tcp/usrreq: No need to hold ssb_rcv token to call soreserve()
[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. Neither the name of the University nor the names of its contributors
47  *    may be used to endorse or promote products derived from this software
48  *    without specific prior written permission.
49  *
50  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
51  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
52  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
53  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
54  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
55  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
56  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
57  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
58  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
59  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
60  * SUCH DAMAGE.
61  *
62  *      From: @(#)tcp_usrreq.c  8.2 (Berkeley) 1/3/94
63  * $FreeBSD: src/sys/netinet/tcp_usrreq.c,v 1.51.2.17 2002/10/11 11:46:44 ume Exp $
64  */
65
66 #include "opt_ipsec.h"
67 #include "opt_inet.h"
68 #include "opt_inet6.h"
69 #include "opt_tcpdebug.h"
70
71 #include <sys/param.h>
72 #include <sys/systm.h>
73 #include <sys/kernel.h>
74 #include <sys/malloc.h>
75 #include <sys/sysctl.h>
76 #include <sys/globaldata.h>
77 #include <sys/thread.h>
78
79 #include <sys/mbuf.h>
80 #ifdef INET6
81 #include <sys/domain.h>
82 #endif /* INET6 */
83 #include <sys/socket.h>
84 #include <sys/socketvar.h>
85 #include <sys/socketops.h>
86 #include <sys/protosw.h>
87
88 #include <sys/thread2.h>
89 #include <sys/msgport2.h>
90 #include <sys/socketvar2.h>
91
92 #include <net/if.h>
93 #include <net/netisr.h>
94 #include <net/route.h>
95
96 #include <net/netmsg2.h>
97 #include <net/netisr2.h>
98
99 #include <netinet/in.h>
100 #include <netinet/in_systm.h>
101 #ifdef INET6
102 #include <netinet/ip6.h>
103 #endif
104 #include <netinet/in_pcb.h>
105 #ifdef INET6
106 #include <netinet6/in6_pcb.h>
107 #endif
108 #include <netinet/in_var.h>
109 #include <netinet/ip_var.h>
110 #ifdef INET6
111 #include <netinet6/ip6_var.h>
112 #include <netinet6/tcp6_var.h>
113 #endif
114 #include <netinet/tcp.h>
115 #include <netinet/tcp_fsm.h>
116 #include <netinet/tcp_seq.h>
117 #include <netinet/tcp_timer.h>
118 #include <netinet/tcp_timer2.h>
119 #include <netinet/tcp_var.h>
120 #include <netinet/tcpip.h>
121 #ifdef TCPDEBUG
122 #include <netinet/tcp_debug.h>
123 #endif
124
125 #ifdef IPSEC
126 #include <netinet6/ipsec.h>
127 #endif /*IPSEC*/
128
129 /*
130  * TCP protocol interface to socket abstraction.
131  */
132 extern  char *tcpstates[];      /* XXX ??? */
133
134 static int      tcp_attach (struct socket *, struct pru_attach_info *);
135 static void     tcp_connect (netmsg_t msg);
136 #ifdef INET6
137 static void     tcp6_connect (netmsg_t msg);
138 static int      tcp6_connect_oncpu(struct tcpcb *tp, int flags,
139                                 struct mbuf **mp,
140                                 struct sockaddr_in6 *sin6,
141                                 struct in6_addr *addr6);
142 #endif /* INET6 */
143 static struct tcpcb *
144                 tcp_disconnect (struct tcpcb *);
145 static struct tcpcb *
146                 tcp_usrclosed (struct tcpcb *);
147
148 #ifdef TCPDEBUG
149 #define TCPDEBUG0       int ostate = 0
150 #define TCPDEBUG1()     ostate = tp ? tp->t_state : 0
151 #define TCPDEBUG2(req)  if (tp && (so->so_options & SO_DEBUG)) \
152                                 tcp_trace(TA_USER, ostate, tp, 0, 0, req)
153 #else
154 #define TCPDEBUG0
155 #define TCPDEBUG1()
156 #define TCPDEBUG2(req)
157 #endif
158
159 static int      tcp_lport_extension = 1;
160 SYSCTL_INT(_net_inet_tcp, OID_AUTO, lportext, CTLFLAG_RW,
161     &tcp_lport_extension, 0, "");
162
163 /*
164  * For some ill optimized programs, which try to use TCP_NOPUSH
165  * to improve performance, will have small amount of data sits
166  * in the sending buffer.  These small amount of data will _not_
167  * be pushed into the network until more data are written into
168  * the socket or the socket write side is shutdown.
169  */ 
170 static int      tcp_disable_nopush = 1;
171 SYSCTL_INT(_net_inet_tcp, OID_AUTO, disable_nopush, CTLFLAG_RW,
172     &tcp_disable_nopush, 0, "TCP_NOPUSH socket option will have no effect");
173
174 /*
175  * Allocate socket buffer space.
176  */
177 static int
178 tcp_usr_preattach(struct socket *so, int proto __unused,
179     struct pru_attach_info *ai)
180 {
181         int error;
182
183         if (so->so_snd.ssb_hiwat == 0 || so->so_rcv.ssb_hiwat == 0) {
184                 error = soreserve(so, tcp_sendspace, tcp_recvspace,
185                                   ai->sb_rlimit);
186                 if (error)
187                         return (error);
188         }
189         atomic_set_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE | SSB_PREALLOC);
190         atomic_set_int(&so->so_snd.ssb_flags, SSB_AUTOSIZE | SSB_PREALLOC);
191
192         return 0;
193 }
194
195 /*
196  * TCP attaches to socket via pru_attach(), reserving space,
197  * and an internet control block.  This socket may move to
198  * other CPU later when we bind/connect.
199  */
200 static void
201 tcp_usr_attach(netmsg_t msg)
202 {
203         struct socket *so = msg->base.nm_so;
204         struct pru_attach_info *ai = msg->attach.nm_ai;
205         int error;
206         struct inpcb *inp;
207         struct tcpcb *tp = NULL;
208         TCPDEBUG0;
209
210         soreference(so);
211         inp = so->so_pcb;
212         TCPDEBUG1();
213         if (inp) {
214                 error = EISCONN;
215                 goto out;
216         }
217
218         error = tcp_attach(so, ai);
219         if (error)
220                 goto out;
221
222         if ((so->so_options & SO_LINGER) && so->so_linger == 0)
223                 so->so_linger = TCP_LINGERTIME;
224         tp = sototcpcb(so);
225 out:
226         sofree(so);             /* from ref above */
227         TCPDEBUG2(PRU_ATTACH);
228         lwkt_replymsg(&msg->lmsg, error);
229 }
230
231 /*
232  * pru_detach() detaches the TCP protocol from the socket.
233  * If the protocol state is non-embryonic, then can't
234  * do this directly: have to initiate a pru_disconnect(),
235  * which may finish later; embryonic TCB's can just
236  * be discarded here.
237  */
238 static void
239 tcp_usr_detach(netmsg_t msg)
240 {
241         struct socket *so = msg->base.nm_so;
242         int error = 0;
243         struct inpcb *inp;
244         struct tcpcb *tp;
245         TCPDEBUG0;
246
247         inp = so->so_pcb;
248
249         /*
250          * If the inp is already detached or never attached, it may have
251          * been due to an async close or async attach failure.  Just return
252          * as if no error occured.
253          *
254          * It's possible for the tcpcb (tp) to disconnect from the inp due
255          * to tcp_drop()->tcp_close() being called.  This may occur *after*
256          * the detach message has been queued so we may find a NULL tp here.
