tcp: Allow listen(2) to be called on the same socket for any number of times
[dragonfly.git] / sys / netinet / tcp_usrreq.c
1 /*
2  * Copyright (c) 2003, 2004 Jeffrey M. Hsu.  All rights reserved.
3  * Copyright (c) 2003, 2004 The DragonFly Project.  All rights reserved.
4  *
5  * This code is derived from software contributed to The DragonFly Project
6  * by Jeffrey M. Hsu.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. Neither the name of The DragonFly Project nor the names of its
17  *    contributors may be used to endorse or promote products derived
18  *    from this software without specific, prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
23  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
24  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
25  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
26  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
27  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
28  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
29  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
30  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  */
33
34 /*
35  * Copyright (c) 1982, 1986, 1988, 1993
36  *      The Regents of the University of California.  All rights reserved.
37  *
38  * Redistribution and use in source and binary forms, with or without
39  * modification, are permitted provided that the following conditions
40  * are met:
41  * 1. Redistributions of source code must retain the above copyright
42  *    notice, this list of conditions and the following disclaimer.
43  * 2. Redistributions in binary form must reproduce the above copyright
44  *    notice, this list of conditions and the following disclaimer in the
45  *    documentation and/or other materials provided with the distribution.
46  * 3. All advertising materials mentioning features or use of this software
47  *    must display the following acknowledgement:
48  *      This product includes software developed by the University of
49  *      California, Berkeley and its contributors.
50  * 4. Neither the name of the University nor the names of its contributors
51  *    may be used to endorse or promote products derived from this software
52  *    without specific prior written permission.
53  *
54  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
55  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
56  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
57  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
58  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
59  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
60  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
61  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
62  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
63  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
64  * SUCH DAMAGE.
65  *
66  *      From: @(#)tcp_usrreq.c  8.2 (Berkeley) 1/3/94
67  * $FreeBSD: src/sys/netinet/tcp_usrreq.c,v 1.51.2.17 2002/10/11 11:46:44 ume Exp $
68  * $DragonFly: src/sys/netinet/tcp_usrreq.c,v 1.51 2008/09/29 20:52:23 dillon Exp $
69  */
70
71 #include "opt_ipsec.h"
72 #include "opt_inet.h"
73 #include "opt_inet6.h"
74 #include "opt_tcpdebug.h"
75
76 #include <sys/param.h>
77 #include <sys/systm.h>
78 #include <sys/kernel.h>
79 #include <sys/malloc.h>
80 #include <sys/sysctl.h>
81 #include <sys/globaldata.h>
82 #include <sys/thread.h>
83
84 #include <sys/mbuf.h>
85 #ifdef INET6
86 #include <sys/domain.h>
87 #endif /* INET6 */
88 #include <sys/socket.h>
89 #include <sys/socketvar.h>
90 #include <sys/protosw.h>
91
92 #include <sys/thread2.h>
93 #include <sys/msgport2.h>
94 #include <sys/socketvar2.h>
95
96 #include <net/if.h>
97 #include <net/netisr.h>
98 #include <net/route.h>
99
100 #include <net/netmsg2.h>
101
102 #include <netinet/in.h>
103 #include <netinet/in_systm.h>
104 #ifdef INET6
105 #include <netinet/ip6.h>
106 #endif
107 #include <netinet/in_pcb.h>
108 #ifdef INET6
109 #include <netinet6/in6_pcb.h>
110 #endif
111 #include <netinet/in_var.h>
112 #include <netinet/ip_var.h>
113 #ifdef INET6
114 #include <netinet6/ip6_var.h>
115 #include <netinet6/tcp6_var.h>
116 #endif
117 #include <netinet/tcp.h>
118 #include <netinet/tcp_fsm.h>
119 #include <netinet/tcp_seq.h>
120 #include <netinet/tcp_timer.h>
121 #include <netinet/tcp_timer2.h>
122 #include <netinet/tcp_var.h>
123 #include <netinet/tcpip.h>
124 #ifdef TCPDEBUG
125 #include <netinet/tcp_debug.h>
126 #endif
127
128 #ifdef IPSEC
129 #include <netinet6/ipsec.h>
130 #endif /*IPSEC*/
131
132 /*
133  * TCP protocol interface to socket abstraction.
134  */
135 extern  char *tcpstates[];      /* XXX ??? */
136
137 static int      tcp_attach (struct socket *, struct pru_attach_info *);
138 static void     tcp_connect (netmsg_t msg);
139 #ifdef INET6
140 static void     tcp6_connect (netmsg_t msg);
141 static int      tcp6_connect_oncpu(struct tcpcb *tp, int flags,
142                                 struct mbuf **mp,
143                                 struct sockaddr_in6 *sin6,
144                                 struct in6_addr *addr6);
145 #endif /* INET6 */
146 static struct tcpcb *
147                 tcp_disconnect (struct tcpcb *);
148 static struct tcpcb *
149                 tcp_usrclosed (struct tcpcb *);
150
151 #ifdef TCPDEBUG
152 #define TCPDEBUG0       int ostate = 0
153 #define TCPDEBUG1()     ostate = tp ? tp->t_state : 0
154 #define TCPDEBUG2(req)  if (tp && (so->so_options & SO_DEBUG)) \
155                                 tcp_trace(TA_USER, ostate, tp, 0, 0, req)
156 #else
157 #define TCPDEBUG0
158 #define TCPDEBUG1()
159 #define TCPDEBUG2(req)
160 #endif
161
162 /*
163  * TCP attaches to socket via pru_attach(), reserving space,
164  * and an internet control block.  This is likely occuring on
165  * cpu0 and may have to move later when we bind/connect.
166  */
167 static void
168 tcp_usr_attach(netmsg_t msg)
169 {
170         struct socket *so = msg->base.nm_so;
171         struct pru_attach_info *ai = msg->attach.nm_ai;
172         int error;
173         struct inpcb *inp;
174         struct tcpcb *tp = 0;
175         TCPDEBUG0;
176
177         soreference(so);
178         inp = so->so_pcb;
179         TCPDEBUG1();
180         if (inp) {
181                 error = EISCONN;
182                 goto out;
183         }
184
185         error = tcp_attach(so, ai);
186         if (error)
187                 goto out;
188
189         if ((so->so_options & SO_LINGER) && so->so_linger == 0)
190                 so->so_linger = TCP_LINGERTIME;
191         tp = sototcpcb(so);
192 out:
193         sofree(so);             /* from ref above */
194         TCPDEBUG2(PRU_ATTACH);
195         lwkt_replymsg(&msg->lmsg, error);
196 }
197
198 /*
199  * pru_detach() detaches the TCP protocol from the socket.
200  * If the protocol state is non-embryonic, then can't
201  * do this directly: have to initiate a pru_disconnect(),
202  * which may finish later; embryonic TCB's can just
203  * be discarded here.
204  */
205 static void
206 tcp_usr_detach(netmsg_t msg)
207 {
208         struct socket *so = msg->base.nm_so;
209         int error = 0;
210         struct inpcb *inp;
211         struct tcpcb *tp;
212         TCPDEBUG0;
213
214         inp = so->so_pcb;
215
216         /*
217          * If the inp is already detached it may have been due to an async
218          * close.  Just return as if no error occured.
