create a template for use in transmissions on this connection.
507  * Enter SYN_SENT state, and mark socket as connecting.
508  * Start keep-alive timer, and seed output sequence space.
509  * Send initial segment on connection.
510  */
511 static void
512 tcp_usr_connect(netmsg_t msg)
513 {
514 	struct socket *so = msg->connect.base.nm_so;
515 	struct sockaddr *nam = msg->connect.nm_nam;
516 	struct thread *td = msg->connect.nm_td;
517 	int error = 0;
518 	struct inpcb *inp;
519 	struct tcpcb *tp;
520 	struct sockaddr_in *sinp;
521 
522 	COMMON_START(so, inp, 0);
523 
524 	/*
525 	 * Must disallow TCP ``connections'' to multicast addresses.
526 	 */
527 	sinp = (struct sockaddr_in *)nam;
528 	if (sinp->sin_family == AF_INET
529 	    && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
530 		error = EAFNOSUPPORT;
531 		goto out;
532 	}
533 
534 	if (!prison_remote_ip(td, (struct sockaddr*)sinp)) {
535 		error = EAFNOSUPPORT; /* IPv6 only jail */
536 		goto out;
537 	}
538 
539 	tcp_connect(msg);
540 	/* msg is invalid now */
541 	return;
542 out:
543 	if (msg->connect.nm_m) {
544 		m_freem(msg->connect.nm_m);
545 		msg->connect.nm_m = NULL;
546 	}
547 	lwkt_replymsg(&msg->lmsg, error);
548 }
549 
550 #ifdef INET6
551 
552 static void
553 tcp6_usr_connect(netmsg_t msg)
554 {
555 	struct socket *so = msg->connect.base.nm_so;
556 	struct sockaddr *nam = msg->connect.nm_nam;
557 	struct thread *td = msg->connect.nm_td;
558 	int error = 0;
559 	struct inpcb *inp;
560 	struct tcpcb *tp;
561 	struct sockaddr_in6 *sin6p;
562 
563 	COMMON_START(so, inp, 0);
564 
565 	/*
566 	 * Must disallow TCP ``connections'' to multicast addresses.
567 	 */
568 	sin6p = (struct sockaddr_in6 *)nam;
569 	if (sin6p->sin6_family == AF_INET6
570 	    && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
571 		error = EAFNOSUPPORT;
572 		goto out;
573 	}
574 
575 	if (!prison_remote_ip(td, nam)) {
576 		error = EAFNOSUPPORT; /* IPv4 only jail */
577 		goto out;
578 	}
579 
580 	if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
581 		struct sockaddr_in *sinp;
582 
583 		if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
584 			error = EINVAL;
585 			goto out;
586 		}
587 		sinp = kmalloc(sizeof(*sinp), M_LWKTMSG, M_INTWAIT);
588 		in6_sin6_2_sin(sinp, sin6p);
589 		inp->inp_vflag |= INP_IPV4;
590 		inp->inp_vflag &= ~INP_IPV6;
591 		msg->connect.nm_nam = (struct sockaddr *)sinp;
592 		msg->connect.nm_reconnect |= NMSG_RECONNECT_NAMALLOC;
593 		tcp_connect(msg);
594 		/* msg is invalid now */
595 		return;
596 	}
597 	inp->inp_vflag &= ~INP_IPV4;
598 	inp->inp_vflag |= INP_IPV6;
599 	inp->inp_inc.inc_isipv6 = 1;
600 
601 	msg->connect.nm_reconnect |= NMSG_RECONNECT_FALLBACK;
602 	tcp6_connect(msg);
603 	/* msg is invalid now */
604 	return;
605 out:
606 	if (msg->connect.nm_m) {
607 		m_freem(msg->connect.nm_m);
608 		msg->connect.nm_m = NULL;
609 	}
610 	lwkt_replymsg(&msg->lmsg, error);
611 }
612 
613 #endif /* INET6 */
614 
615 /*
616  * Initiate disconnect from peer.
617  * If connection never passed embryonic stage, just drop;
618  * else if don't need to let data drain, then can just drop anyways,
619  * else have to begin TCP shutdown process: mark socket disconnecting,
620  * drain unread data, state switch to reflect user close, and
621  * send segment (e.g. FIN) to peer.  Socket will be really disconnected
622  * when peer sends FIN and acks ours.
623  *
624  * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
625  */
626 static void
627 tcp_usr_disconnect(netmsg_t msg)
628 {
629 	struct socket *so = msg->disconnect.base.nm_so;
630 	int error = 0;
631 	struct inpcb *inp;
632 	struct tcpcb *tp;
633 
634 	COMMON_START(so, inp, 1);
635 	tp = tcp_disconnect(tp);
636 	COMMON_END(PRU_DISCONNECT);
637 }
638 
639 /*
640  * Accept a connection.  Essentially all the work is
641  * done at higher levels; just return the address
642  * of the peer, storing through addr.