257          */
258         if (inp) {
259                 if ((tp = intotcpcb(inp)) != NULL) {
260                         TCPDEBUG1();
261                         tp = tcp_disconnect(tp);
262                         TCPDEBUG2(PRU_DETACH);
263                 }
264         }
265         lwkt_replymsg(&msg->lmsg, error);
266 }
267
268 /*
269  * NOTE: ignore_error is non-zero for certain disconnection races
270  * which we want to silently allow, otherwise close() may return
271  * an unexpected error.
272  *
273  * NOTE: The variables (msg) and (tp) are assumed.
274  */
275 #define COMMON_START(so, inp, ignore_error)                     \
276         TCPDEBUG0;                                              \
277                                                                 \
278         inp = so->so_pcb;                                       \
279         do {                                                    \
280                 if (inp == NULL) {                              \
281                         error = ignore_error ? 0 : EINVAL;      \
282                         tp = NULL;                              \
283                         goto out;                               \
284                 }                                               \
285                 tp = intotcpcb(inp);                            \
286                 TCPDEBUG1();                                    \
287         } while(0)
288
289 #define COMMON_END1(req, noreply)                               \
290         out: do {                                               \
291                 TCPDEBUG2(req);                                 \
292                 if (!(noreply))                                 \
293                         lwkt_replymsg(&msg->lmsg, error);       \
294                 return;                                         \
295         } while(0)
296
297 #define COMMON_END(req)         COMMON_END1((req), 0)
298
299 /*
300  * Give the socket an address.
301  */
302 static void
303 tcp_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_in *sinp;
312
313         COMMON_START(so, inp, 0);
314
315         /*
316          * Must check for multicast addresses and disallow binding
317          * to them.
318          */
319         sinp = (struct sockaddr_in *)nam;
320         if (sinp->sin_family == AF_INET &&
321             IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
322                 error = EAFNOSUPPORT;
323                 goto out;
324         }
325         error = in_pcbbind(inp, nam, td);
326         if (error)
327                 goto out;
328
329         COMMON_END(PRU_BIND);
330 }
331
332 #ifdef INET6
333
334 static void
335 tcp6_usr_bind(netmsg_t msg)
336 {
337         struct socket *so = msg->bind.base.nm_so;
338         struct sockaddr *nam = msg->bind.nm_nam;
339         struct thread *td = msg->bind.nm_td;
340         int error = 0;
341         struct inpcb *inp;
342         struct tcpcb *tp;
343         struct sockaddr_in6 *sin6p;
344
345         COMMON_START(so, inp, 0);
346
347         /*
348          * Must check for multicast addresses and disallow binding
349          * to them.
350          */
351         sin6p = (struct sockaddr_in6 *)nam;
352         if (sin6p->sin6_family == AF_INET6 &&
353             IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
354                 error = EAFNOSUPPORT;
355                 goto out;
356         }
357         error = in6_pcbbind(inp, nam, td);
358         if (error)
359                 goto out;
360         COMMON_END(PRU_BIND);
361 }
362 #endif /* INET6 */
363
364 struct netmsg_inswildcard {
365         struct netmsg_base      base;
366         struct inpcb            *nm_inp;
367 };
368
369 static void
370 in_pcbinswildcardhash_handler(netmsg_t msg)
371 {
372         struct netmsg_inswildcard *nm = (struct netmsg_inswildcard *)msg;
373         int cpu = mycpuid, nextcpu;
374
375         in_pcbinswildcardhash_oncpu(nm->nm_inp, &tcbinfo[cpu]);
376
377         nextcpu = cpu + 1;
378         if (nextcpu < ncpus2)
379                 lwkt_forwardmsg(netisr_cpuport(nextcpu), &nm->base.lmsg);
380         else
381                 lwkt_replymsg(&nm->base.lmsg, 0);
382 }
383
384 static void
385 tcp_sosetport(struct lwkt_msg *msg, lwkt_port_t port)
386 {
387         sosetport(((struct netmsg_base *)msg)->nm_so, port);
388 }
389
390 /*
391  * Prepare to accept connections.
392  */
393 static void
394 tcp_usr_listen(netmsg_t msg)
395 {
396         struct socket *so = msg->listen.base.nm_so;
397         struct thread *td = msg->listen.nm_td;
398         int error = 0;
399         struct inpcb *inp;
400         struct tcpcb *tp;
401         struct netmsg_inswildcard nm;
402         lwkt_port_t port0 = netisr_cpuport(0);
403
404         COMMON_START(so, inp, 0);
405
406         if (&curthread->td_msgport != port0) {
407                 lwkt_msg_t lmsg = &msg->listen.base.lmsg;
408
409                 KASSERT((msg->listen.nm_flags & PRUL_RELINK) == 0,
410                     ("already asked to relink"));
411
412                 in_pcbunlink(so->so_pcb, &tcbinfo[mycpuid]);
413                 msg->listen.nm_flags |= PRUL_RELINK;
414
415                 /* See the related comment in tcp_connect() */
416                 lwkt_setmsg_receipt(lmsg, tcp_sosetport);
417                 lwkt_forwardmsg(port0, lmsg);
418                 /* msg invalid now */
419                 return;
420         }
421         KASSERT(so->so_port == port0, ("so_port is not netisr0"));
422
423         if (msg->listen.nm_flags & PRUL_RELINK) {
424                 msg->listen.nm_flags &= ~PRUL_RELINK;
425                 in_pcblink(so->so_pcb, &tcbinfo[mycpuid]);
426         }
427         KASSERT(inp->inp_pcbinfo == &tcbinfo[0], ("pcbinfo is not tcbinfo0"));
428
429         if (tp->t_flags & TF_LISTEN)
430                 goto out;
431
432         if (inp->inp_lport == 0) {
433                 error = in_pcbbind(inp, NULL, td);
434                 if (error)
435                         goto out;
436         }
437
438         tp->t_state = TCPS_LISTEN;
439         tp->t_flags |= TF_LISTEN;
440         tp->tt_msg = NULL; /* Catch any invalid timer usage */
441
442         if (ncpus2 > 1) {
443                 /*
444                  * Put this inpcb into wildcard hash on other cpus.
445                  */
446                 ASSERT_INP_NOTINHASH(inp);
447                 netmsg_init(&nm.base, NULL, &curthread->td_msgport,
448                             MSGF_PRIORITY, in_pcbinswildcardhash_handler);
449                 nm.nm_inp = inp;
450                 lwkt_domsg(netisr_cpuport(1), &nm.base.lmsg, 0);
451         }
452         in_pcbinswildcardhash(inp);
453         COMMON_END(PRU_LISTEN);
454 }
455
456 #ifdef INET6
457
458 static void
459 tcp6_usr_listen(netmsg_t msg)
460 {
461         struct socket *so = msg->listen.base.nm_so;
462         struct thread *td = msg->listen.nm_td;
463         int error = 0;
464         struct inpcb *inp;
465         struct tcpcb *tp;
466         struct netmsg_inswildcard nm;
467
468         COMMON_START(so, inp, 0);
469
470         if (tp->t_flags & TF_LISTEN)
471                 goto out;
472
473         if (inp->inp_lport == 0) {
474                 error = in6_pcbbind(inp, NULL, td);
475                 if (error)
476                         goto out;
477         }
478
479         tp->t_state = TCPS_LISTEN;
480         tp->t_flags |= TF_LISTEN;
481         tp->tt_msg = NULL; /* Catch any invalid timer usage */
482
483         if (ncpus2 > 1) {
484                 /*
485                  * Put this inpcb into wildcard hash on other cpus.
