5f930f44c00125164257422bb1408e76cd9045e8
[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 static int      tcp_lport_extension = 1;
163
164 SYSCTL_INT(_net_inet_tcp, OID_AUTO, lportext, CTLFLAG_RW,
165     &tcp_lport_extension, 0, "");
166
167 /*
168  * TCP attaches to socket via pru_attach(), reserving space,
169  * and an internet control block.  This is likely occuring on
170  * cpu0 and may have to move later when we bind/connect.
171  */
172 static void
173 tcp_usr_attach(netmsg_t msg)
174 {
175         struct socket *so = msg->base.nm_so;
176         struct pru_attach_info *ai = msg->attach.nm_ai;
177         int error;
178         struct inpcb *inp;
179         struct tcpcb *tp = 0;
180         TCPDEBUG0;
181
182         soreference(so);
183         inp = so->so_pcb;
184         TCPDEBUG1();
185         if (inp) {
186                 error = EISCONN;
187                 goto out;
188         }
189
190         error = tcp_attach(so, ai);
191         if (error)
192                 goto out;
193
194         if ((so->so_options & SO_LINGER) && so->so_linger == 0)
195                 so->so_linger = TCP_LINGERTIME;
196         tp = sototcpcb(so);
197 out:
198         sofree(so);             /* from ref above */
199         TCPDEBUG2(PRU_ATTACH);
200         lwkt_replymsg(&msg->lmsg, error);
201 }
202
203 /*
204  * pru_detach() detaches the TCP protocol from the socket.
205  * If the protocol state is non-embryonic, then can't
206  * do this directly: have to initiate a pru_disconnect(),
207  * which may finish later; embryonic TCB's can just
208  * be discarded here.
209  */
210 static void
211 tcp_usr_detach(netmsg_t msg)
212 {
213         struct socket *so = msg->base.nm_so;
214         int error = 0;
215         struct inpcb *inp;
216         struct tcpcb *tp;
217         TCPDEBUG0;
218
219         inp = so->so_pcb;
220
221         /*
222          * If the inp is already detached it may have been due to an async
223          * close.  Just return as if no error occured.
224          *
225          * It's possible for the tcpcb (tp) to disconnect from the inp due
226          * to tcp_drop()->tcp_close() being called.  This may occur *after*
227          * the detach message has been queued so we may find a NULL tp here.
228          */
229         if (inp) {
230                 if ((tp = intotcpcb(inp)) != NULL) {
231                         TCPDEBUG1();
232                         tp = tcp_disconnect(tp);
233                         TCPDEBUG2(PRU_DETACH);
234                 }
235         }
236         lwkt_replymsg(&msg->lmsg, error);
237 }
238
239 /*
240  * NOTE: ignore_error is non-zero for certain disconnection races
241  * which we want to silently allow, otherwise close() may return
242  * an unexpected error.
243  *
244  * NOTE: The variables (msg) and (tp) are assumed.
245  */
246 #define COMMON_START(so, inp, ignore_error)                     \
247         TCPDEBUG0;                                              \
248                                                                 \
249         inp = so->so_pcb;                                       \
250         do {                                                    \
251                  if (inp == NULL) {                             \
252                         error = ignore_error ? 0 : EINVAL;      \
253                         tp = NULL;                              \
254                         goto out;                               \
255                  }                                              \
256                  tp = intotcpcb(inp);                           \
257                  TCPDEBUG1();                                   \
258         } while(0)
259
260 #define COMMON_END(req)                                         \
261         out: do {                                               \
262                 TCPDEBUG2(req);                                 \
263                 lwkt_replymsg(&msg->lmsg, error);               \
264                 return;                                         \
265         } while(0)
266
267 /*
268  * Give the socket an address.
269  */
270 static void
271 tcp_usr_bind(netmsg_t msg)
272 {
273         struct socket *so = msg->bind.base.nm_so;
274         struct sockaddr *nam = msg->bind.nm_nam;
275         struct thread *td = msg->bind.nm_td;
276         int error = 0;
277         struct inpcb *inp;
278         struct tcpcb *tp;
279         struct sockaddr_in *sinp;
280
281         COMMON_START(so, inp, 0);
282
283         /*
284          * Must check for multicast addresses and disallow binding
285          * to them.
286          */
287         sinp = (struct sockaddr_in *)nam;
288         if (sinp->sin_family == AF_INET &&
289             IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
290                 error = EAFNOSUPPORT;
291                 goto out;
292         }
293         error = in_pcbbind(inp, nam, td);
294         if (error)
295                 goto out;
296         COMMON_END(PRU_BIND);
297
298 }
299
300 #ifdef INET6
301
302 static void
303 tcp6_usr_bind(netmsg_t msg)
304 {
305         struct socket *so = msg->bind.base.nm_so;
306         struct sockaddr *nam = msg->bind.nm_nam;
307         struct thread *td = msg->bind.nm_td;
308         int error = 0;
309         struct inpcb *inp;
310         struct tcpcb *tp;
311         struct sockaddr_in6 *sin6p;
312
313         COMMON_START(so, inp, 0);
314
315         /*
316          * Must check for multicast addresses and disallow binding
317          * to them.
318          */
319         sin6p = (struct sockaddr_in6 *)nam;
320         if (sin6p->sin6_family == AF_INET6 &&
321             IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
322                 error = EAFNOSUPPORT;
323                 goto out;
324         }
325         inp->inp_vflag &= ~INP_IPV4;
326         inp->inp_vflag |= INP_IPV6;
327         if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
328                 if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr))
329                         inp->inp_vflag |= INP_IPV4;
330                 else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
331                         struct sockaddr_in sin;
332
333                         in6_sin6_2_sin(&sin, sin6p);
334                         inp->inp_vflag |= INP_IPV4;
335                         inp->inp_vflag &= ~INP_IPV6;
336                         error = in_pcbbind(inp, (struct sockaddr *)&sin, td);
337                         goto out;
338                 }
339         }
340         error = in6_pcbbind(inp, nam, td);
341         if (error)
342                 goto out;
343         COMMON_END(PRU_BIND);
344 }
345 #endif /* INET6 */
346
347 #ifdef SMP
348
349 struct netmsg_inswildcard {
350         struct netmsg_base      base;
351         struct inpcb            *nm_inp;
352 };
353
354 static void
355 in_pcbinswildcardhash_handler(netmsg_t msg)
356 {
357         struct netmsg_inswildcard *nm = (struct netmsg_inswildcard *)msg;
358         int cpu = mycpuid, nextcpu;
359
360         in_pcbinswildcardhash_oncpu(nm->nm_inp, &tcbinfo[cpu]);
361
362         nextcpu = cpu + 1;
363         if (nextcpu < ncpus2)
364                 lwkt_forwardmsg(cpu_portfn(nextcpu), &nm->base.lmsg);
365         else
366                 lwkt_replymsg(&nm->base.lmsg, 0);
367 }
368
369 #endif
370
371 /*
372  * Prepare to accept connections.