219          *
220          * It's possible for the tcpcb (tp) to disconnect from the inp due
221          * to tcp_drop()->tcp_close() being called.  This may occur *after*
222          * the detach message has been queued so we may find a NULL tp here.
223          */
224         if (inp) {
225                 if ((tp = intotcpcb(inp)) != NULL) {
226                         TCPDEBUG1();
227                         tp = tcp_disconnect(tp);
228                         TCPDEBUG2(PRU_DETACH);
229                 }
230         }
231         lwkt_replymsg(&msg->lmsg, error);
232 }
233
234 /*
235  * NOTE: ignore_error is non-zero for certain disconnection races
236  * which we want to silently allow, otherwise close() may return
237  * an unexpected error.
238  *
239  * NOTE: The variables (msg) and (tp) are assumed.
240  */
241 #define COMMON_START(so, inp, ignore_error)                     \
242         TCPDEBUG0;                                              \
243                                                                 \
244         inp = so->so_pcb;                                       \
245         do {                                                    \
246                  if (inp == NULL) {                             \
247                         error = ignore_error ? 0 : EINVAL;      \
248                         tp = NULL;                              \
249                         goto out;                               \
250                  }                                              \
251                  tp = intotcpcb(inp);                           \
252                  TCPDEBUG1();                                   \
253         } while(0)
254
255 #define COMMON_END(req)                                         \
256         out: do {                                               \
257                 TCPDEBUG2(req);                                 \
258                 lwkt_replymsg(&msg->lmsg, error);               \
259                 return;                                         \
260         } while(0)
261
262 /*
263  * Give the socket an address.
264  */
265 static void
266 tcp_usr_bind(netmsg_t msg)
267 {
268         struct socket *so = msg->bind.base.nm_so;
269         struct sockaddr *nam = msg->bind.nm_nam;
270         struct thread *td = msg->bind.nm_td;
271         int error = 0;
272         struct inpcb *inp;
273         struct tcpcb *tp;
274         struct sockaddr_in *sinp;
275
276         COMMON_START(so, inp, 0);
277
278         /*
279          * Must check for multicast addresses and disallow binding
280          * to them.
281          */
282         sinp = (struct sockaddr_in *)nam;
283         if (sinp->sin_family == AF_INET &&
284             IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
285                 error = EAFNOSUPPORT;
286                 goto out;
287         }
288         error = in_pcbbind(inp, nam, td);
289         if (error)
290                 goto out;
291         COMMON_END(PRU_BIND);
292
293 }
294
295 #ifdef INET6
296
297 static void
298 tcp6_usr_bind(netmsg_t msg)
299 {
300         struct socket *so = msg->bind.base.nm_so;
301         struct sockaddr *nam = msg->bind.nm_nam;
302         struct thread *td = msg->bind.nm_td;
303         int error = 0;
304         struct inpcb *inp;
305         struct tcpcb *tp;
306         struct sockaddr_in6 *sin6p;
307
308         COMMON_START(so, inp, 0);
309
310         /*
311          * Must check for multicast addresses and disallow binding
312          * to them.
313          */
314         sin6p = (struct sockaddr_in6 *)nam;
315         if (sin6p->sin6_family == AF_INET6 &&
316             IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
317                 error = EAFNOSUPPORT;
318                 goto out;
319         }
320         inp->inp_vflag &= ~INP_IPV4;
321         inp->inp_vflag |= INP_IPV6;
322         if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
323                 if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr))
324                         inp->inp_vflag |= INP_IPV4;
325                 else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
326                         struct sockaddr_in sin;
327
328                         in6_sin6_2_sin(&sin, sin6p);
329                         inp->inp_vflag |= INP_IPV4;
330                         inp->inp_vflag &= ~INP_IPV6;
331                         error = in_pcbbind(inp, (struct sockaddr *)&sin, td);
332                         goto out;
333                 }
334         }
335         error = in6_pcbbind(inp, nam, td);
336         if (error)
337                 goto out;
338         COMMON_END(PRU_BIND);
339 }
340 #endif /* INET6 */
341
342 #ifdef SMP
343
344 struct netmsg_inswildcard {
345         struct netmsg_base      base;
346         struct inpcb            *nm_inp;
347 };
348
349 static void
350 in_pcbinswildcardhash_handler(netmsg_t msg)
351 {
352         struct netmsg_inswildcard *nm = (struct netmsg_inswildcard *)msg;
353         int cpu = mycpuid, nextcpu;
354
355         in_pcbinswildcardhash_oncpu(nm->nm_inp, &tcbinfo[cpu]);
356
357         nextcpu = cpu + 1;
358         if (nextcpu < ncpus2)
359                 lwkt_forwardmsg(cpu_portfn(nextcpu), &nm->base.lmsg);
360         else
361                 lwkt_replymsg(&nm->base.lmsg, 0);
362 }
363
364 #endif
365
366 /*
367  * Prepare to accept connections.
368  */
369 static void
370 tcp_usr_listen(netmsg_t msg)
371 {
372         struct socket *so = msg->listen.base.nm_so;
373         struct thread *td = msg->listen.nm_td;
374         int error = 0;
375         struct inpcb *inp;
376         struct tcpcb *tp;
377 #ifdef SMP
378         struct netmsg_inswildcard nm;
379 #endif
380
381         COMMON_START(so, inp, 0);
382
383         if (tp->t_flags & TF_LISTEN)
384                 goto out;
385
386         if (inp->inp_lport == 0) {
387                 error = in_pcbbind(inp, NULL, td);
388                 if (error)
389                         goto out;
390         }
391
392         tp->t_state = TCPS_LISTEN;
393         tp->t_flags |= TF_LISTEN;
394         tp->tt_msg = NULL; /* Catch any invalid timer usage */
395
396 #ifdef SMP
397         if (ncpus > 1) {
398                 /*
399                  * We have to set the flag because we can't have other cpus
400                  * messing with our inp's flags.
