xref: /dflybsd-src/sys/netinet/tcp_usrreq.c (revision 73029d086fecb97607aec6269511b65ab8a3131d)
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. Neither the name of the University nor the names of its contributors
47  *    may be used to endorse or promote products derived from this software
48  *    without specific prior written permission.
49  *
50  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
51  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
52  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
53  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
54  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
55  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
56  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
57  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
58  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
59  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
60  * SUCH DAMAGE.
61  *
62  *	From: @(#)tcp_usrreq.c	8.2 (Berkeley) 1/3/94
63  * $FreeBSD: src/sys/netinet/tcp_usrreq.c,v 1.51.2.17 2002/10/11 11:46:44 ume Exp $
64  */
65 
66 #include "opt_ipsec.h"
67 #include "opt_inet.h"
68 #include "opt_inet6.h"
69 #include "opt_tcpdebug.h"
70 
71 #include <sys/param.h>
72 #include <sys/systm.h>
73 #include <sys/kernel.h>
74 #include <sys/malloc.h>
75 #include <sys/sysctl.h>
76 #include <sys/globaldata.h>
77 #include <sys/thread.h>
78 
79 #include <sys/mbuf.h>
80 #ifdef INET6
81 #include <sys/domain.h>
82 #endif /* INET6 */
83 #include <sys/socket.h>
84 #include <sys/socketvar.h>
85 #include <sys/socketops.h>
86 #include <sys/protosw.h>
87 
88 #include <sys/thread2.h>
89 #include <sys/msgport2.h>
90 #include <sys/socketvar2.h>
91 
92 #include <net/if.h>
93 #include <net/netisr.h>
94 #include <net/route.h>
95 
96 #include <net/netmsg2.h>
97 #include <net/netisr2.h>
98 
99 #include <netinet/in.h>
100 #include <netinet/in_systm.h>
101 #ifdef INET6
102 #include <netinet/ip6.h>
103 #endif
104 #include <netinet/in_pcb.h>
105 #ifdef INET6
106 #include <netinet6/in6_pcb.h>
107 #endif
108 #include <netinet/in_var.h>
109 #include <netinet/ip_var.h>
110 #ifdef INET6
111 #include <netinet6/ip6_var.h>
112 #include <netinet6/tcp6_var.h>
113 #endif
114 #include <netinet/tcp.h>
115 #include <netinet/tcp_fsm.h>
116 #include <netinet/tcp_seq.h>
117 #include <netinet/tcp_timer.h>
118 #include <netinet/tcp_timer2.h>
119 #include <netinet/tcp_var.h>
120 #include <netinet/tcpip.h>
121 #ifdef TCPDEBUG
122 #include <netinet/tcp_debug.h>
123 #endif
124 
125 #ifdef IPSEC
126 #include <netinet6/ipsec.h>
127 #endif /*IPSEC*/
128 
129 /*
130  * TCP protocol interface to socket abstraction.
131  */
132 extern	char *tcpstates[];	/* XXX ??? */
133 
134 static int	tcp_attach (struct socket *, struct pru_attach_info *);
135 static void	tcp_connect (netmsg_t msg);
136 #ifdef INET6
137 static void	tcp6_connect (netmsg_t msg);
138 static int	tcp6_connect_oncpu(struct tcpcb *tp, int flags,
139 				struct mbuf **mp,
140 				struct sockaddr_in6 *sin6,
141 				struct in6_addr *addr6);
142 #endif /* INET6 */
143 static struct tcpcb *
144 		tcp_disconnect (struct tcpcb *);
145 static struct tcpcb *
146 		tcp_usrclosed (struct tcpcb *);
147 
148 #ifdef TCPDEBUG
149 #define	TCPDEBUG0	int ostate = 0
150 #define	TCPDEBUG1()	ostate = tp ? tp->t_state : 0
151 #define	TCPDEBUG2(req)	if (tp && (so->so_options & SO_DEBUG)) \
152 				tcp_trace(TA_USER, ostate, tp, 0, 0, req)
153 #else
154 #define	TCPDEBUG0
155 #define	TCPDEBUG1()
156 #define	TCPDEBUG2(req)
157 #endif
158 
159 static int	tcp_lport_extension = 1;
160 SYSCTL_INT(_net_inet_tcp, OID_AUTO, lportext, CTLFLAG_RW,
161     &tcp_lport_extension, 0, "");
162 
163 /*
164  * For some ill optimized programs, which try to use TCP_NOPUSH
165  * to improve performance, will have small amount of data sits
166  * in the sending buffer.  These small amount of data will _not_
167  * be pushed into the network until more data are written into
168  * the socket or the socket write side is shutdown.
169  */
170 static int	tcp_disable_nopush = 1;
171 SYSCTL_INT(_net_inet_tcp, OID_AUTO, disable_nopush, CTLFLAG_RW,
172     &tcp_disable_nopush, 0, "TCP_NOPUSH socket option will have no effect");
173 
174 /*
175  * TCP attaches to socket via pru_attach(), reserving space,
176  * and an internet control block.  This is likely occuring on
177  * cpu0 and may have to move later when we bind/connect.
178  */
179 static void
180 tcp_usr_attach(netmsg_t msg)
181 {
182 	struct socket *so = msg->base.nm_so;
183 	struct pru_attach_info *ai = msg->attach.nm_ai;
184 	int error;
185 	struct inpcb *inp;
186 	struct tcpcb *tp = NULL;
187 	TCPDEBUG0;
188 
189 	soreference(so);
190 	inp = so->so_pcb;
191 	TCPDEBUG1();
192 	if (inp) {
193 		error = EISCONN;
194 		goto out;
195 	}
196 
197 	error = tcp_attach(so, ai);
198 	if (error)
199 		goto out;
200 
201 	if ((so->so_options & SO_LINGER) && so->so_linger == 0)
202 		so->so_linger = TCP_LINGERTIME;
203 	tp = sototcpcb(so);
204 out:
205 	sofree(so);		/* from ref above */
206 	TCPDEBUG2(PRU_ATTACH);
207 	lwkt_replymsg(&msg->lmsg, error);
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(netmsg_t msg)
219 {
220 	struct socket *so = msg->base.nm_so;
221 	int error = 0;
222 	struct inpcb *inp;
223 	struct tcpcb *tp;
224 	TCPDEBUG0;
225 
226 	inp = so->so_pcb;
227 
228 	/*
229 	 * If the inp is already detached it may have been due to an async
230 	 * close.  Just return as if no error occured.
231 	 *
232 	 * It's possible for the tcpcb (tp) to disconnect from the inp due
233 	 * to tcp_drop()->tcp_close() being called.  This may occur *after*
234 	 * the detach message has been queued so we may find a NULL tp here.
235 	 */
236 	if (inp) {
237 		if ((tp = intotcpcb(inp)) != NULL) {
238 			TCPDEBUG1();
239 			tp = tcp_disconnect(tp);
240 			TCPDEBUG2(PRU_DETACH);
241 		}
242 	}
243 	lwkt_replymsg(&msg->lmsg, error);
244 }
245 
246 /*
247  * NOTE: ignore_error is non-zero for certain disconnection races
248  * which we want to silently allow, otherwise close() may return
249  * an unexpected error.
250  *
251  * NOTE: The variables (msg) and (tp) are assumed.
252  */
253 #define	COMMON_START(so, inp, ignore_error)			\
254 	TCPDEBUG0; 						\
255 								\
256 	inp = so->so_pcb; 					\
257 	do {							\
258 		if (inp == NULL) {				\
259 			error = ignore_error ? 0 : EINVAL;	\
260 			tp = NULL;				\
261 			goto out;				\
262 		}						\
263 		tp = intotcpcb(inp);				\
264 		TCPDEBUG1();					\
265 	} while(0)
266 
267 #define COMMON_END1(req, noreply)				\
268 	out: do {						\
269 		TCPDEBUG2(req);					\
270 		if (!(noreply))					\
271 			lwkt_replymsg(&msg->lmsg, error);	\
272 		return;						\
273 	} while(0)
274 
275 #define COMMON_END(req)		COMMON_END1((req), 0)
276 
277 /*
278  * Give the socket an address.