373  */
374 static void
375 tcp_usr_listen(netmsg_t msg)
376 {
377         struct socket *so = msg->listen.base.nm_so;
378         struct thread *td = msg->listen.nm_td;
379         int error = 0;
380         struct inpcb *inp;
381         struct tcpcb *tp;
382 #ifdef SMP
383         struct netmsg_inswildcard nm;
384 #endif
385
386         COMMON_START(so, inp, 0);
387
388         if (tp->t_flags & TF_LISTEN)
389                 goto out;
390
391         if (inp->inp_lport == 0) {
392                 error = in_pcbbind(inp, NULL, td);
393                 if (error)
394                         goto out;
395         }
396
397         tp->t_state = TCPS_LISTEN;
398         tp->t_flags |= TF_LISTEN;
399         tp->tt_msg = NULL; /* Catch any invalid timer usage */
400
401 #ifdef SMP
402         if (ncpus > 1) {
403                 /*
404                  * We have to set the flag because we can't have other cpus
405                  * messing with our inp's flags.
406                  */
407                 KASSERT(!(inp->inp_flags & INP_CONNECTED),
408                         ("already on connhash\n"));
409                 KASSERT(!(inp->inp_flags & INP_WILDCARD),
410                         ("already on wildcardhash\n"));
411                 KASSERT(!(inp->inp_flags & INP_WILDCARD_MP),
412                         ("already on MP wildcardhash\n"));
413                 inp->inp_flags |= INP_WILDCARD_MP;
414
415                 KKASSERT(so->so_port == cpu_portfn(0));
416                 KKASSERT(&curthread->td_msgport == cpu_portfn(0));
417                 KKASSERT(inp->inp_pcbinfo == &tcbinfo[0]);
418
419                 netmsg_init(&nm.base, NULL, &curthread->td_msgport,
420                             MSGF_PRIORITY, in_pcbinswildcardhash_handler);
421                 nm.nm_inp = inp;
422                 lwkt_domsg(cpu_portfn(1), &nm.base.lmsg, 0);
423         }
424 #endif
425         in_pcbinswildcardhash(inp);
426         COMMON_END(PRU_LISTEN);
427 }
428
429 #ifdef INET6
430
431 static void
432 tcp6_usr_listen(netmsg_t msg)
433 {
434         struct socket *so = msg->listen.base.nm_so;
435         struct thread *td = msg->listen.nm_td;
436         int error = 0;
437         struct inpcb *inp;
438         struct tcpcb *tp;
439 #ifdef SMP
440         struct netmsg_inswildcard nm;
441 #endif
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 #ifdef SMP
463         if (ncpus > 1) {
464                 /*
465                  * We have to set the flag because we can't have other cpus
466                  * messing with our inp's flags.
467                  */
468                 KASSERT(!(inp->inp_flags & INP_CONNECTED),
469                         ("already on connhash\n"));
470                 KASSERT(!(inp->inp_flags & INP_WILDCARD),
471                         ("already on wildcardhash\n"));
472                 KASSERT(!(inp->inp_flags & INP_WILDCARD_MP),
473                         ("already on MP wildcardhash\n"));
474                 inp->inp_flags |= INP_WILDCARD_MP;
475
476                 KKASSERT(so->so_port == cpu_portfn(0));
477                 KKASSERT(&curthread->td_msgport == cpu_portfn(0));
478                 KKASSERT(inp->inp_pcbinfo == &tcbinfo[0]);
479
480                 netmsg_init(&nm.base, NULL, &curthread->td_msgport,
481                             MSGF_PRIORITY, in_pcbinswildcardhash_handler);
482                 nm.nm_inp = inp;
483                 lwkt_domsg(cpu_portfn(1), &nm.base.lmsg, 0);
484         }
485 #endif
486         in_pcbinswildcardhash(inp);
487         COMMON_END(PRU_LISTEN);
488 }
489 #endif /* INET6 */
490
491 /*
492  * Initiate connection to peer.
493  * Create a template for use in transmissions on this connection.
494  * Enter SYN_SENT state, and mark socket as connecting.
495  * Start keep-alive timer, and seed output sequence space.
496  * Send initial segment on connection.
497  */
498 static void
499 tcp_usr_connect(netmsg_t msg)
500 {
501         struct socket *so = msg->connect.base.nm_so;
502         struct sockaddr *nam = msg->connect.nm_nam;
503         struct thread *td = msg->connect.nm_td;
504         int error = 0;
505         struct inpcb *inp;
506         struct tcpcb *tp;
507         struct sockaddr_in *sinp;
508
509         COMMON_START(so, inp, 0);
510
511         /*
512          * Must disallow TCP ``connections'' to multicast addresses.
513          */
514         sinp = (struct sockaddr_in *)nam;
515         if (sinp->sin_family == AF_INET
516             && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
517                 error = EAFNOSUPPORT;
518                 goto out;
519         }
520
521         if (!prison_remote_ip(td, (struct sockaddr*)sinp)) {
522                 error = EAFNOSUPPORT; /* IPv6 only jail */
523                 goto out;
524         }
525
526         tcp_connect(msg);
527         /* msg is invalid now */
528         return;
529 out:
530         if (msg->connect.nm_m) {
531                 m_freem(msg->connect.nm_m);
532                 msg->connect.nm_m = NULL;
533         }
534         lwkt_replymsg(&msg->lmsg, error);
535 }
536
537 #ifdef INET6
538
539 static void
540 tcp6_usr_connect(netmsg_t msg)
541 {
542         struct socket *so = msg->connect.base.nm_so;
543         struct sockaddr *nam = msg->connect.nm_nam;
544         struct thread *td = msg->connect.nm_td;
545         int error = 0;
546         struct inpcb *inp;
547         struct tcpcb *tp;
548         struct sockaddr_in6 *sin6p;
549
550         COMMON_START(so, inp, 0);
551
552         /*
553          * Must disallow TCP ``connections'' to multicast addresses.
