tcp: Allow per-tcpcb keepintvl and keepcnt
[dragonfly.git] / sys / netinet / tcp_usrreq.c
CommitLineData
984263bc 1/*
66d6c637
JH
2 * Copyright (c) 2003, 2004 Jeffrey M. Hsu. All rights reserved.
3 * Copyright (c) 2003, 2004 The DragonFly Project. All rights reserved.
f23061d4 4 *
66d6c637
JH
5 * This code is derived from software contributed to The DragonFly Project
6 * by Jeffrey M. Hsu.
f23061d4 7 *
66d6c637
JH
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.
f23061d4 19 *
66d6c637
JH
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/*
984263bc
MD
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 $
69891757 68 * $DragonFly: src/sys/netinet/tcp_usrreq.c,v 1.51 2008/09/29 20:52:23 dillon Exp $
984263bc
MD
69 */
70
71#include "opt_ipsec.h"
b1992928 72#include "opt_inet.h"
984263bc
MD
73#include "opt_inet6.h"
74#include "opt_tcpdebug.h"
75
76#include <sys/param.h>
77#include <sys/systm.h>
78#include <sys/kernel.h>
7fe56515 79#include <sys/malloc.h>
984263bc 80#include <sys/sysctl.h>
d371a63a
JH
81#include <sys/globaldata.h>
82#include <sys/thread.h>
83
984263bc
MD
84#include <sys/mbuf.h>
85#ifdef INET6
86#include <sys/domain.h>
87#endif /* INET6 */
88#include <sys/socket.h>
89#include <sys/socketvar.h>
90#include <sys/protosw.h>
91
1cae611f 92#include <sys/thread2.h>
7fe56515 93#include <sys/msgport2.h>
6cef7136 94#include <sys/socketvar2.h>
7fe56515 95
984263bc 96#include <net/if.h>
7fe56515 97#include <net/netisr.h>
984263bc
MD
98#include <net/route.h>
99
4599cf19
MD
100#include <net/netmsg2.h>
101
984263bc
MD
102#include <netinet/in.h>
103#include <netinet/in_systm.h>
104#ifdef INET6
105#include <netinet/ip6.h>
106#endif
107#include <netinet/in_pcb.h>
108#ifdef INET6
109#include <netinet6/in6_pcb.h>
110#endif
111#include <netinet/in_var.h>
112#include <netinet/ip_var.h>
113#ifdef INET6
114#include <netinet6/ip6_var.h>
65f3e756 115#include <netinet6/tcp6_var.h>
984263bc
MD
116#endif
117#include <netinet/tcp.h>
118#include <netinet/tcp_fsm.h>
119#include <netinet/tcp_seq.h>
120#include <netinet/tcp_timer.h>
a48c5dd5 121#include <netinet/tcp_timer2.h>
984263bc
MD
122#include <netinet/tcp_var.h>
123#include <netinet/tcpip.h>
124#ifdef TCPDEBUG
125#include <netinet/tcp_debug.h>
126#endif
127
128#ifdef IPSEC
129#include <netinet6/ipsec.h>
130#endif /*IPSEC*/
131
132/*
133 * TCP protocol interface to socket abstraction.
134 */
135extern char *tcpstates[]; /* XXX ??? */
136
e4700d00 137static int tcp_attach (struct socket *, struct pru_attach_info *);
002c1265 138static void tcp_connect (netmsg_t msg);
984263bc 139#ifdef INET6
002c1265
MD
140static void tcp6_connect (netmsg_t msg);
141static int tcp6_connect_oncpu(struct tcpcb *tp, int flags,
142 struct mbuf **mp,
48e7b118 143 struct sockaddr_in6 *sin6,
65f3e756 144 struct in6_addr *addr6);
984263bc
MD
145#endif /* INET6 */
146static struct tcpcb *
42a7fc75 147 tcp_disconnect (struct tcpcb *);
984263bc 148static struct tcpcb *
42a7fc75 149 tcp_usrclosed (struct tcpcb *);
984263bc
MD
150
151#ifdef TCPDEBUG
152#define TCPDEBUG0 int ostate = 0
153#define TCPDEBUG1() ostate = tp ? tp->t_state : 0
154#define TCPDEBUG2(req) if (tp && (so->so_options & SO_DEBUG)) \
155 tcp_trace(TA_USER, ostate, tp, 0, 0, req)
156#else
157#define TCPDEBUG0
158#define TCPDEBUG1()
159#define TCPDEBUG2(req)
160#endif
161
f12e9fad 162static int tcp_lport_extension = 1;
db26865d
SZ
163
164SYSCTL_INT(_net_inet_tcp, OID_AUTO, lportext, CTLFLAG_RW,
165 &tcp_lport_extension, 0, "");
166
984263bc
MD
167/*
168 * TCP attaches to socket via pru_attach(), reserving space,
0ce0603e
MD
169 * and an internet control block. This is likely occuring on
170 * cpu0 and may have to move later when we bind/connect.