279  */
280 static void
281 tcp_usr_bind(netmsg_t msg)
282 {
283 	struct socket *so = msg->bind.base.nm_so;
284 	struct sockaddr *nam = msg->bind.nm_nam;
285 	struct thread *td = msg->bind.nm_td;
286 	int error = 0;
287 	struct inpcb *inp;
288 	struct tcpcb *tp;
289 	struct sockaddr_in *sinp;
290 
291 	COMMON_START(so, inp, 0);
292 
293 	/*
294 	 * Must check for multicast addresses and disallow binding
295 	 * to them.
296 	 */
297 	sinp = (struct sockaddr_in *)nam;
298 	if (sinp->sin_family == AF_INET &&
299 	    IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
300 		error = EAFNOSUPPORT;
301 		goto out;
302 	}
303 	error = in_pcbbind(inp, nam, td);
304 	if (error)
305 		goto out;
306 	COMMON_END(PRU_BIND);
307 
308 }
309 
310 #ifdef INET6
311 
312 static void
313 tcp6_usr_bind(netmsg_t msg)
314 {
315 	struct socket *so = msg->bind.base.nm_so;
316 	struct sockaddr *nam = msg->bind.nm_nam;
317 	struct thread *td = msg->bind.nm_td;
318 	int error = 0;
319 	struct inpcb *inp;
320 	struct tcpcb *tp;
321 	struct sockaddr_in6 *sin6p;
322 
323 	COMMON_START(so, inp, 0);
324 
325 	/*
326 	 * Must check for multicast addresses and disallow binding
327 	 * to them.
328 	 */
329 	sin6p = (struct sockaddr_in6 *)nam;
330 	if (sin6p->sin6_family == AF_INET6 &&
331 	    IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
332 		error = EAFNOSUPPORT;
333 		goto out;
334 	}
335 	inp->inp_vflag &= ~INP_IPV4;
336 	inp->inp_vflag |= INP_IPV6;
337 	if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
338 		if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr))
339 			inp->inp_vflag |= INP_IPV4;
340 		else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
341 			struct sockaddr_in sin;
342 
343 			in6_sin6_2_sin(&sin, sin6p);
344 			inp->inp_vflag |= INP_IPV4;
345 			inp->inp_vflag &= ~INP_IPV6;
346 			error = in_pcbbind(inp, (struct sockaddr *)&sin, td);
347 			goto out;
348 		}
349 	}
350 	error = in6_pcbbind(inp, nam, td);
351 	if (error)
352 		goto out;
353 	COMMON_END(PRU_BIND);
354 }
355 #endif /* INET6 */
356 
357 struct netmsg_inswildcard {
358 	struct netmsg_base	base;
359 	struct inpcb		*nm_inp;
360 };
361 
362 static void
363 in_pcbinswildcardhash_handler(netmsg_t msg)
364 {
365 	struct netmsg_inswildcard *nm = (struct netmsg_inswildcard *)msg;
366 	int cpu = mycpuid, nextcpu;
367 
368 	in_pcbinswildcardhash_oncpu(nm->nm_inp, &tcbinfo[cpu]);
369 
370 	nextcpu = cpu + 1;
371 	if (nextcpu < ncpus2)
372 		lwkt_forwardmsg(netisr_cpuport(nextcpu), &nm->base.lmsg);
373 	else
374 		lwkt_replymsg(&nm->base.lmsg, 0);
375 }
376 
377 /*
378  * Prepare to accept connections.
379  */
380 static void
381 tcp_usr_listen(netmsg_t msg)
382 {
383 	struct socket *so = msg->listen.base.nm_so;
384 	struct thread *td = msg->listen.nm_td;
385 	int error = 0;
386 	struct inpcb *inp;
387 	struct tcpcb *tp;
388 	struct netmsg_inswildcard nm;
389 	lwkt_port_t port0 = netisr_cpuport(0);
390 
391 	COMMON_START(so, inp, 0);
392 
393 	if (&curthread->td_msgport != port0) {
394 		KASSERT((msg->listen.nm_flags & PRUL_RELINK) == 0,
395 		    ("already asked to relink"));
396 
397 		in_pcbunlink(so->so_pcb, &tcbinfo[mycpuid]);
398 		sosetport(so, port0);
399 		msg->listen.nm_flags |= PRUL_RELINK;
400 
401 		lwkt_forwardmsg(port0, &msg->listen.base.lmsg);
402 		/* msg invalid now */
403 		return;
404 	}
405 	KASSERT(so->so_port == port0, ("so_port is not netisr0"));
406 
407 	if (msg->listen.nm_flags & PRUL_RELINK) {
408 		msg->listen.nm_flags &= ~PRUL_RELINK;
409 		in_pcblink(so->so_pcb, &tcbinfo[mycpuid]);
410 	}
411 	KASSERT(inp->inp_pcbinfo == &tcbinfo[0], ("pcbinfo is not tcbinfo0"));
412 
413 	if (tp->t_flags & TF_LISTEN)
414 		goto out;
415 
416 	if (inp->inp_lport == 0) {
417 		error = in_pcbbind(inp, NULL, td);
418 		if (error)
419 			goto out;
420 	}
421 
422 	tp->t_state = TCPS_LISTEN;
423 	tp->t_flags |= TF_LISTEN;
424 	tp->tt_msg = NULL; /* Catch any invalid timer usage */
425 
426 	if (ncpus > 1) {
427 		/*
428 		 * We have to set the flag because we can't have other cpus
429 		 * messing with our inp's flags.
430 		 */
431 		KASSERT(!(inp->inp_flags & INP_CONNECTED),
432 			("already on connhash"));
433 		KASSERT(!(inp->inp_flags & INP_WILDCARD),
434 			("already on wildcardhash"));
435 		KASSERT(!(inp->inp_flags & INP_WILDCARD_MP),
436 			("already on MP wildcardhash"));
437 		inp->inp_flags |= INP_WILDCARD_MP;
438 
439 		netmsg_init(&nm.base, NULL, &curthread->td_msgport,
440 			    MSGF_PRIORITY, in_pcbinswildcardhash_handler);
441 		nm.nm_inp = inp;
442 		lwkt_domsg(netisr_cpuport(1), &nm.base.lmsg, 0);
443 	}
444 	in_pcbinswildcardhash(inp);
445 	COMMON_END(PRU_LISTEN);
446 }
447 
448 #ifdef INET6
449 
450 static void
451 tcp6_usr_listen(netmsg_t msg)
452 {
453 	struct socket *so = msg->listen.base.nm_so;
454 	struct thread *td = msg->listen.nm_td;
455 	int error = 0;
456 	struct inpcb *inp;
457 	struct tcpcb *tp;
458 	struct netmsg_inswildcard nm;
459 
460 	COMMON_START(so, inp, 0);
461 
462 	if (tp->t_flags & TF_LISTEN)
463 		goto out;
464 
465 	if (inp->inp_lport == 0) {
466 		if (!(inp->inp_flags & IN6P_IPV6_V6ONLY))
467 			inp->inp_vflag |= INP_IPV4;
468 		else
469 			inp->inp_vflag &= ~INP_IPV4;
470 		error = in6_pcbbind(inp, NULL, td);
471 		if (error)
472 			goto out;
473 	}
474 
475 	tp->t_state = TCPS_LISTEN;
476 	tp->t_flags |= TF_LISTEN;
477 	tp->tt_msg = NULL; /* Catch any invalid timer usage */
478 
479 	if (ncpus > 1) {
480 		/*
481 		 * We have to set the flag because we can't have other cpus
482 		 * messing with our inp's flags.