554          */
555         sin6p = (struct sockaddr_in6 *)nam;
556         if (sin6p->sin6_family == AF_INET6
557             && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
558                 error = EAFNOSUPPORT;
559                 goto out;
560         }
561
562         if (!prison_remote_ip(td, nam)) {
563                 error = EAFNOSUPPORT; /* IPv4 only jail */
564                 goto out;
565         }
566
567         if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
568                 struct sockaddr_in *sinp;
569
570                 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
571                         error = EINVAL;
572                         goto out;
573                 }
574                 sinp = kmalloc(sizeof(*sinp), M_LWKTMSG, M_INTWAIT);
575                 in6_sin6_2_sin(sinp, sin6p);
576                 inp->inp_vflag |= INP_IPV4;
577                 inp->inp_vflag &= ~INP_IPV6;
578                 msg->connect.nm_nam = (struct sockaddr *)sinp;
579                 msg->connect.nm_reconnect |= NMSG_RECONNECT_NAMALLOC;
580                 tcp_connect(msg);
581                 /* msg is invalid now */
582                 return;
583         }
584         inp->inp_vflag &= ~INP_IPV4;
585         inp->inp_vflag |= INP_IPV6;
586         inp->inp_inc.inc_isipv6 = 1;
587
588         msg->connect.nm_reconnect |= NMSG_RECONNECT_FALLBACK;
589         tcp6_connect(msg);
590         /* msg is invalid now */
591         return;
592 out:
593         if (msg->connect.nm_m) {
594                 m_freem(msg->connect.nm_m);
595                 msg->connect.nm_m = NULL;
596         }
597         lwkt_replymsg(&msg->lmsg, error);
598 }
599
600 #endif /* INET6 */
601
602 /*
603  * Initiate disconnect from peer.
604  * If connection never passed embryonic stage, just drop;
605  * else if don't need to let data drain, then can just drop anyways,
606  * else have to begin TCP shutdown process: mark socket disconnecting,
607  * drain unread data, state switch to reflect user close, and
608  * send segment (e.g. FIN) to peer.  Socket will be really disconnected
609  * when peer sends FIN and acks ours.
610  *
611  * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
612  */
613 static void
614 tcp_usr_disconnect(netmsg_t msg)
615 {
616         struct socket *so = msg->disconnect.base.nm_so;
617         int error = 0;
618         struct inpcb *inp;
619         struct tcpcb *tp;
620
621         COMMON_START(so, inp, 1);
622         tp = tcp_disconnect(tp);
623         COMMON_END(PRU_DISCONNECT);
624 }
625
626 /*
627  * Accept a connection.  Essentially all the work is
628  * done at higher levels; just return the address
629  * of the peer, storing through addr.
630  */
631 static void
632 tcp_usr_accept(netmsg_t msg)
633 {
634         struct socket *so = msg->accept.base.nm_so;
635         struct sockaddr **nam = msg->accept.nm_nam;
636         int error = 0;
637         struct inpcb *inp;
638         struct tcpcb *tp = NULL;
639         TCPDEBUG0;
640
641         inp = so->so_pcb;
642         if (so->so_state & SS_ISDISCONNECTED) {
643                 error = ECONNABORTED;
644                 goto out;
645         }
646         if (inp == 0) {
647                 error = EINVAL;
648                 goto out;
649         }
650
651         tp = intotcpcb(inp);
652         TCPDEBUG1();
653         in_setpeeraddr(so, nam);
654         COMMON_END(PRU_ACCEPT);
655 }
656
657 #ifdef INET6
658 static void
659 tcp6_usr_accept(netmsg_t msg)
660 {
661         struct socket *so = msg->accept.base.nm_so;
662         struct sockaddr **nam = msg->accept.nm_nam;
663         int error = 0;
664         struct inpcb *inp;
665         struct tcpcb *tp = NULL;
666         TCPDEBUG0;
667
668         inp = so->so_pcb;
669
670         if (so->so_state & SS_ISDISCONNECTED) {
671                 error = ECONNABORTED;
672                 goto out;
673         }
674         if (inp == 0) {
675                 error = EINVAL;
676                 goto out;
677         }
678         tp = intotcpcb(inp);
679         TCPDEBUG1();
680         in6_mapped_peeraddr(so, nam);
681         COMMON_END(PRU_ACCEPT);
682 }
683 #endif /* INET6 */
684 /*
685  * Mark the connection as being incapable of further output.
686  */
687 static void
688 tcp_usr_shutdown(netmsg_t msg)
689 {
690         struct socket *so = msg->shutdown.base.nm_so;
691         int error = 0;
692         struct inpcb *inp;
693         struct tcpcb *tp;
694
695         COMMON_START(so, inp, 0);
696         socantsendmore(so);
697         tp = tcp_usrclosed(tp);
698         if (tp)
699                 error = tcp_output(tp);
700         COMMON_END(PRU_SHUTDOWN);
701 }
702
703 /*
704  * After a receive, possibly send window update to peer.
705  */
706 static void
707 tcp_usr_rcvd(netmsg_t msg)
708 {
709         struct socket *so = msg->rcvd.base.nm_so;
710         int error = 0;
711         struct inpcb *inp;
712         struct tcpcb *tp;
713
714         COMMON_START(so, inp, 0);
715         tcp_output(tp);
716         COMMON_END(PRU_RCVD);
717 }
718
719 /*
720  * Do a send by putting data in output queue and updating urgent
721  * marker if URG set.  Possibly send more data.  Unlike the other
722  * pru_*() routines, the mbuf chains are our responsibility.  We
723  * must either enqueue them or free them.  The other pru_* routines
724  * generally are caller-frees.
725  */
726 static void
727 tcp_usr_send(netmsg_t msg)
728 {
729         struct socket *so = msg->send.base.nm_so;
730         int flags = msg->send.nm_flags;
731         struct mbuf *m = msg->send.nm_m;
732         struct sockaddr *nam = msg->send.nm_addr;
733         struct mbuf *control = msg->send.nm_control;
734         struct thread *td = msg->send.nm_td;
735         int error = 0;
736         struct inpcb *inp;
737         struct tcpcb *tp;
738 #ifdef INET6
739         int isipv6;
740 #endif
741         TCPDEBUG0;
742
743         inp = so->so_pcb;
744
745         if (inp == NULL) {
746                 /*
747                  * OOPS! we lost a race, the TCP session got reset after
748                  * we checked SS_CANTSENDMORE, eg: while doing uiomove or a
749                  * network interrupt in the non-critical section of sosend().