984263bc 171 */
002c1265
MD
172static void
173tcp_usr_attach(netmsg_t msg)
984263bc 174{
002c1265
MD
175 struct socket *so = msg->base.nm_so;
176 struct pru_attach_info *ai = msg->attach.nm_ai;
984263bc 177 int error;
1cae611f 178 struct inpcb *inp;
984263bc
MD
179 struct tcpcb *tp = 0;
180 TCPDEBUG0;
181
6cef7136 182 soreference(so);
1cae611f 183 inp = so->so_pcb;
984263bc
MD
184 TCPDEBUG1();
185 if (inp) {
186 error = EISCONN;
187 goto out;
188 }
189
e4700d00 190 error = tcp_attach(so, ai);
984263bc
MD
191 if (error)
192 goto out;
193
194 if ((so->so_options & SO_LINGER) && so->so_linger == 0)
195 so->so_linger = TCP_LINGERTIME;
196 tp = sototcpcb(so);
197out:
6cef7136 198 sofree(so); /* from ref above */
984263bc 199 TCPDEBUG2(PRU_ATTACH);
002c1265 200 lwkt_replymsg(&msg->lmsg, error);
984263bc
MD
201}
202
203/*
204 * pru_detach() detaches the TCP protocol from the socket.
205 * If the protocol state is non-embryonic, then can't
206 * do this directly: have to initiate a pru_disconnect(),
207 * which may finish later; embryonic TCB's can just
208 * be discarded here.
209 */
002c1265
MD
210static void
211tcp_usr_detach(netmsg_t msg)
984263bc 212{
002c1265 213 struct socket *so = msg->base.nm_so;
984263bc 214 int error = 0;
1cae611f 215 struct inpcb *inp;
984263bc
MD
216 struct tcpcb *tp;
217 TCPDEBUG0;
218
1cae611f 219 inp = so->so_pcb;
69891757
MD
220
221 /*
222 * If the inp is already detached it may have been due to an async
223 * close. Just return as if no error occured.
002c1265 224 *
621989fd
MD
225 * It's possible for the tcpcb (tp) to disconnect from the inp due
226 * to tcp_drop()->tcp_close() being called. This may occur *after*
227 * the detach message has been queued so we may find a NULL tp here.
228 */
002c1265
MD
229 if (inp) {
230 if ((tp = intotcpcb(inp)) != NULL) {
231 TCPDEBUG1();
232 tp = tcp_disconnect(tp);
233 TCPDEBUG2(PRU_DETACH);
234 }
621989fd 235 }
002c1265 236 lwkt_replymsg(&msg->lmsg, error);
984263bc
MD
237}
238
69891757 239/*
002c1265 240 * NOTE: ignore_error is non-zero for certain disconnection races
69891757
MD
241 * which we want to silently allow, otherwise close() may return
242 * an unexpected error.
002c1265
MD
243 *
244 * NOTE: The variables (msg) and (tp) are assumed.
69891757
MD
245 */
246#define COMMON_START(so, inp, ignore_error) \
002c1265
MD
247 TCPDEBUG0; \
248 \
249 inp = so->so_pcb; \
250 do { \
69891757 251 if (inp == NULL) { \
002c1265
MD
252 error = ignore_error ? 0 : EINVAL; \
253 tp = NULL; \
254 goto out; \
69891757
MD
255 } \
256 tp = intotcpcb(inp); \
257 TCPDEBUG1(); \
258 } while(0)
f23061d4 259
002c1265
MD
260#define COMMON_END(req) \
261 out: do { \
262 TCPDEBUG2(req); \
263 lwkt_replymsg(&msg->lmsg, error); \
264 return; \
265 } while(0)
984263bc
MD
266
267/*
268 * Give the socket an address.