486                  */
487                 KKASSERT(so->so_port == netisr_cpuport(0));
488                 ASSERT_IN_NETISR(0);
489                 KKASSERT(inp->inp_pcbinfo == &tcbinfo[0]);
490                 ASSERT_INP_NOTINHASH(inp);
491
492                 netmsg_init(&nm.base, NULL, &curthread->td_msgport,
493                             MSGF_PRIORITY, in_pcbinswildcardhash_handler);
494                 nm.nm_inp = inp;
495                 lwkt_domsg(netisr_cpuport(1), &nm.base.lmsg, 0);
496         }
497         in_pcbinswildcardhash(inp);
498         COMMON_END(PRU_LISTEN);
499 }
500 #endif /* INET6 */
501
502 /*
503  * Initiate connection to peer.
504  * Create a template for use in transmissions on this connection.
505  * Enter SYN_SENT state, and mark socket as connecting.
506  * Start keep-alive timer, and seed output sequence space.
507  * Send initial segment on connection.
508  */
509 static void
510 tcp_usr_connect(netmsg_t msg)
511 {
512         struct socket *so = msg->connect.base.nm_so;
513         struct sockaddr *nam = msg->connect.nm_nam;
514         struct thread *td = msg->connect.nm_td;
515         int error = 0;
516         struct inpcb *inp;
517         struct tcpcb *tp;
518         struct sockaddr_in *sinp;
519
520         COMMON_START(so, inp, 0);
521
522         /*
523          * Must disallow TCP ``connections'' to multicast addresses.
524          */
525         sinp = (struct sockaddr_in *)nam;
526         if (sinp->sin_family == AF_INET
527             && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
528                 error = EAFNOSUPPORT;
529                 goto out;
530         }
531
532         if (!prison_remote_ip(td, (struct sockaddr*)sinp)) {
533                 error = EAFNOSUPPORT; /* IPv6 only jail */
534                 goto out;
535         }
536
537         tcp_connect(msg);
538         /* msg is invalid now */
539         return;
540 out:
541         if (msg->connect.nm_m) {
542                 m_freem(msg->connect.nm_m);
543                 msg->connect.nm_m = NULL;
544         }
545         if (msg->connect.nm_flags & PRUC_HELDTD)
546                 lwkt_rele(td);
547         if (error && (msg->connect.nm_flags & PRUC_ASYNC)) {
548                 so->so_error = error;
549                 soisdisconnected(so);
550         }
551         lwkt_replymsg(&msg->lmsg, error);
552 }
553
554 #ifdef INET6
555
556 static void
557 tcp6_usr_connect(netmsg_t msg)
558 {
559         struct socket *so = msg->connect.base.nm_so;
560         struct sockaddr *nam = msg->connect.nm_nam;
561         struct thread *td = msg->connect.nm_td;
562         int error = 0;
563         struct inpcb *inp;
564         struct tcpcb *tp;
565         struct sockaddr_in6 *sin6p;
566
567         COMMON_START(so, inp, 0);
568
569         /*
570          * Must disallow TCP ``connections'' to multicast addresses.
571          */
572         sin6p = (struct sockaddr_in6 *)nam;
573         if (sin6p->sin6_family == AF_INET6
574             && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
575                 error = EAFNOSUPPORT;
576                 goto out;
577         }
578
579         if (!prison_remote_ip(td, nam)) {
580                 error = EAFNOSUPPORT; /* IPv4 only jail */
581                 goto out;
582         }
583
584         /* Reject v4-mapped address */
585         if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
586                 error = EADDRNOTAVAIL;
587                 goto out;
588         }
589
590         inp->inp_inc.inc_isipv6 = 1;
591         tcp6_connect(msg);
592         /* msg is invalid now */
593         return;
594 out:
595         if (msg->connect.nm_m) {
596                 m_freem(msg->connect.nm_m);
597                 msg->connect.nm_m = NULL;
598         }
599         lwkt_replymsg(&msg->lmsg, error);
600 }
601
602 #endif /* INET6 */
603
604 /*
605  * Initiate disconnect from peer.
606  * If connection never passed embryonic stage, just drop;
607  * else if don't need to let data drain, then can just drop anyways,
608  * else have to begin TCP shutdown process: mark socket disconnecting,
609  * drain unread data, state switch to reflect user close, and
610  * send segment (e.g. FIN) to peer.  Socket will be really disconnected
611  * when peer sends FIN and acks ours.
612  *
613  * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
614  */
615 static void
616 tcp_usr_disconnect(netmsg_t msg)
617 {
618         struct socket *so = msg->disconnect.base.nm_so;
619         int error = 0;
620         struct inpcb *inp;
621         struct tcpcb *tp;
622
623         COMMON_START(so, inp, 1);
624         tp = tcp_disconnect(tp);
625         COMMON_END(PRU_DISCONNECT);
626 }
627
628 /*
629  * Accept a connection.  Essentially all the work is
630  * done at higher levels; just return the address
631  * of the peer, storing through addr.
632  */
633 static void
634 tcp_usr_accept(netmsg_t msg)
635 {
636         struct socket *so = msg->accept.base.nm_so;
637         struct sockaddr **nam = msg->accept.nm_nam;
638         int error = 0;
639         struct inpcb *inp;
640         struct tcpcb *tp = NULL;
641         TCPDEBUG0;
642
643         inp = so->so_pcb;
644         if (so->so_state & SS_ISDISCONNECTED) {
645                 error = ECONNABORTED;
646                 goto out;
647         }
648         if (inp == NULL) {
649                 error = EINVAL;
650                 goto out;
651         }
652
653         tp = intotcpcb(inp);
654         TCPDEBUG1();
655         in_setpeeraddr(so, nam);
656         COMMON_END(PRU_ACCEPT);
657 }
658
659 #ifdef INET6
660 static void
661 tcp6_usr_accept(netmsg_t msg)
662 {
663         struct socket *so = msg->accept.base.nm_so;
664         struct sockaddr **nam = msg->accept.nm_nam;
665         int error = 0;
666         struct inpcb *inp;
667         struct tcpcb *tp = NULL;
668         TCPDEBUG0;
669
670         inp = so->so_pcb;
671
672         if (so->so_state & SS_ISDISCONNECTED) {
673                 error = ECONNABORTED;
674                 goto out;
675         }
676         if (inp == NULL) {
677                 error = EINVAL;
678                 goto out;
679         }
680         tp = intotcpcb(inp);
681         TCPDEBUG1();
682         in6_setpeeraddr(so, nam);
683         COMMON_END(PRU_ACCEPT);
684 }
685 #endif /* INET6 */
686
687 /*
688  * Mark the connection as being incapable of further output.
689  */
690 static void
691 tcp_usr_shutdown(netmsg_t msg)
692 {
693         struct socket *so = msg->shutdown.base.nm_so;
694         int error = 0;
695         struct inpcb *inp;
696         struct tcpcb *tp;
697
698         COMMON_START(so, inp, 0);
699         socantsendmore(so);
700         tp = tcp_usrclosed(tp);
701         if (tp)
702                 error = tcp_output(tp);
703         COMMON_END(PRU_SHUTDOWN);
704 }
705
706 /*
707  * After a receive, possibly send window update to peer.
708  */
709 static void
710 tcp_usr_rcvd(netmsg_t msg)
711 {
712         struct socket *so = msg->rcvd.base.nm_so;
713         int error = 0, noreply = 0;
714         struct inpcb *inp;
715         struct tcpcb *tp;
716
717         COMMON_START(so, inp, 0);
718
719         if (msg->rcvd.nm_pru_flags & PRUR_ASYNC) {
720                 noreply = 1;
721                 so_async_rcvd_reply(so);
722         }
723         tcp_output(tp);
724
725         COMMON_END1(PRU_RCVD, noreply);
726 }
727
728 /*
729  * Do a send by putting data in output queue and updating urgent
730  * marker if URG set.  Possibly send more data.  Unlike the other
731  * pru_*() routines, the mbuf chains are our responsibility.  We
732  * must either enqueue them or free them.  The other pru_* routines
733  * generally are caller-frees.