401                  */
402                 KASSERT(!(inp->inp_flags & INP_CONNECTED),
403                         ("already on connhash\n"));
404                 KASSERT(!(inp->inp_flags & INP_WILDCARD),
405                         ("already on wildcardhash\n"));
406                 KASSERT(!(inp->inp_flags & INP_WILDCARD_MP),
407                         ("already on MP wildcardhash\n"));
408                 inp->inp_flags |= INP_WILDCARD_MP;
409
410                 KKASSERT(so->so_port == cpu_portfn(0));
411                 KKASSERT(&curthread->td_msgport == cpu_portfn(0));
412                 KKASSERT(inp->inp_pcbinfo == &tcbinfo[0]);
413
414                 netmsg_init(&nm.base, NULL, &curthread->td_msgport,
415                             MSGF_PRIORITY, in_pcbinswildcardhash_handler);
416                 nm.nm_inp = inp;
417                 lwkt_domsg(cpu_portfn(1), &nm.base.lmsg, 0);
418         }
419 #endif
420         in_pcbinswildcardhash(inp);
421         COMMON_END(PRU_LISTEN);
422 }
423
424 #ifdef INET6
425
426 static void
427 tcp6_usr_listen(netmsg_t msg)
428 {
429         struct socket *so = msg->listen.base.nm_so;
430         struct thread *td = msg->listen.nm_td;
431         int error = 0;
432         struct inpcb *inp;
433         struct tcpcb *tp;
434 #ifdef SMP
435         struct netmsg_inswildcard nm;
436 #endif
437
438         COMMON_START(so, inp, 0);
439
440         if (tp->t_flags & TF_LISTEN)
441                 goto out;
442
443         if (inp->inp_lport == 0) {
444                 if (!(inp->inp_flags & IN6P_IPV6_V6ONLY))
445                         inp->inp_vflag |= INP_IPV4;
446                 else
447                         inp->inp_vflag &= ~INP_IPV4;
448                 error = in6_pcbbind(inp, NULL, td);
449                 if (error)
450                         goto out;
451         }
452
453         tp->t_state = TCPS_LISTEN;
454         tp->t_flags |= TF_LISTEN;
455         tp->tt_msg = NULL; /* Catch any invalid timer usage */
456
457 #ifdef SMP
458         if (ncpus > 1) {
459                 /*
460                  * We have to set the flag because we can't have other cpus
461                  * messing with our inp's flags.
462                  */
463                 KASSERT(!(inp->inp_flags & INP_CONNECTED),
464                         ("already on connhash\n"));
465                 KASSERT(!(inp->inp_flags & INP_WILDCARD),
466                         ("already on wildcardhash\n"));
467                 KASSERT(!(inp->inp_flags & INP_WILDCARD_MP),
468                         ("already on MP wildcardhash\n"));
469                 inp->inp_flags |= INP_WILDCARD_MP;
470
471                 KKASSERT(so->so_port == cpu_portfn(0));
472                 KKASSERT(&curthread->td_msgport == cpu_portfn(0));
473                 KKASSERT(inp->inp_pcbinfo == &tcbinfo[0]);
474
475                 netmsg_init(&nm.base, NULL, &curthread->td_msgport,
476                             MSGF_PRIORITY, in_pcbinswildcardhash_handler);
477                 nm.nm_inp = inp;
478                 lwkt_domsg(cpu_portfn(1), &nm.base.lmsg, 0);
479         }
480 #endif
481         in_pcbinswildcardhash(inp);
482         COMMON_END(PRU_LISTEN);
483 }
484 #endif /* INET6 */
485
486 /*
487  * Initiate connection to peer.
488  * Create a template for use in transmissions on this connection.
489  * Enter SYN_SENT state, and mark socket as connecting.
490  * Start keep-alive timer, and seed output sequence space.
491  * Send initial segment on connection.
492  */
493 static void
494 tcp_usr_connect(netmsg_t msg)
495 {
496         struct socket *so = msg->connect.base.nm_so;
497         struct sockaddr *nam = msg->connect.nm_nam;
498         struct thread *td = msg->connect.nm_td;
499         int error = 0;
500         struct inpcb *inp;
501         struct tcpcb *tp;
502         struct sockaddr_in *sinp;
503
504         COMMON_START(so, inp, 0);
505
506         /*
507          * Must disallow TCP ``connections'' to multicast addresses.
508          */
509         sinp = (struct sockaddr_in *)nam;
510         if (sinp->sin_family == AF_INET
511             && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
512                 error = EAFNOSUPPORT;
513                 goto out;
514         }
515
516         if (!prison_remote_ip(td, (struct sockaddr*)sinp)) {
517                 error = EAFNOSUPPORT; /* IPv6 only jail */
518                 goto out;
519         }
520
521         tcp_connect(msg);
522         /* msg is invalid now */
523         return;
524 out:
525         if (msg->connect.nm_m) {
526                 m_freem(msg->connect.nm_m);
527                 msg->connect.nm_m = NULL;
528         }
529         lwkt_replymsg(&msg->lmsg, error);
530 }
531
532 #ifdef INET6
533
534 static void
535 tcp6_usr_connect(netmsg_t msg)
536 {
537         struct socket *so = msg->connect.base.nm_so;
538         struct sockaddr *nam = msg->connect.nm_nam;
539         struct thread *td = msg->connect.nm_td;
540         int error = 0;
541         struct inpcb *inp;
542         struct tcpcb *tp;
543         struct sockaddr_in6 *sin6p;
544
545         COMMON_START(so, inp, 0);
546
547         /*
548          * Must disallow TCP ``connections'' to multicast addresses.
549          */
550         sin6p = (struct sockaddr_in6 *)nam;
551         if (sin6p->sin6_family == AF_INET6
552             && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
553                 error = EAFNOSUPPORT;
554                 goto out;
555         }
556
557         if (!prison_remote_ip(td, nam)) {
558                 error = EAFNOSUPPORT; /* IPv4 only jail */
559                 goto out;
560         }
561
562         if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
563                 struct sockaddr_in *sinp;
564
565                 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
566                         error = EINVAL;
567                         goto out;
568                 }
569                 sinp = kmalloc(sizeof(*sinp), M_LWKTMSG, M_INTWAIT);
570                 in6_sin6_2_sin(sinp, sin6p);
571                 inp->inp_vflag |= INP_IPV4;
572                 inp->inp_vflag &= ~INP_IPV6;
573                 msg->connect.nm_nam = (struct sockaddr *)sinp;
574                 msg->connect.nm_reconnect |= NMSG_RECONNECT_NAMALLOC;
575                 tcp_connect(msg);
576                 /* msg is invalid now */
577                 return;
578         }
579         inp->inp_vflag &= ~INP_IPV4;
580         inp->inp_vflag |= INP_IPV6;
581         inp->inp_inc.inc_isipv6 = 1;
582
583         msg->connect.nm_reconnect |= NMSG_RECONNECT_FALLBACK;
584         tcp6_connect(msg);
585         /* msg is invalid now */
586         return;
587 out:
588         if (msg->connect.nm_m) {
589                 m_freem(msg->connect.nm_m);
590                 msg->connect.nm_m = NULL;
591         }
592         lwkt_replymsg(&msg->lmsg, error);
593 }
594
595 #endif /* INET6 */
596
597 /*
598  * Initiate disconnect from peer.
599  * If connection never passed embryonic stage, just drop;
600  * else if don't need to let data drain, then can just drop anyways,
601  * else have to begin TCP shutdown process: mark socket disconnecting,
602  * drain unread data, state switch to reflect user close, and
603  * send segment (e.g. FIN) to peer.  Socket will be really disconnected
604  * when peer sends FIN and acks ours.
605  *
606  * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
607  */
608 static void
609 tcp_usr_disconnect(netmsg_t msg)
610 {
611         struct socket *so = msg->disconnect.base.nm_so;
612         int error = 0;
613         struct inpcb *inp;
614         struct tcpcb *tp;
615
616         COMMON_START(so, inp, 1);
617         tp = tcp_disconnect(tp);
618         COMMON_END(PRU_DISCONNECT);
619 }
620
621 /*
622  * Accept a connection.  Essentially all the work is
623  * done at higher levels; just return the address
624  * of the peer, storing through addr.