750                  */
751                 m_freem(m);
752                 if (control)
753                         m_freem(control);
754                 error = ECONNRESET;     /* XXX EPIPE? */
755                 tp = NULL;
756                 TCPDEBUG1();
757                 goto out;
758         }
759 #ifdef INET6
760         isipv6 = nam && nam->sa_family == AF_INET6;
761 #endif /* INET6 */
762         tp = intotcpcb(inp);
763         TCPDEBUG1();
764         if (control) {
765                 /* TCP doesn't do control messages (rights, creds, etc) */
766                 if (control->m_len) {
767                         m_freem(control);
768                         m_freem(m);
769                         error = EINVAL;
770                         goto out;
771                 }
772                 m_freem(control);       /* empty control, just free it */
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
786         /*
787          * Do implied connect if not yet connected.  Any data sent
788          * with the connect is handled by tcp_connect() and friends.
789          *
790          * NOTE!  PROTOCOL THREAD MAY BE CHANGED BY THE CONNECT!
791          */
792         if (nam && tp->t_state < TCPS_SYN_SENT) {
793                 kprintf("implied fallback\n");
794                 msg->connect.nm_nam = nam;
795                 msg->connect.nm_td = td;
796                 msg->connect.nm_m = m;
797                 msg->connect.nm_flags = flags;
798                 msg->connect.nm_reconnect = NMSG_RECONNECT_FALLBACK;
799 #ifdef INET6
800                 if (isipv6)
801                         tcp6_connect(msg);
802                 else
803 #endif /* INET6 */
804                         tcp_connect(msg);
805                 /* msg invalid now */
806                 return;
807         }
808
809         /*
810          * Pump the data into the socket.
811          */
812         if (m)
813                 ssb_appendstream(&so->so_snd, m);
814         if (flags & PRUS_OOB) {
815                 /*
816                  * According to RFC961 (Assigned Protocols),
817                  * the urgent pointer points to the last octet
818                  * of urgent data.  We continue, however,
819                  * to consider it to indicate the first octet
820                  * of data past the urgent section.
821                  * Otherwise, snd_up should be one lower.
822                  */
823                 tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
824                 tp->t_flags |= TF_FORCE;
825                 error = tcp_output(tp);
826                 tp->t_flags &= ~TF_FORCE;
827         } else {
828                 if (flags & PRUS_EOF) {
829                         /*
830                          * Close the send side of the connection after
831                          * the data is sent.
832                          */
833                         socantsendmore(so);
834                         tp = tcp_usrclosed(tp);
835                 }
836                 if (tp != NULL) {
837                         if (flags & PRUS_MORETOCOME)
838                                 tp->t_flags |= TF_MORETOCOME;
839                         error = tcp_output(tp);
840                         if (flags & PRUS_MORETOCOME)
841                                 tp->t_flags &= ~TF_MORETOCOME;
842                 }
843         }
844         COMMON_END((flags & PRUS_OOB) ? PRU_SENDOOB :
845                    ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
846 }
847
848 /*
849  * NOTE: (so) is referenced from soabort*() and netmsg_pru_abort()
850  *       will sofree() it when we return.
851  */
852 static void
853 tcp_usr_abort(netmsg_t msg)
854 {
855         struct socket *so = msg->abort.base.nm_so;
856         int error = 0;
857         struct inpcb *inp;
858         struct tcpcb *tp;
859
860         COMMON_START(so, inp, 1);
861         tp = tcp_drop(tp, ECONNABORTED);
862         COMMON_END(PRU_ABORT);
863 }
864
865 /*
866  * Receive out-of-band data.
867  */
868 static void
869 tcp_usr_rcvoob(netmsg_t msg)
870 {
871         struct socket *so = msg->rcvoob.base.nm_so;
872         struct mbuf *m = msg->rcvoob.nm_m;
873         int flags = msg->rcvoob.nm_flags;
874         int error = 0;
875         struct inpcb *inp;
876         struct tcpcb *tp;
877
878         COMMON_START(so, inp, 0);
879         if ((so->so_oobmark == 0 &&
880              (so->so_state & SS_RCVATMARK) == 0) ||
881             so->so_options & SO_OOBINLINE ||
882             tp->t_oobflags & TCPOOB_HADDATA) {
883                 error = EINVAL;
884                 goto out;
885         }
886         if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
887                 error = EWOULDBLOCK;
888                 goto out;
889         }
890         m->m_len = 1;
891         *mtod(m, caddr_t) = tp->t_iobc;
892         if ((flags & MSG_PEEK) == 0)
893                 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
894         COMMON_END(PRU_RCVOOB);
895 }
896
897 /* xxx - should be const */
898 struct pr_usrreqs tcp_usrreqs = {
899         .pru_abort = tcp_usr_abort,
900         .pru_accept = tcp_usr_accept,
901         .pru_attach = tcp_usr_attach,
902         .pru_bind = tcp_usr_bind,
903         .pru_connect = tcp_usr_connect,
904         .pru_connect2 = pr_generic_notsupp,
905         .pru_control = in_control_dispatch,
906         .pru_detach = tcp_usr_detach,
907         .pru_disconnect = tcp_usr_disconnect,
908         .pru_listen = tcp_usr_listen,
909         .pru_peeraddr = in_setpeeraddr_dispatch,
910         .pru_rcvd = tcp_usr_rcvd,
911         .pru_rcvoob = tcp_usr_rcvoob,
912         .pru_send = tcp_usr_send,
913         .pru_sense = pru_sense_null,
914         .pru_shutdown = tcp_usr_shutdown,
915         .pru_sockaddr = in_setsockaddr_dispatch,
916         .pru_sosend = sosend,
917         .pru_soreceive = soreceive
918 };
919
920 #ifdef INET6
921 struct pr_usrreqs tcp6_usrreqs = {
922         .pru_abort = tcp_usr_abort,
923         .pru_accept = tcp6_usr_accept,
924         .pru_attach = tcp_usr_attach,
925         .pru_bind = tcp6_usr_bind,
926         .pru_connect = tcp6_usr_connect,
927         .pru_connect2 = pr_generic_notsupp,
928         .pru_control = in6_control_dispatch,
929         .pru_detach = tcp_usr_detach,
930         .pru_disconnect = tcp_usr_disconnect,
931         .pru_listen = tcp6_usr_listen,
932         .pru_peeraddr = in6_mapped_peeraddr_dispatch,
933         .pru_rcvd = tcp_usr_rcvd,
934         .pru_rcvoob = tcp_usr_rcvoob,
935         .pru_send = tcp_usr_send,
936         .pru_sense = pru_sense_null,
937         .pru_shutdown = tcp_usr_shutdown,
938         .pru_sockaddr = in6_mapped_sockaddr_dispatch,
939         .pru_sosend = sosend,
940         .pru_soreceive = soreceive
941 };
942 #endif /* INET6 */
943
944 static int
945 tcp_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf *m,
946                   struct sockaddr_in *sin, struct sockaddr_in *if_sin)
947 {
948         struct inpcb *inp = tp->t_inpcb, *oinp;
949         struct socket *so = inp->inp_socket;
950         struct route *ro = &inp->inp_route;
951
952         oinp = in_pcblookup_hash(&tcbinfo[mycpu->gd_cpuid],
953                                  sin->sin_addr, sin->sin_port,
954                                  (inp->inp_laddr.s_addr != INADDR_ANY ?
