xref: /dflybsd-src/sys/netinet/tcp_usrreq.c (revision e368a6e95e2cd9556a3e0fc43167d2dcf3a8253f)
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 
390 	COMMON_START(so, inp, 0);
391 
392 	if (tp->t_flags & TF_LISTEN)
393 		goto out;
394 
395 	if (inp->inp_lport == 0) {
396 		error = in_pcbbind(inp, NULL, td);
397 		if (error)
398 			goto out;
399 	}
400 
401 	tp->t_state = TCPS_LISTEN;
402 	tp->t_flags |= TF_LISTEN;
403 	tp->tt_msg = NULL; /* Catch any invalid timer usage */
404 
405 	if (ncpus > 1) {
406 		/*
407 		 * We have to set the flag because we can't have other cpus
408 		 * messing with our inp's flags.
409 		 */
410 		KASSERT(!(inp->inp_flags & INP_CONNECTED),
411 			("already on connhash"));
412 		KASSERT(!(inp->inp_flags & INP_WILDCARD),
413 			("already on wildcardhash"));
414 		KASSERT(!(inp->inp_flags & INP_WILDCARD_MP),
415 			("already on MP wildcardhash"));
416 		inp->inp_flags |= INP_WILDCARD_MP;
417 
418 		KKASSERT(so->so_port == netisr_cpuport(0));
419 		KKASSERT(&curthread->td_msgport == netisr_cpuport(0));
420 		KKASSERT(inp->inp_pcbinfo == &tcbinfo[0]);
421 
422 		netmsg_init(&nm.base, NULL, &curthread->td_msgport,
423 			    MSGF_PRIORITY, in_pcbinswildcardhash_handler);
424 		nm.nm_inp = inp;
425 		lwkt_domsg(netisr_cpuport(1), &nm.base.lmsg, 0);
426 	}
427 	in_pcbinswildcardhash(inp);
428 	COMMON_END(PRU_LISTEN);
429 }
430 
431 #ifdef INET6
432 
433 static void
434 tcp6_usr_listen(netmsg_t msg)
435 {
436 	struct socket *so = msg->listen.base.nm_so;
437 	struct thread *td = msg->listen.nm_td;
438 	int error = 0;
439 	struct inpcb *inp;
440 	struct tcpcb *tp;
441 	struct netmsg_inswildcard nm;
442 
443 	COMMON_START(so, inp, 0);
444 
445 	if (tp->t_flags & TF_LISTEN)
446 		goto out;
447 
448 	if (inp->inp_lport == 0) {
449 		if (!(inp->inp_flags & IN6P_IPV6_V6ONLY))
450 			inp->inp_vflag |= INP_IPV4;
451 		else
452 			inp->inp_vflag &= ~INP_IPV4;
453 		error = in6_pcbbind(inp, NULL, td);
454 		if (error)
455 			goto out;
456 	}
457 
458 	tp->t_state = TCPS_LISTEN;
459 	tp->t_flags |= TF_LISTEN;
460 	tp->tt_msg = NULL; /* Catch any invalid timer usage */
461 
462 	if (ncpus > 1) {
463 		/*
464 		 * We have to set the flag because we can't have other cpus
465 		 * messing with our inp's flags.
466 		 */
467 		KASSERT(!(inp->inp_flags & INP_CONNECTED),
468 			("already on connhash"));
469 		KASSERT(!(inp->inp_flags & INP_WILDCARD),
470 			("already on wildcardhash"));
471 		KASSERT(!(inp->inp_flags & INP_WILDCARD_MP),
472 			("already on MP wildcardhash"));
473 		inp->inp_flags |= INP_WILDCARD_MP;
474 
475 		KKASSERT(so->so_port == netisr_cpuport(0));
476 		KKASSERT(&curthread->td_msgport == netisr_cpuport(0));
477 		KKASSERT(inp->inp_pcbinfo == &tcbinfo[0]);
478 
479 		netmsg_init(&nm.base, NULL, &curthread->td_msgport,
480 			    MSGF_PRIORITY, in_pcbinswildcardhash_handler);
481 		nm.nm_inp = inp;
482 		lwkt_domsg(netisr_cpuport(1), &nm.base.lmsg, 0);
483 	}
484 	in_pcbinswildcardhash(inp);
485 	COMMON_END(PRU_LISTEN);
486 }
487 #endif /* INET6 */
488 
489 /*
490  * Initiate connection to peer.
491  * Create a template for use in transmissions on this connection.
492  * Enter SYN_SENT state, and mark socket as connecting.
493  * Start keep-alive timer, and seed output sequence space.
494  * Send initial segment on connection.
495  */
496 static void
497 tcp_usr_connect(netmsg_t msg)
498 {
499 	struct socket *so = msg->connect.base.nm_so;
500 	struct sockaddr *nam = msg->connect.nm_nam;
501 	struct thread *td = msg->connect.nm_td;
502 	int error = 0;
503 	struct inpcb *inp;
504 	struct tcpcb *tp;
505 	struct sockaddr_in *sinp;
506 
507 	COMMON_START(so, inp, 0);
508 
509 	/*
510 	 * Must disallow TCP ``connections'' to multicast addresses.
511 	 */
512 	sinp = (struct sockaddr_in *)nam;
513 	if (sinp->sin_family == AF_INET
514 	    && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
515 		error = EAFNOSUPPORT;
516 		goto out;
517 	}
518 
519 	if (!prison_remote_ip(td, (struct sockaddr*)sinp)) {
520 		error = EAFNOSUPPORT; /* IPv6 only jail */
521 		goto out;
522 	}
523 
524 	tcp_connect(msg);
525 	/* msg is invalid now */
526 	return;
527 out:
528 	if (msg->connect.nm_m) {
529 		m_freem(msg->connect.nm_m);
530 		msg->connect.nm_m = NULL;
531 	}
532 	if (msg->connect.nm_flags & PRUC_HELDTD)
533 		lwkt_rele(td);
534 	if (error && (msg->connect.nm_flags & PRUC_ASYNC)) {
535 		so->so_error = error;
536 		soisdisconnected(so);
537 	}
538 	lwkt_replymsg(&msg->lmsg, error);
539 }
540 
541 #ifdef INET6
542 
543 static void
544 tcp6_usr_connect(netmsg_t msg)
545 {
546 	struct socket *so = msg->connect.base.nm_so;
547 	struct sockaddr *nam = msg->connect.nm_nam;
548 	struct thread *td = msg->connect.nm_td;
549 	int error = 0;
550 	struct inpcb *inp;
551 	struct tcpcb *tp;
552 	struct sockaddr_in6 *sin6p;
553 
554 	COMMON_START(so, inp, 0);
555 
556 	/*
557 	 * Must disallow TCP ``connections'' to multicast addresses.