269 */
002c1265
MD
270static void
271tcp_usr_bind(netmsg_t msg)
984263bc 272{
002c1265
MD
273 struct socket *so = msg->bind.base.nm_so;
274 struct sockaddr *nam = msg->bind.nm_nam;
275 struct thread *td = msg->bind.nm_td;
984263bc 276 int error = 0;
1cae611f 277 struct inpcb *inp;
984263bc
MD
278 struct tcpcb *tp;
279 struct sockaddr_in *sinp;
280
69891757 281 COMMON_START(so, inp, 0);
984263bc
MD
282
283 /*
284 * Must check for multicast addresses and disallow binding
285 * to them.
286 */
287 sinp = (struct sockaddr_in *)nam;
288 if (sinp->sin_family == AF_INET &&
289 IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
290 error = EAFNOSUPPORT;
291 goto out;
292 }
dadab5e9 293 error = in_pcbbind(inp, nam, td);
984263bc
MD
294 if (error)
295 goto out;
296 COMMON_END(PRU_BIND);
297
298}
299
300#ifdef INET6
002c1265
MD
301
302static void
303tcp6_usr_bind(netmsg_t msg)
984263bc 304{
002c1265
MD
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;
984263bc 308 int error = 0;
1cae611f 309 struct inpcb *inp;
984263bc
MD
310 struct tcpcb *tp;
311 struct sockaddr_in6 *sin6p;
312
69891757 313 COMMON_START(so, inp, 0);
984263bc
MD
314
315 /*
316 * Must check for multicast addresses and disallow binding
317 * to them.
318 */
319 sin6p = (struct sockaddr_in6 *)nam;
320 if (sin6p->sin6_family == AF_INET6 &&
321 IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
322 error = EAFNOSUPPORT;
323 goto out;
324 }
325 inp->inp_vflag &= ~INP_IPV4;
326 inp->inp_vflag |= INP_IPV6;
327 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
328 if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr))
329 inp->inp_vflag |= INP_IPV4;
330 else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
331 struct sockaddr_in sin;
332
333 in6_sin6_2_sin(&sin, sin6p);
334 inp->inp_vflag |= INP_IPV4;
335 inp->inp_vflag &= ~INP_IPV6;
dadab5e9 336 error = in_pcbbind(inp, (struct sockaddr *)&sin, td);
984263bc
MD
337 goto out;
338 }
339 }
dadab5e9 340 error = in6_pcbbind(inp, nam, td);
984263bc
MD
341 if (error)
342 goto out;
343 COMMON_END(PRU_BIND);
344}
345#endif /* INET6 */
346
8affadf8 347#ifdef SMP
002c1265 348
8affadf8 349struct netmsg_inswildcard {
002c1265 350 struct netmsg_base base;
8affadf8 351 struct inpcb *nm_inp;
8affadf8
JH
352};
353
4599cf19 354static void
002c1265 355in_pcbinswildcardhash_handler(netmsg_t msg)
8affadf8 356{
002c1265 357 struct netmsg_inswildcard *nm = (struct netmsg_inswildcard *)msg;
8f9b2aed 358 int cpu = mycpuid, nextcpu;
8affadf8 359
8f9b2aed
SZ
360 in_pcbinswildcardhash_oncpu(nm->nm_inp, &tcbinfo[cpu]);
361
362 nextcpu = cpu + 1;
363 if (nextcpu < ncpus2)
364 lwkt_forwardmsg(cpu_portfn(nextcpu), &nm->base.lmsg);
365 else
366 lwkt_replymsg(&nm->base.lmsg, 0);
8affadf8 367}
002c1265 368
8affadf8
JH
369#endif
370
984263bc
MD
371/*
372 * Prepare to accept connections.
373 */
002c1265
MD
374static void
375tcp_usr_listen(netmsg_t msg)
984263bc 376{
002c1265
MD
377 struct socket *so = msg->listen.base.nm_so;
378 struct thread *td = msg->listen.nm_td;
984263bc 379 int error = 0;
1cae611f 380 struct inpcb *inp;
984263bc 381 struct tcpcb *tp;
8affadf8 382#ifdef SMP
8f9b2aed 383 struct netmsg_inswildcard nm;
8affadf8 384#endif
984263bc 385
69891757 386 COMMON_START(so, inp, 0);
f7b29de5
SZ
387
388 if (tp->t_flags & TF_LISTEN)
389 goto out;
390
8affadf8
JH
391 if (inp->inp_lport == 0) {
392 error = in_pcbbind(inp, NULL, td);
88d239e0 393 if (error)
8affadf8
JH
394 goto out;
395 }
396
397 tp->t_state = TCPS_LISTEN;
f7b29de5 398 tp->t_flags |= TF_LISTEN;
0f758523 399 tp->tt_msg = NULL; /* Catch any invalid timer usage */
8f9b2aed 400
8affadf8 401#ifdef SMP
985b2cb0
MD
402 if (ncpus > 1) {
403 /*
404 * We have to set the flag because we can't have other cpus
405 * messing with our inp's flags.