483 		 */
484 		KASSERT(!(inp->inp_flags & INP_CONNECTED),
485 			("already on connhash"));
486 		KASSERT(!(inp->inp_flags & INP_WILDCARD),
487 			("already on wildcardhash"));
488 		KASSERT(!(inp->inp_flags & INP_WILDCARD_MP),
489 			("already on MP wildcardhash"));
490 		inp->inp_flags |= INP_WILDCARD_MP;
491 
492 		KKASSERT(so->so_port == netisr_cpuport(0));
493 		KKASSERT(&curthread->td_msgport == netisr_cpuport(0));
494 		KKASSERT(inp->inp_pcbinfo == &tcbinfo[0]);
495 
496 		netmsg_init(&nm.base, NULL, &curthread->td_msgport,
497 			    MSGF_PRIORITY, in_pcbinswildcardhash_handler);
498 		nm.nm_inp = inp;
499 		lwkt_domsg(netisr_cpuport(1), &nm.base.lmsg, 0);
500 	}
501 	in_pcbinswildcardhash(inp);
502 	COMMON_END(PRU_LISTEN);
503 }
504 #endif /* INET6 */
505 
506 /*
507  * Initiate connection to peer.
508  * Create a template for use in transmissions on this connection.
509  * Enter SYN_SENT state, and mark socket as connecting.
510  * Start keep-alive timer, and seed output sequence space.
511  * Send initial segment on connection.
512  */
513 static void
514 tcp_usr_connect(netmsg_t msg)
515 {
516 	struct socket *so = msg->connect.base.nm_so;
517 	struct sockaddr *nam = msg->connect.nm_nam;
518 	struct thread *td = msg->connect.nm_td;
519 	int error = 0;
520 	struct inpcb *inp;
521 	struct tcpcb *tp;
522 	struct sockaddr_in *sinp;
523 
524 	COMMON_START(so, inp, 0);
525 
526 	/*
527 	 * Must disallow TCP ``connections'' to multicast addresses.
528 	 */
529 	sinp = (struct sockaddr_in *)nam;
530 	if (sinp->sin_family == AF_INET
531 	    && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
532 		error = EAFNOSUPPORT;
533 		goto out;
534 	}
535 
536 	if (!prison_remote_ip(td, (struct sockaddr*)sinp)) {
537 		error = EAFNOSUPPORT; /* IPv6 only jail */
538 		goto out;
539 	}
540 
541 	tcp_connect(msg);
542 	/* msg is invalid now */
543 	return;
544 out:
545 	if (msg->connect.nm_m) {
546 		m_freem(msg->connect.nm_m);
547 		msg->connect.nm_m = NULL;
548 	}
549 	if (msg->connect.nm_flags & PRUC_HELDTD)
550 		lwkt_rele(td);
551 	if (error && (msg->connect.nm_flags & PRUC_ASYNC)) {
552 		so->so_error = error;
553 		soisdisconnected(so);
554 	}
555 	lwkt_replymsg(&msg->lmsg, error);
556 }
557 
558 #ifdef INET6
559 
560 static void
561 tcp6_usr_connect(netmsg_t msg)
562 {
563 	struct socket *so = msg->connect.base.nm_so;
564 	struct sockaddr *nam = msg->connect.nm_nam;
565 	struct thread *td = msg->connect.nm_td;
566 	int error = 0;
567 	struct inpcb *inp;
568 	struct tcpcb *tp;
569 	struct sockaddr_in6 *sin6p;
570 
571 	COMMON_START(so, inp, 0);
572 
573 	/*
574 	 * Must disallow TCP ``connections'' to multicast addresses.
575 	 */
576 	sin6p = (struct sockaddr_in6 *)nam;
577 	if (sin6p->sin6_family == AF_INET6
578 	    && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
579 		error = EAFNOSUPPORT;
580 		goto out;
581 	}
582 
583 	if (!prison_remote_ip(td, nam)) {
584 		error = EAFNOSUPPORT; /* IPv4 only jail */
585 		goto out;
586 	}
587 
588 	if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
589 		struct sockaddr_in *sinp;
590 
591 		if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
592 			error = EINVAL;
593 			goto out;
594 		}
595 		sinp = kmalloc(sizeof(*sinp), M_LWKTMSG, M_INTWAIT);
596 		in6_sin6_2_sin(sinp, sin6p);
597 		inp->inp_vflag |= INP_IPV4;
598 		inp->inp_vflag &= ~INP_IPV6;
599 		msg->connect.nm_nam = (struct sockaddr *)sinp;
600 		msg->connect.nm_flags |= PRUC_NAMALLOC;
601 		tcp_connect(msg);
602 		/* msg is invalid now */
603 		return;
604 	}
605 	inp->inp_vflag &= ~INP_IPV4;
606 	inp->inp_vflag |= INP_IPV6;
607 	inp->inp_inc.inc_isipv6 = 1;
608 
609 	msg->connect.nm_flags |= PRUC_FALLBACK;
610 	tcp6_connect(msg);
611 	/* msg is invalid now */
612 	return;
613 out:
614 	if (msg->connect.nm_m) {
615 		m_freem(msg->connect.nm_m);
616 		msg->connect.nm_m = NULL;
617 	}
618 	lwkt_replymsg(&msg->lmsg, error);
619 }
620 
621 #endif /* INET6 */
622 
623 /*
624  * Initiate disconnect from peer.
625  * If connection never passed embryonic stage, just drop;
626  * else if don't need to let data drain, then can just drop anyways,
627  * else have to begin TCP shutdown process: mark socket disconnecting,
628  * drain unread data, state switch to reflect user close, and
629  * send segment (e.g. FIN) to peer.  Socket will be really disconnected
630  * when peer sends FIN and acks ours.
631  *
632  * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
633  */
634 static void
635 tcp_usr_disconnect(netmsg_t msg)
636 {
637 	struct socket *so = msg->disconnect.base.nm_so;
638 	int error = 0;
639 	struct inpcb *inp;
640 	struct tcpcb *tp;
641 
642 	COMMON_START(so, inp, 1);
643 	tp = tcp_disconnect(tp);
644 	COMMON_END(PRU_DISCONNECT);
645 }
646 
647 /*
648  * Accept a connection.  Essentially all the work is
649  * done at higher levels; just return the address
650  * of the peer, storing through addr.
651  */
652 static void
653 tcp_usr_accept(netmsg_t msg)
654 {
655 	struct socket *so = msg->accept.base.nm_so;
656 	struct sockaddr **nam = msg->accept.nm_nam;
657 	int error = 0;
658 	struct inpcb *inp;
659 	struct tcpcb *tp = NULL;
660 	TCPDEBUG0;
661 
662 	inp = so->so_pcb;
663 	if (so->so_state & SS_ISDISCONNECTED) {
664 		error = ECONNABORTED;
665 		goto out;
666 	}
667 	if (inp == 0) {
668 		error = EINVAL;
669 		goto out;
670 	}
671 
672 	tp = intotcpcb(inp);
673 	TCPDEBUG1();
674 	in_setpeeraddr(so, nam);
675 	COMMON_END(PRU_ACCEPT);
676 }
677 
678 #ifdef INET6
679 static void
680 tcp6_usr_accept(netmsg_t msg)
681 {
682 	struct socket *so = msg->accept.base.nm_so;
683 	struct sockaddr **nam = msg->accept.nm_nam;
684 	int error = 0;
685 	struct inpcb *inp;
686 	struct tcpcb *tp = NULL;
687 	TCPDEBUG0;
688 
689 	inp = so->so_pcb;
690 
691 	if (so->so_state & SS_ISDISCONNECTED) {
692 		error = ECONNABORTED;
693 		goto out;
694 	}
695 	if (inp == 0) {
696 		error = EINVAL;
697 		goto out;
698 	}
699 	tp = intotcpcb(inp);
700 	TCPDEBUG1();
701 	in6_mapped_peeraddr(so, nam);
702 	COMMON_END(PRU_ACCEPT);
703 }
704 #endif /* INET6 */
705 /*
706  * Mark the connection as being incapable of further output.