625  */
626 static void
627 tcp_usr_accept(netmsg_t msg)
628 {
629         struct socket *so = msg->accept.base.nm_so;
630         struct sockaddr **nam = msg->accept.nm_nam;
631         int error = 0;
632         struct inpcb *inp;
633         struct tcpcb *tp = NULL;
634         TCPDEBUG0;
635
636         inp = so->so_pcb;
637         if (so->so_state & SS_ISDISCONNECTED) {
638                 error = ECONNABORTED;
639                 goto out;
640         }
641         if (inp == 0) {
642                 error = EINVAL;
643                 goto out;
644         }
645
646         tp = intotcpcb(inp);
647         TCPDEBUG1();
648         in_setpeeraddr(so, nam);
649         COMMON_END(PRU_ACCEPT);
650 }
651
652 #ifdef INET6
653 static void
654 tcp6_usr_accept(netmsg_t msg)
655 {
656         struct socket *so = msg->accept.base.nm_so;
657         struct sockaddr **nam = msg->accept.nm_nam;
658         int error = 0;
659         struct inpcb *inp;
660         struct tcpcb *tp = NULL;
661         TCPDEBUG0;
662
663         inp = so->so_pcb;
664
665         if (so->so_state & SS_ISDISCONNECTED) {
666                 error = ECONNABORTED;
667                 goto out;
668         }
669         if (inp == 0) {
670                 error = EINVAL;
671                 goto out;
672         }
673         tp = intotcpcb(inp);
674         TCPDEBUG1();
675         in6_mapped_peeraddr(so, nam);
676         COMMON_END(PRU_ACCEPT);
677 }
678 #endif /* INET6 */
679 /*
680  * Mark the connection as being incapable of further output.
681  */
682 static void
683 tcp_usr_shutdown(netmsg_t msg)
684 {
685         struct socket *so = msg->shutdown.base.nm_so;
686         int error = 0;
687         struct inpcb *inp;
688         struct tcpcb *tp;
689
690         COMMON_START(so, inp, 0);
691         socantsendmore(so);
692         tp = tcp_usrclosed(tp);
693         if (tp)
694                 error = tcp_output(tp);
695         COMMON_END(PRU_SHUTDOWN);
696 }
697
698 /*
699  * After a receive, possibly send window update to peer.
700  */
701 static void
702 tcp_usr_rcvd(netmsg_t msg)
703 {
704         struct socket *so = msg->rcvd.base.nm_so;
705         int error = 0;
706         struct inpcb *inp;
707         struct tcpcb *tp;
708
709         COMMON_START(so, inp, 0);
710         tcp_output(tp);
711         COMMON_END(PRU_RCVD);
712 }
713
714 /*
715  * Do a send by putting data in output queue and updating urgent
716  * marker if URG set.  Possibly send more data.  Unlike the other
717  * pru_*() routines, the mbuf chains are our responsibility.  We
718  * must either enqueue them or free them.  The other pru_* routines
719  * generally are caller-frees.
720  */
721 static void
722 tcp_usr_send(netmsg_t msg)
723 {
724         struct socket *so = msg->send.base.nm_so;
725         int flags = msg->send.nm_flags;
726         struct mbuf *m = msg->send.nm_m;
727         struct sockaddr *nam = msg->send.nm_addr;
728         struct mbuf *control = msg->send.nm_control;
729         struct thread *td = msg->send.nm_td;
730         int error = 0;
731         struct inpcb *inp;
732         struct tcpcb *tp;
733 #ifdef INET6
734         int isipv6;
735 #endif
736         TCPDEBUG0;
737
738         inp = so->so_pcb;
739
740         if (inp == NULL) {
741                 /*
742                  * OOPS! we lost a race, the TCP session got reset after
743                  * we checked SS_CANTSENDMORE, eg: while doing uiomove or a
744                  * network interrupt in the non-critical section of sosend().
745                  */
746                 m_freem(m);
747                 if (control)
748                         m_freem(control);
749                 error = ECONNRESET;     /* XXX EPIPE? */
750                 tp = NULL;
751                 TCPDEBUG1();
752                 goto out;
753         }
754 #ifdef INET6
755         isipv6 = nam && nam->sa_family == AF_INET6;
756 #endif /* INET6 */
757         tp = intotcpcb(inp);
758         TCPDEBUG1();
759         if (control) {
760                 /* TCP doesn't do control messages (rights, creds, etc) */
761                 if (control->m_len) {
762                         m_freem(control);
763                         m_freem(m);
764                         error = EINVAL;
765                         goto out;
766                 }
767                 m_freem(control);       /* empty control, just free it */
768         }
769
770         /*
771          * Don't let too much OOB data build up
772          */
773         if (flags & PRUS_OOB) {
774                 if (ssb_space(&so->so_snd) < -512) {
775                         m_freem(m);
776                         error = ENOBUFS;
777                         goto out;
778                 }
779         }
780
781         /*
782          * Do implied connect if not yet connected.  Any data sent
783          * with the connect is handled by tcp_connect() and friends.
784          *
785          * NOTE!  PROTOCOL THREAD MAY BE CHANGED BY THE CONNECT!
786          */
787         if (nam && tp->t_state < TCPS_SYN_SENT) {
788                 kprintf("implied fallback\n");
789                 msg->connect.nm_nam = nam;
790                 msg->connect.nm_td = td;
791                 msg->connect.nm_m = m;
792                 msg->connect.nm_flags = flags;
793                 msg->connect.nm_reconnect = NMSG_RECONNECT_FALLBACK;
794 #ifdef INET6
795                 if (isipv6)
796                         tcp6_connect(msg);
797                 else
798 #endif /* INET6 */
799                         tcp_connect(msg);
800                 /* msg invalid now */
801                 return;
802         }
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) {
832                         if (flags & PRUS_MORETOCOME)
833                                 tp->t_flags |= TF_MORETOCOME;
834                         error = tcp_output(tp);
835                         if (flags & PRUS_MORETOCOME)
836                                 tp->t_flags &= ~TF_MORETOCOME;
837                 }
838         }
839         COMMON_END((flags & PRUS_OOB) ? PRU_SENDOOB :
840                    ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
841 }
842
843 /*
844  * NOTE: (so) is referenced from soabort*() and netmsg_pru_abort()
845  *       will sofree() it when we return.
846  */
847 static void
848 tcp_usr_abort(netmsg_t msg)
849 {
850         struct socket *so = msg->abort.base.nm_so;
851         int error = 0;
852         struct inpcb *inp;
853         struct tcpcb *tp;
854
855         COMMON_START(so, inp, 1);
856         tp = tcp_drop(tp, ECONNABORTED);
857         COMMON_END(PRU_ABORT);
858 }
859
860 /*
861  * Receive out-of-band data.