955                                   inp->inp_laddr : if_sin->sin_addr),
956                                 inp->inp_lport, 0, NULL);
957         if (oinp != NULL) {
958                 m_freem(m);
959                 return (EADDRINUSE);
960         }
961         if (inp->inp_laddr.s_addr == INADDR_ANY)
962                 inp->inp_laddr = if_sin->sin_addr;
963         inp->inp_faddr = sin->sin_addr;
964         inp->inp_fport = sin->sin_port;
965         inp->inp_cpcbinfo = &tcbinfo[mycpu->gd_cpuid];
966         in_pcbinsconnhash(inp);
967
968         /*
969          * We are now on the inpcb's owner CPU, if the cached route was
970          * freed because the rtentry's owner CPU is not the current CPU
971          * (e.g. in tcp_connect()), then we try to reallocate it here with
972          * the hope that a rtentry may be cloned from a RTF_PRCLONING
973          * rtentry.
974          */
975         if (!(inp->inp_socket->so_options & SO_DONTROUTE) && /*XXX*/
976             ro->ro_rt == NULL) {
977                 bzero(&ro->ro_dst, sizeof(struct sockaddr_in));
978                 ro->ro_dst.sa_family = AF_INET;
979                 ro->ro_dst.sa_len = sizeof(struct sockaddr_in);
980                 ((struct sockaddr_in *)&ro->ro_dst)->sin_addr =
981                         sin->sin_addr;
982                 rtalloc(ro);
983         }
984
985         /*
986          * Now that no more errors can occur, change the protocol processing
987          * port to the current thread (which is the correct thread).
988          *
989          * Create TCP timer message now; we are on the tcpcb's owner
990          * CPU/thread.
991          */
992         tcp_create_timermsg(tp, &curthread->td_msgport);
993
994         /*
995          * Compute window scaling to request.  Use a larger scaling then
996          * needed for the initial receive buffer in case the receive buffer
997          * gets expanded.
998          */
999         if (tp->request_r_scale < TCP_MIN_WINSHIFT)
1000                 tp->request_r_scale = TCP_MIN_WINSHIFT;
1001         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1002                (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat
1003         ) {
1004                 tp->request_r_scale++;
1005         }
1006
1007         soisconnecting(so);
1008         tcpstat.tcps_connattempt++;
1009         tp->t_state = TCPS_SYN_SENT;
1010         tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
1011         tp->iss = tcp_new_isn(tp);
1012         tcp_sendseqinit(tp);
1013         if (m) {
1014                 ssb_appendstream(&so->so_snd, m);
1015                 m = NULL;
1016                 if (flags & PRUS_OOB)
1017                         tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
1018         }
1019
1020         /*
1021          * Close the send side of the connection after
1022          * the data is sent if flagged.
1023          */
1024         if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
1025                 socantsendmore(so);
1026                 tp = tcp_usrclosed(tp);
1027         }
1028         return (tcp_output(tp));
1029 }
1030
1031 /*
1032  * Common subroutine to open a TCP connection to remote host specified
1033  * by struct sockaddr_in in mbuf *nam.  Call in_pcbbind to assign a local
1034  * port number if needed.  Call in_pcbladdr to do the routing and to choose
1035  * a local host address (interface).
1036  * Initialize connection parameters and enter SYN-SENT state.
1037  */
1038 static void
1039 tcp_connect(netmsg_t msg)
1040 {
1041         struct socket *so = msg->connect.base.nm_so;
1042         struct sockaddr *nam = msg->connect.nm_nam;
1043         struct thread *td = msg->connect.nm_td;
1044         struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1045         struct sockaddr_in *if_sin;
1046         struct inpcb *inp;
1047         struct tcpcb *tp;
1048         int error, calc_laddr = 1;
1049 #ifdef SMP
1050         lwkt_port_t port;
1051 #endif
1052
1053         COMMON_START(so, inp, 0);
1054
1055         /*
1056          * Reconnect our pcb if we have to
1057          */
1058         if (msg->connect.nm_reconnect & NMSG_RECONNECT_RECONNECT) {
1059                 msg->connect.nm_reconnect &= ~NMSG_RECONNECT_RECONNECT;
1060                 in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1061         }
1062
1063         /*
1064          * Bind if we have to
1065          */
1066         if (inp->inp_lport == 0) {
1067                 if (tcp_lport_extension) {
1068                         KKASSERT(inp->inp_laddr.s_addr == INADDR_ANY);
1069
1070                         error = in_pcbladdr(inp, nam, &if_sin, td);
1071                         if (error)
1072                                 goto out;
1073                         inp->inp_laddr.s_addr = if_sin->sin_addr.s_addr;
1074
1075                         error = in_pcbconn_bind(inp, nam, td);
1076                         if (error)
1077                                 goto out;
1078
1079                         calc_laddr = 0;
1080                 } else {
1081                         error = in_pcbbind(inp, NULL, td);
1082                         if (error)
1083                                 goto out;
1084                 }
1085         }
1086
1087         if (calc_laddr) {
1088                 /*
1089                  * Calculate the correct protocol processing thread.  The
1090                  * connect operation must run there.  Set the forwarding
1091                  * port before we forward the message or it will get bounced
1092                  * right back to us.
1093                  */
1094                 error = in_pcbladdr(inp, nam, &if_sin, td);
1095                 if (error)
1096                         goto out;
1097         }
1098         KKASSERT(inp->inp_socket == so);
1099
1100 #ifdef SMP
1101         port = tcp_addrport(sin->sin_addr.s_addr, sin->sin_port,
1102                             (inp->inp_laddr.s_addr ?
1103                              inp->inp_laddr.s_addr : if_sin->sin_addr.s_addr),
1104                             inp->inp_lport);
1105
1106         if (port != &curthread->td_msgport) {
1107                 struct route *ro = &inp->inp_route;
1108
1109                 /*
1110                  * in_pcbladdr() may have allocated a route entry for us
1111                  * on the current CPU, but we need a route entry on the
1112                  * inpcb's owner CPU, so free it here.
