53fc1208dfc529763dd3803b55bba6ef5f69dd04
[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  */
69
70 #include "opt_ipsec.h"
71 #include "opt_inet.h"
72 #include "opt_inet6.h"
73 #include "opt_tcpdebug.h"
74
75 #include <sys/param.h>
76 #include <sys/systm.h>
77 #include <sys/kernel.h>
78 #include <sys/malloc.h>
79 #include <sys/sysctl.h>
80 #include <sys/globaldata.h>
81 #include <sys/thread.h>
82
83 #include <sys/mbuf.h>
84 #ifdef INET6
85 #include <sys/domain.h>
86 #endif /* INET6 */
87 #include <sys/socket.h>
88 #include <sys/socketvar.h>
89 #include <sys/protosw.h>
90
91 #include <sys/thread2.h>
92 #include <sys/msgport2.h>
93 #include <sys/socketvar2.h>
94
95 #include <net/if.h>
96 #include <net/netisr.h>
97 #include <net/route.h>
98
99 #include <net/netmsg2.h>
100
101 #include <netinet/in.h>
102 #include <netinet/in_systm.h>
103 #ifdef INET6
104 #include <netinet/ip6.h>
105 #endif
106 #include <netinet/in_pcb.h>
107 #ifdef INET6
108 #include <netinet6/in6_pcb.h>
109 #endif
110 #include <netinet/in_var.h>
111 #include <netinet/ip_var.h>
112 #ifdef INET6
113 #include <netinet6/ip6_var.h>
114 #include <netinet6/tcp6_var.h>
115 #endif
116 #include <netinet/tcp.h>
117 #include <netinet/tcp_fsm.h>
118 #include <netinet/tcp_seq.h>
119 #include <netinet/tcp_timer.h>
120 #include <netinet/tcp_timer2.h>
121 #include <netinet/tcp_var.h>
122 #include <netinet/tcpip.h>
123 #ifdef TCPDEBUG
124 #include <netinet/tcp_debug.h>
125 #endif
126
127 #ifdef IPSEC
128 #include <netinet6/ipsec.h>
129 #endif /*IPSEC*/
130
131 /*
132  * TCP protocol interface to socket abstraction.
133  */
134 extern  char *tcpstates[];      /* XXX ??? */
135
136 static int      tcp_attach (struct socket *, struct pru_attach_info *);
137 static void     tcp_connect (netmsg_t msg);
138 #ifdef INET6
139 static void     tcp6_connect (netmsg_t msg);
140 static int      tcp6_connect_oncpu(struct tcpcb *tp, int flags,
141                                 struct mbuf **mp,
142                                 struct sockaddr_in6 *sin6,
143                                 struct in6_addr *addr6);
144 #endif /* INET6 */
145 static struct tcpcb *
146                 tcp_disconnect (struct tcpcb *);
147 static struct tcpcb *
148                 tcp_usrclosed (struct tcpcb *);
149
150 #ifdef TCPDEBUG
151 #define TCPDEBUG0       int ostate = 0
152 #define TCPDEBUG1()     ostate = tp ? tp->t_state : 0
153 #define TCPDEBUG2(req)  if (tp && (so->so_options & SO_DEBUG)) \
154                                 tcp_trace(TA_USER, ostate, tp, 0, 0, req)
155 #else
156 #define TCPDEBUG0
157 #define TCPDEBUG1()
158 #define TCPDEBUG2(req)
159 #endif
160
161 static int      tcp_lport_extension = 1;
162 SYSCTL_INT(_net_inet_tcp, OID_AUTO, lportext, CTLFLAG_RW,
163     &tcp_lport_extension, 0, "");
164
165 /*
166  * For some ill optimized programs, which try to use TCP_NOPUSH
167  * to improve performance, will have small amount of data sits
168  * in the sending buffer.  These small amount of data will _not_
169  * be pushed into the network until more data are written into
170  * the socket or the socket write side is shutdown.
171  */ 
172 static int      tcp_disable_nopush = 1;
173 SYSCTL_INT(_net_inet_tcp, OID_AUTO, disable_nopush, CTLFLAG_RW,
174     &tcp_disable_nopush, 0, "TCP_NOPUSH socket option will have no effect");
175
176 /*
177  * TCP attaches to socket via pru_attach(), reserving space,
178  * and an internet control block.  This is likely occuring on
179  * cpu0 and may have to move later when we bind/connect.
180  */
181 static void
182 tcp_usr_attach(netmsg_t msg)
183 {
184         struct socket *so = msg->base.nm_so;
185         struct pru_attach_info *ai = msg->attach.nm_ai;
186         int error;
187         struct inpcb *inp;
188         struct tcpcb *tp = NULL;
189         TCPDEBUG0;
190
191         soreference(so);
192         inp = so->so_pcb;
193         TCPDEBUG1();
194         if (inp) {
195                 error = EISCONN;
196                 goto out;
197         }
198
199         error = tcp_attach(so, ai);
200         if (error)
201                 goto out;
202
203         if ((so->so_options & SO_LINGER) && so->so_linger == 0)
204                 so->so_linger = TCP_LINGERTIME;
205         tp = sototcpcb(so);
206 out:
207         sofree(so);             /* from ref above */
208         TCPDEBUG2(PRU_ATTACH);
209         lwkt_replymsg(&msg->lmsg, error);
210 }
211
212 /*
213  * pru_detach() detaches the TCP protocol from the socket.
214  * If the protocol state is non-embryonic, then can't
215  * do this directly: have to initiate a pru_disconnect(),
216  * which may finish later; embryonic TCB's can just
217  * be discarded here.
218  */
219 static void
220 tcp_usr_detach(netmsg_t msg)
221 {
222         struct socket *so = msg->base.nm_so;
223         int error = 0;
224         struct inpcb *inp;
225         struct tcpcb *tp;
226         TCPDEBUG0;
227
228         inp = so->so_pcb;
229
230         /*
231          * If the inp is already detached it may have been due to an async
232          * close.  Just return as if no error occured.
233          *
234          * It's possible for the tcpcb (tp) to disconnect from the inp due
235          * to tcp_drop()->tcp_close() being called.  This may occur *after*
236          * the detach message has been queued so we may find a NULL tp here.
237          */
238         if (inp) {
239                 if ((tp = intotcpcb(inp)) != NULL) {
240                         TCPDEBUG1();
241                         tp = tcp_disconnect(tp);
242                         TCPDEBUG2(PRU_DETACH);
243                 }
244         }
245         lwkt_replymsg(&msg->lmsg, error);
246 }
247
248 /*
249  * NOTE: ignore_error is non-zero for certain disconnection races
250  * which we want to silently allow, otherwise close() may return
251  * an unexpected error.
252  *
253  * NOTE: The variables (msg) and (tp) are assumed.
254  */
255 #define COMMON_START(so, inp, ignore_error)                     \
256         TCPDEBUG0;                                              \
257                                                                 \
258         inp = so->so_pcb;                                       \
259         do {                                                    \
260                  if (inp == NULL) {                             \
261                         error = ignore_error ? 0 : EINVAL;      \
262                         tp = NULL;                              \
263                         goto out;                               \
264                  }                                              \
265                  tp = intotcpcb(inp);                           \
266                  TCPDEBUG1();                                   \
267         } while(0)
268
269 #define COMMON_END1(req, noreply)                               \
270         out: do {                                               \
271                 TCPDEBUG2(req);                                 \
272                 if (!(noreply))                                 \
273                         lwkt_replymsg(&msg->lmsg, error);       \
274                 return;                                         \
275         } while(0)
276
277 #define COMMON_END(req)         COMMON_END1((req), 0)
278
279 /*
280  * Give the socket an address.