643  */
644 static void
645 tcp_usr_accept(netmsg_t msg)
646 {
647 	struct socket *so = msg->accept.base.nm_so;
648 	struct sockaddr **nam = msg->accept.nm_nam;
649 	int error = 0;
650 	struct inpcb *inp;
651 	struct tcpcb *tp = NULL;
652 	TCPDEBUG0;
653 
654 	inp = so->so_pcb;
655 	if (so->so_state & SS_ISDISCONNECTED) {
656 		error = ECONNABORTED;
657 		goto out;
658 	}
659 	if (inp == 0) {
660 		error = EINVAL;
661 		goto out;
662 	}
663 
664 	tp = intotcpcb(inp);
665 	TCPDEBUG1();
666 	in_setpeeraddr(so, nam);
667 	COMMON_END(PRU_ACCEPT);
668 }
669 
670 #ifdef INET6
671 static void
672 tcp6_usr_accept(netmsg_t msg)
673 {
674 	struct socket *so = msg->accept.base.nm_so;
675 	struct sockaddr **nam = msg->accept.nm_nam;
676 	int error = 0;
677 	struct inpcb *inp;
678 	struct tcpcb *tp = NULL;
679 	TCPDEBUG0;
680 
681 	inp = so->so_pcb;
682 
683 	if (so->so_state & SS_ISDISCONNECTED) {
684 		error = ECONNABORTED;
685 		goto out;
686 	}
687 	if (inp == 0) {
688 		error = EINVAL;
689 		goto out;
690 	}
691 	tp = intotcpcb(inp);
692 	TCPDEBUG1();
693 	in6_mapped_peeraddr(so, nam);
694 	COMMON_END(PRU_ACCEPT);
695 }
696 #endif /* INET6 */
697 /*
698  * Mark the connection as being incapable of further output.
699  */
700 static void
701 tcp_usr_shutdown(netmsg_t msg)
702 {
703 	struct socket *so = msg->shutdown.base.nm_so;
704 	int error = 0;
705 	struct inpcb *inp;
706 	struct tcpcb *tp;
707 
708 	COMMON_START(so, inp, 0);
709 	socantsendmore(so);
710 	tp = tcp_usrclosed(tp);
711 	if (tp)
712 		error = tcp_output(tp);
713 	COMMON_END(PRU_SHUTDOWN);
714 }
715 
716 /*
717  * After a receive, possibly send window update to peer.
718  */
719 static void
720 tcp_usr_rcvd(netmsg_t msg)
721 {
722 	struct socket *so = msg->rcvd.base.nm_so;
723 	int error = 0, noreply = 0;
724 	struct inpcb *inp;
725 	struct tcpcb *tp;
726 
727 	COMMON_START(so, inp, 0);
728 
729 	if (msg->rcvd.nm_pru_flags & PRUR_ASYNC) {
730 		noreply = 1;
731 		so_async_rcvd_reply(so);
732 	}
733 	tcp_output(tp);
734 
735 	COMMON_END1(PRU_RCVD, noreply);
736 }
737 
738 /*
739  * Do a send by putting data in output queue and updating urgent
740  * marker if URG set.  Possibly send more data.  Unlike the other
741  * pru_*() routines, the mbuf chains are our responsibility.  We
742  * must either enqueue them or free them.  The other pru_* routines
743  * generally are caller-frees.
744  */
745 static void
746 tcp_usr_send(netmsg_t msg)
747 {
748 	struct socket *so = msg->send.base.nm_so;
749 	int flags = msg->send.nm_flags;
750 	struct mbuf *m = msg->send.nm_m;
751 	int error = 0;
752 	struct inpcb *inp;
753 	struct tcpcb *tp;
754 	TCPDEBUG0;
755 
756 	KKASSERT(msg->send.nm_control == NULL);
757 	KKASSERT(msg->send.nm_addr == NULL);
758 	KKASSERT((flags & PRUS_FREEADDR) == 0);
759 
760 	inp = so->so_pcb;
761 
762 	if (inp == NULL) {
763 		/*
764 		 * OOPS! we lost a race, the TCP session got reset after
765 		 * we checked SS_CANTSENDMORE, eg: while doing uiomove or a
766 		 * network interrupt in the non-critical section of sosend().
767 		 */
768 		m_freem(m);
769 		error = ECONNRESET;	/* XXX EPIPE? */
770 		tp = NULL;
771 		TCPDEBUG1();
772 		goto out;
773 	}
774 	tp = intotcpcb(inp);
775 	TCPDEBUG1();
776 
777 #ifdef foo
778 	/*
779 	 * This is no longer necessary, since:
780 	 * - sosendtcp() has already checked it for us
781 	 * - It does not work with asynchronized send
782 	 */
783 
784 	/*
785 	 * Don't let too much OOB data build up
786 	 */
787 	if (flags & PRUS_OOB) {
788 		if (ssb_space(&so->so_snd) < -512) {
789 			m_freem(m);
790 			error = ENOBUFS;
791 			goto out;
792 		}
793 	}
794 #endif
795 
796 	/*
797 	 * Pump the data into the socket.
798 	 */
799 	if (m)
800 		ssb_appendstream(&so->so_snd, m);
801 	if (flags & PRUS_OOB) {
802 		/*
803 		 * According to RFC961 (Assigned Protocols),
804 		 * the urgent pointer points to the last octet
805 		 * of urgent data.  We continue, however,
806 		 * to consider it to indicate the first octet
807 		 * of data past the urgent section.
808 		 * Otherwise, snd_up should be one lower.
809 		 */
810 		tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
811 		tp->t_flags |= TF_FORCE;
812 		error = tcp_output(tp);
813 		tp->t_flags &= ~TF_FORCE;
814 	} else {
815 		if (flags & PRUS_EOF) {
816 			/*
817 			 * Close the send side of the connection after
818 			 * the data is sent.
819 			 */
820 			socantsendmore(so);
821 			tp = tcp_usrclosed(tp);
822 		}
823 		if (tp != NULL) {
824 			if (flags & PRUS_MORETOCOME)
825 				tp->t_flags |= TF_MORETOCOME;
826 			error = tcp_output(tp);
827 			if (flags & PRUS_MORETOCOME)
828 				tp->t_flags &= ~TF_MORETOCOME;
829 		}
830 	}
831 	COMMON_END1((flags & PRUS_OOB) ? PRU_SENDOOB :
832 		   ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND),
833 		   (flags & PRUS_NOREPLY));
834 }
835 
836 /*
837  * NOTE: (so) is referenced from soabort*() and netmsg_pru_abort()
838  *	 will sofree() it when we return.