707  */
708 static void
709 tcp_usr_shutdown(netmsg_t msg)
710 {
711 	struct socket *so = msg->shutdown.base.nm_so;
712 	int error = 0;
713 	struct inpcb *inp;
714 	struct tcpcb *tp;
715 
716 	COMMON_START(so, inp, 0);
717 	socantsendmore(so);
718 	tp = tcp_usrclosed(tp);
719 	if (tp)
720 		error = tcp_output(tp);
721 	COMMON_END(PRU_SHUTDOWN);
722 }
723 
724 /*
725  * After a receive, possibly send window update to peer.
726  */
727 static void
728 tcp_usr_rcvd(netmsg_t msg)
729 {
730 	struct socket *so = msg->rcvd.base.nm_so;
731 	int error = 0, noreply = 0;
732 	struct inpcb *inp;
733 	struct tcpcb *tp;
734 
735 	COMMON_START(so, inp, 0);
736 
737 	if (msg->rcvd.nm_pru_flags & PRUR_ASYNC) {
738 		noreply = 1;
739 		so_async_rcvd_reply(so);
740 	}
741 	tcp_output(tp);
742 
743 	COMMON_END1(PRU_RCVD, noreply);
744 }
745 
746 /*
747  * Do a send by putting data in output queue and updating urgent
748  * marker if URG set.  Possibly send more data.  Unlike the other
749  * pru_*() routines, the mbuf chains are our responsibility.  We
750  * must either enqueue them or free them.  The other pru_* routines
751  * generally are caller-frees.
752  */
753 static void
754 tcp_usr_send(netmsg_t msg)
755 {
756 	struct socket *so = msg->send.base.nm_so;
757 	int flags = msg->send.nm_flags;
758 	struct mbuf *m = msg->send.nm_m;
759 	int error = 0;
760 	struct inpcb *inp;
761 	struct tcpcb *tp;
762 	TCPDEBUG0;
763 
764 	KKASSERT(msg->send.nm_control == NULL);
765 	KKASSERT(msg->send.nm_addr == NULL);
766 	KKASSERT((flags & PRUS_FREEADDR) == 0);
767 
768 	inp = so->so_pcb;
769 
770 	if (inp == NULL) {
771 		/*
772 		 * OOPS! we lost a race, the TCP session got reset after
773 		 * we checked SS_CANTSENDMORE, eg: while doing uiomove or a
774 		 * network interrupt in the non-critical section of sosend().
775 		 */
776 		m_freem(m);
777 		error = ECONNRESET;	/* XXX EPIPE? */
778 		tp = NULL;
779 		TCPDEBUG1();
780 		goto out;
781 	}
782 	tp = intotcpcb(inp);
783 	TCPDEBUG1();
784 
785 #ifdef foo
786 	/*
787 	 * This is no longer necessary, since:
788 	 * - sosendtcp() has already checked it for us
789 	 * - It does not work with asynchronized send
790 	 */
791 
792 	/*
793 	 * Don't let too much OOB data build up
794 	 */
795 	if (flags & PRUS_OOB) {
796 		if (ssb_space(&so->so_snd) < -512) {
797 			m_freem(m);
798 			error = ENOBUFS;
799 			goto out;
800 		}
801 	}
802 #endif
803 
804 	/*
805 	 * Pump the data into the socket.
806 	 */
807 	if (m)
808 		ssb_appendstream(&so->so_snd, m);
809 	if (flags & PRUS_OOB) {
810 		/*
811 		 * According to RFC961 (Assigned Protocols),
812 		 * the urgent pointer points to the last octet
813 		 * of urgent data.  We continue, however,
814 		 * to consider it to indicate the first octet
815 		 * of data past the urgent section.
816 		 * Otherwise, snd_up should be one lower.
817 		 */
818 		tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
819 		tp->t_flags |= TF_FORCE;
820 		error = tcp_output(tp);
821 		tp->t_flags &= ~TF_FORCE;
822 	} else {
823 		if (flags & PRUS_EOF) {
824 			/*
825 			 * Close the send side of the connection after
826 			 * the data is sent.
827 			 */
828 			socantsendmore(so);
829 			tp = tcp_usrclosed(tp);
830 		}
831 		if (tp != NULL && !tcp_output_pending(tp)) {
832 			if (flags & PRUS_MORETOCOME)
833 				tp->t_flags |= TF_MORETOCOME;
834 			error = tcp_output_fair(tp);
835 			if (flags & PRUS_MORETOCOME)
836 				tp->t_flags &= ~TF_MORETOCOME;
837 		}
838 	}
839 	COMMON_END1((flags & PRUS_OOB) ? PRU_SENDOOB :
840 		   ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND),
841 		   (flags & PRUS_NOREPLY));
842 }
843 
844 /*
845  * NOTE: (so) is referenced from soabort*() and netmsg_pru_abort()
846  *	 will sofree() it when we return.
847  */
848 static void
849 tcp_usr_abort(netmsg_t msg)
850 {
851 	struct socket *so = msg->abort.base.nm_so;
852 	int error = 0;
853 	struct inpcb *inp;
854 	struct tcpcb *tp;
855 
856 	COMMON_START(so, inp, 1);
857 	tp = tcp_drop(tp, ECONNABORTED);
858 	COMMON_END(PRU_ABORT);
859 }
860 
861 /*
862  * Receive out-of-band data.
863  */
864 static void
865 tcp_usr_rcvoob(netmsg_t msg)
866 {
867 	struct socket *so = msg->rcvoob.base.nm_so;
868 	struct mbuf *m = msg->rcvoob.nm_m;
869 	int flags = msg->rcvoob.nm_flags;
870 	int error = 0;
871 	struct inpcb *inp;
872 	struct tcpcb *tp;
873 
874 	COMMON_START(so, inp, 0);
875 	if ((so->so_oobmark == 0 &&
876 	     (so->so_state & SS_RCVATMARK) == 0) ||
877 	    so->so_options & SO_OOBINLINE ||
878 	    tp->t_oobflags & TCPOOB_HADDATA) {
879 		error = EINVAL;
880 		goto out;
881 	}
882 	if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
883 		error = EWOULDBLOCK;
884 		goto out;
885 	}
886 	m->m_len = 1;
887 	*mtod(m, caddr_t) = tp->t_iobc;
888 	if ((flags & MSG_PEEK) == 0)
889 		tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
890 	COMMON_END(PRU_RCVOOB);
891 }
892 
893 static void
894 tcp_usr_savefaddr(struct socket *so, const struct sockaddr *faddr)
895 {
896 	in_savefaddr(so, faddr);
897 }
898 
899 #ifdef INET6
900 static void
901 tcp6_usr_savefaddr(struct socket *so, const struct sockaddr *faddr)
902 {
903 	in6_mapped_savefaddr(so, faddr);
904 }
905 #endif
906 
907 static int
908 tcp_usr_preconnect(struct socket *so, const struct sockaddr *nam,
909     struct thread *td __unused)
910 {
911 	const struct sockaddr_in *sinp;
912 
913 	sinp = (const struct sockaddr_in *)nam;
914 	if (sinp->sin_family == AF_INET &&
915 	    IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
916 		return EAFNOSUPPORT;
917 
918 	soisconnecting(so);
919 	return 0;
920 }
921 
922 /* xxx - should be const */
923 struct pr_usrreqs tcp_usrreqs = {
924 	.pru_abort = tcp_usr_abort,
925 	.pru_accept = tcp_usr_accept,
926 	.pru_attach = tcp_usr_attach,
927 	.pru_bind = tcp_usr_bind,
928 	.pru_connect = tcp_usr_connect,
929 	.pru_connect2 = pr_generic_notsupp,
930 	.pru_control = in_control_dispatch,
931 	.pru_detach = tcp_usr_detach,
932 	.pru_disconnect = tcp_usr_disconnect,
933 	.pru_listen = tcp_usr_listen,
934 	.pru_peeraddr = in_setpeeraddr_dispatch,
935 	.pru_rcvd = tcp_usr_rcvd,