558 	 */
559 	sin6p = (struct sockaddr_in6 *)nam;
560 	if (sin6p->sin6_family == AF_INET6
561 	    && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
562 		error = EAFNOSUPPORT;
563 		goto out;
564 	}
565 
566 	if (!prison_remote_ip(td, nam)) {
567 		error = EAFNOSUPPORT; /* IPv4 only jail */
568 		goto out;
569 	}
570 
571 	if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
572 		struct sockaddr_in *sinp;
573 
574 		if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
575 			error = EINVAL;
576 			goto out;
577 		}
578 		sinp = kmalloc(sizeof(*sinp), M_LWKTMSG, M_INTWAIT);
579 		in6_sin6_2_sin(sinp, sin6p);
580 		inp->inp_vflag |= INP_IPV4;
581 		inp->inp_vflag &= ~INP_IPV6;
582 		msg->connect.nm_nam = (struct sockaddr *)sinp;
583 		msg->connect.nm_flags |= PRUC_NAMALLOC;
584 		tcp_connect(msg);
585 		/* msg is invalid now */
586 		return;
587 	}
588 	inp->inp_vflag &= ~INP_IPV4;
589 	inp->inp_vflag |= INP_IPV6;
590 	inp->inp_inc.inc_isipv6 = 1;
591 
592 	msg->connect.nm_flags |= PRUC_FALLBACK;
593 	tcp6_connect(msg);
594 	/* msg is invalid now */
595 	return;
596 out:
597 	if (msg->connect.nm_m) {
598 		m_freem(msg->connect.nm_m);
599 		msg->connect.nm_m = NULL;
600 	}
601 	lwkt_replymsg(&msg->lmsg, error);
602 }
603 
604 #endif /* INET6 */
605 
606 /*
607  * Initiate disconnect from peer.
608  * If connection never passed embryonic stage, just drop;
609  * else if don't need to let data drain, then can just drop anyways,
610  * else have to begin TCP shutdown process: mark socket disconnecting,
611  * drain unread data, state switch to reflect user close, and
612  * send segment (e.g. FIN) to peer.  Socket will be really disconnected
613  * when peer sends FIN and acks ours.
614  *
615  * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
616  */
617 static void
618 tcp_usr_disconnect(netmsg_t msg)
619 {
620 	struct socket *so = msg->disconnect.base.nm_so;
621 	int error = 0;
622 	struct inpcb *inp;
623 	struct tcpcb *tp;
624 
625 	COMMON_START(so, inp, 1);
626 	tp = tcp_disconnect(tp);
627 	COMMON_END(PRU_DISCONNECT);
628 }
629 
630 /*
631  * Accept a connection.  Essentially all the work is
632  * done at higher levels; just return the address
633  * of the peer, storing through addr.
634  */
635 static void
636 tcp_usr_accept(netmsg_t msg)
637 {
638 	struct socket *so = msg->accept.base.nm_so;
639 	struct sockaddr **nam = msg->accept.nm_nam;
640 	int error = 0;
641 	struct inpcb *inp;
642 	struct tcpcb *tp = NULL;
643 	TCPDEBUG0;
644 
645 	inp = so->so_pcb;
646 	if (so->so_state & SS_ISDISCONNECTED) {
647 		error = ECONNABORTED;
648 		goto out;
649 	}
650 	if (inp == 0) {
651 		error = EINVAL;
652 		goto out;
653 	}
654 
655 	tp = intotcpcb(inp);
656 	TCPDEBUG1();
657 	in_setpeeraddr(so, nam);
658 	COMMON_END(PRU_ACCEPT);
659 }
660 
661 #ifdef INET6
662 static void
663 tcp6_usr_accept(netmsg_t msg)
664 {
665 	struct socket *so = msg->accept.base.nm_so;
666 	struct sockaddr **nam = msg->accept.nm_nam;
667 	int error = 0;
668 	struct inpcb *inp;
669 	struct tcpcb *tp = NULL;
670 	TCPDEBUG0;
671 
672 	inp = so->so_pcb;
673 
674 	if (so->so_state & SS_ISDISCONNECTED) {
675 		error = ECONNABORTED;
676 		goto out;
677 	}
678 	if (inp == 0) {
679 		error = EINVAL;
680 		goto out;
681 	}
682 	tp = intotcpcb(inp);
683 	TCPDEBUG1();
684 	in6_mapped_peeraddr(so, nam);
685 	COMMON_END(PRU_ACCEPT);
686 }
687 #endif /* INET6 */
688 /*
689  * Mark the connection as being incapable of further output.
690  */
691 static void
692 tcp_usr_shutdown(netmsg_t msg)
693 {
694 	struct socket *so = msg->shutdown.base.nm_so;
695 	int error = 0;
696 	struct inpcb *inp;
697 	struct tcpcb *tp;
698 
699 	COMMON_START(so, inp, 0);
700 	socantsendmore(so);
701 	tp = tcp_usrclosed(tp);
702 	if (tp)
703 		error = tcp_output(tp);
704 	COMMON_END(PRU_SHUTDOWN);
705 }
706 
707 /*
708  * After a receive, possibly send window update to peer.
709  */
710 static void
711 tcp_usr_rcvd(netmsg_t msg)
712 {
713 	struct socket *so = msg->rcvd.base.nm_so;
714 	int error = 0, noreply = 0;
715 	struct inpcb *inp;
716 	struct tcpcb *tp;
717 
718 	COMMON_START(so, inp, 0);
719 
720 	if (msg->rcvd.nm_pru_flags & PRUR_ASYNC) {
721 		noreply = 1;
722 		so_async_rcvd_reply(so);
723 	}
724 	tcp_output(tp);
725 
726 	COMMON_END1(PRU_RCVD, noreply);
727 }
728 
729 /*
730  * Do a send by putting data in output queue and updating urgent
731  * marker if URG set.  Possibly send more data.  Unlike the other
732  * pru_*() routines, the mbuf chains are our responsibility.  We
733  * must either enqueue them or free them.  The other pru_* routines
734  * generally are caller-frees.