406 */
f78f8f4a
SZ
407 KASSERT(!(inp->inp_flags & INP_CONNECTED),
408 ("already on connhash\n"));
409 KASSERT(!(inp->inp_flags & INP_WILDCARD),
410 ("already on wildcardhash\n"));
411 KASSERT(!(inp->inp_flags & INP_WILDCARD_MP),
412 ("already on MP wildcardhash\n"));
985b2cb0
MD
413 inp->inp_flags |= INP_WILDCARD_MP;
414
415 KKASSERT(so->so_port == cpu_portfn(0));
416 KKASSERT(&curthread->td_msgport == cpu_portfn(0));
417 KKASSERT(inp->inp_pcbinfo == &tcbinfo[0]);
418
419 netmsg_init(&nm.base, NULL, &curthread->td_msgport,
420 MSGF_PRIORITY, in_pcbinswildcardhash_handler);
421 nm.nm_inp = inp;
422 lwkt_domsg(cpu_portfn(1), &nm.base.lmsg, 0);
423 }
8affadf8 424#endif
8f9b2aed 425 in_pcbinswildcardhash(inp);
984263bc
MD
426 COMMON_END(PRU_LISTEN);
427}
428
429#ifdef INET6
002c1265
MD
430
431static void
432tcp6_usr_listen(netmsg_t msg)
984263bc 433{
002c1265
MD
434 struct socket *so = msg->listen.base.nm_so;
435 struct thread *td = msg->listen.nm_td;
984263bc 436 int error = 0;
1cae611f 437 struct inpcb *inp;
984263bc 438 struct tcpcb *tp;
00638fa3 439#ifdef SMP
8f9b2aed 440 struct netmsg_inswildcard nm;
00638fa3 441#endif
984263bc 442
69891757 443 COMMON_START(so, inp, 0);
f7b29de5
SZ
444
445 if (tp->t_flags & TF_LISTEN)
446 goto out;
447
984263bc 448 if (inp->inp_lport == 0) {
00638fa3 449 if (!(inp->inp_flags & IN6P_IPV6_V6ONLY))
984263bc 450 inp->inp_vflag |= INP_IPV4;
00638fa3
JH
451 else
452 inp->inp_vflag &= ~INP_IPV4;
60233e58 453 error = in6_pcbbind(inp, NULL, td);
88d239e0
SZ
454 if (error)
455 goto out;
984263bc 456 }
88d239e0
SZ
457
458 tp->t_state = TCPS_LISTEN;
f7b29de5 459 tp->t_flags |= TF_LISTEN;
88d239e0
SZ
460 tp->tt_msg = NULL; /* Catch any invalid timer usage */
461
00638fa3 462#ifdef SMP
985b2cb0
MD
463 if (ncpus > 1) {
464 /*
465 * We have to set the flag because we can't have other cpus
466 * messing with our inp's flags.
467 */
f78f8f4a
SZ
468 KASSERT(!(inp->inp_flags & INP_CONNECTED),
469 ("already on connhash\n"));
470 KASSERT(!(inp->inp_flags & INP_WILDCARD),
471 ("already on wildcardhash\n"));
472 KASSERT(!(inp->inp_flags & INP_WILDCARD_MP),
473 ("already on MP wildcardhash\n"));
985b2cb0
MD
474 inp->inp_flags |= INP_WILDCARD_MP;
475
476 KKASSERT(so->so_port == cpu_portfn(0));
477 KKASSERT(&curthread->td_msgport == cpu_portfn(0));
478 KKASSERT(inp->inp_pcbinfo == &tcbinfo[0]);
479
480 netmsg_init(&nm.base, NULL, &curthread->td_msgport,
481 MSGF_PRIORITY, in_pcbinswildcardhash_handler);
482 nm.nm_inp = inp;
483 lwkt_domsg(cpu_portfn(1), &nm.base.lmsg, 0);
484 }
00638fa3 485#endif
8f9b2aed 486 in_pcbinswildcardhash(inp);
984263bc
MD
487 COMMON_END(PRU_LISTEN);
488}
489#endif /* INET6 */
490
491/*
492 * Initiate connection to peer.