936 	.pru_rcvoob = tcp_usr_rcvoob,
937 	.pru_send = tcp_usr_send,
938 	.pru_sense = pru_sense_null,
939 	.pru_shutdown = tcp_usr_shutdown,
940 	.pru_sockaddr = in_setsockaddr_dispatch,
941 	.pru_sosend = sosendtcp,
942 	.pru_soreceive = sorecvtcp,
943 	.pru_savefaddr = tcp_usr_savefaddr,
944 	.pru_preconnect = tcp_usr_preconnect
945 };
946 
947 #ifdef INET6
948 struct pr_usrreqs tcp6_usrreqs = {
949 	.pru_abort = tcp_usr_abort,
950 	.pru_accept = tcp6_usr_accept,
951 	.pru_attach = tcp_usr_attach,
952 	.pru_bind = tcp6_usr_bind,
953 	.pru_connect = tcp6_usr_connect,
954 	.pru_connect2 = pr_generic_notsupp,
955 	.pru_control = in6_control_dispatch,
956 	.pru_detach = tcp_usr_detach,
957 	.pru_disconnect = tcp_usr_disconnect,
958 	.pru_listen = tcp6_usr_listen,
959 	.pru_peeraddr = in6_mapped_peeraddr_dispatch,
960 	.pru_rcvd = tcp_usr_rcvd,
961 	.pru_rcvoob = tcp_usr_rcvoob,
962 	.pru_send = tcp_usr_send,
963 	.pru_sense = pru_sense_null,
964 	.pru_shutdown = tcp_usr_shutdown,
965 	.pru_sockaddr = in6_mapped_sockaddr_dispatch,
966 	.pru_sosend = sosendtcp,
967 	.pru_soreceive = sorecvtcp,
968 	.pru_savefaddr = tcp6_usr_savefaddr
969 };
970 #endif /* INET6 */
971 
972 static int
973 tcp_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf *m,
974 		  struct sockaddr_in *sin, struct sockaddr_in *if_sin)
975 {
976 	struct inpcb *inp = tp->t_inpcb, *oinp;
977 	struct socket *so = inp->inp_socket;
978 	struct route *ro = &inp->inp_route;
979 
980 	KASSERT(inp->inp_pcbinfo == &tcbinfo[mycpu->gd_cpuid],
981 	    ("pcbinfo mismatch"));
982 
983 	oinp = in_pcblookup_hash(inp->inp_pcbinfo,
984 				 sin->sin_addr, sin->sin_port,
985 				 (inp->inp_laddr.s_addr != INADDR_ANY ?
986 				  inp->inp_laddr : if_sin->sin_addr),
987 				inp->inp_lport, 0, NULL);
988 	if (oinp != NULL) {
989 		m_freem(m);
990 		return (EADDRINUSE);
991 	}
992 	if (inp->inp_laddr.s_addr == INADDR_ANY)
993 		inp->inp_laddr = if_sin->sin_addr;
994 	inp->inp_faddr = sin->sin_addr;
995 	inp->inp_fport = sin->sin_port;
996 	in_pcbinsconnhash(inp);
997 
998 	/*
999 	 * We are now on the inpcb's owner CPU, if the cached route was
1000 	 * freed because the rtentry's owner CPU is not the current CPU
1001 	 * (e.g. in tcp_connect()), then we try to reallocate it here with
1002 	 * the hope that a rtentry may be cloned from a RTF_PRCLONING
1003 	 * rtentry.
1004 	 */
1005 	if (!(inp->inp_socket->so_options & SO_DONTROUTE) && /*XXX*/
1006 	    ro->ro_rt == NULL) {
1007 		bzero(&ro->ro_dst, sizeof(struct sockaddr_in));
1008 		ro->ro_dst.sa_family = AF_INET;
1009 		ro->ro_dst.sa_len = sizeof(struct sockaddr_in);
1010 		((struct sockaddr_in *)&ro->ro_dst)->sin_addr =
1011 			sin->sin_addr;
1012 		rtalloc(ro);
1013 	}
1014 
1015 	/*
1016 	 * Now that no more errors can occur, change the protocol processing
1017 	 * port to the current thread (which is the correct thread).
1018 	 *
1019 	 * Create TCP timer message now; we are on the tcpcb's owner
1020 	 * CPU/thread.
1021 	 */
1022 	tcp_create_timermsg(tp, &curthread->td_msgport);
1023 
1024 	/*
1025 	 * Compute window scaling to request.  Use a larger scaling then
1026 	 * needed for the initial receive buffer in case the receive buffer
1027 	 * gets expanded.
1028 	 */
1029 	if (tp->request_r_scale < TCP_MIN_WINSHIFT)
1030 		tp->request_r_scale = TCP_MIN_WINSHIFT;
1031 	while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1032 	       (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat
1033 	) {
1034 		tp->request_r_scale++;
1035 	}
1036 
1037 	soisconnecting(so);
1038 	tcpstat.tcps_connattempt++;
1039 	tp->t_state = TCPS_SYN_SENT;
1040 	tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
1041 	tp->iss = tcp_new_isn(tp);
1042 	tcp_sendseqinit(tp);
1043 	if (m) {
1044 		ssb_appendstream(&so->so_snd, m);
1045 		m = NULL;
1046 		if (flags & PRUS_OOB)
1047 			tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
1048 	}
1049 
1050 	/*
1051 	 * Close the send side of the connection after
1052 	 * the data is sent if flagged.
1053 	 */
1054 	if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
1055 		socantsendmore(so);
1056 		tp = tcp_usrclosed(tp);
1057 	}
1058 	return (tcp_output(tp));
1059 }
1060 
1061 /*
1062  * Common subroutine to open a TCP connection to remote host specified
1063  * by struct sockaddr_in in mbuf *nam.  Call in_pcbbind to assign a local
1064  * port number if needed.  Call in_pcbladdr to do the routing and to choose
1065  * a local host address (interface).
1066  * Initialize connection parameters and enter SYN-SENT state.
1067  */
1068 static void
1069 tcp_connect(netmsg_t msg)
1070 {
1071 	struct socket *so = msg->connect.base.nm_so;
1072 	struct sockaddr *nam = msg->connect.nm_nam;
1073 	struct thread *td = msg->connect.nm_td;
1074 	struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1075 	struct sockaddr_in *if_sin = NULL;
1076 	struct inpcb *inp;
1077 	struct tcpcb *tp;
1078 	int error;
1079 	lwkt_port_t port;
1080 
1081 	COMMON_START(so, inp, 0);
1082 
1083 	/*
1084 	 * Reconnect our pcb if we have to
1085 	 */
1086 	if (msg->connect.nm_flags & PRUC_RECONNECT) {
1087 		msg->connect.nm_flags &= ~PRUC_RECONNECT;
1088 		in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1089 	}
1090 
1091 	/*
1092 	 * Bind if we have to
1093 	 */
1094 	if (inp->inp_lport == 0) {
1095 		if (tcp_lport_extension) {
1096 			KKASSERT(inp->inp_laddr.s_addr == INADDR_ANY);
1097 
1098 			error = in_pcbladdr(inp, nam, &if_sin, td);
1099 			if (error)
1100 				goto out;
1101 			inp->inp_laddr.s_addr = if_sin->sin_addr.s_addr;
1102 
1103 			error = in_pcbbind_remote(inp, nam, td);
1104 			if (error)
1105 				goto out;
1106 
1107 			msg->connect.nm_flags |= PRUC_HASLADDR;
1108 		} else {
1109 			error = in_pcbbind(inp, NULL, td);
1110 			if (error)
1111 				goto out;
1112 		}
1113 	}
1114 
1115 	if ((msg->connect.nm_flags & PRUC_HASLADDR) == 0) {
1116 		/*
1117 		 * Calculate the correct protocol processing thread.  The
1118 		 * connect operation must run there.  Set the forwarding
1119 		 * port before we forward the message or it will get bounced
1120 		 * right back to us.