862  */
863 static void
864 tcp_usr_rcvoob(netmsg_t msg)
865 {
866         struct socket *so = msg->rcvoob.base.nm_so;
867         struct mbuf *m = msg->rcvoob.nm_m;
868         int flags = msg->rcvoob.nm_flags;
869         int error = 0;
870         struct inpcb *inp;
871         struct tcpcb *tp;
872
873         COMMON_START(so, inp, 0);
874         if ((so->so_oobmark == 0 &&
875              (so->so_state & SS_RCVATMARK) == 0) ||
876             so->so_options & SO_OOBINLINE ||
877             tp->t_oobflags & TCPOOB_HADDATA) {
878                 error = EINVAL;
879                 goto out;
880         }
881         if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
882                 error = EWOULDBLOCK;
883                 goto out;
884         }
885         m->m_len = 1;
886         *mtod(m, caddr_t) = tp->t_iobc;
887         if ((flags & MSG_PEEK) == 0)
888                 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
889         COMMON_END(PRU_RCVOOB);
890 }
891
892 /* xxx - should be const */
893 struct pr_usrreqs tcp_usrreqs = {
894         .pru_abort = tcp_usr_abort,
895         .pru_accept = tcp_usr_accept,
896         .pru_attach = tcp_usr_attach,
897         .pru_bind = tcp_usr_bind,
898         .pru_connect = tcp_usr_connect,
899         .pru_connect2 = pr_generic_notsupp,
900         .pru_control = in_control_dispatch,
901         .pru_detach = tcp_usr_detach,
902         .pru_disconnect = tcp_usr_disconnect,
903         .pru_listen = tcp_usr_listen,
904         .pru_peeraddr = in_setpeeraddr_dispatch,
905         .pru_rcvd = tcp_usr_rcvd,
906         .pru_rcvoob = tcp_usr_rcvoob,
907         .pru_send = tcp_usr_send,
908         .pru_sense = pru_sense_null,
909         .pru_shutdown = tcp_usr_shutdown,
910         .pru_sockaddr = in_setsockaddr_dispatch,
911         .pru_sosend = sosend,
912         .pru_soreceive = soreceive
913 };
914
915 #ifdef INET6
916 struct pr_usrreqs tcp6_usrreqs = {
917         .pru_abort = tcp_usr_abort,
918         .pru_accept = tcp6_usr_accept,
919         .pru_attach = tcp_usr_attach,
920         .pru_bind = tcp6_usr_bind,
921         .pru_connect = tcp6_usr_connect,
922         .pru_connect2 = pr_generic_notsupp,
923         .pru_control = in6_control_dispatch,
924         .pru_detach = tcp_usr_detach,
925         .pru_disconnect = tcp_usr_disconnect,
926         .pru_listen = tcp6_usr_listen,
927         .pru_peeraddr = in6_mapped_peeraddr_dispatch,
928         .pru_rcvd = tcp_usr_rcvd,
929         .pru_rcvoob = tcp_usr_rcvoob,
930         .pru_send = tcp_usr_send,
931         .pru_sense = pru_sense_null,
932         .pru_shutdown = tcp_usr_shutdown,
933         .pru_sockaddr = in6_mapped_sockaddr_dispatch,
934         .pru_sosend = sosend,
935         .pru_soreceive = soreceive
936 };
937 #endif /* INET6 */
938
939 static int
940 tcp_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf *m,
941                   struct sockaddr_in *sin, struct sockaddr_in *if_sin)
942 {
943         struct inpcb *inp = tp->t_inpcb, *oinp;
944         struct socket *so = inp->inp_socket;
945         struct route *ro = &inp->inp_route;
946
947         oinp = in_pcblookup_hash(&tcbinfo[mycpu->gd_cpuid],
948                                  sin->sin_addr, sin->sin_port,
949                                  (inp->inp_laddr.s_addr != INADDR_ANY ?
950                                   inp->inp_laddr : if_sin->sin_addr),
951                                 inp->inp_lport, 0, NULL);
952         if (oinp != NULL) {
953                 m_freem(m);
954                 return (EADDRINUSE);
955         }
956         if (inp->inp_laddr.s_addr == INADDR_ANY)
957                 inp->inp_laddr = if_sin->sin_addr;
958         inp->inp_faddr = sin->sin_addr;
959         inp->inp_fport = sin->sin_port;
960         inp->inp_cpcbinfo = &tcbinfo[mycpu->gd_cpuid];
961         in_pcbinsconnhash(inp);
962
963         /*
964          * We are now on the inpcb's owner CPU, if the cached route was
965          * freed because the rtentry's owner CPU is not the current CPU
966          * (e.g. in tcp_connect()), then we try to reallocate it here with
967          * the hope that a rtentry may be cloned from a RTF_PRCLONING
968          * rtentry.
969          */
970         if (!(inp->inp_socket->so_options & SO_DONTROUTE) && /*XXX*/
971             ro->ro_rt == NULL) {
972                 bzero(&ro->ro_dst, sizeof(struct sockaddr_in));
973                 ro->ro_dst.sa_family = AF_INET;
974                 ro->ro_dst.sa_len = sizeof(struct sockaddr_in);
975                 ((struct sockaddr_in *)&ro->ro_dst)->sin_addr =
976                         sin->sin_addr;
977                 rtalloc(ro);
978         }
979
980         /*
981          * Now that no more errors can occur, change the protocol processing
982          * port to the current thread (which is the correct thread).
983          *
984          * Create TCP timer message now; we are on the tcpcb's owner
985          * CPU/thread.
986          */
987         tcp_create_timermsg(tp, &curthread->td_msgport);
988
989         /*
990          * Compute window scaling to request.  Use a larger scaling then
991          * needed for the initial receive buffer in case the receive buffer
992          * gets expanded.
993          */
994         if (tp->request_r_scale < TCP_MIN_WINSHIFT)
995                 tp->request_r_scale = TCP_MIN_WINSHIFT;
996         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
997                (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat
998         ) {
999                 tp->request_r_scale++;
1000         }
1001
1002         soisconnecting(so);
1003         tcpstat.tcps_connattempt++;
1004         tp->t_state = TCPS_SYN_SENT;
1005         tcp_callout_reset(tp, tp->tt_keep, tcp_keepinit, tcp_timer_keep);
1006         tp->iss = tcp_new_isn(tp);
1007         tcp_sendseqinit(tp);
1008         if (m) {
1009                 ssb_appendstream(&so->so_snd, m);
1010                 m = NULL;
1011                 if (flags & PRUS_OOB)
1012                         tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
1013         }
1014
1015         /*
1016          * Close the send side of the connection after
1017          * the data is sent if flagged.
1018          */
1019         if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
1020                 socantsendmore(so);
1021                 tp = tcp_usrclosed(tp);
1022         }
1023         return (tcp_output(tp));
1024 }
1025
1026 /*
1027  * Common subroutine to open a TCP connection to remote host specified
1028  * by struct sockaddr_in in mbuf *nam.  Call in_pcbbind to assign a local
1029  * port number if needed.  Call in_pcbladdr to do the routing and to choose
1030  * a local host address (interface).
1031  * Initialize connection parameters and enter SYN-SENT state.