735  */
736 static void
737 tcp_usr_send(netmsg_t msg)
738 {
739 	struct socket *so = msg->send.base.nm_so;
740 	int flags = msg->send.nm_flags;
741 	struct mbuf *m = msg->send.nm_m;
742 	int error = 0;
743 	struct inpcb *inp;
744 	struct tcpcb *tp;
745 	TCPDEBUG0;
746 
747 	KKASSERT(msg->send.nm_control == NULL);
748 	KKASSERT(msg->send.nm_addr == NULL);
749 	KKASSERT((flags & PRUS_FREEADDR) == 0);
750 
751 	inp = so->so_pcb;
752 
753 	if (inp == NULL) {
754 		/*
755 		 * OOPS! we lost a race, the TCP session got reset after
756 		 * we checked SS_CANTSENDMORE, eg: while doing uiomove or a
757 		 * network interrupt in the non-critical section of sosend().
758 		 */
759 		m_freem(m);
760 		error = ECONNRESET;	/* XXX EPIPE? */
761 		tp = NULL;
762 		TCPDEBUG1();
763 		goto out;
764 	}
765 	tp = intotcpcb(inp);
766 	TCPDEBUG1();
767 
768 #ifdef foo
769 	/*
770 	 * This is no longer necessary, since:
771 	 * - sosendtcp() has already checked it for us
772 	 * - It does not work with asynchronized send
773 	 */
774 
775 	/*
776 	 * Don't let too much OOB data build up
777 	 */
778 	if (flags & PRUS_OOB) {
779 		if (ssb_space(&so->so_snd) < -512) {
780 			m_freem(m);
781 			error = ENOBUFS;
782 			goto out;
783 		}
784 	}
785 #endif
786 
787 	/*
788 	 * Pump the data into the socket.
789 	 */
790 	if (m)
791 		ssb_appendstream(&so->so_snd, m);
792 	if (flags & PRUS_OOB) {
793 		/*
794 		 * According to RFC961 (Assigned Protocols),
795 		 * the urgent pointer points to the last octet
796 		 * of urgent data.  We continue, however,
797 		 * to consider it to indicate the first octet
798 		 * of data past the urgent section.
799 		 * Otherwise, snd_up should be one lower.
800 		 */
801 		tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
802 		tp->t_flags |= TF_FORCE;
803 		error = tcp_output(tp);
804 		tp->t_flags &= ~TF_FORCE;
805 	} else {
806 		if (flags & PRUS_EOF) {
807 			/*
808 			 * Close the send side of the connection after
809 			 * the data is sent.
810 			 */
811 			socantsendmore(so);
812 			tp = tcp_usrclosed(tp);
813 		}
814 		if (tp != NULL && !tcp_output_pending(tp)) {
815 			if (flags & PRUS_MORETOCOME)
816 				tp->t_flags |= TF_MORETOCOME;
817 			error = tcp_output_fair(tp);
818 			if (flags & PRUS_MORETOCOME)
819 				tp->t_flags &= ~TF_MORETOCOME;
820 		}
821 	}
822 	COMMON_END1((flags & PRUS_OOB) ? PRU_SENDOOB :
823 		   ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND),
824 		   (flags & PRUS_NOREPLY));
825 }
826 
827 /*
828  * NOTE: (so) is referenced from soabort*() and netmsg_pru_abort()
829  *	 will sofree() it when we return.
830  */
831 static void
832 tcp_usr_abort(netmsg_t msg)
833 {
834 	struct socket *so = msg->abort.base.nm_so;
835 	int error = 0;
836 	struct inpcb *inp;
837 	struct tcpcb *tp;
838 
839 	COMMON_START(so, inp, 1);
840 	tp = tcp_drop(tp, ECONNABORTED);
841 	COMMON_END(PRU_ABORT);
842 }
843 
844 /*
845  * Receive out-of-band data.
846  */
847 static void
848 tcp_usr_rcvoob(netmsg_t msg)
849 {
850 	struct socket *so = msg->rcvoob.base.nm_so;
851 	struct mbuf *m = msg->rcvoob.nm_m;
852 	int flags = msg->rcvoob.nm_flags;
853 	int error = 0;
854 	struct inpcb *inp;
855 	struct tcpcb *tp;
856 
857 	COMMON_START(so, inp, 0);
858 	if ((so->so_oobmark == 0 &&
859 	     (so->so_state & SS_RCVATMARK) == 0) ||
860 	    so->so_options & SO_OOBINLINE ||
861 	    tp->t_oobflags & TCPOOB_HADDATA) {
862 		error = EINVAL;
863 		goto out;
864 	}
865 	if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
866 		error = EWOULDBLOCK;
867 		goto out;
868 	}
869 	m->m_len = 1;
870 	*mtod(m, caddr_t) = tp->t_iobc;
871 	if ((flags & MSG_PEEK) == 0)
872 		tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
873 	COMMON_END(PRU_RCVOOB);
874 }
875 
876 static void
877 tcp_usr_savefaddr(struct socket *so, const struct sockaddr *faddr)
878 {
879 	in_savefaddr(so, faddr);
880 }
881 
882 #ifdef INET6
883 static void
884 tcp6_usr_savefaddr(struct socket *so, const struct sockaddr *faddr)
885 {
886 	in6_mapped_savefaddr(so, faddr);
887 }
888 #endif
889 
890 static int
891 tcp_usr_preconnect(struct socket *so, const struct sockaddr *nam,
892     struct thread *td __unused)
893 {
894 	const struct sockaddr_in *sinp;
895 
896 	sinp = (const struct sockaddr_in *)nam;
897 	if (sinp->sin_family == AF_INET &&
898 	    IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
899 		return EAFNOSUPPORT;
900 
901 	soisconnecting(so);
902 	return 0;
903 }
904 
905 /* xxx - should be const */
906 struct pr_usrreqs tcp_usrreqs = {
907 	.pru_abort = tcp_usr_abort,
908 	.pru_accept = tcp_usr_accept,
909 	.pru_attach = tcp_usr_attach,
910 	.pru_bind = tcp_usr_bind,
911 	.pru_connect = tcp_usr_connect,