1121 		 */
1122 		error = in_pcbladdr(inp, nam, &if_sin, td);
1123 		if (error)
1124 			goto out;
1125 	}
1126 	KKASSERT(inp->inp_socket == so);
1127 
1128 	port = tcp_addrport(sin->sin_addr.s_addr, sin->sin_port,
1129 			    (inp->inp_laddr.s_addr != INADDR_ANY ?
1130 			     inp->inp_laddr.s_addr : if_sin->sin_addr.s_addr),
1131 			    inp->inp_lport);
1132 
1133 	if (port != &curthread->td_msgport) {
1134 		struct route *ro = &inp->inp_route;
1135 
1136 		/*
1137 		 * in_pcbladdr() may have allocated a route entry for us
1138 		 * on the current CPU, but we need a route entry on the
1139 		 * inpcb's owner CPU, so free it here.
1140 		 */
1141 		if (ro->ro_rt != NULL)
1142 			RTFREE(ro->ro_rt);
1143 		bzero(ro, sizeof(*ro));
1144 
1145 		/*
1146 		 * We are moving the protocol processing port the socket
1147 		 * is on, we have to unlink here and re-link on the
1148 		 * target cpu.
1149 		 */
1150 		in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1151 		sosetport(so, port);
1152 		msg->connect.nm_flags |= PRUC_RECONNECT;
1153 		msg->connect.base.nm_dispatch = tcp_connect;
1154 
1155 		lwkt_forwardmsg(port, &msg->connect.base.lmsg);
1156 		/* msg invalid now */
1157 		return;
1158 	} else if (msg->connect.nm_flags & PRUC_HELDTD) {
1159 		/*
1160 		 * The original thread is no longer needed; release it.
1161 		 */
1162 		lwkt_rele(td);
1163 		msg->connect.nm_flags &= ~PRUC_HELDTD;
1164 	}
1165 	error = tcp_connect_oncpu(tp, msg->connect.nm_sndflags,
1166 				  msg->connect.nm_m, sin, if_sin);
1167 	msg->connect.nm_m = NULL;
1168 out:
1169 	if (msg->connect.nm_m) {
1170 		m_freem(msg->connect.nm_m);
1171 		msg->connect.nm_m = NULL;
1172 	}
1173 	if (msg->connect.nm_flags & PRUC_NAMALLOC) {
1174 		kfree(msg->connect.nm_nam, M_LWKTMSG);
1175 		msg->connect.nm_nam = NULL;
1176 	}
1177 	if (msg->connect.nm_flags & PRUC_HELDTD)
1178 		lwkt_rele(td);
1179 	if (error && (msg->connect.nm_flags & PRUC_ASYNC)) {
1180 		so->so_error = error;
1181 		soisdisconnected(so);
1182 	}
1183 	lwkt_replymsg(&msg->connect.base.lmsg, error);
1184 	/* msg invalid now */
1185 }
1186 
1187 #ifdef INET6
1188 
1189 static void
1190 tcp6_connect(netmsg_t msg)
1191 {
1192 	struct tcpcb *tp;
1193 	struct socket *so = msg->connect.base.nm_so;
1194 	struct sockaddr *nam = msg->connect.nm_nam;
1195 	struct thread *td = msg->connect.nm_td;
1196 	struct inpcb *inp;
1197 	struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam;
1198 	struct in6_addr *addr6;
1199 	lwkt_port_t port;
1200 	int error;
1201 
1202 	COMMON_START(so, inp, 0);
1203 
1204 	/*
1205 	 * Reconnect our pcb if we have to
1206 	 */
1207 	if (msg->connect.nm_flags & PRUC_RECONNECT) {
1208 		msg->connect.nm_flags &= ~PRUC_RECONNECT;
1209 		in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1210 	}
1211 
1212 	/*
1213 	 * Bind if we have to
1214 	 */
1215 	if (inp->inp_lport == 0) {
1216 		error = in6_pcbbind(inp, NULL, td);
1217 		if (error)
1218 			goto out;
1219 	}
1220 
1221 	/*
1222 	 * Cannot simply call in_pcbconnect, because there might be an
1223 	 * earlier incarnation of this same connection still in
1224 	 * TIME_WAIT state, creating an ADDRINUSE error.
1225 	 */
1226 	error = in6_pcbladdr(inp, nam, &addr6, td);
1227 	if (error)
1228 		goto out;
1229 
1230 	port = tcp6_addrport();	/* XXX hack for now, always cpu0 */
1231 
1232 	if (port != &curthread->td_msgport) {
1233 		struct route *ro = &inp->inp_route;
1234 
1235 		/*
1236 		 * in_pcbladdr() may have allocated a route entry for us
1237 		 * on the current CPU, but we need a route entry on the
1238 		 * inpcb's owner CPU, so free it here.
1239 		 */
1240 		if (ro->ro_rt != NULL)
1241 			RTFREE(ro->ro_rt);
1242 		bzero(ro, sizeof(*ro));
1243 
1244 		in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1245 		sosetport(so, port);
1246 		msg->connect.nm_flags |= PRUC_RECONNECT;
1247 		msg->connect.base.nm_dispatch = tcp6_connect;
1248 
1249 		lwkt_forwardmsg(port, &msg->connect.base.lmsg);
1250 		/* msg invalid now */
1251 		return;
1252 	}
1253 	error = tcp6_connect_oncpu(tp, msg->connect.nm_sndflags,
1254 				   &msg->connect.nm_m, sin6, addr6);
1255 	/* nm_m may still be intact */
1256 out:
1257 	if (error && (msg->connect.nm_flags & PRUC_FALLBACK)) {
1258 		tcp_connect(msg);
1259 		/* msg invalid now */
1260 	} else {
1261 		if (msg->connect.nm_m) {
1262 			m_freem(msg->connect.nm_m);
1263 			msg->connect.nm_m = NULL;
1264 		}
1265 		if (msg->connect.nm_flags & PRUC_NAMALLOC) {
1266 			kfree(msg->connect.nm_nam, M_LWKTMSG);
1267 			msg->connect.nm_nam = NULL;
1268 		}
1269 		lwkt_replymsg(&msg->connect.base.lmsg, error);
1270 		/* msg invalid now */
1271 	}
1272 }
1273 
1274 static int
1275 tcp6_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf **mp,
1276 		   struct sockaddr_in6 *sin6, struct in6_addr *addr6)
1277 {
1278 	struct mbuf *m = *mp;
1279 	struct inpcb *inp = tp->t_inpcb;
1280 	struct socket *so = inp->inp_socket;
1281 	struct inpcb *oinp;
1282 
1283 	/*
1284 	 * Cannot simply call in_pcbconnect, because there might be an
1285 	 * earlier incarnation of this same connection still in
1286 	 * TIME_WAIT state, creating an ADDRINUSE error.
1287 	 */
1288 	oinp = in6_pcblookup_hash(inp->inp_pcbinfo,
1289 				  &sin6->sin6_addr, sin6->sin6_port,
1290 				  (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) ?
1291 				      addr6 : &inp->in6p_laddr),
1292 				  inp->inp_lport,  0, NULL);
1293 	if (oinp)
1294 		return (EADDRINUSE);
1295 
1296 	if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
1297 		inp->in6p_laddr = *addr6;
1298 	inp->in6p_faddr = sin6->sin6_addr;
1299 	inp->inp_fport = sin6->sin6_port;
1300 	if ((sin6->sin6_flowinfo & IPV6_FLOWINFO_MASK) != 0)
1301 		inp->in6p_flowinfo = sin6->sin6_flowinfo;
1302 	in_pcbinsconnhash(inp);
1303 
1304 	/*
1305 	 * Now that no more errors can occur, change the protocol processing
1306 	 * port to the current thread (which is the correct thread).