734  */
735 static void
736 tcp_usr_send(netmsg_t msg)
737 {
738         struct socket *so = msg->send.base.nm_so;
739         int flags = msg->send.nm_flags;
740         struct mbuf *m = msg->send.nm_m;
741         int error = 0;
742         struct inpcb *inp;
743         struct tcpcb *tp;
744         TCPDEBUG0;
745
746         KKASSERT(msg->send.nm_control == NULL);
747         KKASSERT(msg->send.nm_addr == NULL);
748         KKASSERT((flags & PRUS_FREEADDR) == 0);
749
750         inp = so->so_pcb;
751
752         if (inp == NULL) {
753                 /*
754                  * OOPS! we lost a race, the TCP session got reset after
755                  * we checked SS_CANTSENDMORE, eg: while doing uiomove or a
756                  * network interrupt in the non-critical section of sosend().
757                  */
758                 m_freem(m);
759                 error = ECONNRESET;     /* XXX EPIPE? */
760                 tp = NULL;
761                 TCPDEBUG1();
762                 goto out;
763         }
764         tp = intotcpcb(inp);
765         TCPDEBUG1();
766
767 #ifdef foo
768         /*
769          * This is no longer necessary, since:
770          * - sosendtcp() has already checked it for us
771          * - It does not work with asynchronized send
772          */
773
774         /*
775          * Don't let too much OOB data build up
776          */
777         if (flags & PRUS_OOB) {
778                 if (ssb_space(&so->so_snd) < -512) {
779                         m_freem(m);
780                         error = ENOBUFS;
781                         goto out;
782                 }
783         }
784 #endif
785
786         /*
787          * Pump the data into the socket.
788          */
789         if (m) {
790                 ssb_appendstream(&so->so_snd, m);
791                 sowwakeup(so);
792         }
793         if (flags & PRUS_OOB) {
794                 /*
795                  * According to RFC961 (Assigned Protocols),
796                  * the urgent pointer points to the last octet
797                  * of urgent data.  We continue, however,
798                  * to consider it to indicate the first octet
799                  * of data past the urgent section.
800                  * Otherwise, snd_up should be one lower.
801                  */
802                 tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
803                 tp->t_flags |= TF_FORCE;
804                 error = tcp_output(tp);
805                 tp->t_flags &= ~TF_FORCE;
806         } else {
807                 if (flags & PRUS_EOF) {
808                         /*
809                          * Close the send side of the connection after
810                          * the data is sent.
811                          */
812                         socantsendmore(so);
813                         tp = tcp_usrclosed(tp);
814                 }
815                 if (tp != NULL && !tcp_output_pending(tp)) {
816                         if (flags & PRUS_MORETOCOME)
817                                 tp->t_flags |= TF_MORETOCOME;
818                         error = tcp_output_fair(tp);
819                         if (flags & PRUS_MORETOCOME)
820                                 tp->t_flags &= ~TF_MORETOCOME;
821                 }
822         }
823         COMMON_END1((flags & PRUS_OOB) ? PRU_SENDOOB :
824                    ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND),
825                    (flags & PRUS_NOREPLY));
826 }
827
828 /*
829  * NOTE: (so) is referenced from soabort*() and netmsg_pru_abort()
830  *       will sofree() it when we return.
831  */
832 static void
833 tcp_usr_abort(netmsg_t msg)
834 {
835         struct socket *so = msg->abort.base.nm_so;
836         int error = 0;
837         struct inpcb *inp;
838         struct tcpcb *tp;
839
840         COMMON_START(so, inp, 1);
841         tp = tcp_drop(tp, ECONNABORTED);
842         COMMON_END(PRU_ABORT);
843 }
844
845 /*
846  * Receive out-of-band data.
847  */
848 static void
849 tcp_usr_rcvoob(netmsg_t msg)
850 {
851         struct socket *so = msg->rcvoob.base.nm_so;
852         struct mbuf *m = msg->rcvoob.nm_m;
853         int flags = msg->rcvoob.nm_flags;
854         int error = 0;
855         struct inpcb *inp;
856         struct tcpcb *tp;
857
858         COMMON_START(so, inp, 0);
859         if ((so->so_oobmark == 0 &&
860              (so->so_state & SS_RCVATMARK) == 0) ||
861             so->so_options & SO_OOBINLINE ||
862             tp->t_oobflags & TCPOOB_HADDATA) {
863                 error = EINVAL;
864                 goto out;
865         }
866         if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
867                 error = EWOULDBLOCK;
868                 goto out;
869         }
870         m->m_len = 1;
871         *mtod(m, caddr_t) = tp->t_iobc;
872         if ((flags & MSG_PEEK) == 0)
873                 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
874         COMMON_END(PRU_RCVOOB);
875 }
876
877 static void
878 tcp_usr_savefaddr(struct socket *so, const struct sockaddr *faddr)
879 {
880         in_savefaddr(so, faddr);
881 }
882
883 #ifdef INET6
884 static void
885 tcp6_usr_savefaddr(struct socket *so, const struct sockaddr *faddr)
886 {
887         in6_savefaddr(so, faddr);
888 }
889 #endif
890
891 static int
892 tcp_usr_preconnect(struct socket *so, const struct sockaddr *nam,
893     struct thread *td __unused)
894 {
895         const struct sockaddr_in *sinp;
896
897         sinp = (const struct sockaddr_in *)nam;
898         if (sinp->sin_family == AF_INET &&
899             IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
900                 return EAFNOSUPPORT;
901
902         soisconnecting(so);
903         return 0;
904 }
905
906 /* xxx - should be const */
907 struct pr_usrreqs tcp_usrreqs = {
908         .pru_abort = tcp_usr_abort,
909         .pru_accept = tcp_usr_accept,
910         .pru_attach = tcp_usr_attach,
911         .pru_bind = tcp_usr_bind,
912         .pru_connect = tcp_usr_connect,
913         .pru_connect2 = pr_generic_notsupp,
914         .pru_control = in_control_dispatch,
915         .pru_detach = tcp_usr_detach,
916         .pru_disconnect = tcp_usr_disconnect,
917         .pru_listen = tcp_usr_listen,
918         .pru_peeraddr = in_setpeeraddr_dispatch,
919         .pru_rcvd = tcp_usr_rcvd,
920         .pru_rcvoob = tcp_usr_rcvoob,
921         .pru_send = tcp_usr_send,
922         .pru_sense = pru_sense_null,
923         .pru_shutdown = tcp_usr_shutdown,
924         .pru_sockaddr = in_setsockaddr_dispatch,
925         .pru_sosend = sosendtcp,
926         .pru_soreceive = sorecvtcp,
927         .pru_savefaddr = tcp_usr_savefaddr,
928         .pru_preconnect = tcp_usr_preconnect,
929         .pru_preattach = tcp_usr_preattach
930 };
931
932 #ifdef INET6
933 struct pr_usrreqs tcp6_usrreqs = {
934         .pru_abort = tcp_usr_abort,
935         .pru_accept = tcp6_usr_accept,
936         .pru_attach = tcp_usr_attach,
937         .pru_bind = tcp6_usr_bind,
938         .pru_connect = tcp6_usr_connect,
939         .pru_connect2 = pr_generic_notsupp,
940         .pru_control = in6_control_dispatch,
941         .pru_detach = tcp_usr_detach,
942         .pru_disconnect = tcp_usr_disconnect,
943         .pru_listen = tcp6_usr_listen,
944         .pru_peeraddr = in6_setpeeraddr_dispatch,
945         .pru_rcvd = tcp_usr_rcvd,
946         .pru_rcvoob = tcp_usr_rcvoob,
947         .pru_send = tcp_usr_send,
948         .pru_sense = pru_sense_null,
949         .pru_shutdown = tcp_usr_shutdown,
950         .pru_sockaddr = in6_setsockaddr_dispatch,
951         .pru_sosend = sosendtcp,
952         .pru_soreceive = sorecvtcp,
953         .pru_savefaddr = tcp6_usr_savefaddr
954 };
955 #endif /* INET6 */
956
957 static int
958 tcp_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf *m,
959                   struct sockaddr_in *sin, struct sockaddr_in *if_sin)
960 {
961         struct inpcb *inp = tp->t_inpcb, *oinp;
962         struct socket *so = inp->inp_socket;
963         struct route *ro = &inp->inp_route;
964
965         KASSERT(inp->inp_pcbinfo == &tcbinfo[mycpu->gd_cpuid],
966             ("pcbinfo mismatch"));
967
968         oinp = in_pcblookup_hash(inp->inp_pcbinfo,
969                                  sin->sin_addr, sin->sin_port,
970                                  (inp->inp_laddr.s_addr != INADDR_ANY ?