912 	.pru_connect2 = pr_generic_notsupp,
913 	.pru_control = in_control_dispatch,
914 	.pru_detach = tcp_usr_detach,
915 	.pru_disconnect = tcp_usr_disconnect,
916 	.pru_listen = tcp_usr_listen,
917 	.pru_peeraddr = in_setpeeraddr_dispatch,
918 	.pru_rcvd = tcp_usr_rcvd,
919 	.pru_rcvoob = tcp_usr_rcvoob,
920 	.pru_send = tcp_usr_send,
921 	.pru_sense = pru_sense_null,
922 	.pru_shutdown = tcp_usr_shutdown,
923 	.pru_sockaddr = in_setsockaddr_dispatch,
924 	.pru_sosend = sosendtcp,
925 	.pru_soreceive = sorecvtcp,
926 	.pru_savefaddr = tcp_usr_savefaddr,
927 	.pru_preconnect = tcp_usr_preconnect
928 };
929 
930 #ifdef INET6
931 struct pr_usrreqs tcp6_usrreqs = {
932 	.pru_abort = tcp_usr_abort,
933 	.pru_accept = tcp6_usr_accept,
934 	.pru_attach = tcp_usr_attach,
935 	.pru_bind = tcp6_usr_bind,
936 	.pru_connect = tcp6_usr_connect,
937 	.pru_connect2 = pr_generic_notsupp,
938 	.pru_control = in6_control_dispatch,
939 	.pru_detach = tcp_usr_detach,
940 	.pru_disconnect = tcp_usr_disconnect,
941 	.pru_listen = tcp6_usr_listen,
942 	.pru_peeraddr = in6_mapped_peeraddr_dispatch,
943 	.pru_rcvd = tcp_usr_rcvd,
944 	.pru_rcvoob = tcp_usr_rcvoob,
945 	.pru_send = tcp_usr_send,
946 	.pru_sense = pru_sense_null,
947 	.pru_shutdown = tcp_usr_shutdown,
948 	.pru_sockaddr = in6_mapped_sockaddr_dispatch,
949 	.pru_sosend = sosendtcp,
950 	.pru_soreceive = sorecvtcp,
951 	.pru_savefaddr = tcp6_usr_savefaddr
952 };
953 #endif /* INET6 */
954 
955 static int
956 tcp_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf *m,
957 		  struct sockaddr_in *sin, struct sockaddr_in *if_sin)
958 {
959 	struct inpcb *inp = tp->t_inpcb, *oinp;
960 	struct socket *so = inp->inp_socket;
961 	struct route *ro = &inp->inp_route;
962 
963 	oinp = in_pcblookup_hash(&tcbinfo[mycpu->gd_cpuid],
964 				 sin->sin_addr, sin->sin_port,
965 				 (inp->inp_laddr.s_addr != INADDR_ANY ?
966 				  inp->inp_laddr : if_sin->sin_addr),
967 				inp->inp_lport, 0, NULL);
968 	if (oinp != NULL) {
969 		m_freem(m);
970 		return (EADDRINUSE);
971 	}
972 	if (inp->inp_laddr.s_addr == INADDR_ANY)
973 		inp->inp_laddr = if_sin->sin_addr;
974 	inp->inp_faddr = sin->sin_addr;
975 	inp->inp_fport = sin->sin_port;
976 	inp->inp_cpcbinfo = &tcbinfo[mycpu->gd_cpuid];
977 	in_pcbinsconnhash(inp);
978 
979 	/*
980 	 * We are now on the inpcb's owner CPU, if the cached route was
981 	 * freed because the rtentry's owner CPU is not the current CPU
982 	 * (e.g. in tcp_connect()), then we try to reallocate it here with
983 	 * the hope that a rtentry may be cloned from a RTF_PRCLONING
984 	 * rtentry.
985 	 */
986 	if (!(inp->inp_socket->so_options & SO_DONTROUTE) && /*XXX*/
987 	    ro->ro_rt == NULL) {
988 		bzero(&ro->ro_dst, sizeof(struct sockaddr_in));
989 		ro->ro_dst.sa_family = AF_INET;
990 		ro->ro_dst.sa_len = sizeof(struct sockaddr_in);
991 		((struct sockaddr_in *)&ro->ro_dst)->sin_addr =
992 			sin->sin_addr;
993 		rtalloc(ro);
994 	}
995 
996 	/*
997 	 * Now that no more errors can occur, change the protocol processing
998 	 * port to the current thread (which is the correct thread).
999 	 *
1000 	 * Create TCP timer message now; we are on the tcpcb's owner
1001 	 * CPU/thread.
1002 	 */
1003 	tcp_create_timermsg(tp, &curthread->td_msgport);
1004 
1005 	/*
1006 	 * Compute window scaling to request.  Use a larger scaling then
1007 	 * needed for the initial receive buffer in case the receive buffer
1008 	 * gets expanded.
1009 	 */
1010 	if (tp->request_r_scale < TCP_MIN_WINSHIFT)
1011 		tp->request_r_scale = TCP_MIN_WINSHIFT;
1012 	while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1013 	       (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat
1014 	) {
1015 		tp->request_r_scale++;
1016 	}
1017 
1018 	soisconnecting(so);
1019 	tcpstat.tcps_connattempt++;
1020 	tp->t_state = TCPS_SYN_SENT;
1021 	tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
1022 	tp->iss = tcp_new_isn(tp);
1023 	tcp_sendseqinit(tp);
1024 	if (m) {
1025 		ssb_appendstream(&so->so_snd, m);
1026 		m = NULL;
1027 		if (flags & PRUS_OOB)
1028 			tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
1029 	}
1030 
1031 	/*
1032 	 * Close the send side of the connection after
1033 	 * the data is sent if flagged.
1034 	 */
1035 	if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
1036 		socantsendmore(so);
1037 		tp = tcp_usrclosed(tp);
1038 	}
1039 	return (tcp_output(tp));
1040 }
1041 
1042 /*
1043  * Common subroutine to open a TCP connection to remote host specified
1044  * by struct sockaddr_in in mbuf *nam.  Call in_pcbbind to assign a local
1045  * port number if needed.  Call in_pcbladdr to do the routing and to choose
1046  * a local host address (interface).