1307 	 *
1308 	 * Create TCP timer message now; we are on the tcpcb's owner
1309 	 * CPU/thread.
1310 	 */
1311 	tcp_create_timermsg(tp, &curthread->td_msgport);
1312 
1313 	/* Compute window scaling to request.  */
1314 	if (tp->request_r_scale < TCP_MIN_WINSHIFT)
1315 		tp->request_r_scale = TCP_MIN_WINSHIFT;
1316 	while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1317 	    (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat) {
1318 		tp->request_r_scale++;
1319 	}
1320 
1321 	soisconnecting(so);
1322 	tcpstat.tcps_connattempt++;
1323 	tp->t_state = TCPS_SYN_SENT;
1324 	tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
1325 	tp->iss = tcp_new_isn(tp);
1326 	tcp_sendseqinit(tp);
1327 	if (m) {
1328 		ssb_appendstream(&so->so_snd, m);
1329 		*mp = NULL;
1330 		if (flags & PRUS_OOB)
1331 			tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
1332 	}
1333 
1334 	/*
1335 	 * Close the send side of the connection after
1336 	 * the data is sent if flagged.
1337 	 */
1338 	if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
1339 		socantsendmore(so);
1340 		tp = tcp_usrclosed(tp);
1341 	}
1342 	return (tcp_output(tp));
1343 }
1344 
1345 #endif /* INET6 */
1346 
1347 /*
1348  * The new sockopt interface makes it possible for us to block in the
1349  * copyin/out step (if we take a page fault).  Taking a page fault while
1350  * in a critical section is probably a Bad Thing.  (Since sockets and pcbs
1351  * both now use TSM, there probably isn't any need for this function to
1352  * run in a critical section any more.  This needs more examination.)
1353  */
1354 void
1355 tcp_ctloutput(netmsg_t msg)
1356 {
1357 	struct socket *so = msg->base.nm_so;
1358 	struct sockopt *sopt = msg->ctloutput.nm_sopt;
1359 	int	error, opt, optval, opthz;
1360 	struct	inpcb *inp;
1361 	struct	tcpcb *tp;
1362 
1363 	error = 0;
1364 	inp = so->so_pcb;
1365 	if (inp == NULL) {
1366 		error = ECONNRESET;
1367 		goto done;
1368 	}
1369 
1370 	if (sopt->sopt_level != IPPROTO_TCP) {
1371 		switch (sopt->sopt_name) {
1372 		case IP_MULTICAST_IF:
1373 		case IP_MULTICAST_VIF:
1374 		case IP_MULTICAST_TTL:
1375 		case IP_MULTICAST_LOOP:
1376 		case IP_ADD_MEMBERSHIP:
1377 		case IP_DROP_MEMBERSHIP:
1378 			/*
1379 			 * Multicast does not make sense on TCP sockets.
1380 			 */
1381 			error = EOPNOTSUPP;
1382 			goto done;
1383 		}
1384 #ifdef INET6
1385 		if (INP_CHECK_SOCKAF(so, AF_INET6))
1386 			ip6_ctloutput_dispatch(msg);
1387 		else
1388 #endif /* INET6 */
1389 		ip_ctloutput(msg);
1390 		/* msg invalid now */
1391 		return;
1392 	}
1393 	tp = intotcpcb(inp);
1394 
1395 	switch (sopt->sopt_dir) {
1396 	case SOPT_SET:
1397 		error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1398 				      sizeof optval);
1399 		if (error)
1400 			break;
1401 		switch (sopt->sopt_name) {
1402 		case TCP_FASTKEEP:
1403 			if (optval > 0)
1404 				tp->t_keepidle = tp->t_keepintvl;
1405 			else
1406 				tp->t_keepidle = tcp_keepidle;
1407 			tcp_timer_keep_activity(tp, 0);
1408 			break;
1409 #ifdef TCP_SIGNATURE
1410 		case TCP_SIGNATURE_ENABLE:
1411 			if (tp->t_state == TCPS_CLOSED) {
1412 				/*
1413 				 * This is the only safe state that this
1414 				 * option could be changed.  Some segments
1415 				 * could already have been sent in other
1416 				 * states.
1417 				 */
1418 				if (optval > 0)
1419 					tp->t_flags |= TF_SIGNATURE;
1420 				else
1421 					tp->t_flags &= ~TF_SIGNATURE;
1422 			} else {
1423 				error = EOPNOTSUPP;
1424 			}
1425 			break;
1426 #endif /* TCP_SIGNATURE */
1427 		case TCP_NODELAY:
1428 		case TCP_NOOPT:
1429 			switch (sopt->sopt_name) {
1430 			case TCP_NODELAY:
1431 				opt = TF_NODELAY;
1432 				break;
1433 			case TCP_NOOPT:
1434 				opt = TF_NOOPT;
1435 				break;
1436 			default:
1437 				opt = 0; /* dead code to fool gcc */
1438 				break;
1439 			}
1440 
1441 			if (optval)
1442 				tp->t_flags |= opt;
1443 			else
1444 				tp->t_flags &= ~opt;
1445 			break;
1446 
1447 		case TCP_NOPUSH:
1448 			if (tcp_disable_nopush)
1449 				break;
1450 			if (optval)
1451 				tp->t_flags |= TF_NOPUSH;
1452 			else {
1453 				tp->t_flags &= ~TF_NOPUSH;
1454 				error = tcp_output(tp);
1455 			}
1456 			break;
1457 
1458 		case TCP_MAXSEG:
1459 			/*
1460 			 * Must be between 0 and maxseg.  If the requested
1461 			 * maxseg is too small to satisfy the desired minmss,
1462 			 * pump it up (silently so sysctl modifications of
1463 			 * minmss do not create unexpected program failures).
1464 			 * Handle degenerate cases.
1465 			 */
1466 			if (optval > 0 && optval <= tp->t_maxseg) {
1467 				if (optval + 40 < tcp_minmss) {
1468 					optval = tcp_minmss - 40;
1469 					if (optval < 0)
1470 						optval = 1;
1471 				}
1472 				tp->t_maxseg = optval;
1473 			} else {
1474 				error = EINVAL;
1475 			}
1476 			break;
1477 
1478 		case TCP_KEEPINIT:
1479 			opthz = ((int64_t)optval * hz) / 1000;
1480 			if (opthz >= 1)
1481 				tp->t_keepinit = opthz;
1482 			else
1483 				error = EINVAL;
1484 			break;
1485 
1486 		case TCP_KEEPIDLE:
1487 			opthz = ((int64_t)optval * hz) / 1000;
1488 			if (opthz >= 1) {
1489 				tp->t_keepidle = opthz;
1490 				tcp_timer_keep_activity(tp, 0);
1491 			} else {
1492 				error = EINVAL;
1493 			}
1494 			break;
1495 
1496 		case TCP_KEEPINTVL:
1497 			opthz = ((int64_t)optval * hz) / 1000;
1498 			if (opthz >= 1) {
1499 				tp->t_keepintvl = opthz;
1500 				tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt;
1501 			} else {
1502 				error = EINVAL;
1503 			}
1504 			break;
1505 
1506 		case TCP_KEEPCNT:
1507 			if (optval > 0) {
1508 				tp->t_keepcnt = optval;
1509 				tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt;
1510 			} else {
1511 				error = EINVAL;
1512 			}
1513 			break;
1514 
1515 		default:
1516 			error = ENOPROTOOPT;
1517 			break;
1518 		}
1519 		break;
1520 
1521 	case SOPT_GET:
1522 		switch (sopt->sopt_name) {
1523 #ifdef TCP_SIGNATURE
1524 		case TCP_SIGNATURE_ENABLE:
1525 			optval = (tp->t_flags & TF_SIGNATURE) ? 1 : 0;
1526 			break;
1527 #endif /* TCP_SIGNATURE */
1528 		case TCP_NODELAY:
1529 			optval = tp->t_flags & TF_NODELAY;
1530 			break;
1531 		case TCP_MAXSEG:
1532 			optval = tp->t_maxseg;
1533 			break;
1534 		case TCP_NOOPT:
1535 			optval = tp->t_flags & TF_NOOPT;
1536 			break;
1537 		case TCP_NOPUSH:
1538 			optval = tp->t_flags & TF_NOPUSH;
1539 			break;
1540 		case TCP_KEEPINIT:
1541 			optval = ((int64_t)tp->t_keepinit * 1000) / hz;
1542 			break;
1543 		case TCP_KEEPIDLE:
1544 			optval = ((int64_t)tp->t_keepidle * 1000) / hz;
1545 			break;
1546 		case TCP_KEEPINTVL:
1547 			optval = ((int64_t)tp->t_keepintvl * 1000) / hz;
1548 			break;
1549 		case TCP_KEEPCNT:
1550 			optval = tp->t_keepcnt;
1551 			break;
1552 		default:
1553 			error = ENOPROTOOPT;
1554 			break;
1555 		}
1556 		if (error == 0)
1557 			soopt_from_kbuf(sopt, &optval, sizeof optval);
1558 		break;
1559 	}
1560 done:
1561 	lwkt_replymsg(&msg->lmsg, error);
1562 }
1563 
1564 /*
1565  * tcp_sendspace and tcp_recvspace are the default send and receive window
1566  * sizes, respectively.  These are obsolescent (this information should
1567  * be set by the route).