1113                  */
1114                 if (ro->ro_rt != NULL)
1115                         RTFREE(ro->ro_rt);
1116                 bzero(ro, sizeof(*ro));
1117
1118                 /*
1119                  * We are moving the protocol processing port the socket
1120                  * is on, we have to unlink here and re-link on the
1121                  * target cpu.
1122                  */
1123                 in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1124                 sosetport(so, port);
1125                 msg->connect.nm_reconnect |= NMSG_RECONNECT_RECONNECT;
1126                 msg->connect.base.nm_dispatch = tcp_connect;
1127
1128                 lwkt_forwardmsg(port, &msg->connect.base.lmsg);
1129                 /* msg invalid now */
1130                 return;
1131         }
1132 #else
1133         KKASSERT(so->so_port == &curthread->td_msgport);
1134 #endif
1135         error = tcp_connect_oncpu(tp, msg->connect.nm_flags,
1136                                   msg->connect.nm_m, sin, if_sin);
1137         msg->connect.nm_m = NULL;
1138 out:
1139         if (msg->connect.nm_m) {
1140                 m_freem(msg->connect.nm_m);
1141                 msg->connect.nm_m = NULL;
1142         }
1143         if (msg->connect.nm_reconnect & NMSG_RECONNECT_NAMALLOC) {
1144                 kfree(msg->connect.nm_nam, M_LWKTMSG);
1145                 msg->connect.nm_nam = NULL;
1146         }
1147         lwkt_replymsg(&msg->connect.base.lmsg, error);
1148         /* msg invalid now */
1149 }
1150
1151 #ifdef INET6
1152
1153 static void
1154 tcp6_connect(netmsg_t msg)
1155 {
1156         struct tcpcb *tp;
1157         struct socket *so = msg->connect.base.nm_so;
1158         struct sockaddr *nam = msg->connect.nm_nam;
1159         struct thread *td = msg->connect.nm_td;
1160         struct inpcb *inp;
1161         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam;
1162         struct in6_addr *addr6;
1163 #ifdef SMP
1164         lwkt_port_t port;
1165 #endif
1166         int error;
1167
1168         COMMON_START(so, inp, 0);
1169
1170         /*
1171          * Reconnect our pcb if we have to
1172          */
1173         if (msg->connect.nm_reconnect & NMSG_RECONNECT_RECONNECT) {
1174                 msg->connect.nm_reconnect &= ~NMSG_RECONNECT_RECONNECT;
1175                 in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1176         }
1177
1178         /*
1179          * Bind if we have to
1180          */
1181         if (inp->inp_lport == 0) {
1182                 error = in6_pcbbind(inp, NULL, td);
1183                 if (error)
1184                         goto out;
1185         }
1186
1187         /*
1188          * Cannot simply call in_pcbconnect, because there might be an
1189          * earlier incarnation of this same connection still in
1190          * TIME_WAIT state, creating an ADDRINUSE error.
1191          */
1192         error = in6_pcbladdr(inp, nam, &addr6, td);
1193         if (error)
1194                 goto out;
1195
1196 #ifdef SMP
1197         port = tcp6_addrport(); /* XXX hack for now, always cpu0 */
1198
1199         if (port != &curthread->td_msgport) {
1200                 struct route *ro = &inp->inp_route;
1201
1202                 /*
1203                  * in_pcbladdr() may have allocated a route entry for us
1204                  * on the current CPU, but we need a route entry on the
1205                  * inpcb's owner CPU, so free it here.
1206                  */
1207                 if (ro->ro_rt != NULL)
1208                         RTFREE(ro->ro_rt);
1209                 bzero(ro, sizeof(*ro));
1210
1211                 in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1212                 sosetport(so, port);
1213                 msg->connect.nm_reconnect |= NMSG_RECONNECT_RECONNECT;
1214                 msg->connect.base.nm_dispatch = tcp6_connect;
1215
1216                 lwkt_forwardmsg(port, &msg->connect.base.lmsg);
1217                 /* msg invalid now */
1218                 return;
1219         }
1220 #endif
1221         error = tcp6_connect_oncpu(tp, msg->connect.nm_flags,
1222                                    &msg->connect.nm_m, sin6, addr6);
1223         /* nm_m may still be intact */
1224 out:
1225         if (error && (msg->connect.nm_reconnect & NMSG_RECONNECT_FALLBACK)) {
1226                 tcp_connect(msg);
1227                 /* msg invalid now */
1228         } else {
1229                 if (msg->connect.nm_m) {
1230                         m_freem(msg->connect.nm_m);
1231                         msg->connect.nm_m = NULL;
1232                 }
1233                 if (msg->connect.nm_reconnect & NMSG_RECONNECT_NAMALLOC) {
1234                         kfree(msg->connect.nm_nam, M_LWKTMSG);
1235                         msg->connect.nm_nam = NULL;
1236                 }
1237                 lwkt_replymsg(&msg->connect.base.lmsg, error);
1238                 /* msg invalid now */
1239         }
1240 }
1241
1242 static int
1243 tcp6_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf **mp,
1244                    struct sockaddr_in6 *sin6, struct in6_addr *addr6)
1245 {
1246         struct mbuf *m = *mp;
1247         struct inpcb *inp = tp->t_inpcb;
1248         struct socket *so = inp->inp_socket;
1249         struct inpcb *oinp;
1250
1251         /*
1252          * Cannot simply call in_pcbconnect, because there might be an
1253          * earlier incarnation of this same connection still in
1254          * TIME_WAIT state, creating an ADDRINUSE error.
1255          */
1256         oinp = in6_pcblookup_hash(inp->inp_cpcbinfo,
1257                                   &sin6->sin6_addr, sin6->sin6_port,
1258                                   (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) ?
1259                                       addr6 : &inp->in6p_laddr),
1260                                   inp->inp_lport,  0, NULL);
1261         if (oinp)
1262                 return (EADDRINUSE);
1263
1264         if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
1265                 inp->in6p_laddr = *addr6;
1266         inp->in6p_faddr = sin6->sin6_addr;
1267         inp->inp_fport = sin6->sin6_port;
1268         if ((sin6->sin6_flowinfo & IPV6_FLOWINFO_MASK) != 0)
1269                 inp->in6p_flowinfo = sin6->sin6_flowinfo;
1270         in_pcbinsconnhash(inp);
1271
1272         /*
1273          * Now that no more errors can occur, change the protocol processing
1274          * port to the current thread (which is the correct thread).
1275          *
1276          * Create TCP timer message now; we are on the tcpcb's owner
1277          * CPU/thread.