971                                   inp->inp_laddr : if_sin->sin_addr),
972                                 inp->inp_lport, 0, NULL);
973         if (oinp != NULL) {
974                 m_freem(m);
975                 return (EADDRINUSE);
976         }
977         if (inp->inp_laddr.s_addr == INADDR_ANY)
978                 inp->inp_laddr = if_sin->sin_addr;
979         inp->inp_faddr = sin->sin_addr;
980         inp->inp_fport = sin->sin_port;
981         in_pcbinsconnhash(inp);
982
983         /*
984          * We are now on the inpcb's owner CPU, if the cached route was
985          * freed because the rtentry's owner CPU is not the current CPU
986          * (e.g. in tcp_connect()), then we try to reallocate it here with
987          * the hope that a rtentry may be cloned from a RTF_PRCLONING
988          * rtentry.
989          */
990         if (!(inp->inp_socket->so_options & SO_DONTROUTE) && /*XXX*/
991             ro->ro_rt == NULL) {
992                 bzero(&ro->ro_dst, sizeof(struct sockaddr_in));
993                 ro->ro_dst.sa_family = AF_INET;
994                 ro->ro_dst.sa_len = sizeof(struct sockaddr_in);
995                 ((struct sockaddr_in *)&ro->ro_dst)->sin_addr =
996                         sin->sin_addr;
997                 rtalloc(ro);
998         }
999
1000         /*
1001          * Now that no more errors can occur, change the protocol processing
1002          * port to the current thread (which is the correct thread).
1003          *
1004          * Create TCP timer message now; we are on the tcpcb's owner
1005          * CPU/thread.
1006          */
1007         tcp_create_timermsg(tp, &curthread->td_msgport);
1008
1009         /*
1010          * Compute window scaling to request.  Use a larger scaling then
1011          * needed for the initial receive buffer in case the receive buffer
1012          * gets expanded.
1013          */
1014         if (tp->request_r_scale < TCP_MIN_WINSHIFT)
1015                 tp->request_r_scale = TCP_MIN_WINSHIFT;
1016         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1017                (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat
1018         ) {
1019                 tp->request_r_scale++;
1020         }
1021
1022         soisconnecting(so);
1023         tcpstat.tcps_connattempt++;
1024         tp->t_state = TCPS_SYN_SENT;
1025         tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
1026         tp->iss = tcp_new_isn(tp);
1027         tcp_sendseqinit(tp);
1028         if (m) {
1029                 ssb_appendstream(&so->so_snd, m);
1030                 m = NULL;
1031                 if (flags & PRUS_OOB)
1032                         tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
1033         }
1034
1035         /*
1036          * Close the send side of the connection after
1037          * the data is sent if flagged.
1038          */
1039         if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
1040                 socantsendmore(so);
1041                 tp = tcp_usrclosed(tp);
1042         }
1043         return (tcp_output(tp));
1044 }
1045
1046 /*
1047  * Common subroutine to open a TCP connection to remote host specified
1048  * by struct sockaddr_in in mbuf *nam.  Call in_pcbbind to assign a local
1049  * port number if needed.  Call in_pcbladdr to do the routing and to choose
1050  * a local host address (interface).
1051  * Initialize connection parameters and enter SYN-SENT state.
1052  */
1053 static void
1054 tcp_connect(netmsg_t msg)
1055 {
1056         struct socket *so = msg->connect.base.nm_so;
1057         struct sockaddr *nam = msg->connect.nm_nam;
1058         struct thread *td = msg->connect.nm_td;
1059         struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1060         struct sockaddr_in *if_sin = NULL;
1061         struct inpcb *inp;
1062         struct tcpcb *tp;
1063         int error;
1064         lwkt_port_t port;
1065
1066         COMMON_START(so, inp, 0);
1067
1068         /*
1069          * Reconnect our pcb if we have to
1070          */
1071         if (msg->connect.nm_flags & PRUC_RECONNECT) {
1072                 msg->connect.nm_flags &= ~PRUC_RECONNECT;
1073                 in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1074         }
1075
1076         /*
1077          * Bind if we have to
1078          */
1079         if (inp->inp_lport == 0) {
1080                 if (tcp_lport_extension) {
1081                         KKASSERT(inp->inp_laddr.s_addr == INADDR_ANY);
1082
1083                         error = in_pcbladdr(inp, nam, &if_sin, td);
1084                         if (error)
1085                                 goto out;
1086                         inp->inp_laddr.s_addr = if_sin->sin_addr.s_addr;
1087
1088                         error = in_pcbbind_remote(inp, nam, td);
1089                         if (error)
1090                                 goto out;
1091
1092                         msg->connect.nm_flags |= PRUC_HASLADDR;
1093                 } else {
1094                         error = in_pcbbind(inp, NULL, td);
1095                         if (error)
1096                                 goto out;
1097                 }
1098         }
1099
1100         if ((msg->connect.nm_flags & PRUC_HASLADDR) == 0) {
1101                 /*
1102                  * Calculate the correct protocol processing thread.  The
1103                  * connect operation must run there.  Set the forwarding
1104                  * port before we forward the message or it will get bounced
1105                  * right back to us.
1106                  */
1107                 error = in_pcbladdr(inp, nam, &if_sin, td);
1108                 if (error)
1109                         goto out;
1110         }
1111         KKASSERT(inp->inp_socket == so);
1112
1113         port = tcp_addrport(sin->sin_addr.s_addr, sin->sin_port,
1114                             (inp->inp_laddr.s_addr != INADDR_ANY ?
1115                              inp->inp_laddr.s_addr : if_sin->sin_addr.s_addr),
1116                             inp->inp_lport);
1117
1118         if (port != &curthread->td_msgport) {
1119                 lwkt_msg_t lmsg = &msg->connect.base.lmsg;
1120
1121                 /*
1122                  * in_pcbladdr() may have allocated a route entry for us
1123                  * on the current CPU, but we need a route entry on the
1124                  * inpcb's owner CPU, so free it here.
1125                  */
1126                 in_pcbresetroute(inp);
1127
1128                 /*
1129                  * We are moving the protocol processing port the socket
1130                  * is on, we have to unlink here and re-link on the
1131                  * target cpu.