1047  * Initialize connection parameters and enter SYN-SENT state.
1048  */
1049 static void
1050 tcp_connect(netmsg_t msg)
1051 {
1052 	struct socket *so = msg->connect.base.nm_so;
1053 	struct sockaddr *nam = msg->connect.nm_nam;
1054 	struct thread *td = msg->connect.nm_td;
1055 	struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1056 	struct sockaddr_in *if_sin;
1057 	struct inpcb *inp;
1058 	struct tcpcb *tp;
1059 	int error, calc_laddr = 1;
1060 	lwkt_port_t port;
1061 
1062 	COMMON_START(so, inp, 0);
1063 
1064 	/*
1065 	 * Reconnect our pcb if we have to
1066 	 */
1067 	if (msg->connect.nm_flags & PRUC_RECONNECT) {
1068 		msg->connect.nm_flags &= ~PRUC_RECONNECT;
1069 		in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1070 	}
1071 
1072 	/*
1073 	 * Bind if we have to
1074 	 */
1075 	if (inp->inp_lport == 0) {
1076 		if (tcp_lport_extension) {
1077 			KKASSERT(inp->inp_laddr.s_addr == INADDR_ANY);
1078 
1079 			error = in_pcbladdr(inp, nam, &if_sin, td);
1080 			if (error)
1081 				goto out;
1082 			inp->inp_laddr.s_addr = if_sin->sin_addr.s_addr;
1083 
1084 			error = in_pcbconn_bind(inp, nam, td);
1085 			if (error)
1086 				goto out;
1087 
1088 			calc_laddr = 0;
1089 		} else {
1090 			error = in_pcbbind(inp, NULL, td);
1091 			if (error)
1092 				goto out;
1093 		}
1094 	}
1095 
1096 	if (calc_laddr) {
1097 		/*
1098 		 * Calculate the correct protocol processing thread.  The
1099 		 * connect operation must run there.  Set the forwarding
1100 		 * port before we forward the message or it will get bounced
1101 		 * right back to us.
1102 		 */
1103 		error = in_pcbladdr(inp, nam, &if_sin, td);
1104 		if (error)
1105 			goto out;
1106 	}
1107 	KKASSERT(inp->inp_socket == so);
1108 
1109 	port = tcp_addrport(sin->sin_addr.s_addr, sin->sin_port,
1110 			    (inp->inp_laddr.s_addr ?
1111 			     inp->inp_laddr.s_addr : if_sin->sin_addr.s_addr),
1112 			    inp->inp_lport);
1113 
1114 	if (port != &curthread->td_msgport) {
1115 		struct route *ro = &inp->inp_route;
1116 
1117 		/*
1118 		 * in_pcbladdr() may have allocated a route entry for us
1119 		 * on the current CPU, but we need a route entry on the
1120 		 * inpcb's owner CPU, so free it here.
1121 		 */
1122 		if (ro->ro_rt != NULL)
1123 			RTFREE(ro->ro_rt);
1124 		bzero(ro, sizeof(*ro));
1125 
1126 		/*
1127 		 * We are moving the protocol processing port the socket
1128 		 * is on, we have to unlink here and re-link on the
1129 		 * target cpu.
1130 		 */
1131 		in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1132 		sosetport(so, port);
1133 		msg->connect.nm_flags |= PRUC_RECONNECT;
1134 		msg->connect.base.nm_dispatch = tcp_connect;
1135 
1136 		lwkt_forwardmsg(port, &msg->connect.base.lmsg);
1137 		/* msg invalid now */
1138 		return;
1139 	} else if (msg->connect.nm_flags & PRUC_HELDTD) {
1140 		/*
1141 		 * The original thread is no longer needed; release it.
1142 		 */
1143 		lwkt_rele(td);
1144 		msg->connect.nm_flags &= ~PRUC_HELDTD;
1145 	}
1146 	error = tcp_connect_oncpu(tp, msg->connect.nm_sndflags,
1147 				  msg->connect.nm_m, sin, if_sin);
1148 	msg->connect.nm_m = NULL;
1149 out:
1150 	if (msg->connect.nm_m) {
1151 		m_freem(msg->connect.nm_m);
1152 		msg->connect.nm_m = NULL;
1153 	}
1154 	if (msg->connect.nm_flags & PRUC_NAMALLOC) {
1155 		kfree(msg->connect.nm_nam, M_LWKTMSG);
1156 		msg->connect.nm_nam = NULL;
1157 	}
1158 	if (msg->connect.nm_flags & PRUC_HELDTD)
1159 		lwkt_rele(td);
1160 	if (error && (msg->connect.nm_flags & PRUC_ASYNC)) {
1161 		so->so_error = error;
1162 		soisdisconnected(so);
1163 	}
1164 	lwkt_replymsg(&msg->connect.base.lmsg, error);
1165 	/* msg invalid now */
1166 }
1167 
1168 #ifdef INET6
1169 
1170 static void
1171 tcp6_connect(netmsg_t msg)
1172 {
1173 	struct tcpcb *tp;
1174 	struct socket *so = msg->connect.base.nm_so;
1175 	struct sockaddr *nam = msg->connect.nm_nam;
1176 	struct thread *td = msg->connect.nm_td;
1177 	struct inpcb *inp;
1178 	struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam;
1179 	struct in6_addr *addr6;
1180 	lwkt_port_t port;
1181 	int error;
1182 
1183 	COMMON_START(so, inp, 0);
1184 
1185 	/*
1186 	 * Reconnect our pcb if we have to
1187 	 */
1188 	if (msg->connect.nm_flags & PRUC_RECONNECT) {
1189 		msg->connect.nm_flags &= ~PRUC_RECONNECT;
1190 		in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1191 	}
1192 
1193 	/*
1194 	 * Bind if we have to
1195 	 */
1196 	if (inp->inp_lport == 0) {
1197 		error = in6_pcbbind(inp, NULL, td);
1198 		if (error)
1199 			goto out;
1200 	}
1201 
1202 	/*
1203 	 * Cannot simply call in_pcbconnect, because there might be an
1204 	 * earlier incarnation of this same connection still in
1205 	 * TIME_WAIT state, creating an ADDRINUSE error.