281  */
282 static void
283 tcp_usr_bind(netmsg_t msg)
284 {
285         struct socket *so = msg->bind.base.nm_so;
286         struct sockaddr *nam = msg->bind.nm_nam;
287         struct thread *td = msg->bind.nm_td;
288         int error = 0;
289         struct inpcb *inp;
290         struct tcpcb *tp;
291         struct sockaddr_in *sinp;
292
293         COMMON_START(so, inp, 0);
294
295         /*
296          * Must check for multicast addresses and disallow binding
297          * to them.
298          */
299         sinp = (struct sockaddr_in *)nam;
300         if (sinp->sin_family == AF_INET &&
301             IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
302                 error = EAFNOSUPPORT;
303                 goto out;
304         }
305         error = in_pcbbind(inp, nam, td);
306         if (error)
307                 goto out;
308         COMMON_END(PRU_BIND);
309
310 }
311
312 #ifdef INET6
313
314 static void
315 tcp6_usr_bind(netmsg_t msg)
316 {
317         struct socket *so = msg->bind.base.nm_so;
318         struct sockaddr *nam = msg->bind.nm_nam;
319         struct thread *td = msg->bind.nm_td;
320         int error = 0;
321         struct inpcb *inp;
322         struct tcpcb *tp;
323         struct sockaddr_in6 *sin6p;
324
325         COMMON_START(so, inp, 0);
326
327         /*
328          * Must check for multicast addresses and disallow binding
329          * to them.
330          */
331         sin6p = (struct sockaddr_in6 *)nam;
332         if (sin6p->sin6_family == AF_INET6 &&
333             IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
334                 error = EAFNOSUPPORT;
335                 goto out;
336         }
337         inp->inp_vflag &= ~INP_IPV4;
338         inp->inp_vflag |= INP_IPV6;
339         if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
340                 if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr))
341                         inp->inp_vflag |= INP_IPV4;
342                 else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
343                         struct sockaddr_in sin;
344
345                         in6_sin6_2_sin(&sin, sin6p);
346                         inp->inp_vflag |= INP_IPV4;
347                         inp->inp_vflag &= ~INP_IPV6;
348                         error = in_pcbbind(inp, (struct sockaddr *)&sin, td);
349                         goto out;
350                 }
351         }
352         error = in6_pcbbind(inp, nam, td);
353         if (error)
354                 goto out;
355         COMMON_END(PRU_BIND);
356 }
357 #endif /* INET6 */
358
359 #ifdef SMP
360
361 struct netmsg_inswildcard {
362         struct netmsg_base      base;
363         struct inpcb            *nm_inp;
364 };
365
366 static void
367 in_pcbinswildcardhash_handler(netmsg_t msg)
368 {
369         struct netmsg_inswildcard *nm = (struct netmsg_inswildcard *)msg;
370         int cpu = mycpuid, nextcpu;
371
372         in_pcbinswildcardhash_oncpu(nm->nm_inp, &tcbinfo[cpu]);
373
374         nextcpu = cpu + 1;
375         if (nextcpu < ncpus2)
376                 lwkt_forwardmsg(cpu_portfn(nextcpu), &nm->base.lmsg);
377         else
378                 lwkt_replymsg(&nm->base.lmsg, 0);
379 }
380
381 #endif
382
383 /*
384  * Prepare to accept connections.
385  */
386 static void
387 tcp_usr_listen(netmsg_t msg)
388 {
389         struct socket *so = msg->listen.base.nm_so;
390         struct thread *td = msg->listen.nm_td;
391         int error = 0;
392         struct inpcb *inp;
393         struct tcpcb *tp;
394 #ifdef SMP
395         struct netmsg_inswildcard nm;
396 #endif
397
398         COMMON_START(so, inp, 0);
399
400         if (tp->t_flags & TF_LISTEN)
401                 goto out;
402
403         if (inp->inp_lport == 0) {
404                 error = in_pcbbind(inp, NULL, td);
405                 if (error)
406                         goto out;
407         }
408
409         tp->t_state = TCPS_LISTEN;
410         tp->t_flags |= TF_LISTEN;
411         tp->tt_msg = NULL; /* Catch any invalid timer usage */
412
413 #ifdef SMP
414         if (ncpus > 1) {
415                 /*
416                  * We have to set the flag because we can't have other cpus
417                  * messing with our inp's flags.
418                  */
419                 KASSERT(!(inp->inp_flags & INP_CONNECTED),
420                         ("already on connhash"));
421                 KASSERT(!(inp->inp_flags & INP_WILDCARD),
422                         ("already on wildcardhash"));
423                 KASSERT(!(inp->inp_flags & INP_WILDCARD_MP),
424                         ("already on MP wildcardhash"));
425                 inp->inp_flags |= INP_WILDCARD_MP;
426
427                 KKASSERT(so->so_port == cpu_portfn(0));
428                 KKASSERT(&curthread->td_msgport == cpu_portfn(0));
429                 KKASSERT(inp->inp_pcbinfo == &tcbinfo[0]);
430
431                 netmsg_init(&nm.base, NULL, &curthread->td_msgport,
432                             MSGF_PRIORITY, in_pcbinswildcardhash_handler);
433                 nm.nm_inp = inp;
434                 lwkt_domsg(cpu_portfn(1), &nm.base.lmsg, 0);
435         }
436 #endif
437         in_pcbinswildcardhash(inp);
438         COMMON_END(PRU_LISTEN);
439 }
440
441 #ifdef INET6
442
443 static void
444 tcp6_usr_listen(netmsg_t msg)
445 {
446         struct socket *so = msg->listen.base.nm_so;
447         struct thread *td = msg->listen.nm_td;
448         int error = 0;
449         struct inpcb *inp;
450         struct tcpcb *tp;
451 #ifdef SMP
452         struct netmsg_inswildcard nm;
453 #endif
454
455         COMMON_START(so, inp, 0);
456
457         if (tp->t_flags & TF_LISTEN)
458                 goto out;
459
460         if (inp->inp_lport == 0) {
461                 if (!(inp->inp_flags & IN6P_IPV6_V6ONLY))
462                         inp->inp_vflag |= INP_IPV4;
463                 else
464                         inp->inp_vflag &= ~INP_IPV4;
465                 error = in6_pcbbind(inp, NULL, td);
466                 if (error)
467                         goto out;
468         }
469
470         tp->t_state = TCPS_LISTEN;
471         tp->t_flags |= TF_LISTEN;
472         tp->tt_msg = NULL; /* Catch any invalid timer usage */
473
474 #ifdef SMP
475         if (ncpus > 1) {
476                 /*
477                  * We have to set the flag because we can't have other cpus
478                  * messing with our inp's flags.