839  */
840 static void
841 tcp_usr_abort(netmsg_t msg)
842 {
843 	struct socket *so = msg->abort.base.nm_so;
844 	int error = 0;
845 	struct inpcb *inp;
846 	struct tcpcb *tp;
847 
848 	COMMON_START(so, inp, 1);
849 	tp = tcp_drop(tp, ECONNABORTED);
850 	COMMON_END(PRU_ABORT);
851 }
852 
853 /*
854  * Receive out-of-band data.
855  */
856 static void
857 tcp_usr_rcvoob(netmsg_t msg)
858 {
859 	struct socket *so = msg->rcvoob.base.nm_so;
860 	struct mbuf *m = msg->rcvoob.nm_m;
861 	int flags = msg->rcvoob.nm_flags;
862 	int error = 0;
863 	struct inpcb *inp;
864 	struct tcpcb *tp;
865 
866 	COMMON_START(so, inp, 0);
867 	if ((so->so_oobmark == 0 &&
868 	     (so->so_state & SS_RCVATMARK) == 0) ||
869 	    so->so_options & SO_OOBINLINE ||
870 	    tp->t_oobflags & TCPOOB_HADDATA) {
871 		error = EINVAL;
872 		goto out;
873 	}
874 	if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
875 		error = EWOULDBLOCK;
876 		goto out;
877 	}
878 	m->m_len = 1;
879 	*mtod(m, caddr_t) = tp->t_iobc;
880 	if ((flags & MSG_PEEK) == 0)
881 		tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
882 	COMMON_END(PRU_RCVOOB);
883 }
884 
885 static void
886 tcp_usr_savefaddr(struct socket *so, const struct sockaddr *faddr)
887 {
888 	in_savefaddr(so, faddr);
889 }
890 
891 #ifdef INET6
892 static void
893 tcp6_usr_savefaddr(struct socket *so, const struct sockaddr *faddr)
894 {
895 	in6_mapped_savefaddr(so, faddr);
896 }
897 #endif
898 
899 /* xxx - should be const */
900 struct pr_usrreqs tcp_usrreqs = {
901 	.pru_abort = tcp_usr_abort,
902 	.pru_accept = tcp_usr_accept,
903 	.pru_attach = tcp_usr_attach,
904 	.pru_bind = tcp_usr_bind,
905 	.pru_connect = tcp_usr_connect,
906 	.pru_connect2 = pr_generic_notsupp,
907 	.pru_control = in_control_dispatch,
908 	.pru_detach = tcp_usr_detach,
909 	.pru_disconnect = tcp_usr_disconnect,
910 	.pru_listen = tcp_usr_listen,
911 	.pru_peeraddr = in_setpeeraddr_dispatch,
912 	.pru_rcvd = tcp_usr_rcvd,
913 	.pru_rcvoob = tcp_usr_rcvoob,
914 	.pru_send = tcp_usr_send,
915 	.pru_sense = pru_sense_null,
916 	.pru_shutdown = tcp_usr_shutdown,
917 	.pru_sockaddr = in_setsockaddr_dispatch,
918 	.pru_sosend = sosendtcp,
919 	.pru_soreceive = sorecvtcp,
920 	.pru_savefaddr = tcp_usr_savefaddr
921 };
922 
923 #ifdef INET6
924 struct pr_usrreqs tcp6_usrreqs = {
925 	.pru_abort = tcp_usr_abort,
926 	.pru_accept = tcp6_usr_accept,
927 	.pru_attach = tcp_usr_attach,
928 	.pru_bind = tcp6_usr_bind,
929 	.pru_connect = tcp6_usr_connect,
930 	.pru_connect2 = pr_generic_notsupp,
931 	.pru_control = in6_control_dispatch,
932 	.pru_detach = tcp_usr_detach,
933 	.pru_disconnect = tcp_usr_disconnect,
934 	.pru_listen = tcp6_usr_listen,
935 	.pru_peeraddr = in6_mapped_peeraddr_dispatch,
936 	.pru_rcvd = tcp_usr_rcvd,
937 	.pru_rcvoob = tcp_usr_rcvoob,
938 	.pru_send = tcp_usr_send,
939 	.pru_sense = pru_sense_null,
940 	.pru_shutdown = tcp_usr_shutdown,
941 	.pru_sockaddr = in6_mapped_sockaddr_dispatch,
942 	.pru_sosend = sosendtcp,
943 	.pru_soreceive = sorecvtcp,
944 	.pru_savefaddr = tcp6_usr_savefaddr
945 };
946 #endif /* INET6 */
947 
948 static int
949 tcp_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf *m,
950 		  struct sockaddr_in *sin, struct sockaddr_in *if_sin)
951 {
952 	struct inpcb *inp = tp->t_inpcb, *oinp;
953 	struct socket *so = inp->inp_socket;
954 	struct route *ro = &inp->inp_route;
955 
956 	oinp = in_pcblookup_hash(&tcbinfo[mycpu->gd_cpuid],
957 				 sin->sin_addr, sin->sin_port,
958 				 (inp->inp_laddr.s_addr != INADDR_ANY ?