1278          */
1279         tcp_create_timermsg(tp, &curthread->td_msgport);
1280
1281         /* Compute window scaling to request.  */
1282         if (tp->request_r_scale < TCP_MIN_WINSHIFT)
1283                 tp->request_r_scale = TCP_MIN_WINSHIFT;
1284         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1285             (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat) {
1286                 tp->request_r_scale++;
1287         }
1288
1289         soisconnecting(so);
1290         tcpstat.tcps_connattempt++;
1291         tp->t_state = TCPS_SYN_SENT;
1292         tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
1293         tp->iss = tcp_new_isn(tp);
1294         tcp_sendseqinit(tp);
1295         if (m) {
1296                 ssb_appendstream(&so->so_snd, m);
1297                 *mp = NULL;
1298                 if (flags & PRUS_OOB)
1299                         tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
1300         }
1301
1302         /*
1303          * Close the send side of the connection after
1304          * the data is sent if flagged.
1305          */
1306         if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
1307                 socantsendmore(so);
1308                 tp = tcp_usrclosed(tp);
1309         }
1310         return (tcp_output(tp));
1311 }
1312
1313 #endif /* INET6 */
1314
1315 /*
1316  * The new sockopt interface makes it possible for us to block in the
1317  * copyin/out step (if we take a page fault).  Taking a page fault while
1318  * in a critical section is probably a Bad Thing.  (Since sockets and pcbs
1319  * both now use TSM, there probably isn't any need for this function to 
1320  * run in a critical section any more.  This needs more examination.)
1321  */
1322 void
1323 tcp_ctloutput(netmsg_t msg)
1324 {
1325         struct socket *so = msg->base.nm_so;
1326         struct sockopt *sopt = msg->ctloutput.nm_sopt;
1327         int     error, opt, optval, opthz;
1328         struct  inpcb *inp;
1329         struct  tcpcb *tp;
1330
1331         error = 0;
1332         inp = so->so_pcb;
1333         if (inp == NULL) {
1334                 error = ECONNRESET;
1335                 goto done;
1336         }
1337
1338         if (sopt->sopt_level != IPPROTO_TCP) {
1339 #ifdef INET6
1340                 if (INP_CHECK_SOCKAF(so, AF_INET6))
1341                         ip6_ctloutput_dispatch(msg);
1342                 else
1343 #endif /* INET6 */
1344                 ip_ctloutput(msg);
1345                 /* msg invalid now */
1346                 return;
1347         }
1348         tp = intotcpcb(inp);
1349
1350         switch (sopt->sopt_dir) {
1351         case SOPT_SET:
1352                 error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1353                                       sizeof optval);
1354                 if (error)
1355                         break;
1356                 switch (sopt->sopt_name) {
1357                 case TCP_FASTKEEP:
1358                         if (optval > 0) {
1359                                 if ((tp->t_flags & TF_FASTKEEP) == 0) {
1360                                         tp->t_flags |= TF_FASTKEEP;
1361                                         tcp_timer_keep_activity(tp, 0);
1362                                 }
1363                         } else {
1364                                 tp->t_flags &= ~TF_FASTKEEP;
1365                         }
1366                         break;
1367 #ifdef TCP_SIGNATURE
1368                 case TCP_SIGNATURE_ENABLE:
1369                         if (optval > 0)
1370                                 tp->t_flags |= TF_SIGNATURE;
1371                         else
1372                                 tp->t_flags &= ~TF_SIGNATURE;
1373                         break;
1374 #endif /* TCP_SIGNATURE */
1375                 case TCP_NODELAY:
1376                 case TCP_NOOPT:
1377                         switch (sopt->sopt_name) {
1378                         case TCP_NODELAY:
1379                                 opt = TF_NODELAY;
1380                                 break;
1381                         case TCP_NOOPT:
1382                                 opt = TF_NOOPT;
1383                                 break;
1384                         default:
1385                                 opt = 0; /* dead code to fool gcc */
1386                                 break;
1387                         }
1388
1389                         if (optval)
1390                                 tp->t_flags |= opt;
1391                         else
1392                                 tp->t_flags &= ~opt;
1393                         break;
1394
1395                 case TCP_NOPUSH:
1396                         if (optval)
1397                                 tp->t_flags |= TF_NOPUSH;
1398                         else {
1399                                 tp->t_flags &= ~TF_NOPUSH;
1400                                 error = tcp_output(tp);
1401                         }
1402                         break;
1403
1404                 case TCP_MAXSEG:
1405                         /*
1406                          * Must be between 0 and maxseg.  If the requested
1407                          * maxseg is too small to satisfy the desired minmss,
1408                          * pump it up (silently so sysctl modifications of
1409                          * minmss do not create unexpected program failures).
1410                          * Handle degenerate cases.
1411                          */
1412                         if (optval > 0 && optval <= tp->t_maxseg) {
1413                                 if (optval + 40 < tcp_minmss) {
1414                                         optval = tcp_minmss - 40;
1415                                         if (optval < 0)
1416                                                 optval = 1;
1417                                 }
1418                                 tp->t_maxseg = optval;
1419                         } else {
1420                                 error = EINVAL;
1421                         }
1422                         break;
1423
1424                 case TCP_KEEPINIT:
1425                         opthz = ((int64_t)optval * hz) / 1000;
1426                         if (opthz >= 1)
1427                                 tp->t_keepinit = opthz;
1428                         else
1429                                 error = EINVAL;
1430                         break;
1431
1432                 case TCP_KEEPIDLE:
1433                         opthz = ((int64_t)optval * hz) / 1000;
1434                         if (opthz >= 1)
1435                                 tp->t_keepidle = opthz;
1436                         else
1437                                 error = EINVAL;
1438                         break;
1439
1440                 default:
1441                         error = ENOPROTOOPT;
1442                         break;
1443                 }
1444                 break;
1445
1446         case SOPT_GET:
1447                 switch (sopt->sopt_name) {
1448 #ifdef TCP_SIGNATURE
1449                 case TCP_SIGNATURE_ENABLE:
1450                         optval = (tp->t_flags & TF_SIGNATURE) ? 1 : 0;
1451                         break;
1452 #endif /* TCP_SIGNATURE */
1453                 case TCP_NODELAY:
1454                         optval = tp->t_flags & TF_NODELAY;
1455                         break;
1456                 case TCP_MAXSEG:
1457                         optval = tp->t_maxseg;
1458                         break;
1459                 case TCP_NOOPT:
1460                         optval = tp->t_flags & TF_NOOPT;
1461                         break;
1462                 case TCP_NOPUSH:
1463                         optval = tp->t_flags & TF_NOPUSH;
1464                         break;
1465                 case TCP_KEEPINIT:
1466                         optval = ((int64_t)tp->t_keepinit * 1000) / hz;
1467                         break;
1468                 case TCP_KEEPIDLE:
1469                         optval = ((int64_t)tp->t_keepidle * 1000) / hz;
1470                         break;
1471                 default:
1472                         error = ENOPROTOOPT;
1473                         break;
1474                 }
1475                 if (error == 0)
1476                         soopt_from_kbuf(sopt, &optval, sizeof optval);
1477                 break;
1478         }
1479 done:
1480         lwkt_replymsg(&msg->lmsg, error);
1481 }
1482
1483 /*
1484  * tcp_sendspace and tcp_recvspace are the default send and receive window
1485  * sizes, respectively.  These are obsolescent (this information should
1486  * be set by the route).