479                  */
480                 KASSERT(!(inp->inp_flags & INP_CONNECTED),
481                         ("already on connhash"));
482                 KASSERT(!(inp->inp_flags & INP_WILDCARD),
483                         ("already on wildcardhash"));
484                 KASSERT(!(inp->inp_flags & INP_WILDCARD_MP),
485                         ("already on MP wildcardhash"));
486                 inp->inp_flags |= INP_WILDCARD_MP;
487
488                 KKASSERT(so->so_port == cpu_portfn(0));
489                 KKASSERT(&curthread->td_msgport == cpu_portfn(0));
490                 KKASSERT(inp->inp_pcbinfo == &tcbinfo[0]);
491
492                 netmsg_init(&nm.base, NULL, &curthread->td_msgport,
493                             MSGF_PRIORITY, in_pcbinswildcardhash_handler);
494                 nm.nm_inp = inp;
495                 lwkt_domsg(cpu_portfn(1), &nm.base.lmsg, 0);
496         }
497 #endif
498         in_pcbinswildcardhash(inp);
499         COMMON_END(PRU_LISTEN);
500 }
501 #endif /* INET6 */
502
503 /*
504  * Initiate connection to peer.
505  * Create a template for use in transmissions on this connection.
506  * Enter SYN_SENT state, and mark socket as connecting.
507  * Start keep-alive timer, and seed output sequence space.
508  * Send initial segment on connection.
509  */
510 static void
511 tcp_usr_connect(netmsg_t msg)
512 {
513         struct socket *so = msg->connect.base.nm_so;
514         struct sockaddr *nam = msg->connect.nm_nam;
515         struct thread *td = msg->connect.nm_td;
516         int error = 0;
517         struct inpcb *inp;
518         struct tcpcb *tp;
519         struct sockaddr_in *sinp;
520
521         COMMON_START(so, inp, 0);
522
523         /*
524          * Must disallow TCP ``connections'' to multicast addresses.
525          */
526         sinp = (struct sockaddr_in *)nam;
527         if (sinp->sin_family == AF_INET
528             && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
529                 error = EAFNOSUPPORT;
530                 goto out;
531         }
532
533         if (!prison_remote_ip(td, (struct sockaddr*)sinp)) {
534                 error = EAFNOSUPPORT; /* IPv6 only jail */
535                 goto out;
536         }
537
538         tcp_connect(msg);
539         /* msg is invalid now */
540         return;
541 out:
542         if (msg->connect.nm_m) {
543                 m_freem(msg->connect.nm_m);
544                 msg->connect.nm_m = NULL;
545         }
546         lwkt_replymsg(&msg->lmsg, error);
547 }
548
549 #ifdef INET6
550
551 static void
552 tcp6_usr_connect(netmsg_t msg)
553 {
554         struct socket *so = msg->connect.base.nm_so;
555         struct sockaddr *nam = msg->connect.nm_nam;
556         struct thread *td = msg->connect.nm_td;
557         int error = 0;
558         struct inpcb *inp;
559         struct tcpcb *tp;
560         struct sockaddr_in6 *sin6p;
561
562         COMMON_START(so, inp, 0);
563
564         /*
565          * Must disallow TCP ``connections'' to multicast addresses.
566          */
567         sin6p = (struct sockaddr_in6 *)nam;
568         if (sin6p->sin6_family == AF_INET6
569             && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
570                 error = EAFNOSUPPORT;
571                 goto out;
572         }
573
574         if (!prison_remote_ip(td, nam)) {
575                 error = EAFNOSUPPORT; /* IPv4 only jail */
576                 goto out;
577         }
578
579         if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
580                 struct sockaddr_in *sinp;
581
582                 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
583                         error = EINVAL;
584                         goto out;
585                 }
586                 sinp = kmalloc(sizeof(*sinp), M_LWKTMSG, M_INTWAIT);
587                 in6_sin6_2_sin(sinp, sin6p);
588                 inp->inp_vflag |= INP_IPV4;
589                 inp->inp_vflag &= ~INP_IPV6;
590                 msg->connect.nm_nam = (struct sockaddr *)sinp;
591                 msg->connect.nm_reconnect |= NMSG_RECONNECT_NAMALLOC;
592                 tcp_connect(msg);
593                 /* msg is invalid now */
594                 return;
595         }
596         inp->inp_vflag &= ~INP_IPV4;
597         inp->inp_vflag |= INP_IPV6;
598         inp->inp_inc.inc_isipv6 = 1;
599
600         msg->connect.nm_reconnect |= NMSG_RECONNECT_FALLBACK;
601         tcp6_connect(msg);
602         /* msg is invalid now */
603         return;
604 out:
605         if (msg->connect.nm_m) {
606                 m_freem(msg->connect.nm_m);
607                 msg->connect.nm_m = NULL;
608         }
609         lwkt_replymsg(&msg->lmsg, error);
610 }
611
612 #endif /* INET6 */
613
614 /*
615  * Initiate disconnect from peer.
616  * If connection never passed embryonic stage, just drop;
617  * else if don't need to let data drain, then can just drop anyways,
618  * else have to begin TCP shutdown process: mark socket disconnecting,
619  * drain unread data, state switch to reflect user close, and
620  * send segment (e.g. FIN) to peer.  Socket will be really disconnected
621  * when peer sends FIN and acks ours.
622  *
623  * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
624  */
625 static void
626 tcp_usr_disconnect(netmsg_t msg)
627 {
628         struct socket *so = msg->disconnect.base.nm_so;
629         int error = 0;
630         struct inpcb *inp;
631         struct tcpcb *tp;
632
633         COMMON_START(so, inp, 1);
634         tp = tcp_disconnect(tp);
635         COMMON_END(PRU_DISCONNECT);
636 }
637
638 /*
639  * Accept a connection.  Essentially all the work is
640  * done at higher levels; just return the address
641  * of the peer, storing through addr.
642  */
643 static void
644 tcp_usr_accept(netmsg_t msg)
645 {
646         struct socket *so = msg->accept.base.nm_so;
647         struct sockaddr **nam = msg->accept.nm_nam;
648         int error = 0;
649         struct inpcb *inp;
650         struct tcpcb *tp = NULL;
651         TCPDEBUG0;
652
653         inp = so->so_pcb;
654         if (so->so_state & SS_ISDISCONNECTED) {
655                 error = ECONNABORTED;
656                 goto out;
657         }
658         if (inp == 0) {
659                 error = EINVAL;
660                 goto out;
661         }
662
663         tp = intotcpcb(inp);
664         TCPDEBUG1();
665         in_setpeeraddr(so, nam);
666         COMMON_END(PRU_ACCEPT);
667 }
668
669 #ifdef INET6
670 static void
671 tcp6_usr_accept(netmsg_t msg)
672 {
673         struct socket *so = msg->accept.base.nm_so;
674         struct sockaddr **nam = msg->accept.nm_nam;
675         int error = 0;
676         struct inpcb *inp;
677         struct tcpcb *tp = NULL;
678         TCPDEBUG0;
679
680         inp = so->so_pcb;
681
682         if (so->so_state & SS_ISDISCONNECTED) {
683                 error = ECONNABORTED;
684                 goto out;
685         }
686         if (inp == 0) {
687                 error = EINVAL;
688                 goto out;
689         }
690         tp = intotcpcb(inp);
691         TCPDEBUG1();
692         in6_mapped_peeraddr(so, nam);
693         COMMON_END(PRU_ACCEPT);
694 }
695 #endif /* INET6 */
696 /*
697  * Mark the connection as being incapable of further output.