959 				  inp->inp_laddr : if_sin->sin_addr),
960 				inp->inp_lport, 0, NULL);
961 	if (oinp != NULL) {
962 		m_freem(m);
963 		return (EADDRINUSE);
964 	}
965 	if (inp->inp_laddr.s_addr == INADDR_ANY)
966 		inp->inp_laddr = if_sin->sin_addr;
967 	inp->inp_faddr = sin->sin_addr;
968 	inp->inp_fport = sin->sin_port;
969 	inp->inp_cpcbinfo = &tcbinfo[mycpu->gd_cpuid];
970 	in_pcbinsconnhash(inp);
971 
972 	/*
973 	 * We are now on the inpcb's owner CPU, if the cached route was
974 	 * freed because the rtentry's owner CPU is not the current CPU
975 	 * (e.g. in tcp_connect()), then we try to reallocate it here with
976 	 * the hope that a rtentry may be cloned from a RTF_PRCLONING
977 	 * rtentry.
978 	 */
979 	if (!(inp->inp_socket->so_options & SO_DONTROUTE) && /*XXX*/
980 	    ro->ro_rt == NULL) {
981 		bzero(&ro->ro_dst, sizeof(struct sockaddr_in));
982 		ro->ro_dst.sa_family = AF_INET;
983 		ro->ro_dst.sa_len = sizeof(struct sockaddr_in);
984 		((struct sockaddr_in *)&ro->ro_dst)->sin_addr =
985 			sin->sin_addr;
986 		rtalloc(ro);
987 	}
988 
989 	/*
990 	 * Now that no more errors can occur, change the protocol processing
991 	 * port to the current thread (which is the correct thread).
992 	 *
993 	 * Create TCP timer message now; we are on the tcpcb's owner
994 	 * CPU/thread.
995 	 */
996 	tcp_create_timermsg(tp, &curthread->td_msgport);
997 
998 	/*
999 	 * Compute window scaling to request.  Use a larger scaling then
1000 	 * needed for the initial receive buffer in case the receive buffer
1001 	 * gets expanded.
1002 	 */
1003 	if (tp->request_r_scale < TCP_MIN_WINSHIFT)
1004 		tp->request_r_scale = TCP_MIN_WINSHIFT;
1005 	while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1006 	       (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat
1007 	) {
1008 		tp->request_r_scale++;
1009 	}
1010 
1011 	soisconnecting(so);
1012 	tcpstat.tcps_connattempt++;
1013 	tp->t_state = TCPS_SYN_SENT;
1014 	tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
1015 	tp->iss = tcp_new_isn(tp);
1016 	tcp_sendseqinit(tp);
1017 	if (m) {
1018 		ssb_appendstream(&so->so_snd, m);
1019 		m = NULL;
1020 		if (flags & PRUS_OOB)
1021 			tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
1022 	}
1023 
1024 	/*
1025 	 * Close the send side of the connection after
1026 	 * the data is sent if flagged.
1027 	 */
1028 	if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
1029 		socantsendmore(so);
1030 		tp = tcp_usrclosed(tp);
1031 	}
1032 	return (tcp_output(tp));
1033 }
1034 
1035 /*
1036  * Common subroutine to open a TCP connection to remote host specified
1037  * by struct sockaddr_in in mbuf *nam.  Call in_pcbbind to assign a local
1038  * port number if needed.  Call in_pcbladdr to do the routing and to choose
1039  * a local host address (interface).
1040  * Initialize connection parameters and enter SYN-SENT state.
1041  */
1042 static void
1043 tcp_connect(netmsg_t msg)
1044 {
1045 	struct socket *so = msg->connect.base.nm_so;
1046 	struct sockaddr *nam = msg->connect.nm_nam;
1047 	struct thread *td = msg->connect.nm_td;
1048 	struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1049 	struct sockaddr_in *if_sin;
1050 	struct inpcb *inp;
1051 	struct tcpcb *tp;
1052 	int error, calc_laddr = 1;
1053 #ifdef SMP
1054 	lwkt_port_t port;
1055 #endif
1056 
1057 	COMMON_START(so, inp, 0);
1058 
1059 	/*
1060 	 * Reconnect our pcb if we have to
1061 	 */
1062 	if (msg->connect.nm_reconnect & NMSG_RECONNECT_RECONNECT) {
1063 		msg->connect.nm_reconnect &= ~NMSG_RECONNECT_RECONNECT;
1064 		in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1065 	}
1066 
1067 	/*
1068 	 * Bind if we have to
1069 	 */
1070 	if (inp->inp_lport == 0) {
1071 		if (tcp_lport_extension) {
1072 			KKASSERT(inp->inp_laddr.s_addr == INADDR_ANY);
1073 
1074 			error = in_pcbladdr(inp, nam, &if_sin, td);
1075 			if (error)
1076 				goto out;
1077 			inp->inp_laddr.s_addr = if_sin->sin_addr.s_addr;
1078 
1079 			error = in_pcbconn_bind(inp, nam, td);
1080 			if (error)
1081 				goto out;
1082 
1083 			calc_laddr = 0;
1084 		} else {
1085 			error = in_pcbbind(inp, NULL, td);
1086 			if (error)
1087 				goto out;
1088 		}
1089 	}
1090 
1091 	if (calc_laddr) {
1092 		/*
1093 		 * Calculate the correct protocol processing thread.  The
1094 		 * connect operation must run there.  Set the forwarding
1095 		 * port before we forward the message or it will get bounced
1096 		 * right back to us.
1097 		 */
1098 		error = in_pcbladdr(inp, nam, &if_sin, td);
1099 		if (error)
1100 			goto out;
1101 	}
1102 	KKASSERT(inp->inp_socket == so);
1103 
1104 #ifdef SMP
1105 	port = tcp_addrport(sin->sin_addr.s_addr, sin->sin_port,
1106 			    (inp->inp_laddr.s_addr ?