493 * Create a template for use in transmissions on this connection.
494 * Enter SYN_SENT state, and mark socket as connecting.
495 * Start keep-alive timer, and seed output sequence space.
496 * Send initial segment on connection.
497 */
002c1265
MD
498static void
499tcp_usr_connect(netmsg_t msg)
984263bc 500{
002c1265
MD
501 struct socket *so = msg->connect.base.nm_so;
502 struct sockaddr *nam = msg->connect.nm_nam;
503 struct thread *td = msg->connect.nm_td;
984263bc 504 int error = 0;
1cae611f 505 struct inpcb *inp;
984263bc
MD
506 struct tcpcb *tp;
507 struct sockaddr_in *sinp;
508
69891757 509 COMMON_START(so, inp, 0);
984263bc
MD
510
511 /*
512 * Must disallow TCP ``connections'' to multicast addresses.
513 */
514 sinp = (struct sockaddr_in *)nam;
515 if (sinp->sin_family == AF_INET
516 && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
517 error = EAFNOSUPPORT;
518 goto out;
519 }
520
3e4150ef
VBD
521 if (!prison_remote_ip(td, (struct sockaddr*)sinp)) {
522 error = EAFNOSUPPORT; /* IPv6 only jail */
523 goto out;
524 }
984263bc 525
002c1265
MD
526 tcp_connect(msg);
527 /* msg is invalid now */
528 return;
529out:
530 if (msg->connect.nm_m) {
531 m_freem(msg->connect.nm_m);
532 msg->connect.nm_m = NULL;
533 }
534 lwkt_replymsg(&msg->lmsg, error);
984263bc
MD
535}
536
537#ifdef INET6
002c1265
MD
538
539static void
540tcp6_usr_connect(netmsg_t msg)
984263bc 541{
002c1265
MD
542 struct socket *so = msg->connect.base.nm_so;
543 struct sockaddr *nam = msg->connect.nm_nam;
544 struct thread *td = msg->connect.nm_td;
984263bc 545 int error = 0;
1cae611f 546 struct inpcb *inp;
984263bc
MD
547 struct tcpcb *tp;
548 struct sockaddr_in6 *sin6p;
549
69891757 550 COMMON_START(so, inp, 0);
984263bc
MD
551
552 /*
553 * Must disallow TCP ``connections'' to multicast addresses.
554 */
555 sin6p = (struct sockaddr_in6 *)nam;
556 if (sin6p->sin6_family == AF_INET6
557 && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr)) {
558 error = EAFNOSUPPORT;
559 goto out;
560 }
561
3e4150ef
VBD
562 if (!prison_remote_ip(td, nam)) {
563 error = EAFNOSUPPORT; /* IPv4 only jail */
564 goto out;
565 }
566
984263bc 567 if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
002c1265 568 struct sockaddr_in *sinp;
984263bc
MD
569
570 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
571 error = EINVAL;
572 goto out;
573 }
002c1265
MD
574 sinp = kmalloc(sizeof(*sinp), M_LWKTMSG, M_INTWAIT);
575 in6_sin6_2_sin(sinp, sin6p);
984263bc
MD
576 inp->inp_vflag |= INP_IPV4;
577 inp->inp_vflag &= ~INP_IPV6;
002c1265
MD
578 msg->connect.nm_nam = (struct sockaddr *)sinp;
579 msg->connect.nm_reconnect |= NMSG_RECONNECT_NAMALLOC;
580 tcp_connect(msg);
581 /* msg is invalid now */
582 return;
984263bc
MD
583 }
584 inp->inp_vflag &= ~INP_IPV4;
585 inp->inp_vflag |= INP_IPV6;
586 inp->inp_inc.inc_isipv6 = 1;
002c1265
MD
587
588 msg->connect.nm_reconnect |= NMSG_RECONNECT_FALLBACK;
589 tcp6_connect(msg);
590 /* msg is invalid now */
591 return;
592out:
593 if (msg->connect.nm_m) {
594 m_freem(msg->connect.nm_m);
595 msg->connect.nm_m = NULL;
596 }
597 lwkt_replymsg(&msg->lmsg, error);
984263bc 598}
002c1265 599
984263bc
MD
600#endif /* INET6 */
601
602/*
603 * Initiate disconnect from peer.