698  */
699 static void
700 tcp_usr_shutdown(netmsg_t msg)
701 {
702         struct socket *so = msg->shutdown.base.nm_so;
703         int error = 0;
704         struct inpcb *inp;
705         struct tcpcb *tp;
706
707         COMMON_START(so, inp, 0);
708         socantsendmore(so);
709         tp = tcp_usrclosed(tp);
710         if (tp)
711                 error = tcp_output(tp);
712         COMMON_END(PRU_SHUTDOWN);
713 }
714
715 /*
716  * After a receive, possibly send window update to peer.
717  */
718 static void
719 tcp_usr_rcvd(netmsg_t msg)
720 {
721         struct socket *so = msg->rcvd.base.nm_so;
722         int error = 0;
723         struct inpcb *inp;
724         struct tcpcb *tp;
725
726         COMMON_START(so, inp, 0);
727         tcp_output(tp);
728         COMMON_END(PRU_RCVD);
729 }
730
731 /*
732  * Do a send by putting data in output queue and updating urgent
733  * marker if URG set.  Possibly send more data.  Unlike the other
734  * pru_*() routines, the mbuf chains are our responsibility.  We
735  * must either enqueue them or free them.  The other pru_* routines
736  * generally are caller-frees.
737  */
738 static void
739 tcp_usr_send(netmsg_t msg)
740 {
741         struct socket *so = msg->send.base.nm_so;
742         int flags = msg->send.nm_flags;
743         struct mbuf *m = msg->send.nm_m;
744         int error = 0;
745         struct inpcb *inp;
746         struct tcpcb *tp;
747         TCPDEBUG0;
748
749         KKASSERT(msg->send.nm_control == NULL);
750         KKASSERT(msg->send.nm_addr == NULL);
751         KKASSERT((flags & PRUS_FREEADDR) == 0);
752
753         inp = so->so_pcb;
754
755         if (inp == NULL) {
756                 /*
757                  * OOPS! we lost a race, the TCP session got reset after
758                  * we checked SS_CANTSENDMORE, eg: while doing uiomove or a
759                  * network interrupt in the non-critical section of sosend().
760                  */
761                 m_freem(m);
762                 error = ECONNRESET;     /* XXX EPIPE? */
763                 tp = NULL;
764                 TCPDEBUG1();
765                 goto out;
766         }
767         tp = intotcpcb(inp);
768         TCPDEBUG1();
769
770 #ifdef foo
771         /*
772          * This is no longer necessary, since:
773          * - sosendtcp() has already checked it for us
774          * - It does not work with asynchronized send
775          */
776
777         /*
778          * Don't let too much OOB data build up
779          */
780         if (flags & PRUS_OOB) {
781                 if (ssb_space(&so->so_snd) < -512) {
782                         m_freem(m);
783                         error = ENOBUFS;
784                         goto out;
785                 }
786         }
787 #endif
788
789         /*
790          * Pump the data into the socket.
791          */
792         if (m)
793                 ssb_appendstream(&so->so_snd, m);
794         if (flags & PRUS_OOB) {
795                 /*
796                  * According to RFC961 (Assigned Protocols),
797                  * the urgent pointer points to the last octet
798                  * of urgent data.  We continue, however,
799                  * to consider it to indicate the first octet
800                  * of data past the urgent section.
801                  * Otherwise, snd_up should be one lower.
802                  */
803                 tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
804                 tp->t_flags |= TF_FORCE;
805                 error = tcp_output(tp);
806                 tp->t_flags &= ~TF_FORCE;
807         } else {
808                 if (flags & PRUS_EOF) {
809                         /*
810                          * Close the send side of the connection after
811                          * the data is sent.
812                          */
813                         socantsendmore(so);
814                         tp = tcp_usrclosed(tp);
815                 }
816                 if (tp != NULL) {
817                         if (flags & PRUS_MORETOCOME)
818                                 tp->t_flags |= TF_MORETOCOME;
819                         error = tcp_output(tp);
820                         if (flags & PRUS_MORETOCOME)
821                                 tp->t_flags &= ~TF_MORETOCOME;
822                 }
823         }
824         COMMON_END1((flags & PRUS_OOB) ? PRU_SENDOOB :
825                    ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND),
826                    (flags & PRUS_NOREPLY));
827 }
828
829 /*
830  * NOTE: (so) is referenced from soabort*() and netmsg_pru_abort()
831  *       will sofree() it when we return.
832  */
833 static void
834 tcp_usr_abort(netmsg_t msg)
835 {
836         struct socket *so = msg->abort.base.nm_so;
837         int error = 0;
838         struct inpcb *inp;
839         struct tcpcb *tp;
840
841         COMMON_START(so, inp, 1);
842         tp = tcp_drop(tp, ECONNABORTED);
843         COMMON_END(PRU_ABORT);
844 }
845
846 /*
847  * Receive out-of-band data.
848  */
849 static void
850 tcp_usr_rcvoob(netmsg_t msg)
851 {
852         struct socket *so = msg->rcvoob.base.nm_so;
853         struct mbuf *m = msg->rcvoob.nm_m;
854         int flags = msg->rcvoob.nm_flags;
855         int error = 0;
856         struct inpcb *inp;
857         struct tcpcb *tp;
858
859         COMMON_START(so, inp, 0);
860         if ((so->so_oobmark == 0 &&
861              (so->so_state & SS_RCVATMARK) == 0) ||
862             so->so_options & SO_OOBINLINE ||
863             tp->t_oobflags & TCPOOB_HADDATA) {
864                 error = EINVAL;
865                 goto out;
866         }
867         if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
868                 error = EWOULDBLOCK;
869                 goto out;
870         }
871         m->m_len = 1;
872         *mtod(m, caddr_t) = tp->t_iobc;
873         if ((flags & MSG_PEEK) == 0)
874                 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
875         COMMON_END(PRU_RCVOOB);
876 }
877
878 static void
879 tcp_usr_savefaddr(struct socket *so, const struct sockaddr *faddr)
880 {
881         in_savefaddr(so, faddr);
882 }
883
884 #ifdef INET6
885 static void
886 tcp6_usr_savefaddr(struct socket *so, const struct sockaddr *faddr)
887 {
888         in6_mapped_savefaddr(so, faddr);
889 }
890 #endif
891
892 /* xxx - should be const */
893 struct pr_usrreqs tcp_usrreqs = {
894         .pru_abort = tcp_usr_abort,
895         .pru_accept = tcp_usr_accept,
896         .pru_attach = tcp_usr_attach,
897         .pru_bind = tcp_usr_bind,
898         .pru_connect = tcp_usr_connect,
899         .pru_connect2 = pr_generic_notsupp,
900         .pru_control = in_control_dispatch,
901         .pru_detach = tcp_usr_detach,
902         .pru_disconnect = tcp_usr_disconnect,
903         .pru_listen = tcp_usr_listen,
904         .pru_peeraddr = in_setpeeraddr_dispatch,
905         .pru_rcvd = tcp_usr_rcvd,
906         .pru_rcvoob = tcp_usr_rcvoob,
907         .pru_send = tcp_usr_send,
908         .pru_sense = pru_sense_null,
909         .pru_shutdown = tcp_usr_shutdown,
910         .pru_sockaddr = in_setsockaddr_dispatch,
911         .pru_sosend = sosendtcp,
912         .pru_soreceive = soreceive,
913         .pru_savefaddr = tcp_usr_savefaddr
914 };
915
916 #ifdef INET6
917 struct pr_usrreqs tcp6_usrreqs = {
918         .pru_abort = tcp_usr_abort,
919         .pru_accept = tcp6_usr_accept,
920         .pru_attach = tcp_usr_attach,
921         .pru_bind = tcp6_usr_bind,
922         .pru_connect = tcp6_usr_connect,
923         .pru_connect2 = pr_generic_notsupp,
924         .pru_control = in6_control_dispatch,
925         .pru_detach = tcp_usr_detach,
926         .pru_disconnect = tcp_usr_disconnect,
927         .pru_listen = tcp6_usr_listen,
928         .pru_peeraddr = in6_mapped_peeraddr_dispatch,
929         .pru_rcvd = tcp_usr_rcvd,
930         .pru_rcvoob = tcp_usr_rcvoob,
931         .pru_send = tcp_usr_send,
932         .pru_sense = pru_sense_null,
933         .pru_shutdown = tcp_usr_shutdown,
934         .pru_sockaddr = in6_mapped_sockaddr_dispatch,
935         .pru_sosend = sosendtcp,
936         .pru_soreceive = soreceive,
937         .pru_savefaddr = tcp6_usr_savefaddr
938 };
939 #endif /* INET6 */
940
941 static int
942 tcp_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf *m,
943                   struct sockaddr_in *sin, struct sockaddr_in *if_sin)
944 {
945         struct inpcb *inp = tp->t_inpcb, *oinp;
946         struct socket *so = inp->inp_socket;
947         struct route *ro = &inp->inp_route;
948
949         oinp = in_pcblookup_hash(&tcbinfo[mycpu->gd_cpuid],
950                                  sin->sin_addr, sin->sin_port,
951                                  (inp->inp_laddr.s_addr != INADDR_ANY ?