1032  */
1033 static void
1034 tcp_connect(netmsg_t msg)
1035 {
1036         struct socket *so = msg->connect.base.nm_so;
1037         struct sockaddr *nam = msg->connect.nm_nam;
1038         struct thread *td = msg->connect.nm_td;
1039         struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1040         struct sockaddr_in *if_sin;
1041         struct inpcb *inp;
1042         struct tcpcb *tp;
1043         int error;
1044 #ifdef SMP
1045         lwkt_port_t port;
1046 #endif
1047
1048         COMMON_START(so, inp, 0);
1049
1050         /*
1051          * Reconnect our pcb if we have to
1052          */
1053         if (msg->connect.nm_reconnect & NMSG_RECONNECT_RECONNECT) {
1054                 msg->connect.nm_reconnect &= ~NMSG_RECONNECT_RECONNECT;
1055                 in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1056         }
1057
1058         /*
1059          * Bind if we have to
1060          */
1061         if (inp->inp_lport == 0) {
1062                 error = in_pcbbind(inp, NULL, td);
1063                 if (error)
1064                         goto out;
1065         }
1066         so = inp->inp_socket;
1067         KKASSERT(so);
1068
1069         /*
1070          * Calculate the correct protocol processing thread.  The connect
1071          * operation must run there.  Set the forwarding port before we
1072          * forward the message or it will get bounced right back to us.
1073          */
1074         error = in_pcbladdr(inp, nam, &if_sin, td);
1075         if (error)
1076                 goto out;
1077
1078 #ifdef SMP
1079         port = tcp_addrport(sin->sin_addr.s_addr, sin->sin_port,
1080                             (inp->inp_laddr.s_addr ?
1081                              inp->inp_laddr.s_addr : if_sin->sin_addr.s_addr),
1082                             inp->inp_lport);
1083
1084         if (port != &curthread->td_msgport) {
1085                 struct route *ro = &inp->inp_route;
1086
1087                 /*
1088                  * in_pcbladdr() may have allocated a route entry for us
1089                  * on the current CPU, but we need a route entry on the
1090                  * inpcb's owner CPU, so free it here.
1091                  */
1092                 if (ro->ro_rt != NULL)
1093                         RTFREE(ro->ro_rt);
1094                 bzero(ro, sizeof(*ro));
1095
1096                 /*
1097                  * We are moving the protocol processing port the socket
1098                  * is on, we have to unlink here and re-link on the
1099                  * target cpu.
1100                  */
1101                 in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1102                 sosetport(so, port);
1103                 msg->connect.nm_reconnect |= NMSG_RECONNECT_RECONNECT;
1104                 msg->connect.base.nm_dispatch = tcp_connect;
1105
1106                 lwkt_forwardmsg(port, &msg->connect.base.lmsg);
1107                 /* msg invalid now */
1108                 return;
1109         }
1110 #else
1111         KKASSERT(so->so_port == &curthread->td_msgport);
1112 #endif
1113         error = tcp_connect_oncpu(tp, msg->connect.nm_flags,
1114                                   msg->connect.nm_m, sin, if_sin);
1115         msg->connect.nm_m = NULL;
1116 out:
1117         if (msg->connect.nm_m) {
1118                 m_freem(msg->connect.nm_m);
1119                 msg->connect.nm_m = NULL;
1120         }
1121         if (msg->connect.nm_reconnect & NMSG_RECONNECT_NAMALLOC) {
1122                 kfree(msg->connect.nm_nam, M_LWKTMSG);
1123                 msg->connect.nm_nam = NULL;
1124         }
1125         lwkt_replymsg(&msg->connect.base.lmsg, error);
1126         /* msg invalid now */
1127 }
1128
1129 #ifdef INET6
1130
1131 static void
1132 tcp6_connect(netmsg_t msg)
1133 {
1134         struct tcpcb *tp;
1135         struct socket *so = msg->connect.base.nm_so;
1136         struct sockaddr *nam = msg->connect.nm_nam;
1137         struct thread *td = msg->connect.nm_td;
1138         struct inpcb *inp;
1139         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam;
1140         struct in6_addr *addr6;
1141 #ifdef SMP
1142         lwkt_port_t port;
1143 #endif
1144         int error;
1145
1146         COMMON_START(so, inp, 0);
1147
1148         /*
1149          * Reconnect our pcb if we have to
1150          */
1151         if (msg->connect.nm_reconnect & NMSG_RECONNECT_RECONNECT) {
1152                 msg->connect.nm_reconnect &= ~NMSG_RECONNECT_RECONNECT;
1153                 in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1154         }
1155
1156         /*
1157          * Bind if we have to
1158          */
1159         if (inp->inp_lport == 0) {
1160                 error = in6_pcbbind(inp, NULL, td);
1161                 if (error)
1162                         goto out;
1163         }
1164
1165         /*
1166          * Cannot simply call in_pcbconnect, because there might be an
1167          * earlier incarnation of this same connection still in
1168          * TIME_WAIT state, creating an ADDRINUSE error.
1169          */
1170         error = in6_pcbladdr(inp, nam, &addr6, td);
1171         if (error)
1172                 goto out;
1173
1174 #ifdef SMP
1175         port = tcp6_addrport(); /* XXX hack for now, always cpu0 */
1176
1177         if (port != &curthread->td_msgport) {
1178                 struct route *ro = &inp->inp_route;
1179
1180                 /*
1181                  * in_pcbladdr() may have allocated a route entry for us
1182                  * on the current CPU, but we need a route entry on the
1183                  * inpcb's owner CPU, so free it here.
1184                  */
1185                 if (ro->ro_rt != NULL)
1186                         RTFREE(ro->ro_rt);
1187                 bzero(ro, sizeof(*ro));
1188
1189                 in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1190                 sosetport(so, port);
1191                 msg->connect.nm_reconnect |= NMSG_RECONNECT_RECONNECT;
1192                 msg->connect.base.nm_dispatch = tcp6_connect;
1193
1194                 lwkt_forwardmsg(port, &msg->connect.base.lmsg);
1195                 /* msg invalid now */
1196                 return;
1197         }
1198 #endif
1199         error = tcp6_connect_oncpu(tp, msg->connect.nm_flags,
1200                                    &msg->connect.nm_m, sin6, addr6);
1201         /* nm_m may still be intact */
1202 out:
1203         if (error && (msg->connect.nm_reconnect & NMSG_RECONNECT_FALLBACK)) {
1204                 tcp_connect(msg);
1205                 /* msg invalid now */
1206         } else {
1207                 if (msg->connect.nm_m) {
1208                         m_freem(msg->connect.nm_m);
1209                         msg->connect.nm_m = NULL;
1210                 }
1211                 if (msg->connect.nm_reconnect & NMSG_RECONNECT_NAMALLOC) {
1212                         kfree(msg->connect.nm_nam, M_LWKTMSG);
1213                         msg->connect.nm_nam = NULL;
1214                 }
1215                 lwkt_replymsg(&msg->connect.base.lmsg, error);
1216                 /* msg invalid now */
1217         }
1218 }
1219
1220 static int
1221 tcp6_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf **mp,
1222                    struct sockaddr_in6 *sin6, struct in6_addr *addr6)
1223 {
1224         struct mbuf *m = *mp;
1225         struct inpcb *inp = tp->t_inpcb;
1226         struct socket *so = inp->inp_socket;
1227         struct inpcb *oinp;
1228
1229         /*
1230          * Cannot simply call in_pcbconnect, because there might be an
1231          * earlier incarnation of this same connection still in
1232          * TIME_WAIT state, creating an ADDRINUSE error.