1107 			     inp->inp_laddr.s_addr : if_sin->sin_addr.s_addr),
1108 			    inp->inp_lport);
1109 
1110 	if (port != &curthread->td_msgport) {
1111 		struct route *ro = &inp->inp_route;
1112 
1113 		/*
1114 		 * in_pcbladdr() may have allocated a route entry for us
1115 		 * on the current CPU, but we need a route entry on the
1116 		 * inpcb's owner CPU, so free it here.
1117 		 */
1118 		if (ro->ro_rt != NULL)
1119 			RTFREE(ro->ro_rt);
1120 		bzero(ro, sizeof(*ro));
1121 
1122 		/*
1123 		 * We are moving the protocol processing port the socket
1124 		 * is on, we have to unlink here and re-link on the
1125 		 * target cpu.
1126 		 */
1127 		in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1128 		sosetport(so, port);
1129 		msg->connect.nm_reconnect |= NMSG_RECONNECT_RECONNECT;
1130 		msg->connect.base.nm_dispatch = tcp_connect;
1131 
1132 		lwkt_forwardmsg(port, &msg->connect.base.lmsg);
1133 		/* msg invalid now */
1134 		return;
1135 	}
1136 #else
1137 	KKASSERT(so->so_port == &curthread->td_msgport);
1138 #endif
1139 	error = tcp_connect_oncpu(tp, msg->connect.nm_flags,
1140 				  msg->connect.nm_m, sin, if_sin);
1141 	msg->connect.nm_m = NULL;
1142 out:
1143 	if (msg->connect.nm_m) {
1144 		m_freem(msg->connect.nm_m);
1145 		msg->connect.nm_m = NULL;
1146 	}
1147 	if (msg->connect.nm_reconnect & NMSG_RECONNECT_NAMALLOC) {
1148 		kfree(msg->connect.nm_nam, M_LWKTMSG);
1149 		msg->connect.nm_nam = NULL;
1150 	}
1151 	lwkt_replymsg(&msg->connect.base.lmsg, error);
1152 	/* msg invalid now */
1153 }
1154 
1155 #ifdef INET6
1156 
1157 static void
1158 tcp6_connect(netmsg_t msg)
1159 {
1160 	struct tcpcb *tp;
1161 	struct socket *so = msg->connect.base.nm_so;
1162 	struct sockaddr *nam = msg->connect.nm_nam;
1163 	struct thread *td = msg->connect.nm_td;
1164 	struct inpcb *inp;
1165 	struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam;
1166 	struct in6_addr *addr6;
1167 #ifdef SMP
1168 	lwkt_port_t port;
1169 #endif
1170 	int error;
1171 
1172 	COMMON_START(so, inp, 0);
1173 
1174 	/*
1175 	 * Reconnect our pcb if we have to
1176 	 */
1177 	if (msg->connect.nm_reconnect & NMSG_RECONNECT_RECONNECT) {
1178 		msg->connect.nm_reconnect &= ~NMSG_RECONNECT_RECONNECT;
1179 		in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1180 	}
1181 
1182 	/*
1183 	 * Bind if we have to
1184 	 */
1185 	if (inp->inp_lport == 0) {
1186 		error = in6_pcbbind(inp, NULL, td);
1187 		if (error)
1188 			goto out;
1189 	}
1190 
1191 	/*
1192 	 * Cannot simply call in_pcbconnect, because there might be an
1193 	 * earlier incarnation of this same connection still in
1194 	 * TIME_WAIT state, creating an ADDRINUSE error.
1195 	 */
1196 	error = in6_pcbladdr(inp, nam, &addr6, td);
1197 	if (error)
1198 		goto out;
1199 
1200 #ifdef SMP
1201 	port = tcp6_addrport();	/* XXX hack for now, always cpu0 */
1202 
1203 	if (port != &curthread->td_msgport) {
1204 		struct route *ro = &inp->inp_route;
1205 
1206 		/*
1207 		 * in_pcbladdr() may have allocated a route entry for us
1208 		 * on the current CPU, but we need a route entry on the
1209 		 * inpcb's owner CPU, so free it here.
1210 		 */
1211 		if (ro->ro_rt != NULL)
1212 			RTFREE(ro->ro_rt);
1213 		bzero(ro, sizeof(*ro));
1214 
1215 		in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1216 		sosetport(so, port);
1217 		msg->connect.nm_reconnect |= NMSG_RECONNECT_RECONNECT;
1218 		msg->connect.base.nm_dispatch = tcp6_connect;
1219 
1220 		lwkt_forwardmsg(port, &msg->connect.base.lmsg);
1221 		/* msg invalid now */
1222 		return;
1223 	}
1224 #endif
1225 	error = tcp6_connect_oncpu(tp, msg->connect.nm_flags,
1226 				   &msg->connect.nm_m, sin6, addr6);
1227 	/* nm_m may still be intact */
1228 out:
1229 	if (error && (msg->connect.nm_reconnect & NMSG_RECONNECT_FALLBACK)) {
1230 		tcp_connect(msg);
1231 		/* msg invalid now */
1232 	} else {
1233 		if (msg->connect.nm_m) {
1234 			m_freem(msg->connect.nm_m);
1235 			msg->connect.nm_m = NULL;
1236 		}
1237 		if (msg->connect.nm_reconnect & NMSG_RECONNECT_NAMALLOC) {
1238 			kfree(msg->connect.nm_nam, M_LWKTMSG);
1239 			msg->connect.nm_nam = NULL;
1240 		}
1241 		lwkt_replymsg(&msg->connect.base.lmsg, error);
1242 		/* msg invalid now */
1243 	}
1244 }
1245 
1246 static int
1247 tcp6_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf **mp,
1248 		   struct sockaddr_in6 *sin6, struct in6_addr *addr6)
1249 {
1250 	struct mbuf *m = *mp;
1251 	struct inpcb *inp = tp->t_inpcb;
1252 	struct socket *so = inp->inp_socket;
1253 	struct inpcb *oinp;
1254 
1255 	/*
1256 	 * Cannot simply call in_pcbconnect, because there might be an
1257 	 * earlier incarnation of this same connection still in
1258 	 * TIME_WAIT state, creating an ADDRINUSE error.