952                                   inp->inp_laddr : if_sin->sin_addr),
953                                 inp->inp_lport, 0, NULL);
954         if (oinp != NULL) {
955                 m_freem(m);
956                 return (EADDRINUSE);
957         }
958         if (inp->inp_laddr.s_addr == INADDR_ANY)
959                 inp->inp_laddr = if_sin->sin_addr;
960         inp->inp_faddr = sin->sin_addr;
961         inp->inp_fport = sin->sin_port;
962         inp->inp_cpcbinfo = &tcbinfo[mycpu->gd_cpuid];
963         in_pcbinsconnhash(inp);
964
965         /*
966          * We are now on the inpcb's owner CPU, if the cached route was
967          * freed because the rtentry's owner CPU is not the current CPU
968          * (e.g. in tcp_connect()), then we try to reallocate it here with
969          * the hope that a rtentry may be cloned from a RTF_PRCLONING
970          * rtentry.
971          */
972         if (!(inp->inp_socket->so_options & SO_DONTROUTE) && /*XXX*/
973             ro->ro_rt == NULL) {
974                 bzero(&ro->ro_dst, sizeof(struct sockaddr_in));
975                 ro->ro_dst.sa_family = AF_INET;
976                 ro->ro_dst.sa_len = sizeof(struct sockaddr_in);
977                 ((struct sockaddr_in *)&ro->ro_dst)->sin_addr =
978                         sin->sin_addr;
979                 rtalloc(ro);
980         }
981
982         /*
983          * Now that no more errors can occur, change the protocol processing
984          * port to the current thread (which is the correct thread).
985          *
986          * Create TCP timer message now; we are on the tcpcb's owner
987          * CPU/thread.
988          */
989         tcp_create_timermsg(tp, &curthread->td_msgport);
990
991         /*
992          * Compute window scaling to request.  Use a larger scaling then
993          * needed for the initial receive buffer in case the receive buffer
994          * gets expanded.
995          */
996         if (tp->request_r_scale < TCP_MIN_WINSHIFT)
997                 tp->request_r_scale = TCP_MIN_WINSHIFT;
998         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
999                (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat
1000         ) {
1001                 tp->request_r_scale++;
1002         }
1003
1004         soisconnecting(so);
1005         tcpstat.tcps_connattempt++;
1006         tp->t_state = TCPS_SYN_SENT;
1007         tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
1008         tp->iss = tcp_new_isn(tp);
1009         tcp_sendseqinit(tp);
1010         if (m) {
1011                 ssb_appendstream(&so->so_snd, m);
1012                 m = NULL;
1013                 if (flags & PRUS_OOB)
1014                         tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
1015         }
1016
1017         /*
1018          * Close the send side of the connection after
1019          * the data is sent if flagged.
1020          */
1021         if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
1022                 socantsendmore(so);
1023                 tp = tcp_usrclosed(tp);
1024         }
1025         return (tcp_output(tp));
1026 }
1027
1028 /*
1029  * Common subroutine to open a TCP connection to remote host specified
1030  * by struct sockaddr_in in mbuf *nam.  Call in_pcbbind to assign a local
1031  * port number if needed.  Call in_pcbladdr to do the routing and to choose
1032  * a local host address (interface).
1033  * Initialize connection parameters and enter SYN-SENT state.
1034  */
1035 static void
1036 tcp_connect(netmsg_t msg)
1037 {
1038         struct socket *so = msg->connect.base.nm_so;
1039         struct sockaddr *nam = msg->connect.nm_nam;
1040         struct thread *td = msg->connect.nm_td;
1041         struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1042         struct sockaddr_in *if_sin;
1043         struct inpcb *inp;
1044         struct tcpcb *tp;
1045         int error, calc_laddr = 1;
1046 #ifdef SMP
1047         lwkt_port_t port;
1048 #endif
1049
1050         COMMON_START(so, inp, 0);
1051
1052         /*
1053          * Reconnect our pcb if we have to
1054          */
1055         if (msg->connect.nm_reconnect & NMSG_RECONNECT_RECONNECT) {
1056                 msg->connect.nm_reconnect &= ~NMSG_RECONNECT_RECONNECT;
1057                 in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1058         }
1059
1060         /*
1061          * Bind if we have to
1062          */
1063         if (inp->inp_lport == 0) {
1064                 if (tcp_lport_extension) {
1065                         KKASSERT(inp->inp_laddr.s_addr == INADDR_ANY);
1066
1067                         error = in_pcbladdr(inp, nam, &if_sin, td);
1068                         if (error)
1069                                 goto out;
1070                         inp->inp_laddr.s_addr = if_sin->sin_addr.s_addr;
1071
1072                         error = in_pcbconn_bind(inp, nam, td);
1073                         if (error)
1074                                 goto out;
1075
1076                         calc_laddr = 0;
1077                 } else {
1078                         error = in_pcbbind(inp, NULL, td);
1079                         if (error)
1080                                 goto out;
1081                 }
1082         }
1083
1084         if (calc_laddr) {
1085                 /*
1086                  * Calculate the correct protocol processing thread.  The
1087                  * connect operation must run there.  Set the forwarding
1088                  * port before we forward the message or it will get bounced
1089                  * right back to us.