1233          */
1234         oinp = in6_pcblookup_hash(inp->inp_cpcbinfo,
1235                                   &sin6->sin6_addr, sin6->sin6_port,
1236                                   (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) ?
1237                                       addr6 : &inp->in6p_laddr),
1238                                   inp->inp_lport,  0, NULL);
1239         if (oinp)
1240                 return (EADDRINUSE);
1241
1242         if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
1243                 inp->in6p_laddr = *addr6;
1244         inp->in6p_faddr = sin6->sin6_addr;
1245         inp->inp_fport = sin6->sin6_port;
1246         if ((sin6->sin6_flowinfo & IPV6_FLOWINFO_MASK) != 0)
1247                 inp->in6p_flowinfo = sin6->sin6_flowinfo;
1248         in_pcbinsconnhash(inp);
1249
1250         /*
1251          * Now that no more errors can occur, change the protocol processing
1252          * port to the current thread (which is the correct thread).
1253          *
1254          * Create TCP timer message now; we are on the tcpcb's owner
1255          * CPU/thread.
1256          */
1257         tcp_create_timermsg(tp, &curthread->td_msgport);
1258
1259         /* Compute window scaling to request.  */
1260         if (tp->request_r_scale < TCP_MIN_WINSHIFT)
1261                 tp->request_r_scale = TCP_MIN_WINSHIFT;
1262         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1263             (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat) {
1264                 tp->request_r_scale++;
1265         }
1266
1267         soisconnecting(so);
1268         tcpstat.tcps_connattempt++;
1269         tp->t_state = TCPS_SYN_SENT;
1270         tcp_callout_reset(tp, tp->tt_keep, tcp_keepinit, tcp_timer_keep);
1271         tp->iss = tcp_new_isn(tp);
1272         tcp_sendseqinit(tp);
1273         if (m) {
1274                 ssb_appendstream(&so->so_snd, m);
1275                 *mp = NULL;
1276                 if (flags & PRUS_OOB)
1277                         tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
1278         }
1279
1280         /*
1281          * Close the send side of the connection after
1282          * the data is sent if flagged.
1283          */
1284         if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
1285                 socantsendmore(so);
1286                 tp = tcp_usrclosed(tp);
1287         }
1288         return (tcp_output(tp));
1289 }
1290
1291 #endif /* INET6 */
1292
1293 /*
1294  * The new sockopt interface makes it possible for us to block in the
1295  * copyin/out step (if we take a page fault).  Taking a page fault while
1296  * in a critical section is probably a Bad Thing.  (Since sockets and pcbs
1297  * both now use TSM, there probably isn't any need for this function to 
1298  * run in a critical section any more.  This needs more examination.)
1299  */
1300 void
1301 tcp_ctloutput(netmsg_t msg)
1302 {
1303         struct socket *so = msg->base.nm_so;
1304         struct sockopt *sopt = msg->ctloutput.nm_sopt;
1305         int     error, opt, optval;
1306         struct  inpcb *inp;
1307         struct  tcpcb *tp;
1308
1309         error = 0;
1310         inp = so->so_pcb;
1311         if (inp == NULL) {
1312                 error = ECONNRESET;
1313                 goto done;
1314         }
1315
1316         if (sopt->sopt_level != IPPROTO_TCP) {
1317 #ifdef INET6
1318                 if (INP_CHECK_SOCKAF(so, AF_INET6))
1319                         ip6_ctloutput_dispatch(msg);
1320                 else
1321 #endif /* INET6 */
1322                 ip_ctloutput(msg);
1323                 /* msg invalid now */
1324                 return;
1325         }
1326         tp = intotcpcb(inp);
1327
1328         switch (sopt->sopt_dir) {
1329         case SOPT_SET:
1330                 error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1331                                       sizeof optval);
1332                 if (error)
1333                         break;
1334                 switch (sopt->sopt_name) {
1335                 case TCP_FASTKEEP:
1336                         if (optval > 0) {
1337                                 if ((tp->t_flags & TF_FASTKEEP) == 0) {
1338                                         tp->t_flags |= TF_FASTKEEP;
1339                                         tcp_timer_keep_activity(tp, 0);
1340                                 }
1341                         } else {
1342                                 tp->t_flags &= ~TF_FASTKEEP;
1343                         }
1344                         break;
1345 #ifdef TCP_SIGNATURE
1346                 case TCP_SIGNATURE_ENABLE:
1347                         if (optval > 0)
1348                                 tp->t_flags |= TF_SIGNATURE;
1349                         else
1350                                 tp->t_flags &= ~TF_SIGNATURE;
1351                         break;
1352 #endif /* TCP_SIGNATURE */
1353                 case TCP_NODELAY:
1354                 case TCP_NOOPT:
1355                         switch (sopt->sopt_name) {
1356                         case TCP_NODELAY:
1357                                 opt = TF_NODELAY;
1358                                 break;
1359                         case TCP_NOOPT:
1360                                 opt = TF_NOOPT;
1361                                 break;
1362                         default:
1363                                 opt = 0; /* dead code to fool gcc */
1364                                 break;
1365                         }
1366
1367                         if (optval)
1368                                 tp->t_flags |= opt;
1369                         else
1370                                 tp->t_flags &= ~opt;
1371                         break;
1372
1373                 case TCP_NOPUSH:
1374                         if (optval)
1375                                 tp->t_flags |= TF_NOPUSH;
1376                         else {
1377                                 tp->t_flags &= ~TF_NOPUSH;
1378                                 error = tcp_output(tp);
1379                         }
1380                         break;
1381
1382                 case TCP_MAXSEG:
1383                         /*
1384                          * Must be between 0 and maxseg.  If the requested
1385                          * maxseg is too small to satisfy the desired minmss,
1386                          * pump it up (silently so sysctl modifications of
1387                          * minmss do not create unexpected program failures).
1388                          * Handle degenerate cases.