1206 	 */
1207 	error = in6_pcbladdr(inp, nam, &addr6, td);
1208 	if (error)
1209 		goto out;
1210 
1211 	port = tcp6_addrport();	/* XXX hack for now, always cpu0 */
1212 
1213 	if (port != &curthread->td_msgport) {
1214 		struct route *ro = &inp->inp_route;
1215 
1216 		/*
1217 		 * in_pcbladdr() may have allocated a route entry for us
1218 		 * on the current CPU, but we need a route entry on the
1219 		 * inpcb's owner CPU, so free it here.
1220 		 */
1221 		if (ro->ro_rt != NULL)
1222 			RTFREE(ro->ro_rt);
1223 		bzero(ro, sizeof(*ro));
1224 
1225 		in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1226 		sosetport(so, port);
1227 		msg->connect.nm_flags |= PRUC_RECONNECT;
1228 		msg->connect.base.nm_dispatch = tcp6_connect;
1229 
1230 		lwkt_forwardmsg(port, &msg->connect.base.lmsg);
1231 		/* msg invalid now */
1232 		return;
1233 	}
1234 	error = tcp6_connect_oncpu(tp, msg->connect.nm_sndflags,
1235 				   &msg->connect.nm_m, sin6, addr6);
1236 	/* nm_m may still be intact */
1237 out:
1238 	if (error && (msg->connect.nm_flags & PRUC_FALLBACK)) {
1239 		tcp_connect(msg);
1240 		/* msg invalid now */
1241 	} else {
1242 		if (msg->connect.nm_m) {
1243 			m_freem(msg->connect.nm_m);
1244 			msg->connect.nm_m = NULL;
1245 		}
1246 		if (msg->connect.nm_flags & PRUC_NAMALLOC) {
1247 			kfree(msg->connect.nm_nam, M_LWKTMSG);
1248 			msg->connect.nm_nam = NULL;
1249 		}
1250 		lwkt_replymsg(&msg->connect.base.lmsg, error);
1251 		/* msg invalid now */
1252 	}
1253 }
1254 
1255 static int
1256 tcp6_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf **mp,
1257 		   struct sockaddr_in6 *sin6, struct in6_addr *addr6)
1258 {
1259 	struct mbuf *m = *mp;
1260 	struct inpcb *inp = tp->t_inpcb;
1261 	struct socket *so = inp->inp_socket;
1262 	struct inpcb *oinp;
1263 
1264 	/*
1265 	 * Cannot simply call in_pcbconnect, because there might be an
1266 	 * earlier incarnation of this same connection still in
1267 	 * TIME_WAIT state, creating an ADDRINUSE error.
1268 	 */
1269 	oinp = in6_pcblookup_hash(inp->inp_cpcbinfo,
1270 				  &sin6->sin6_addr, sin6->sin6_port,
1271 				  (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) ?
1272 				      addr6 : &inp->in6p_laddr),
1273 				  inp->inp_lport,  0, NULL);
1274 	if (oinp)
1275 		return (EADDRINUSE);
1276 
1277 	if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
1278 		inp->in6p_laddr = *addr6;
1279 	inp->in6p_faddr = sin6->sin6_addr;
1280 	inp->inp_fport = sin6->sin6_port;
1281 	if ((sin6->sin6_flowinfo & IPV6_FLOWINFO_MASK) != 0)
1282 		inp->in6p_flowinfo = sin6->sin6_flowinfo;
1283 	in_pcbinsconnhash(inp);
1284 
1285 	/*
1286 	 * Now that no more errors can occur, change the protocol processing
1287 	 * port to the current thread (which is the correct thread).
1288 	 *
1289 	 * Create TCP timer message now; we are on the tcpcb's owner
1290 	 * CPU/thread.
1291 	 */
1292 	tcp_create_timermsg(tp, &curthread->td_msgport);
1293 
1294 	/* Compute window scaling to request.  */
1295 	if (tp->request_r_scale < TCP_MIN_WINSHIFT)
1296 		tp->request_r_scale = TCP_MIN_WINSHIFT;
1297 	while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1298 	    (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat) {
1299 		tp->request_r_scale++;
1300 	}
1301 
1302 	soisconnecting(so);
1303 	tcpstat.tcps_connattempt++;
1304 	tp->t_state = TCPS_SYN_SENT;
1305 	tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
1306 	tp->iss = tcp_new_isn(tp);
1307 	tcp_sendseqinit(tp);
1308 	if (m) {
1309 		ssb_appendstream(&so->so_snd, m);
1310 		*mp = NULL;
1311 		if (flags & PRUS_OOB)
1312 			tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
1313 	}
1314 
1315 	/*
1316 	 * Close the send side of the connection after
1317 	 * the data is sent if flagged.
1318 	 */
1319 	if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
1320 		socantsendmore(so);
1321 		tp = tcp_usrclosed(tp);
1322 	}
1323 	return (tcp_output(tp));
1324 }
1325 
1326 #endif /* INET6 */
1327 
1328 /*
1329  * The new sockopt interface makes it possible for us to block in the
1330  * copyin/out step (if we take a page fault).  Taking a page fault while
1331  * in a critical section is probably a Bad Thing.  (Since sockets and pcbs
1332  * both now use TSM, there probably isn't any need for this function to
1333  * run in a critical section any more.  This needs more examination.)