1259 	 */
1260 	oinp = in6_pcblookup_hash(inp->inp_cpcbinfo,
1261 				  &sin6->sin6_addr, sin6->sin6_port,
1262 				  (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) ?
1263 				      addr6 : &inp->in6p_laddr),
1264 				  inp->inp_lport,  0, NULL);
1265 	if (oinp)
1266 		return (EADDRINUSE);
1267 
1268 	if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
1269 		inp->in6p_laddr = *addr6;
1270 	inp->in6p_faddr = sin6->sin6_addr;
1271 	inp->inp_fport = sin6->sin6_port;
1272 	if ((sin6->sin6_flowinfo & IPV6_FLOWINFO_MASK) != 0)
1273 		inp->in6p_flowinfo = sin6->sin6_flowinfo;
1274 	in_pcbinsconnhash(inp);
1275 
1276 	/*
1277 	 * Now that no more errors can occur, change the protocol processing
1278 	 * port to the current thread (which is the correct thread).
1279 	 *
1280 	 * Create TCP timer message now; we are on the tcpcb's owner
1281 	 * CPU/thread.
1282 	 */
1283 	tcp_create_timermsg(tp, &curthread->td_msgport);
1284 
1285 	/* Compute window scaling to request.  */
1286 	if (tp->request_r_scale < TCP_MIN_WINSHIFT)
1287 		tp->request_r_scale = TCP_MIN_WINSHIFT;
1288 	while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1289 	    (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat) {
1290 		tp->request_r_scale++;
1291 	}
1292 
1293 	soisconnecting(so);
1294 	tcpstat.tcps_connattempt++;
1295 	tp->t_state = TCPS_SYN_SENT;
1296 	tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
1297 	tp->iss = tcp_new_isn(tp);
1298 	tcp_sendseqinit(tp);
1299 	if (m) {
1300 		ssb_appendstream(&so->so_snd, m);
1301 		*mp = NULL;
1302 		if (flags & PRUS_OOB)
1303 			tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
1304 	}
1305 
1306 	/*
1307 	 * Close the send side of the connection after
1308 	 * the data is sent if flagged.
1309 	 */
1310 	if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
1311 		socantsendmore(so);
1312 		tp = tcp_usrclosed(tp);
1313 	}
1314 	return (tcp_output(tp));
1315 }
1316 
1317 #endif /* INET6 */
1318 
1319 /*
1320  * The new sockopt interface makes it possible for us to block in the
1321  * copyin/out step (if we take a page fault).  Taking a page fault while
1322  * in a critical section is probably a Bad Thing.  (Since sockets and pcbs
1323  * both now use TSM, there probably isn't any need for this function to
1324  * run in a critical section any more.  This needs more examination.)
1325  */
1326 void
1327 tcp_ctloutput(netmsg_t msg)
1328 {
1329 	struct socket *so = msg->base.nm_so;
1330 	struct sockopt *sopt = msg->ctloutput.nm_sopt;
1331 	int	error, opt, optval, opthz;
1332 	struct	inpcb *inp;
1333 	struct	tcpcb *tp;
1334 
1335 	error = 0;
1336 	inp = so->so_pcb;
1337 	if (inp == NULL) {
1338 		error = ECONNRESET;
1339 		goto done;
1340 	}
1341 
1342 	if (sopt->sopt_level != IPPROTO_TCP) {
1343 #ifdef INET6
1344 		if (INP_CHECK_SOCKAF(so, AF_INET6))
1345 			ip6_ctloutput_dispatch(msg);
1346 		else
1347 #endif /* INET6 */
1348 		ip_ctloutput(msg);
1349 		/* msg invalid now */
1350 		return;
1351 	}
1352 	tp = intotcpcb(inp);
1353 
1354 	switch (sopt->sopt_dir) {
1355 	case SOPT_SET:
1356 		error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1357 				      sizeof optval);
1358 		if (error)
1359 			break;
1360 		switch (sopt->sopt_name) {
1361 		case TCP_FASTKEEP:
1362 			if (optval > 0)
1363 				tp->t_keepidle = tp->t_keepintvl;
1364 			else
1365 				tp->t_keepidle = tcp_keepidle;
1366 			tcp_timer_keep_activity(tp, 0);
1367 			break;
1368 #ifdef TCP_SIGNATURE
1369 		case TCP_SIGNATURE_ENABLE:
1370 			if (tp->t_state == TCPS_CLOSED) {
1371 				/*
1372 				 * This is the only safe state that this
1373 				 * option could be changed.  Some segments
1374 				 * could already have been sent in other
1375 				 * states.