1389                          */
1390                         if (optval > 0 && optval <= tp->t_maxseg) {
1391                                 if (optval + 40 < tcp_minmss) {
1392                                         optval = tcp_minmss - 40;
1393                                         if (optval < 0)
1394                                                 optval = 1;
1395                                 }
1396                                 tp->t_maxseg = optval;
1397                         } else {
1398                                 error = EINVAL;
1399                         }
1400                         break;
1401
1402                 default:
1403                         error = ENOPROTOOPT;
1404                         break;
1405                 }
1406                 break;
1407
1408         case SOPT_GET:
1409                 switch (sopt->sopt_name) {
1410 #ifdef TCP_SIGNATURE
1411                 case TCP_SIGNATURE_ENABLE:
1412                         optval = (tp->t_flags & TF_SIGNATURE) ? 1 : 0;
1413                         break;
1414 #endif /* TCP_SIGNATURE */
1415                 case TCP_NODELAY:
1416                         optval = tp->t_flags & TF_NODELAY;
1417                         break;
1418                 case TCP_MAXSEG:
1419                         optval = tp->t_maxseg;
1420                         break;
1421                 case TCP_NOOPT:
1422                         optval = tp->t_flags & TF_NOOPT;
1423                         break;
1424                 case TCP_NOPUSH:
1425                         optval = tp->t_flags & TF_NOPUSH;
1426                         break;
1427                 default:
1428                         error = ENOPROTOOPT;
1429                         break;
1430                 }
1431                 if (error == 0)
1432                         soopt_from_kbuf(sopt, &optval, sizeof optval);
1433                 break;
1434         }
1435 done:
1436         lwkt_replymsg(&msg->lmsg, error);
1437 }
1438
1439 /*
1440  * tcp_sendspace and tcp_recvspace are the default send and receive window
1441  * sizes, respectively.  These are obsolescent (this information should
1442  * be set by the route).
1443  *
1444  * Use a default that does not require tcp window scaling to be turned
1445  * on.  Individual programs or the administrator can increase the default.
1446  */
1447 u_long  tcp_sendspace = 57344;  /* largest multiple of PAGE_SIZE < 64k */
1448 SYSCTL_INT(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW,
1449     &tcp_sendspace , 0, "Maximum outgoing TCP datagram size");
1450 u_long  tcp_recvspace = 57344;  /* largest multiple of PAGE_SIZE < 64k */
1451 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
1452     &tcp_recvspace , 0, "Maximum incoming TCP datagram size");
1453
1454 /*
1455  * Attach TCP protocol to socket, allocating internet protocol control
1456  * block, tcp control block, bufer space, and entering LISTEN state
1457  * if to accept connections.
1458  */
1459 static int
1460 tcp_attach(struct socket *so, struct pru_attach_info *ai)
1461 {
1462         struct tcpcb *tp;
1463         struct inpcb *inp;
1464         int error;
1465         int cpu;
1466 #ifdef INET6
1467         int isipv6 = INP_CHECK_SOCKAF(so, AF_INET6) != 0;
1468 #endif
1469
1470         if (so->so_snd.ssb_hiwat == 0 || so->so_rcv.ssb_hiwat == 0) {
1471                 lwkt_gettoken(&so->so_rcv.ssb_token);
1472                 error = soreserve(so, tcp_sendspace, tcp_recvspace,
1473                                   ai->sb_rlimit);
1474                 lwkt_reltoken(&so->so_rcv.ssb_token);
1475                 if (error)
1476                         return (error);
1477         }
1478         atomic_set_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE);
1479         atomic_set_int(&so->so_snd.ssb_flags, SSB_AUTOSIZE);
1480         cpu = mycpu->gd_cpuid;
1481
1482         /*
1483          * Set the default port for protocol processing. This will likely
1484          * change when we connect.
1485          */
1486         error = in_pcballoc(so, &tcbinfo[cpu]);
1487         if (error)
1488                 return (error);
1489         inp = so->so_pcb;
1490 #ifdef INET6
1491         if (isipv6) {
1492                 inp->inp_vflag |= INP_IPV6;
1493                 inp->in6p_hops = -1;    /* use kernel default */
1494         }
1495         else
1496 #endif
1497         inp->inp_vflag |= INP_IPV4;
1498         tp = tcp_newtcpcb(inp);
1499         if (tp == NULL) {
1500                 /*
1501                  * Make sure the socket is destroyed by the pcbdetach.
1502                  */
1503                 soreference(so);
1504 #ifdef INET6
1505                 if (isipv6)
1506                         in6_pcbdetach(inp);
1507                 else
1508 #endif
1509                 in_pcbdetach(inp);
1510                 sofree(so);     /* from ref above */
1511                 return (ENOBUFS);
1512         }
1513         tp->t_state = TCPS_CLOSED;
1514         return (0);
1515 }
1516
1517 /*
1518  * Initiate (or continue) disconnect.
1519  * If embryonic state, just send reset (once).
1520  * If in ``let data drain'' option and linger null, just drop.
1521  * Otherwise (hard), mark socket disconnecting and drop
1522  * current input data; switch states based on user close, and
1523  * send segment to peer (with FIN).
1524  */
1525 static struct tcpcb *
1526 tcp_disconnect(struct tcpcb *tp)
1527 {
1528         struct socket *so = tp->t_inpcb->inp_socket;
1529
1530         if (tp->t_state < TCPS_ESTABLISHED) {
1531                 tp = tcp_close(tp);
1532         } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
1533                 tp = tcp_drop(tp, 0);
1534         } else {
1535                 lwkt_gettoken(&so->so_rcv.ssb_token);
1536                 soisdisconnecting(so);
1537                 sbflush(&so->so_rcv.sb);
1538                 tp = tcp_usrclosed(tp);
1539                 if (tp)
1540                         tcp_output(tp);
1541                 lwkt_reltoken(&so->so_rcv.ssb_token);
1542         }
1543         return (tp);
1544 }
1545
1546 /*
1547  * User issued close, and wish to trail through shutdown states:
1548  * if never received SYN, just forget it.  If got a SYN from peer,
1549  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
1550  * If already got a FIN from peer, then almost done; go to LAST_ACK
1551  * state.  In all other cases, have already sent FIN to peer (e.g.
1552  * after PRU_SHUTDOWN), and just have to play tedious game waiting
1553  * for peer to send FIN or not respond to keep-alives, etc.
1554  * We can let the user exit from the close as soon as the FIN is acked.
1555  */
1556 static struct tcpcb *
1557 tcp_usrclosed(struct tcpcb *tp)
1558 {
1559
1560         switch (tp->t_state) {
1561
1562         case TCPS_CLOSED:
1563         case TCPS_LISTEN:
1564                 tp->t_state = TCPS_CLOSED;
1565                 tp = tcp_close(tp);
1566                 break;
1567
1568         case TCPS_SYN_SENT:
1569         case TCPS_SYN_RECEIVED:
1570                 tp->t_flags |= TF_NEEDFIN;
1571                 break;
1572
1573         case TCPS_ESTABLISHED:
1574                 tp->t_state = TCPS_FIN_WAIT_1;
1575                 break;
1576
1577         case TCPS_CLOSE_WAIT:
1578                 tp->t_state = TCPS_LAST_ACK;
1579                 break;
1580         }
1581         if (tp && tp->t_state >= TCPS_FIN_WAIT_2) {
1582                 soisdisconnected(tp->t_inpcb->inp_socket);
1583                 /* To prevent the connection hanging in FIN_WAIT_2 forever. */
1584                 if (tp->t_state == TCPS_FIN_WAIT_2) {
1585                         tcp_callout_reset(tp, tp->tt_2msl, tcp_maxidle,
1586                             tcp_timer_2msl);
1587                 }
1588         }
1589         return (tp);
1590 }