1334  */
1335 void
1336 tcp_ctloutput(netmsg_t msg)
1337 {
1338 	struct socket *so = msg->base.nm_so;
1339 	struct sockopt *sopt = msg->ctloutput.nm_sopt;
1340 	int	error, opt, optval, opthz;
1341 	struct	inpcb *inp;
1342 	struct	tcpcb *tp;
1343 
1344 	error = 0;
1345 	inp = so->so_pcb;
1346 	if (inp == NULL) {
1347 		error = ECONNRESET;
1348 		goto done;
1349 	}
1350 
1351 	if (sopt->sopt_level != IPPROTO_TCP) {
1352 #ifdef INET6
1353 		if (INP_CHECK_SOCKAF(so, AF_INET6))
1354 			ip6_ctloutput_dispatch(msg);
1355 		else
1356 #endif /* INET6 */
1357 		ip_ctloutput(msg);
1358 		/* msg invalid now */
1359 		return;
1360 	}
1361 	tp = intotcpcb(inp);
1362 
1363 	switch (sopt->sopt_dir) {
1364 	case SOPT_SET:
1365 		error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1366 				      sizeof optval);
1367 		if (error)
1368 			break;
1369 		switch (sopt->sopt_name) {
1370 		case TCP_FASTKEEP:
1371 			if (optval > 0)
1372 				tp->t_keepidle = tp->t_keepintvl;
1373 			else
1374 				tp->t_keepidle = tcp_keepidle;
1375 			tcp_timer_keep_activity(tp, 0);
1376 			break;
1377 #ifdef TCP_SIGNATURE
1378 		case TCP_SIGNATURE_ENABLE:
1379 			if (tp->t_state == TCPS_CLOSED) {
1380 				/*
1381 				 * This is the only safe state that this
1382 				 * option could be changed.  Some segments
1383 				 * could already have been sent in other
1384 				 * states.
1385 				 */
1386 				if (optval > 0)
1387 					tp->t_flags |= TF_SIGNATURE;
1388 				else
1389 					tp->t_flags &= ~TF_SIGNATURE;
1390 			} else {
1391 				error = EOPNOTSUPP;
1392 			}
1393 			break;
1394 #endif /* TCP_SIGNATURE */
1395 		case TCP_NODELAY:
1396 		case TCP_NOOPT:
1397 			switch (sopt->sopt_name) {
1398 			case TCP_NODELAY:
1399 				opt = TF_NODELAY;
1400 				break;
1401 			case TCP_NOOPT:
1402 				opt = TF_NOOPT;
1403 				break;
1404 			default:
1405 				opt = 0; /* dead code to fool gcc */
1406 				break;
1407 			}
1408 
1409 			if (optval)
1410 				tp->t_flags |= opt;
1411 			else
1412 				tp->t_flags &= ~opt;
1413 			break;
1414 
1415 		case TCP_NOPUSH:
1416 			if (tcp_disable_nopush)
1417 				break;
1418 			if (optval)
1419 				tp->t_flags |= TF_NOPUSH;
1420 			else {
1421 				tp->t_flags &= ~TF_NOPUSH;
1422 				error = tcp_output(tp);
1423 			}
1424 			break;
1425 
1426 		case TCP_MAXSEG:
1427 			/*
1428 			 * Must be between 0 and maxseg.  If the requested
1429 			 * maxseg is too small to satisfy the desired minmss,
1430 			 * pump it up (silently so sysctl modifications of
1431 			 * minmss do not create unexpected program failures).
1432 			 * Handle degenerate cases.
1433 			 */
1434 			if (optval > 0 && optval <= tp->t_maxseg) {
1435 				if (optval + 40 < tcp_minmss) {
1436 					optval = tcp_minmss - 40;
1437 					if (optval < 0)
1438 						optval = 1;
1439 				}
1440 				tp->t_maxseg = optval;
1441 			} else {
1442 				error = EINVAL;
1443 			}
1444 			break;
1445 
1446 		case TCP_KEEPINIT:
1447 			opthz = ((int64_t)optval * hz) / 1000;
1448 			if (opthz >= 1)
1449 				tp->t_keepinit = opthz;
1450 			else
1451 				error = EINVAL;
1452 			break;
1453 
1454 		case TCP_KEEPIDLE:
1455 			opthz = ((int64_t)optval * hz) / 1000;
1456 			if (opthz >= 1) {
1457 				tp->t_keepidle = opthz;
1458 				tcp_timer_keep_activity(tp, 0);
1459 			} else {
1460 				error = EINVAL;
1461 			}
1462 			break;
1463 
1464 		case TCP_KEEPINTVL:
1465 			opthz = ((int64_t)optval * hz) / 1000;
1466 			if (opthz >= 1) {
1467 				tp->t_keepintvl = opthz;
1468 				tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt;
1469 			} else {
1470 				error = EINVAL;
1471 			}
1472 			break;
1473 
1474 		case TCP_KEEPCNT:
1475 			if (optval > 0) {
1476 				tp->t_keepcnt = optval;
1477 				tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt;
1478 			} else {
1479 				error = EINVAL;
1480 			}
1481 			break;
1482 
1483 		default:
1484 			error = ENOPROTOOPT;
1485 			break;
1486 		}
1487 		break;
1488 
1489 	case SOPT_GET:
1490 		switch (sopt->sopt_name) {
1491 #ifdef TCP_SIGNATURE
1492 		case TCP_SIGNATURE_ENABLE:
1493 			optval = (tp->t_flags & TF_SIGNATURE) ? 1 : 0;
1494 			break;
1495 #endif /* TCP_SIGNATURE */
1496 		case TCP_NODELAY:
1497 			optval = tp->t_flags & TF_NODELAY;
1498 			break;
1499 		case TCP_MAXSEG:
1500 			optval = tp->t_maxseg;
1501 			break;
1502 		case TCP_NOOPT:
1503 			optval = tp->t_flags & TF_NOOPT;
1504 			break;
1505 		case TCP_NOPUSH:
1506 			optval = tp->t_flags & TF_NOPUSH;
1507 			break;
1508 		case TCP_KEEPINIT:
1509 			optval = ((int64_t)tp->t_keepinit * 1000) / hz;
1510 			break;
1511 		case TCP_KEEPIDLE:
1512 			optval = ((int64_t)tp->t_keepidle * 1000) / hz;
1513 			break;
1514 		case TCP_KEEPINTVL:
1515 			optval = ((int64_t)tp->t_keepintvl * 1000) / hz;
1516 			break;
1517 		case TCP_KEEPCNT:
1518 			optval = tp->t_keepcnt;
1519 			break;
1520 		default:
1521 			error = ENOPROTOOPT;
1522 			break;
1523 		}
1524 		if (error == 0)
1525 			soopt_from_kbuf(sopt, &optval, sizeof optval);
1526 		break;
1527 	}
1528 done:
1529 	lwkt_replymsg(&msg->lmsg, error);
1530 }
1531 
1532 /*
1533  * tcp_sendspace and tcp_recvspace are the default send and receive window
1534  * sizes, respectively.  These are obsolescent (this information should
1535  * be set by the route).