1487  *
1488  * Use a default that does not require tcp window scaling to be turned
1489  * on.  Individual programs or the administrator can increase the default.
1490  */
1491 u_long  tcp_sendspace = 57344;  /* largest multiple of PAGE_SIZE < 64k */
1492 SYSCTL_INT(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW,
1493     &tcp_sendspace , 0, "Maximum outgoing TCP datagram size");
1494 u_long  tcp_recvspace = 57344;  /* largest multiple of PAGE_SIZE < 64k */
1495 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
1496     &tcp_recvspace , 0, "Maximum incoming TCP datagram size");
1497
1498 /*
1499  * Attach TCP protocol to socket, allocating internet protocol control
1500  * block, tcp control block, bufer space, and entering LISTEN state
1501  * if to accept connections.
1502  */
1503 static int
1504 tcp_attach(struct socket *so, struct pru_attach_info *ai)
1505 {
1506         struct tcpcb *tp;
1507         struct inpcb *inp;
1508         int error;
1509         int cpu;
1510 #ifdef INET6
1511         int isipv6 = INP_CHECK_SOCKAF(so, AF_INET6) != 0;
1512 #endif
1513
1514         if (so->so_snd.ssb_hiwat == 0 || so->so_rcv.ssb_hiwat == 0) {
1515                 lwkt_gettoken(&so->so_rcv.ssb_token);
1516                 error = soreserve(so, tcp_sendspace, tcp_recvspace,
1517                                   ai->sb_rlimit);
1518                 lwkt_reltoken(&so->so_rcv.ssb_token);
1519                 if (error)
1520                         return (error);
1521         }
1522         atomic_set_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE);
1523         atomic_set_int(&so->so_snd.ssb_flags, SSB_AUTOSIZE);
1524         cpu = mycpu->gd_cpuid;
1525
1526         /*
1527          * Set the default port for protocol processing. This will likely
1528          * change when we connect.
1529          */
1530         error = in_pcballoc(so, &tcbinfo[cpu]);
1531         if (error)
1532                 return (error);
1533         inp = so->so_pcb;
1534 #ifdef INET6
1535         if (isipv6) {
1536                 inp->inp_vflag |= INP_IPV6;
1537                 inp->in6p_hops = -1;    /* use kernel default */
1538         }
1539         else
1540 #endif
1541         inp->inp_vflag |= INP_IPV4;
1542         tp = tcp_newtcpcb(inp);
1543         if (tp == NULL) {
1544                 /*
1545                  * Make sure the socket is destroyed by the pcbdetach.
1546                  */
1547                 soreference(so);
1548 #ifdef INET6
1549                 if (isipv6)
1550                         in6_pcbdetach(inp);
1551                 else
1552 #endif
1553                 in_pcbdetach(inp);
1554                 sofree(so);     /* from ref above */
1555                 return (ENOBUFS);
1556         }
1557         tp->t_state = TCPS_CLOSED;
1558         return (0);
1559 }
1560
1561 /*
1562  * Initiate (or continue) disconnect.
1563  * If embryonic state, just send reset (once).
1564  * If in ``let data drain'' option and linger null, just drop.
1565  * Otherwise (hard), mark socket disconnecting and drop
1566  * current input data; switch states based on user close, and
1567  * send segment to peer (with FIN).
1568  */
1569 static struct tcpcb *
1570 tcp_disconnect(struct tcpcb *tp)
1571 {
1572         struct socket *so = tp->t_inpcb->inp_socket;
1573
1574         if (tp->t_state < TCPS_ESTABLISHED) {
1575                 tp = tcp_close(tp);
1576         } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
1577                 tp = tcp_drop(tp, 0);
1578         } else {
1579                 lwkt_gettoken(&so->so_rcv.ssb_token);
1580                 soisdisconnecting(so);
1581                 sbflush(&so->so_rcv.sb);
1582                 tp = tcp_usrclosed(tp);
1583                 if (tp)
1584                         tcp_output(tp);
1585                 lwkt_reltoken(&so->so_rcv.ssb_token);
1586         }
1587         return (tp);
1588 }
1589
1590 /*
1591  * User issued close, and wish to trail through shutdown states:
1592  * if never received SYN, just forget it.  If got a SYN from peer,
1593  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
1594  * If already got a FIN from peer, then almost done; go to LAST_ACK
1595  * state.  In all other cases, have already sent FIN to peer (e.g.
1596  * after PRU_SHUTDOWN), and just have to play tedious game waiting
1597  * for peer to send FIN or not respond to keep-alives, etc.
1598  * We can let the user exit from the close as soon as the FIN is acked.
1599  */
1600 static struct tcpcb *
1601 tcp_usrclosed(struct tcpcb *tp)
1602 {
1603
1604         switch (tp->t_state) {
1605
1606         case TCPS_CLOSED:
1607         case TCPS_LISTEN:
1608                 tp->t_state = TCPS_CLOSED;
1609                 tp = tcp_close(tp);
1610                 break;
1611
1612         case TCPS_SYN_SENT:
1613         case TCPS_SYN_RECEIVED:
1614                 tp->t_flags |= TF_NEEDFIN;
1615                 break;
1616
1617         case TCPS_ESTABLISHED:
1618                 tp->t_state = TCPS_FIN_WAIT_1;
1619                 break;
1620
1621         case TCPS_CLOSE_WAIT:
1622                 tp->t_state = TCPS_LAST_ACK;
1623                 break;
1624         }
1625         if (tp && tp->t_state >= TCPS_FIN_WAIT_2) {
1626                 soisdisconnected(tp->t_inpcb->inp_socket);
1627                 /* To prevent the connection hanging in FIN_WAIT_2 forever. */
1628                 if (tp->t_state == TCPS_FIN_WAIT_2) {
1629                         tcp_callout_reset(tp, tp->tt_2msl, tcp_maxidle,
1630                             tcp_timer_2msl);
1631                 }
1632         }
1633         return (tp);
1634 }