1090                  */
1091                 error = in_pcbladdr(inp, nam, &if_sin, td);
1092                 if (error)
1093                         goto out;
1094         }
1095         KKASSERT(inp->inp_socket == so);
1096
1097 #ifdef SMP
1098         port = tcp_addrport(sin->sin_addr.s_addr, sin->sin_port,
1099                             (inp->inp_laddr.s_addr ?
1100                              inp->inp_laddr.s_addr : if_sin->sin_addr.s_addr),
1101                             inp->inp_lport);
1102
1103         if (port != &curthread->td_msgport) {
1104                 struct route *ro = &inp->inp_route;
1105
1106                 /*
1107                  * in_pcbladdr() may have allocated a route entry for us
1108                  * on the current CPU, but we need a route entry on the
1109                  * inpcb's owner CPU, so free it here.
1110                  */
1111                 if (ro->ro_rt != NULL)
1112                         RTFREE(ro->ro_rt);
1113                 bzero(ro, sizeof(*ro));
1114
1115                 /*
1116                  * We are moving the protocol processing port the socket
1117                  * is on, we have to unlink here and re-link on the
1118                  * target cpu.
1119                  */
1120                 in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1121                 sosetport(so, port);
1122                 msg->connect.nm_reconnect |= NMSG_RECONNECT_RECONNECT;
1123                 msg->connect.base.nm_dispatch = tcp_connect;
1124
1125                 lwkt_forwardmsg(port, &msg->connect.base.lmsg);
1126                 /* msg invalid now */
1127                 return;
1128         }
1129 #else
1130         KKASSERT(so->so_port == &curthread->td_msgport);
1131 #endif
1132         error = tcp_connect_oncpu(tp, msg->connect.nm_flags,
1133                                   msg->connect.nm_m, sin, if_sin);
1134         msg->connect.nm_m = NULL;
1135 out:
1136         if (msg->connect.nm_m) {
1137                 m_freem(msg->connect.nm_m);
1138                 msg->connect.nm_m = NULL;
1139         }
1140         if (msg->connect.nm_reconnect & NMSG_RECONNECT_NAMALLOC) {
1141                 kfree(msg->connect.nm_nam, M_LWKTMSG);
1142                 msg->connect.nm_nam = NULL;
1143         }
1144         lwkt_replymsg(&msg->connect.base.lmsg, error);
1145         /* msg invalid now */
1146 }
1147
1148 #ifdef INET6
1149
1150 static void
1151 tcp6_connect(netmsg_t msg)
1152 {
1153         struct tcpcb *tp;
1154         struct socket *so = msg->connect.base.nm_so;
1155         struct sockaddr *nam = msg->connect.nm_nam;
1156         struct thread *td = msg->connect.nm_td;
1157         struct inpcb *inp;
1158         struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam;
1159         struct in6_addr *addr6;
1160 #ifdef SMP
1161         lwkt_port_t port;
1162 #endif
1163         int error;
1164
1165         COMMON_START(so, inp, 0);
1166
1167         /*
1168          * Reconnect our pcb if we have to
1169          */
1170         if (msg->connect.nm_reconnect & NMSG_RECONNECT_RECONNECT) {
1171                 msg->connect.nm_reconnect &= ~NMSG_RECONNECT_RECONNECT;
1172                 in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1173         }
1174
1175         /*
1176          * Bind if we have to
1177          */
1178         if (inp->inp_lport == 0) {
1179                 error = in6_pcbbind(inp, NULL, td);
1180                 if (error)
1181                         goto out;
1182         }
1183
1184         /*
1185          * Cannot simply call in_pcbconnect, because there might be an
1186          * earlier incarnation of this same connection still in
1187          * TIME_WAIT state, creating an ADDRINUSE error.
1188          */
1189         error = in6_pcbladdr(inp, nam, &addr6, td);
1190         if (error)
1191                 goto out;
1192
1193 #ifdef SMP
1194         port = tcp6_addrport(); /* XXX hack for now, always cpu0 */
1195
1196         if (port != &curthread->td_msgport) {
1197                 struct route *ro = &inp->inp_route;
1198
1199                 /*
1200                  * in_pcbladdr() may have allocated a route entry for us
1201                  * on the current CPU, but we need a route entry on the
1202                  * inpcb's owner CPU, so free it here.
1203                  */
1204                 if (ro->ro_rt != NULL)
1205                         RTFREE(ro->ro_rt);
1206                 bzero(ro, sizeof(*ro));
1207
1208                 in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1209                 sosetport(so, port);
1210                 msg->connect.nm_reconnect |= NMSG_RECONNECT_RECONNECT;
1211                 msg->connect.base.nm_dispatch = tcp6_connect;
1212
1213                 lwkt_forwardmsg(port, &msg->connect.base.lmsg);
1214                 /* msg invalid now */
1215                 return;
1216         }
1217 #endif
1218         error = tcp6_connect_oncpu(tp, msg->connect.nm_flags,
1219                                    &msg->connect.nm_m, sin6, addr6);
1220         /* nm_m may still be intact */
1221 out:
1222         if (error && (msg->connect.nm_reconnect & NMSG_RECONNECT_FALLBACK)) {
1223                 tcp_connect(msg);
1224                 /* msg invalid now */
1225         } else {
1226                 if (msg->connect.nm_m) {
1227                         m_freem(msg->connect.nm_m);
1228                         msg->connect.nm_m = NULL;
1229                 }
1230                 if (msg->connect.nm_reconnect & NMSG_RECONNECT_NAMALLOC) {
1231                         kfree(msg->connect.nm_nam, M_LWKTMSG);
1232                         msg->connect.nm_nam = NULL;
1233                 }
1234                 lwkt_replymsg(&msg->connect.base.lmsg, error);
1235                 /* msg invalid now */
1236         }
1237 }
1238
1239 static int
1240 tcp6_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf **mp,
1241                    struct sockaddr_in6 *sin6, struct in6_addr *addr6)
1242 {
1243         struct mbuf *m = *mp;
1244         struct inpcb *inp = tp->t_inpcb;
1245         struct socket *so = inp->inp_socket;
1246         struct inpcb *oinp;
1247
1248         /*
1249          * Cannot simply call in_pcbconnect, because there might be an
1250          * earlier incarnation of this same connection still in
1251          * TIME_WAIT state, creating an ADDRINUSE error.
1252          */
1253         oinp = in6_pcblookup_hash(inp->inp_cpcbinfo,
1254                                   &sin6->sin6_addr, sin6->sin6_port,
1255                                   (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) ?
1256                                       addr6 : &inp->in6p_laddr),
1257                                   inp->inp_lport,  0, NULL);
1258         if (oinp)
1259                 return (EADDRINUSE);
1260
1261         if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
1262                 inp->in6p_laddr = *addr6;
1263         inp->in6p_faddr = sin6->sin6_addr;
1264         inp->inp_fport = sin6->sin6_port;
1265         if ((sin6->sin6_flowinfo & IPV6_FLOWINFO_MASK) != 0)
1266                 inp->in6p_flowinfo = sin6->sin6_flowinfo;
1267         in_pcbinsconnhash(inp);
1268
1269         /*
1270          * Now that no more errors can occur, change the protocol processing
1271          * port to the current thread (which is the correct thread).
1272          *
1273          * Create TCP timer message now; we are on the tcpcb's owner
1274          * CPU/thread.