1536  *
1537  * Use a default that does not require tcp window scaling to be turned
1538  * on.  Individual programs or the administrator can increase the default.
1539  */
1540 u_long	tcp_sendspace = 57344;	/* largest multiple of PAGE_SIZE < 64k */
1541 SYSCTL_INT(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW,
1542     &tcp_sendspace , 0, "Maximum outgoing TCP datagram size");
1543 u_long	tcp_recvspace = 57344;	/* largest multiple of PAGE_SIZE < 64k */
1544 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
1545     &tcp_recvspace , 0, "Maximum incoming TCP datagram size");
1546 
1547 /*
1548  * Attach TCP protocol to socket, allocating internet protocol control
1549  * block, tcp control block, bufer space, and entering LISTEN state
1550  * if to accept connections.
1551  */
1552 static int
1553 tcp_attach(struct socket *so, struct pru_attach_info *ai)
1554 {
1555 	struct tcpcb *tp;
1556 	struct inpcb *inp;
1557 	int error;
1558 	int cpu;
1559 #ifdef INET6
1560 	int isipv6 = INP_CHECK_SOCKAF(so, AF_INET6) != 0;
1561 #endif
1562 
1563 	if (so->so_snd.ssb_hiwat == 0 || so->so_rcv.ssb_hiwat == 0) {
1564 		lwkt_gettoken(&so->so_rcv.ssb_token);
1565 		error = soreserve(so, tcp_sendspace, tcp_recvspace,
1566 				  ai->sb_rlimit);
1567 		lwkt_reltoken(&so->so_rcv.ssb_token);
1568 		if (error)
1569 			return (error);
1570 	}
1571 	atomic_set_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE);
1572 	atomic_set_int(&so->so_snd.ssb_flags, SSB_AUTOSIZE);
1573 	cpu = mycpu->gd_cpuid;
1574 
1575 	/*
1576 	 * Set the default port for protocol processing. This will likely
1577 	 * change when we connect.
1578 	 */
1579 	error = in_pcballoc(so, &tcbinfo[cpu]);
1580 	if (error)
1581 		return (error);
1582 	inp = so->so_pcb;
1583 #ifdef INET6
1584 	if (isipv6) {
1585 		inp->inp_vflag |= INP_IPV6;
1586 		inp->in6p_hops = -1;	/* use kernel default */
1587 	}
1588 	else
1589 #endif
1590 	inp->inp_vflag |= INP_IPV4;
1591 	tp = tcp_newtcpcb(inp);
1592 	if (tp == NULL) {
1593 		/*
1594 		 * Make sure the socket is destroyed by the pcbdetach.
1595 		 */
1596 		soreference(so);
1597 #ifdef INET6
1598 		if (isipv6)
1599 			in6_pcbdetach(inp);
1600 		else
1601 #endif
1602 		in_pcbdetach(inp);
1603 		sofree(so);	/* from ref above */
1604 		return (ENOBUFS);
1605 	}
1606 	tp->t_state = TCPS_CLOSED;
1607 	/* Keep a reference for asynchronized pru_rcvd */
1608 	soreference(so);
1609 	return (0);
1610 }
1611 
1612 /*
1613  * Initiate (or continue) disconnect.
1614  * If embryonic state, just send reset (once).
1615  * If in ``let data drain'' option and linger null, just drop.
1616  * Otherwise (hard), mark socket disconnecting and drop
1617  * current input data; switch states based on user close, and
1618  * send segment to peer (with FIN).
1619  */
1620 static struct tcpcb *
1621 tcp_disconnect(struct tcpcb *tp)
1622 {
1623 	struct socket *so = tp->t_inpcb->inp_socket;
1624 
1625 	if (tp->t_state < TCPS_ESTABLISHED) {
1626 		tp = tcp_close(tp);
1627 	} else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
1628 		tp = tcp_drop(tp, 0);
1629 	} else {
1630 		lwkt_gettoken(&so->so_rcv.ssb_token);
1631 		soisdisconnecting(so);
1632 		sbflush(&so->so_rcv.sb);
1633 		tp = tcp_usrclosed(tp);
1634 		if (tp)
1635 			tcp_output(tp);
1636 		lwkt_reltoken(&so->so_rcv.ssb_token);
1637 	}
1638 	return (tp);
1639 }
1640 
1641 /*
1642  * User issued close, and wish to trail through shutdown states:
1643  * if never received SYN, just forget it.  If got a SYN from peer,
1644  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
1645  * If already got a FIN from peer, then almost done; go to LAST_ACK
1646  * state.  In all other cases, have already sent FIN to peer (e.g.
1647  * after PRU_SHUTDOWN), and just have to play tedious game waiting
1648  * for peer to send FIN or not respond to keep-alives, etc.
1649  * We can let the user exit from the close as soon as the FIN is acked.
1650  */
1651 static struct tcpcb *
1652 tcp_usrclosed(struct tcpcb *tp)
1653 {
1654 
1655 	switch (tp->t_state) {
1656 
1657 	case TCPS_CLOSED:
1658 	case TCPS_LISTEN:
1659 		tp->t_state = TCPS_CLOSED;
1660 		tp = tcp_close(tp);
1661 		break;
1662 
1663 	case TCPS_SYN_SENT:
1664 	case TCPS_SYN_RECEIVED:
1665 		tp->t_flags |= TF_NEEDFIN;
1666 		break;
1667 
1668 	case TCPS_ESTABLISHED:
1669 		tp->t_state = TCPS_FIN_WAIT_1;
1670 		break;
1671 
1672 	case TCPS_CLOSE_WAIT:
1673 		tp->t_state = TCPS_LAST_ACK;
1674 		break;
1675 	}
1676 	if (tp && tp->t_state >= TCPS_FIN_WAIT_2) {
1677 		soisdisconnected(tp->t_inpcb->inp_socket);
1678 		/* To prevent the connection hanging in FIN_WAIT_2 forever. */
1679 		if (tp->t_state == TCPS_FIN_WAIT_2) {
1680 			tcp_callout_reset(tp, tp->tt_2msl, tp->t_maxidle,
1681 			    tcp_timer_2msl);
1682 		}
1683 	}
1684 	return (tp);
1685 }
1686