1132                  */
1133                 in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1134                 msg->connect.nm_flags |= PRUC_RECONNECT;
1135                 msg->connect.base.nm_dispatch = tcp_connect;
1136
1137                 /*
1138                  * Use message put done receipt to change this socket's
1139                  * so_port, i.e. _after_ this message was put onto the
1140                  * target netisr's msgport but _before_ the message could
1141                  * be pulled from the target netisr's msgport, so that:
1142                  * - The upper half (socket code) will not see the new
1143                  *   msgport before this message reaches the new msgport
1144                  *   and messages for this socket will be ordered.
1145                  * - This message will see the new msgport, when its
1146                  *   handler is called in the target netisr.
1147                  *
1148                  * NOTE:
1149                  * We MUST use messege put done receipt to change this
1150                  * socket's so_port:
1151                  * If we changed the so_port in this netisr after the
1152                  * lwkt_forwardmsg (so messages for this socket will be
1153                  * ordered) and changed the so_port in the target netisr
1154                  * at the very beginning of this message's handler, we
1155                  * would suffer so_port overwritten race, given this
1156                  * message might be forwarded again.
1157                  *
1158                  * NOTE:
1159                  * This mechanism depends on that the netisr's msgport
1160                  * is spin msgport (currently it is :).
1161                  *
1162                  * If the upper half saw the new msgport before this
1163                  * message reached the target netisr's msgport, the
1164                  * messages sent from the upper half could reach the new
1165                  * msgport before this message, thus there would be
1166                  * message reordering.  The worst case could be soclose()
1167                  * saw the new msgport and the detach message could reach
1168                  * the new msgport before this message, i.e. the inpcb
1169                  * could have been destroyed when this message was still
1170                  * pending on or on its way to the new msgport.  Other
1171                  * weird cases could also happen, e.g. inpcb->inp_pcbinfo,
1172                  * since we have unlinked this inpcb from the current
1173                  * pcbinfo first.
1174                  */
1175                 lwkt_setmsg_receipt(lmsg, tcp_sosetport);
1176                 lwkt_forwardmsg(port, lmsg);
1177                 /* msg invalid now */
1178                 return;
1179         } else if (msg->connect.nm_flags & PRUC_HELDTD) {
1180                 /*
1181                  * The original thread is no longer needed; release it.
1182                  */
1183                 lwkt_rele(td);
1184                 msg->connect.nm_flags &= ~PRUC_HELDTD;
1185         }
1186         error = tcp_connect_oncpu(tp, msg->connect.nm_sndflags,
1187                                   msg->connect.nm_m, sin, if_sin);
1188         msg->connect.nm_m = NULL;
1189 out:
1190         if (msg->connect.nm_m) {
1191                 m_freem(msg->connect.nm_m);
1192                 msg->connect.nm_m = NULL;
1193         }
1194         if (msg->connect.nm_flags & PRUC_HELDTD)
1195                 lwkt_rele(td);
1196         if (error && (msg->connect.nm_flags & PRUC_ASYNC)) {
1197                 so->so_error = error;
1198                 soisdisconnected(so);
1199         }
1200         lwkt_replymsg(&msg->connect.base.lmsg, error);
1201         /* msg invalid now */
1202 }
1203
1204 #ifdef INET6
1205
1206 static void
1207 tcp6_connect(netmsg_t msg)
1208 {
1209         struct tcpcb *tp;
1210         struct socket *so = msg->connect.base.nm_so;
1211         struct sockaddr *nam = msg->connect.nm_nam;
1212         struct thread *td = msg->connect.nm_td;
1213         struct inpcb *inp;
1214         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam;
1215         struct in6_addr *addr6;
1216         lwkt_port_t port;
1217         int error;
1218
1219         COMMON_START(so, inp, 0);
1220
1221         /*
1222          * Reconnect our pcb if we have to
1223          */
1224         if (msg->connect.nm_flags & PRUC_RECONNECT) {
1225                 msg->connect.nm_flags &= ~PRUC_RECONNECT;
1226                 in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1227         }
1228
1229         /*
1230          * Bind if we have to
1231          */
1232         if (inp->inp_lport == 0) {
1233                 error = in6_pcbbind(inp, NULL, td);
1234                 if (error)
1235                         goto out;
1236         }
1237
1238         /*
1239          * Cannot simply call in_pcbconnect, because there might be an
1240          * earlier incarnation of this same connection still in
1241          * TIME_WAIT state, creating an ADDRINUSE error.
1242          */
1243         error = in6_pcbladdr(inp, nam, &addr6, td);
1244         if (error)
1245                 goto out;
1246
1247         port = tcp6_addrport(); /* XXX hack for now, always cpu0 */
1248
1249         if (port != &curthread->td_msgport) {
1250                 lwkt_msg_t lmsg = &msg->connect.base.lmsg;
1251
1252                 /*
1253                  * in_pcbladdr() may have allocated a route entry for us
1254                  * on the current CPU, but we need a route entry on the
1255                  * inpcb's owner CPU, so free it here.
1256                  */
1257                 in_pcbresetroute(inp);
1258
1259                 in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1260                 msg->connect.nm_flags |= PRUC_RECONNECT;
1261                 msg->connect.base.nm_dispatch = tcp6_connect;
1262
1263                 /* See the related comment in tcp_connect() */
1264                 lwkt_setmsg_receipt(lmsg, tcp_sosetport);
1265                 lwkt_forwardmsg(port, lmsg);
1266                 /* msg invalid now */
1267                 return;
1268         }
1269         error = tcp6_connect_oncpu(tp, msg->connect.nm_sndflags,
1270                                    &msg->connect.nm_m, sin6, addr6);
1271         /* nm_m may still be intact */
1272 out:
1273         if (msg->connect.nm_m) {
1274                 m_freem(msg->connect.nm_m);
1275                 msg->connect.nm_m = NULL;
1276         }
1277         lwkt_replymsg(&msg->connect.base.lmsg, error);
1278         /* msg invalid now */
1279 }
1280
1281 static int
1282 tcp6_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf **mp,
1283                    struct sockaddr_in6 *sin6, struct in6_addr *addr6)
1284 {
1285         struct mbuf *m = *mp;
1286         struct inpcb *inp = tp->t_inpcb;
1287         struct socket *so = inp->inp_socket;
1288         struct inpcb *oinp;
1289
1290         /*
1291          * Cannot simply call in_pcbconnect, because there might be an
1292          * earlier incarnation of this same connection still in
1293          * TIME_WAIT state, creating an ADDRINUSE error.
1294          */
1295         oinp = in6_pcblookup_hash(inp->inp_pcbinfo,
1296                                   &sin6->sin6_addr, sin6->sin6_port,
1297                                   (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) ?
1298                                       addr6 : &inp->in6p_laddr),
1299                                   inp->inp_lport,  0, NULL);
1300         if (oinp)
1301                 return (EADDRINUSE);
1302
1303         if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
1304                 inp->in6p_laddr = *addr6;
1305         inp->in6p_faddr = sin6->sin6_addr;
1306         inp->inp_fport = sin6->sin6_port;
1307         if ((sin6->sin6_flowinfo & IPV6_FLOWINFO_MASK) != 0)
1308                 inp->in6p_flowinfo = sin6->sin6_flowinfo;
1309         in_pcbinsconnhash(inp);
1310
1311         /*
1312          * Now that no more errors can occur, change the protocol processing
1313          * port to the current thread (which is the correct thread).
1314          *
1315          * Create TCP timer message now; we are on the tcpcb's owner
1316          * CPU/thread.