604 * If connection never passed embryonic stage, just drop;
605 * else if don't need to let data drain, then can just drop anyways,
606 * else have to begin TCP shutdown process: mark socket disconnecting,
607 * drain unread data, state switch to reflect user close, and
608 * send segment (e.g. FIN) to peer. Socket will be really disconnected
609 * when peer sends FIN and acks ours.
610 *
611 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
612 */
002c1265
MD
613static void
614tcp_usr_disconnect(netmsg_t msg)
984263bc 615{
002c1265 616 struct socket *so = msg->disconnect.base.nm_so;
984263bc 617 int error = 0;
1cae611f 618 struct inpcb *inp;
984263bc
MD
619 struct tcpcb *tp;
620
69891757 621 COMMON_START(so, inp, 1);
984263bc
MD
622 tp = tcp_disconnect(tp);
623 COMMON_END(PRU_DISCONNECT);
624}
625
626/*
627 * Accept a connection. Essentially all the work is
628 * done at higher levels; just return the address
629 * of the peer, storing through addr.
630 */
002c1265
MD
631static void
632tcp_usr_accept(netmsg_t msg)
984263bc 633{
002c1265
MD
634 struct socket *so = msg->accept.base.nm_so;
635 struct sockaddr **nam = msg->accept.nm_nam;
984263bc 636 int error = 0;
1cae611f 637 struct inpcb *inp;
984263bc
MD
638 struct tcpcb *tp = NULL;
639 TCPDEBUG0;
640
1cae611f 641 inp = so->so_pcb;
984263bc
MD
642 if (so->so_state & SS_ISDISCONNECTED) {
643 error = ECONNABORTED;
644 goto out;
645 }
002c1265
MD
646 if (inp == 0) {
647 error = EINVAL;
648 goto out;
649 }
6cef7136 650
984263bc
MD
651 tp = intotcpcb(inp);
652 TCPDEBUG1();
653 in_setpeeraddr(so, nam);
654 COMMON_END(PRU_ACCEPT);
655}
656
657#ifdef INET6
002c1265
MD
658static void
659tcp6_usr_accept(netmsg_t msg)
984263bc 660{
002c1265
MD
661 struct socket *so = msg->accept.base.nm_so;
662 struct sockaddr **nam = msg->accept.nm_nam;
984263bc 663 int error = 0;
1cae611f 664 struct inpcb *inp;
984263bc
MD
665 struct tcpcb *tp = NULL;
666 TCPDEBUG0;
667
1cae611f
MD
668 inp = so->so_pcb;
669
984263bc
MD
670 if (so->so_state & SS_ISDISCONNECTED) {
671 error = ECONNABORTED;
672 goto out;
673 }
002c1265
MD
674 if (inp == 0) {
675 error = EINVAL;
676 goto out;
677 }
984263bc
MD
678 tp = intotcpcb(inp);
679 TCPDEBUG1();
680 in6_mapped_peeraddr(so, nam);
681 COMMON_END(PRU_ACCEPT);
682}
683#endif /* INET6 */
684/*
685 * Mark the connection as being incapable of further output.
686 */
002c1265
MD
687static void
688tcp_usr_shutdown(netmsg_t msg)
984263bc 689{
002c1265 690 struct socket *so = msg->shutdown.base.nm_so;
984263bc 691 int error = 0;
1cae611f 692 struct inpcb *inp;
984263bc
MD
693 struct tcpcb *tp;
694
69891757 695 COMMON_START(so, inp, 0);
984263bc
MD
696 socantsendmore(so);
697 tp = tcp_usrclosed(tp);
698 if (tp)
699 error = tcp_output(tp);
700 COMMON_END(PRU_SHUTDOWN);
701}
702
703/*
704 * After a receive, possibly send window update to peer.
705 */
002c1265
MD
706static void
707tcp_usr_rcvd(netmsg_t msg)
984263bc 708{
002c1265 709 struct socket *so = msg->rcvd.base.nm_so;
984263bc 710 int error = 0;
1cae611f 711 struct inpcb *inp;
984263bc
MD
712 struct tcpcb *tp;
713
69891757 714 COMMON_START(so, inp, 0);
984263bc
MD
715 tcp_output(tp);
716 COMMON_END(PRU_RCVD);
717}
718
719/*
720 * Do a send by putting data in output queue and updating urgent
721 * marker if URG set. Possibly send more data. Unlike the other
722 * pru_*() routines, the mbuf chains are our responsibility. We
723 * must either enqueue them or free them. The other pru_* routines
724 * generally are caller-frees.