1568  *
1569  * Use a default that does not require tcp window scaling to be turned
1570  * on.  Individual programs or the administrator can increase the default.
1571  */
1572 u_long	tcp_sendspace = 57344;	/* largest multiple of PAGE_SIZE < 64k */
1573 SYSCTL_INT(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW,
1574     &tcp_sendspace , 0, "Maximum outgoing TCP datagram size");
1575 u_long	tcp_recvspace = 57344;	/* largest multiple of PAGE_SIZE < 64k */
1576 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
1577     &tcp_recvspace , 0, "Maximum incoming TCP datagram size");
1578 
1579 /*
1580  * Attach TCP protocol to socket, allocating internet protocol control
1581  * block, tcp control block, buffer space, and entering CLOSED state.
1582  */
1583 static int
1584 tcp_attach(struct socket *so, struct pru_attach_info *ai)
1585 {
1586 	struct tcpcb *tp;
1587 	struct inpcb *inp;
1588 	int error;
1589 	int cpu;
1590 #ifdef INET6
1591 	int isipv6 = INP_CHECK_SOCKAF(so, AF_INET6) != 0;
1592 #endif
1593 
1594 	if (so->so_snd.ssb_hiwat == 0 || so->so_rcv.ssb_hiwat == 0) {
1595 		lwkt_gettoken(&so->so_rcv.ssb_token);
1596 		error = soreserve(so, tcp_sendspace, tcp_recvspace,
1597 				  ai->sb_rlimit);
1598 		lwkt_reltoken(&so->so_rcv.ssb_token);
1599 		if (error)
1600 			return (error);
1601 	}
1602 	atomic_set_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE);
1603 	atomic_set_int(&so->so_snd.ssb_flags, SSB_AUTOSIZE);
1604 	cpu = mycpu->gd_cpuid;
1605 
1606 	/*
1607 	 * Set the default port for protocol processing. This will likely
1608 	 * change when we connect.
1609 	 */
1610 	error = in_pcballoc(so, &tcbinfo[cpu]);
1611 	if (error)
1612 		return (error);
1613 	inp = so->so_pcb;
1614 #ifdef INET6
1615 	if (isipv6) {
1616 		inp->inp_vflag |= INP_IPV6;
1617 		inp->in6p_hops = -1;	/* use kernel default */
1618 	}
1619 	else
1620 #endif
1621 	inp->inp_vflag |= INP_IPV4;
1622 	tp = tcp_newtcpcb(inp);
1623 	if (tp == NULL) {
1624 		/*
1625 		 * Make sure the socket is destroyed by the pcbdetach.
1626 		 */
1627 		soreference(so);
1628 #ifdef INET6
1629 		if (isipv6)
1630 			in6_pcbdetach(inp);
1631 		else
1632 #endif
1633 		in_pcbdetach(inp);
1634 		sofree(so);	/* from ref above */
1635 		return (ENOBUFS);
1636 	}
1637 	tp->t_state = TCPS_CLOSED;
1638 	/* Keep a reference for asynchronized pru_rcvd */
1639 	soreference(so);
1640 	return (0);
1641 }
1642 
1643 /*
1644  * Initiate (or continue) disconnect.
1645  * If embryonic state, just send reset (once).
1646  * If in ``let data drain'' option and linger null, just drop.
1647  * Otherwise (hard), mark socket disconnecting and drop
1648  * current input data; switch states based on user close, and
1649  * send segment to peer (with FIN).
1650  */
1651 static struct tcpcb *
1652 tcp_disconnect(struct tcpcb *tp)
1653 {
1654 	struct socket *so = tp->t_inpcb->inp_socket;
1655 
1656 	if (tp->t_state < TCPS_ESTABLISHED) {
1657 		tp = tcp_close(tp);
1658 	} else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
1659 		tp = tcp_drop(tp, 0);
1660 	} else {
1661 		lwkt_gettoken(&so->so_rcv.ssb_token);
1662 		soisdisconnecting(so);
1663 		sbflush(&so->so_rcv.sb);
1664 		tp = tcp_usrclosed(tp);
1665 		if (tp)
1666 			tcp_output(tp);
1667 		lwkt_reltoken(&so->so_rcv.ssb_token);
1668 	}
1669 	return (tp);
1670 }
1671 
1672 /*
1673  * User issued close, and wish to trail through shutdown states:
1674  * if never received SYN, just forget it.  If got a SYN from peer,
1675  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
1676  * If already got a FIN from peer, then almost done; go to LAST_ACK
1677  * state.  In all other cases, have already sent FIN to peer (e.g.
1678  * after PRU_SHUTDOWN), and just have to play tedious game waiting
1679  * for peer to send FIN or not respond to keep-alives, etc.
1680  * We can let the user exit from the close as soon as the FIN is acked.
1681  */
1682 static struct tcpcb *
1683 tcp_usrclosed(struct tcpcb *tp)
1684 {
1685 
1686 	switch (tp->t_state) {
1687 
1688 	case TCPS_CLOSED:
1689 	case TCPS_LISTEN:
1690 		tp->t_state = TCPS_CLOSED;
1691 		tp = tcp_close(tp);
1692 		break;
1693 
1694 	case TCPS_SYN_SENT:
1695 	case TCPS_SYN_RECEIVED:
1696 		tp->t_flags |= TF_NEEDFIN;
1697 		break;
1698 
1699 	case TCPS_ESTABLISHED:
1700 		tp->t_state = TCPS_FIN_WAIT_1;
1701 		break;
1702 
1703 	case TCPS_CLOSE_WAIT:
1704 		tp->t_state = TCPS_LAST_ACK;
1705 		break;
1706 	}
1707 	if (tp && tp->t_state >= TCPS_FIN_WAIT_2) {
1708 		soisdisconnected(tp->t_inpcb->inp_socket);
1709 		/* To prevent the connection hanging in FIN_WAIT_2 forever. */
1710 		if (tp->t_state == TCPS_FIN_WAIT_2) {
1711 			tcp_callout_reset(tp, tp->tt_2msl, tp->t_maxidle,
1712 			    tcp_timer_2msl);
1713 		}
1714 	}
1715 	return (tp);
1716 }
1717