1376 				 */
1377 				if (optval > 0)
1378 					tp->t_flags |= TF_SIGNATURE;
1379 				else
1380 					tp->t_flags &= ~TF_SIGNATURE;
1381 			} else {
1382 				error = EOPNOTSUPP;
1383 			}
1384 			break;
1385 #endif /* TCP_SIGNATURE */
1386 		case TCP_NODELAY:
1387 		case TCP_NOOPT:
1388 			switch (sopt->sopt_name) {
1389 			case TCP_NODELAY:
1390 				opt = TF_NODELAY;
1391 				break;
1392 			case TCP_NOOPT:
1393 				opt = TF_NOOPT;
1394 				break;
1395 			default:
1396 				opt = 0; /* dead code to fool gcc */
1397 				break;
1398 			}
1399 
1400 			if (optval)
1401 				tp->t_flags |= opt;
1402 			else
1403 				tp->t_flags &= ~opt;
1404 			break;
1405 
1406 		case TCP_NOPUSH:
1407 			if (tcp_disable_nopush)
1408 				break;
1409 			if (optval)
1410 				tp->t_flags |= TF_NOPUSH;
1411 			else {
1412 				tp->t_flags &= ~TF_NOPUSH;
1413 				error = tcp_output(tp);
1414 			}
1415 			break;
1416 
1417 		case TCP_MAXSEG:
1418 			/*
1419 			 * Must be between 0 and maxseg.  If the requested
1420 			 * maxseg is too small to satisfy the desired minmss,
1421 			 * pump it up (silently so sysctl modifications of
1422 			 * minmss do not create unexpected program failures).
1423 			 * Handle degenerate cases.
1424 			 */
1425 			if (optval > 0 && optval <= tp->t_maxseg) {
1426 				if (optval + 40 < tcp_minmss) {
1427 					optval = tcp_minmss - 40;
1428 					if (optval < 0)
1429 						optval = 1;
1430 				}
1431 				tp->t_maxseg = optval;
1432 			} else {
1433 				error = EINVAL;
1434 			}
1435 			break;
1436 
1437 		case TCP_KEEPINIT:
1438 			opthz = ((int64_t)optval * hz) / 1000;
1439 			if (opthz >= 1)
1440 				tp->t_keepinit = opthz;
1441 			else
1442 				error = EINVAL;
1443 			break;
1444 
1445 		case TCP_KEEPIDLE:
1446 			opthz = ((int64_t)optval * hz) / 1000;
1447 			if (opthz >= 1) {
1448 				tp->t_keepidle = opthz;
1449 				tcp_timer_keep_activity(tp, 0);
1450 			} else {
1451 				error = EINVAL;
1452 			}
1453 			break;
1454 
1455 		case TCP_KEEPINTVL:
1456 			opthz = ((int64_t)optval * hz) / 1000;
1457 			if (opthz >= 1) {
1458 				tp->t_keepintvl = opthz;
1459 				tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt;
1460 			} else {
1461 				error = EINVAL;
1462 			}
1463 			break;
1464 
1465 		case TCP_KEEPCNT:
1466 			if (optval > 0) {
1467 				tp->t_keepcnt = optval;
1468 				tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt;
1469 			} else {
1470 				error = EINVAL;
1471 			}
1472 			break;
1473 
1474 		default:
1475 			error = ENOPROTOOPT;
1476 			break;
1477 		}
1478 		break;
1479 
1480 	case SOPT_GET:
1481 		switch (sopt->sopt_name) {
1482 #ifdef TCP_SIGNATURE
1483 		case TCP_SIGNATURE_ENABLE:
1484 			optval = (tp->t_flags & TF_SIGNATURE) ? 1 : 0;
1485 			break;
1486 #endif /* TCP_SIGNATURE */
1487 		case TCP_NODELAY:
1488 			optval = tp->t_flags & TF_NODELAY;
1489 			break;
1490 		case TCP_MAXSEG:
1491 			optval = tp->t_maxseg;
1492 			break;
1493 		case TCP_NOOPT:
1494 			optval = tp->t_flags & TF_NOOPT;
1495 			break;
1496 		case TCP_NOPUSH:
1497 			optval = tp->t_flags & TF_NOPUSH;
1498 			break;
1499 		case TCP_KEEPINIT:
1500 			optval = ((int64_t)tp->t_keepinit * 1000) / hz;
1501 			break;
1502 		case TCP_KEEPIDLE:
1503 			optval = ((int64_t)tp->t_keepidle * 1000) / hz;
1504 			break;
1505 		case TCP_KEEPINTVL:
1506 			optval = ((int64_t)tp->t_keepintvl * 1000) / hz;
1507 			break;
1508 		case TCP_KEEPCNT:
1509 			optval = tp->t_keepcnt;
1510 			break;
1511 		default:
1512 			error = ENOPROTOOPT;
1513 			break;
1514 		}
1515 		if (error == 0)
1516 			soopt_from_kbuf(sopt, &optval, sizeof optval);
1517 		break;
1518 	}
1519 done:
1520 	lwkt_replymsg(&msg->lmsg, error);
1521 }
1522 
1523 /*
1524  * tcp_sendspace and tcp_recvspace are the default send and receive window
1525  * sizes, respectively.  These are obsolescent (this information should
1526  * be set by the route).
1527  *
1528  * Use a default that does not require tcp window scaling to be turned
1529  * on.  Individual programs or the administrator can increase the default.
1530  */
1531 u_long	tcp_sendspace = 57344;	/* largest multiple of PAGE_SIZE < 64k */
1532 SYSCTL_INT(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW,
1533     &tcp_sendspace , 0, "Maximum outgoing TCP datagram size");
1534 u_long	tcp_recvspace = 57344;	/* largest multiple of PAGE_SIZE < 64k */
1535 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
1536     &tcp_recvspace , 0, "Maximum incoming TCP datagram size");
1537 
1538 /*
1539  * Attach TCP protocol to socket, allocating internet protocol control
1540  * block, tcp control block, bufer space, and entering LISTEN state
1541  * if to accept connections.