1275          */
1276         tcp_create_timermsg(tp, &curthread->td_msgport);
1277
1278         /* Compute window scaling to request.  */
1279         if (tp->request_r_scale < TCP_MIN_WINSHIFT)
1280                 tp->request_r_scale = TCP_MIN_WINSHIFT;
1281         while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1282             (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat) {
1283                 tp->request_r_scale++;
1284         }
1285
1286         soisconnecting(so);
1287         tcpstat.tcps_connattempt++;
1288         tp->t_state = TCPS_SYN_SENT;
1289         tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
1290         tp->iss = tcp_new_isn(tp);
1291         tcp_sendseqinit(tp);
1292         if (m) {
1293                 ssb_appendstream(&so->so_snd, m);
1294                 *mp = NULL;
1295                 if (flags & PRUS_OOB)
1296                         tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
1297         }
1298
1299         /*
1300          * Close the send side of the connection after
1301          * the data is sent if flagged.
1302          */
1303         if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
1304                 socantsendmore(so);
1305                 tp = tcp_usrclosed(tp);
1306         }
1307         return (tcp_output(tp));
1308 }
1309
1310 #endif /* INET6 */
1311
1312 /*
1313  * The new sockopt interface makes it possible for us to block in the
1314  * copyin/out step (if we take a page fault).  Taking a page fault while
1315  * in a critical section is probably a Bad Thing.  (Since sockets and pcbs
1316  * both now use TSM, there probably isn't any need for this function to 
1317  * run in a critical section any more.  This needs more examination.)
1318  */
1319 void
1320 tcp_ctloutput(netmsg_t msg)
1321 {
1322         struct socket *so = msg->base.nm_so;
1323         struct sockopt *sopt = msg->ctloutput.nm_sopt;
1324         int     error, opt, optval, opthz;
1325         struct  inpcb *inp;
1326         struct  tcpcb *tp;
1327
1328         error = 0;
1329         inp = so->so_pcb;
1330         if (inp == NULL) {
1331                 error = ECONNRESET;
1332                 goto done;
1333         }
1334
1335         if (sopt->sopt_level != IPPROTO_TCP) {
1336 #ifdef INET6
1337                 if (INP_CHECK_SOCKAF(so, AF_INET6))
1338                         ip6_ctloutput_dispatch(msg);
1339                 else
1340 #endif /* INET6 */
1341                 ip_ctloutput(msg);
1342                 /* msg invalid now */
1343                 return;
1344         }
1345         tp = intotcpcb(inp);
1346
1347         switch (sopt->sopt_dir) {
1348         case SOPT_SET:
1349                 error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1350                                       sizeof optval);
1351                 if (error)
1352                         break;
1353                 switch (sopt->sopt_name) {
1354                 case TCP_FASTKEEP:
1355                         if (optval > 0)
1356                                 tp->t_keepidle = tp->t_keepintvl;
1357                         else
1358                                 tp->t_keepidle = tcp_keepidle;
1359                         tcp_timer_keep_activity(tp, 0);
1360                         break;
1361 #ifdef TCP_SIGNATURE
1362                 case TCP_SIGNATURE_ENABLE:
1363                         if (tp->t_state == TCPS_CLOSED) {
1364                                 /*
1365                                  * This is the only safe state that this
1366                                  * option could be changed.  Some segments
1367                                  * could already have been sent in other
1368                                  * states.
1369                                  */
1370                                 if (optval > 0)
1371                                         tp->t_flags |= TF_SIGNATURE;
1372                                 else
1373                                         tp->t_flags &= ~TF_SIGNATURE;
1374                         } else {
1375                                 error = EOPNOTSUPP;
1376                         }
1377                         break;
1378 #endif /* TCP_SIGNATURE */
1379                 case TCP_NODELAY:
1380                 case TCP_NOOPT:
1381                         switch (sopt->sopt_name) {
1382                         case TCP_NODELAY:
1383                                 opt = TF_NODELAY;
1384                                 break;
1385                         case TCP_NOOPT:
1386                                 opt = TF_NOOPT;
1387                                 break;
1388                         default:
1389                                 opt = 0; /* dead code to fool gcc */
1390                                 break;
1391                         }
1392
1393                         if (optval)
1394                                 tp->t_flags |= opt;
1395                         else
1396                                 tp->t_flags &= ~opt;
1397                         break;
1398
1399                 case TCP_NOPUSH:
1400                         if (tcp_disable_nopush)
1401                                 break;
1402                         if (optval)
1403                                 tp->t_flags |= TF_NOPUSH;
1404                         else {
1405                                 tp->t_flags &= ~TF_NOPUSH;
1406                                 error = tcp_output(tp);
1407                         }
1408                         break;
1409
1410                 case TCP_MAXSEG:
1411                         /*
1412                          * Must be between 0 and maxseg.  If the requested
1413                          * maxseg is too small to satisfy the desired minmss,
1414                          * pump it up (silently so sysctl modifications of
1415                          * minmss do not create unexpected program failures).
1416                          * Handle degenerate cases.
1417                          */
1418                         if (optval > 0 && optval <= tp->t_maxseg) {
1419                                 if (optval + 40 < tcp_minmss) {
1420                                         optval = tcp_minmss - 40;
1421                                         if (optval < 0)
1422                                                 optval = 1;
1423                                 }
1424                                 tp->t_maxseg = optval;
1425                         } else {
1426                                 error = EINVAL;
1427                         }
1428                         break;
1429
1430                 case TCP_KEEPINIT:
1431                         opthz = ((int64_t)optval * hz) / 1000;
1432                         if (opthz >= 1)
1433                                 tp->t_keepinit = opthz;
1434                         else
1435                                 error = EINVAL;
1436                         break;
1437
1438                 case TCP_KEEPIDLE:
1439                         opthz = ((int64_t)optval * hz) / 1000;
1440                         if (opthz >= 1) {
1441                                 tp->t_keepidle = opthz;
1442                                 tcp_timer_keep_activity(tp, 0);
1443                         } else {
1444                                 error = EINVAL;
1445                         }
1446                         break;
1447
1448                 case TCP_KEEPINTVL:
1449                         opthz = ((int64_t)optval * hz) / 1000;
1450                         if (opthz >= 1) {
1451                                 tp->t_keepintvl = opthz;
1452                                 tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt;
1453                         } else {
1454                                 error = EINVAL;
1455                         }
1456                         break;
1457
1458                 case TCP_KEEPCNT:
1459                         if (optval > 0) {
1460                                 tp->t_keepcnt = optval;
1461                                 tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt;
1462                         } else {
1463                                 error = EINVAL;
1464                         }
1465                         break;
1466
1467                 default:
1468                         error = ENOPROTOOPT;
1469                         break;
1470                 }
1471                 break;
1472
1473         case SOPT_GET:
1474                 switch (sopt->sopt_name) {
1475 #ifdef TCP_SIGNATURE
1476                 case TCP_SIGNATURE_ENABLE:
1477                         optval = (tp->t_flags & TF_SIGNATURE) ? 1 : 0;
1478                         break;
1479 #endif /* TCP_SIGNATURE */
1480                 case TCP_NODELAY:
1481                         optval = tp->t_flags & TF_NODELAY;
1482                         break;
1483                 case TCP_MAXSEG:
1484                         optval = tp->t_maxseg;
1485                         break;
1486                 case TCP_NOOPT:
1487                         optval = tp->t_flags & TF_NOOPT;
1488                         break;
1489                 case TCP_NOPUSH:
1490                         optval = tp->t_flags & TF_NOPUSH;
1491                         break;
1492                 case TCP_KEEPINIT:
1493                         optval = ((int64_t)tp->t_keepinit * 1000) / hz;
1494                         break;
1495                 case TCP_KEEPIDLE:
1496                         optval = ((int64_t)tp->t_keepidle * 1000) / hz;
1497                         break;
1498                 case TCP_KEEPINTVL:
1499                         optval = ((int64_t)tp->t_keepintvl * 1000) / hz;
1500                         break;
1501                 case TCP_KEEPCNT:
1502                         optval = tp->t_keepcnt;
1503                         break;
1504                 default:
1505                         error = ENOPROTOOPT;
1506                         break;
1507                 }
1508                 if (error == 0)
1509                         soopt_from_kbuf(sopt, &optval, sizeof optval);
1510                 break;
1511         }
1512 done:
1513         lwkt_replymsg(&msg->lmsg, error);
1514 }
1515
1516 /*
1517  * tcp_sendspace and tcp_recvspace are the default send and receive window
1518  * sizes, respectively.  These are obsolescent (this information should
1519  * be set by the route).