1317          */
1318         tcp_create_timermsg(tp, &curthread->td_msgport);
1319
1320         /* Compute window scaling to request.  */
1321         if (tp->request_r_scale < TCP_MIN_WINSHIFT)
1322                 tp->request_r_scale = TCP_MIN_WINSHIFT;
1323         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1324             (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat) {
1325                 tp->request_r_scale++;
1326         }
1327
1328         soisconnecting(so);
1329         tcpstat.tcps_connattempt++;
1330         tp->t_state = TCPS_SYN_SENT;
1331         tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
1332         tp->iss = tcp_new_isn(tp);
1333         tcp_sendseqinit(tp);
1334         if (m) {
1335                 ssb_appendstream(&so->so_snd, m);
1336                 *mp = NULL;
1337                 if (flags & PRUS_OOB)
1338                         tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
1339         }
1340
1341         /*
1342          * Close the send side of the connection after
1343          * the data is sent if flagged.
1344          */
1345         if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
1346                 socantsendmore(so);
1347                 tp = tcp_usrclosed(tp);
1348         }
1349         return (tcp_output(tp));
1350 }
1351
1352 #endif /* INET6 */
1353
1354 /*
1355  * The new sockopt interface makes it possible for us to block in the
1356  * copyin/out step (if we take a page fault).  Taking a page fault while
1357  * in a critical section is probably a Bad Thing.  (Since sockets and pcbs
1358  * both now use TSM, there probably isn't any need for this function to 
1359  * run in a critical section any more.  This needs more examination.)
1360  */
1361 void
1362 tcp_ctloutput(netmsg_t msg)
1363 {
1364         struct socket *so = msg->base.nm_so;
1365         struct sockopt *sopt = msg->ctloutput.nm_sopt;
1366         int     error, opt, optval, opthz;
1367         struct  inpcb *inp;
1368         struct  tcpcb *tp;
1369
1370         error = 0;
1371         inp = so->so_pcb;
1372         if (inp == NULL) {
1373                 error = ECONNRESET;
1374                 goto done;
1375         }
1376         tp = intotcpcb(inp);
1377
1378         /* Get socket's owner cpuid hint */
1379         if (sopt->sopt_level == SOL_SOCKET &&
1380             sopt->sopt_dir == SOPT_GET &&
1381             sopt->sopt_name == SO_CPUHINT) {
1382                 if (tp->t_flags & TF_LISTEN) {
1383                         /*
1384                          * Listen sockets owner cpuid is always 0,
1385                          * which does not make sense if SO_REUSEPORT
1386                          * is not set.
1387                          */
1388                         if (so->so_options & SO_REUSEPORT)
1389                                 optval = (inp->inp_lgrpindex & ncpus2_mask);
1390                         else
1391                                 optval = -1; /* no hint */
1392                 } else {
1393                         optval = mycpuid;
1394                 }
1395                 soopt_from_kbuf(sopt, &optval, sizeof(optval));
1396                 goto done;
1397         }
1398
1399         if (sopt->sopt_level != IPPROTO_TCP) {
1400                 if (sopt->sopt_level == IPPROTO_IP) {
1401                         switch (sopt->sopt_name) {
1402                         case IP_MULTICAST_IF:
1403                         case IP_MULTICAST_VIF:
1404                         case IP_MULTICAST_TTL:
1405                         case IP_MULTICAST_LOOP:
1406                         case IP_ADD_MEMBERSHIP:
1407                         case IP_DROP_MEMBERSHIP:
1408                                 /*
1409                                  * Multicast does not make sense on
1410                                  * TCP sockets.
1411                                  */
1412                                 error = EOPNOTSUPP;
1413                                 goto done;
1414                         }
1415                 }
1416 #ifdef INET6
1417                 if (INP_CHECK_SOCKAF(so, AF_INET6))
1418                         ip6_ctloutput_dispatch(msg);
1419                 else
1420 #endif /* INET6 */
1421                 ip_ctloutput(msg);
1422                 /* msg invalid now */
1423                 return;
1424         }
1425
1426         switch (sopt->sopt_dir) {
1427         case SOPT_SET:
1428                 error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1429                                       sizeof optval);
1430                 if (error)
1431                         break;
1432                 switch (sopt->sopt_name) {
1433                 case TCP_FASTKEEP:
1434                         if (optval > 0)
1435                                 tp->t_keepidle = tp->t_keepintvl;
1436                         else
1437                                 tp->t_keepidle = tcp_keepidle;
1438                         tcp_timer_keep_activity(tp, 0);
1439                         break;
1440 #ifdef TCP_SIGNATURE
1441                 case TCP_SIGNATURE_ENABLE:
1442                         if (tp->t_state == TCPS_CLOSED) {
1443                                 /*
1444                                  * This is the only safe state that this
1445                                  * option could be changed.  Some segments
1446                                  * could already have been sent in other
1447                                  * states.
1448                                  */
1449                                 if (optval > 0)
1450                                         tp->t_flags |= TF_SIGNATURE;
1451                                 else
1452                                         tp->t_flags &= ~TF_SIGNATURE;
1453                         } else {
1454                                 error = EOPNOTSUPP;
1455                         }
1456                         break;
1457 #endif /* TCP_SIGNATURE */
1458                 case TCP_NODELAY:
1459                 case TCP_NOOPT:
1460                         switch (sopt->sopt_name) {
1461                         case TCP_NODELAY:
1462                                 opt = TF_NODELAY;
1463                                 break;
1464                         case TCP_NOOPT:
1465                                 opt = TF_NOOPT;
1466                                 break;
1467                         default:
1468                                 opt = 0; /* dead code to fool gcc */
1469                                 break;
1470                         }
1471
1472                         if (optval)
1473                                 tp->t_flags |= opt;
1474                         else
1475                                 tp->t_flags &= ~opt;
1476                         break;
1477
1478                 case TCP_NOPUSH:
1479                         if (tcp_disable_nopush)
1480                                 break;
1481                         if (optval)
1482                                 tp->t_flags |= TF_NOPUSH;
1483                         else {
1484                                 tp->t_flags &= ~TF_NOPUSH;
1485                                 error = tcp_output(tp);
1486                         }
1487                         break;
1488
1489                 case TCP_MAXSEG:
1490                         /*
1491                          * Must be between 0 and maxseg.  If the requested
1492                          * maxseg is too small to satisfy the desired minmss,
1493                          * pump it up (silently so sysctl modifications of
1494                          * minmss do not create unexpected program failures).
1495                          * Handle degenerate cases.