1542  */
1543 static int
1544 tcp_attach(struct socket *so, struct pru_attach_info *ai)
1545 {
1546 	struct tcpcb *tp;
1547 	struct inpcb *inp;
1548 	int error;
1549 	int cpu;
1550 #ifdef INET6
1551 	int isipv6 = INP_CHECK_SOCKAF(so, AF_INET6) != 0;
1552 #endif
1553 
1554 	if (so->so_snd.ssb_hiwat == 0 || so->so_rcv.ssb_hiwat == 0) {
1555 		lwkt_gettoken(&so->so_rcv.ssb_token);
1556 		error = soreserve(so, tcp_sendspace, tcp_recvspace,
1557 				  ai->sb_rlimit);
1558 		lwkt_reltoken(&so->so_rcv.ssb_token);
1559 		if (error)
1560 			return (error);
1561 	}
1562 	atomic_set_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE);
1563 	atomic_set_int(&so->so_snd.ssb_flags, SSB_AUTOSIZE);
1564 	cpu = mycpu->gd_cpuid;
1565 
1566 	/*
1567 	 * Set the default port for protocol processing. This will likely
1568 	 * change when we connect.
1569 	 */
1570 	error = in_pcballoc(so, &tcbinfo[cpu]);
1571 	if (error)
1572 		return (error);
1573 	inp = so->so_pcb;
1574 #ifdef INET6
1575 	if (isipv6) {
1576 		inp->inp_vflag |= INP_IPV6;
1577 		inp->in6p_hops = -1;	/* use kernel default */
1578 	}
1579 	else
1580 #endif
1581 	inp->inp_vflag |= INP_IPV4;
1582 	tp = tcp_newtcpcb(inp);
1583 	if (tp == NULL) {
1584 		/*
1585 		 * Make sure the socket is destroyed by the pcbdetach.
1586 		 */
1587 		soreference(so);
1588 #ifdef INET6
1589 		if (isipv6)
1590 			in6_pcbdetach(inp);
1591 		else
1592 #endif
1593 		in_pcbdetach(inp);
1594 		sofree(so);	/* from ref above */
1595 		return (ENOBUFS);
1596 	}
1597 	tp->t_state = TCPS_CLOSED;
1598 	return (0);
1599 }
1600 
1601 /*
1602  * Initiate (or continue) disconnect.
1603  * If embryonic state, just send reset (once).
1604  * If in ``let data drain'' option and linger null, just drop.
1605  * Otherwise (hard), mark socket disconnecting and drop
1606  * current input data; switch states based on user close, and
1607  * send segment to peer (with FIN).
1608  */
1609 static struct tcpcb *
1610 tcp_disconnect(struct tcpcb *tp)
1611 {
1612 	struct socket *so = tp->t_inpcb->inp_socket;
1613 
1614 	if (tp->t_state < TCPS_ESTABLISHED) {
1615 		tp = tcp_close(tp);
1616 	} else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
1617 		tp = tcp_drop(tp, 0);
1618 	} else {
1619 		lwkt_gettoken(&so->so_rcv.ssb_token);
1620 		soisdisconnecting(so);
1621 		sbflush(&so->so_rcv.sb);
1622 		tp = tcp_usrclosed(tp);
1623 		if (tp)
1624 			tcp_output(tp);
1625 		lwkt_reltoken(&so->so_rcv.ssb_token);
1626 	}
1627 	return (tp);
1628 }
1629 
1630 /*
1631  * User issued close, and wish to trail through shutdown states:
1632  * if never received SYN, just forget it.  If got a SYN from peer,
1633  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
1634  * If already got a FIN from peer, then almost done; go to LAST_ACK
1635  * state.  In all other cases, have already sent FIN to peer (e.g.
1636  * after PRU_SHUTDOWN), and just have to play tedious game waiting
1637  * for peer to send FIN or not respond to keep-alives, etc.
1638  * We can let the user exit from the close as soon as the FIN is acked.
1639  */
1640 static struct tcpcb *
1641 tcp_usrclosed(struct tcpcb *tp)
1642 {
1643 
1644 	switch (tp->t_state) {
1645 
1646 	case TCPS_CLOSED:
1647 	case TCPS_LISTEN:
1648 		tp->t_state = TCPS_CLOSED;
1649 		tp = tcp_close(tp);
1650 		break;
1651 
1652 	case TCPS_SYN_SENT:
1653 	case TCPS_SYN_RECEIVED:
1654 		tp->t_flags |= TF_NEEDFIN;
1655 		break;
1656 
1657 	case TCPS_ESTABLISHED:
1658 		tp->t_state = TCPS_FIN_WAIT_1;
1659 		break;
1660 
1661 	case TCPS_CLOSE_WAIT:
1662 		tp->t_state = TCPS_LAST_ACK;
1663 		break;
1664 	}
1665 	if (tp && tp->t_state >= TCPS_FIN_WAIT_2) {
1666 		soisdisconnected(tp->t_inpcb->inp_socket);
1667 		/* To prevent the connection hanging in FIN_WAIT_2 forever. */
1668 		if (tp->t_state == TCPS_FIN_WAIT_2) {
1669 			tcp_callout_reset(tp, tp->tt_2msl, tp->t_maxidle,
1670 			    tcp_timer_2msl);
1671 		}
1672 	}
1673 	return (tp);
1674 }
1675