1520  *
1521  * Use a default that does not require tcp window scaling to be turned
1522  * on.  Individual programs or the administrator can increase the default.
1523  */
1524 u_long  tcp_sendspace = 57344;  /* largest multiple of PAGE_SIZE < 64k */
1525 SYSCTL_INT(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW,
1526     &tcp_sendspace , 0, "Maximum outgoing TCP datagram size");
1527 u_long  tcp_recvspace = 57344;  /* largest multiple of PAGE_SIZE < 64k */
1528 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
1529     &tcp_recvspace , 0, "Maximum incoming TCP datagram size");
1530
1531 /*
1532  * Attach TCP protocol to socket, allocating internet protocol control
1533  * block, tcp control block, bufer space, and entering LISTEN state
1534  * if to accept connections.
1535  */
1536 static int
1537 tcp_attach(struct socket *so, struct pru_attach_info *ai)
1538 {
1539         struct tcpcb *tp;
1540         struct inpcb *inp;
1541         int error;
1542         int cpu;
1543 #ifdef INET6
1544         int isipv6 = INP_CHECK_SOCKAF(so, AF_INET6) != 0;
1545 #endif
1546
1547         if (so->so_snd.ssb_hiwat == 0 || so->so_rcv.ssb_hiwat == 0) {
1548                 lwkt_gettoken(&so->so_rcv.ssb_token);
1549                 error = soreserve(so, tcp_sendspace, tcp_recvspace,
1550                                   ai->sb_rlimit);
1551                 lwkt_reltoken(&so->so_rcv.ssb_token);
1552                 if (error)
1553                         return (error);
1554         }
1555         atomic_set_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE);
1556         atomic_set_int(&so->so_snd.ssb_flags, SSB_AUTOSIZE);
1557         cpu = mycpu->gd_cpuid;
1558
1559         /*
1560          * Set the default port for protocol processing. This will likely
1561          * change when we connect.
1562          */
1563         error = in_pcballoc(so, &tcbinfo[cpu]);
1564         if (error)
1565                 return (error);
1566         inp = so->so_pcb;
1567 #ifdef INET6
1568         if (isipv6) {
1569                 inp->inp_vflag |= INP_IPV6;
1570                 inp->in6p_hops = -1;    /* use kernel default */
1571         }
1572         else
1573 #endif
1574         inp->inp_vflag |= INP_IPV4;
1575         tp = tcp_newtcpcb(inp);
1576         if (tp == NULL) {
1577                 /*
1578                  * Make sure the socket is destroyed by the pcbdetach.
1579                  */
1580                 soreference(so);
1581 #ifdef INET6
1582                 if (isipv6)
1583                         in6_pcbdetach(inp);
1584                 else
1585 #endif
1586                 in_pcbdetach(inp);
1587                 sofree(so);     /* from ref above */
1588                 return (ENOBUFS);
1589         }
1590         tp->t_state = TCPS_CLOSED;
1591         return (0);
1592 }
1593
1594 /*
1595  * Initiate (or continue) disconnect.
1596  * If embryonic state, just send reset (once).
1597  * If in ``let data drain'' option and linger null, just drop.
1598  * Otherwise (hard), mark socket disconnecting and drop
1599  * current input data; switch states based on user close, and
1600  * send segment to peer (with FIN).
1601  */
1602 static struct tcpcb *
1603 tcp_disconnect(struct tcpcb *tp)
1604 {
1605         struct socket *so = tp->t_inpcb->inp_socket;
1606
1607         if (tp->t_state < TCPS_ESTABLISHED) {
1608                 tp = tcp_close(tp);
1609         } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
1610                 tp = tcp_drop(tp, 0);
1611         } else {
1612                 lwkt_gettoken(&so->so_rcv.ssb_token);
1613                 soisdisconnecting(so);
1614                 sbflush(&so->so_rcv.sb);
1615                 tp = tcp_usrclosed(tp);
1616                 if (tp)
1617                         tcp_output(tp);
1618                 lwkt_reltoken(&so->so_rcv.ssb_token);
1619         }
1620         return (tp);
1621 }
1622
1623 /*
1624  * User issued close, and wish to trail through shutdown states:
1625  * if never received SYN, just forget it.  If got a SYN from peer,
1626  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
1627  * If already got a FIN from peer, then almost done; go to LAST_ACK
1628  * state.  In all other cases, have already sent FIN to peer (e.g.
1629  * after PRU_SHUTDOWN), and just have to play tedious game waiting
1630  * for peer to send FIN or not respond to keep-alives, etc.
1631  * We can let the user exit from the close as soon as the FIN is acked.
1632  */
1633 static struct tcpcb *
1634 tcp_usrclosed(struct tcpcb *tp)
1635 {
1636
1637         switch (tp->t_state) {
1638
1639         case TCPS_CLOSED:
1640         case TCPS_LISTEN:
1641                 tp->t_state = TCPS_CLOSED;
1642                 tp = tcp_close(tp);
1643                 break;
1644
1645         case TCPS_SYN_SENT:
1646         case TCPS_SYN_RECEIVED:
1647                 tp->t_flags |= TF_NEEDFIN;
1648                 break;
1649
1650         case TCPS_ESTABLISHED:
1651                 tp->t_state = TCPS_FIN_WAIT_1;
1652                 break;
1653
1654         case TCPS_CLOSE_WAIT:
1655                 tp->t_state = TCPS_LAST_ACK;
1656                 break;
1657         }
1658         if (tp && tp->t_state >= TCPS_FIN_WAIT_2) {
1659                 soisdisconnected(tp->t_inpcb->inp_socket);
1660                 /* To prevent the connection hanging in FIN_WAIT_2 forever. */
1661                 if (tp->t_state == TCPS_FIN_WAIT_2) {
1662                         tcp_callout_reset(tp, tp->tt_2msl, tp->t_maxidle,
1663                             tcp_timer_2msl);
1664                 }
1665         }
1666         return (tp);
1667 }