1496                          */
1497                         if (optval > 0 && optval <= tp->t_maxseg) {
1498                                 if (optval + 40 < tcp_minmss) {
1499                                         optval = tcp_minmss - 40;
1500                                         if (optval < 0)
1501                                                 optval = 1;
1502                                 }
1503                                 tp->t_maxseg = optval;
1504                         } else {
1505                                 error = EINVAL;
1506                         }
1507                         break;
1508
1509                 case TCP_KEEPINIT:
1510                         opthz = ((int64_t)optval * hz) / 1000;
1511                         if (opthz >= 1)
1512                                 tp->t_keepinit = opthz;
1513                         else
1514                                 error = EINVAL;
1515                         break;
1516
1517                 case TCP_KEEPIDLE:
1518                         opthz = ((int64_t)optval * hz) / 1000;
1519                         if (opthz >= 1) {
1520                                 tp->t_keepidle = opthz;
1521                                 tcp_timer_keep_activity(tp, 0);
1522                         } else {
1523                                 error = EINVAL;
1524                         }
1525                         break;
1526
1527                 case TCP_KEEPINTVL:
1528                         opthz = ((int64_t)optval * hz) / 1000;
1529                         if (opthz >= 1) {
1530                                 tp->t_keepintvl = opthz;
1531                                 tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt;
1532                         } else {
1533                                 error = EINVAL;
1534                         }
1535                         break;
1536
1537                 case TCP_KEEPCNT:
1538                         if (optval > 0) {
1539                                 tp->t_keepcnt = optval;
1540                                 tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt;
1541                         } else {
1542                                 error = EINVAL;
1543                         }
1544                         break;
1545
1546                 default:
1547                         error = ENOPROTOOPT;
1548                         break;
1549                 }
1550                 break;
1551
1552         case SOPT_GET:
1553                 switch (sopt->sopt_name) {
1554 #ifdef TCP_SIGNATURE
1555                 case TCP_SIGNATURE_ENABLE:
1556                         optval = (tp->t_flags & TF_SIGNATURE) ? 1 : 0;
1557                         break;
1558 #endif /* TCP_SIGNATURE */
1559                 case TCP_NODELAY:
1560                         optval = tp->t_flags & TF_NODELAY;
1561                         break;
1562                 case TCP_MAXSEG:
1563                         optval = tp->t_maxseg;
1564                         break;
1565                 case TCP_NOOPT:
1566                         optval = tp->t_flags & TF_NOOPT;
1567                         break;
1568                 case TCP_NOPUSH:
1569                         optval = tp->t_flags & TF_NOPUSH;
1570                         break;
1571                 case TCP_KEEPINIT:
1572                         optval = ((int64_t)tp->t_keepinit * 1000) / hz;
1573                         break;
1574                 case TCP_KEEPIDLE:
1575                         optval = ((int64_t)tp->t_keepidle * 1000) / hz;
1576                         break;
1577                 case TCP_KEEPINTVL:
1578                         optval = ((int64_t)tp->t_keepintvl * 1000) / hz;
1579                         break;
1580                 case TCP_KEEPCNT:
1581                         optval = tp->t_keepcnt;
1582                         break;
1583                 default:
1584                         error = ENOPROTOOPT;
1585                         break;
1586                 }
1587                 if (error == 0)
1588                         soopt_from_kbuf(sopt, &optval, sizeof optval);
1589                 break;
1590         }
1591 done:
1592         lwkt_replymsg(&msg->lmsg, error);
1593 }
1594
1595 /*
1596  * tcp_sendspace and tcp_recvspace are the default send and receive window
1597  * sizes, respectively.  These are obsolescent (this information should
1598  * be set by the route).
1599  *
1600  * Use a default that does not require tcp window scaling to be turned
1601  * on.  Individual programs or the administrator can increase the default.
1602  */
1603 u_long  tcp_sendspace = 57344;  /* largest multiple of PAGE_SIZE < 64k */
1604 SYSCTL_INT(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW,
1605     &tcp_sendspace , 0, "Maximum outgoing TCP datagram size");
1606 u_long  tcp_recvspace = 57344;  /* largest multiple of PAGE_SIZE < 64k */
1607 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
1608     &tcp_recvspace , 0, "Maximum incoming TCP datagram size");
1609
1610 /*
1611  * Attach TCP protocol to socket, allocating internet protocol control
1612  * block, tcp control block, buffer space, and entering CLOSED state.
1613  */
1614 static int
1615 tcp_attach(struct socket *so, struct pru_attach_info *ai)
1616 {
1617         struct tcpcb *tp;
1618         struct inpcb *inp;
1619         int error;
1620         int cpu;
1621 #ifdef INET6
1622         boolean_t isipv6 = INP_CHECK_SOCKAF(so, AF_INET6);
1623 #endif
1624
1625         if (ai != NULL) {
1626                 error = tcp_usr_preattach(so, 0 /* don't care */, ai);
1627                 if (error)
1628                         return (error);
1629         } else {
1630                 /* Post attach; do nothing */
1631         }
1632
1633         cpu = mycpu->gd_cpuid;
1634
1635         /*
1636          * Set the default pcbinfo.  This will likely change when we
1637          * bind/connect.
1638          */
1639         error = in_pcballoc(so, &tcbinfo[cpu]);
1640         if (error)
1641                 return (error);
1642         inp = so->so_pcb;
1643 #ifdef INET6
1644         if (isipv6)
1645                 inp->in6p_hops = -1;    /* use kernel default */
1646 #endif
1647         tp = tcp_newtcpcb(inp);
1648         if (tp == NULL) {
1649                 /*
1650                  * Make sure the socket is destroyed by the pcbdetach.
1651                  */
1652                 soreference(so);
1653 #ifdef INET6
1654                 if (isipv6)
1655                         in6_pcbdetach(inp);
1656                 else
1657 #endif
1658                 in_pcbdetach(inp);
1659                 sofree(so);     /* from ref above */
1660                 return (ENOBUFS);
1661         }
1662         tp->t_state = TCPS_CLOSED;
1663         /* Keep a reference for asynchronized pru_rcvd */
1664         soreference(so);
1665         return (0);
1666 }
1667
1668 /*
1669  * Initiate (or continue) disconnect.
1670  * If embryonic state, just send reset (once).
1671  * If in ``let data drain'' option and linger null, just drop.
1672  * Otherwise (hard), mark socket disconnecting and drop
1673  * current input data; switch states based on user close, and
1674  * send segment to peer (with FIN).
1675  */
1676 static struct tcpcb *
1677 tcp_disconnect(struct tcpcb *tp)
1678 {
1679         struct socket *so = tp->t_inpcb->inp_socket;
1680
1681         if (tp->t_state < TCPS_ESTABLISHED) {
1682                 tp = tcp_close(tp);
1683         } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
1684                 tp = tcp_drop(tp, 0);
1685         } else {
1686                 lwkt_gettoken(&so->so_rcv.ssb_token);
1687                 soisdisconnecting(so);
1688                 sbflush(&so->so_rcv.sb);
1689                 tp = tcp_usrclosed(tp);
1690                 if (tp)
1691                         tcp_output(tp);
1692                 lwkt_reltoken(&so->so_rcv.ssb_token);
1693         }
1694         return (tp);
1695 }
1696
1697 /*
1698  * User issued close, and wish to trail through shutdown states:
1699  * if never received SYN, just forget it.  If got a SYN from peer,
1700  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
1701  * If already got a FIN from peer, then almost done; go to LAST_ACK
1702  * state.  In all other cases, have already sent FIN to peer (e.g.
1703  * after PRU_SHUTDOWN), and just have to play tedious game waiting
1704  * for peer to send FIN or not respond to keep-alives, etc.
1705  * We can let the user exit from the close as soon as the FIN is acked.
1706  */
1707 static struct tcpcb *
1708 tcp_usrclosed(struct tcpcb *tp)
1709 {
1710
1711         switch (tp->t_state) {
1712
1713         case TCPS_CLOSED:
1714         case TCPS_LISTEN:
1715                 tp->t_state = TCPS_CLOSED;
1716                 tp = tcp_close(tp);
1717                 break;
1718
1719         case TCPS_SYN_SENT:
1720         case TCPS_SYN_RECEIVED:
1721                 tp->t_flags |= TF_NEEDFIN;
1722                 break;
1723
1724         case TCPS_ESTABLISHED:
1725                 tp->t_state = TCPS_FIN_WAIT_1;
1726                 break;
1727
1728         case TCPS_CLOSE_WAIT:
1729                 tp->t_state = TCPS_LAST_ACK;
1730                 break;
1731         }
1732         if (tp && tp->t_state >= TCPS_FIN_WAIT_2) {
1733                 soisdisconnected(tp->t_inpcb->inp_socket);
1734                 /* To prevent the connection hanging in FIN_WAIT_2 forever. */
1735                 if (tp->t_state == TCPS_FIN_WAIT_2) {
1736                         tcp_callout_reset(tp, tp->tt_2msl, tp->t_maxidle,
1737                             tcp_timer_2msl);
1738                 }
1739         }
1740         return (tp);
1741 }