725 */
002c1265
MD
726static void
727tcp_usr_send(netmsg_t msg)
984263bc 728{
002c1265
MD
729 struct socket *so = msg->send.base.nm_so;
730 int flags = msg->send.nm_flags;
731 struct mbuf *m = msg->send.nm_m;
732 struct sockaddr *nam = msg->send.nm_addr;
733 struct mbuf *control = msg->send.nm_control;
734 struct thread *td = msg->send.nm_td;
984263bc 735 int error = 0;
1cae611f 736 struct inpcb *inp;
984263bc
MD
737 struct tcpcb *tp;
738#ifdef INET6
739 int isipv6;
740#endif
741 TCPDEBUG0;
742
1cae611f
MD
743 inp = so->so_pcb;
744
984263bc
MD
745 if (inp == NULL) {
746 /*
747 * OOPS! we lost a race, the TCP session got reset after
748 * we checked SS_CANTSENDMORE, eg: while doing uiomove or a
1cae611f 749 * network interrupt in the non-critical section of sosend().
984263bc 750 */
48e7b118 751 m_freem(m);
984263bc
MD
752 if (control)
753 m_freem(control);
754 error = ECONNRESET; /* XXX EPIPE? */
755 tp = NULL;
756 TCPDEBUG1();
757 goto out;
758 }
759#ifdef INET6
760 isipv6 = nam && nam->sa_family == AF_INET6;
761#endif /* INET6 */
762 tp = intotcpcb(inp);
763 TCPDEBUG1();
764 if (control) {
765 /* TCP doesn't do control messages (rights, creds, etc) */
766 if (control->m_len) {
767 m_freem(control);
48e7b118 768 m_freem(m);
984263bc
MD
769 error = EINVAL;
770 goto out;
771 }
772 m_freem(control); /* empty control, just free it */
773 }
984263bc 774
48e7b118
MD
775 /*
776 * Don't let too much OOB data build up
777 */
778 if (flags & PRUS_OOB) {
6d49aa6f 779 if (ssb_space(&so->so_snd) < -512) {
984263bc
MD
780 m_freem(m);
781 error = ENOBUFS;
782 goto out;
783 }
48e7b118
MD
784 }
785
786 /*
787 * Do implied connect if not yet connected. Any data sent
788 * with the connect is handled by tcp_connect() and friends.
789 *
790 * NOTE! PROTOCOL THREAD MAY BE CHANGED BY THE CONNECT!
791 */
792 if (nam && tp->t_state < TCPS_SYN_SENT) {
002c1265
MD
793 kprintf("implied fallback\n");
794 msg->connect.nm_nam = nam;
795 msg->connect.nm_td = td;
796 msg->connect.nm_m = m;
797 msg->connect.nm_flags = flags;
798 msg->connect.nm_reconnect = NMSG_RECONNECT_FALLBACK;
48e7b118
MD
799#ifdef INET6
800 if (isipv6)
002c1265 801 tcp6_connect(msg);
48e7b118
MD
802 else
803#endif /* INET6 */
002c1265
MD
804 tcp_connect(msg);
805 /* msg invalid now */
806 return;
48e7b118
MD
807 }
808
809 /*
810 * Pump the data into the socket.
811 */
812 if (m)
813 ssb_appendstream(&so->so_snd, m);
814 if (flags & PRUS_OOB) {
984263bc
MD
815 /*
816 * According to RFC961 (Assigned Protocols),
817 * the urgent pointer points to the last octet
818 * of urgent data. We continue, however,
819 * to consider it to indicate the first octet
820 * of data past the urgent section.
821 * Otherwise, snd_up should be one lower.
822 */
6d49aa6f 823 tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
c1d3b892 824 tp->t_flags |= TF_FORCE;
984263bc 825 error = tcp_output(tp);
c1d3b892 826 tp->t_flags &= ~TF_FORCE;
48e7b118
MD
827 } else {
828 if (flags & PRUS_EOF) {
829 /*
830 * Close the send side of the connection after
831 * the data is sent.
832 */
833 socantsendmore(so);
834 tp = tcp_usrclosed(tp);
835 }
836 if (tp != NULL) {
837 if (flags & PRUS_MORETOCOME)
838 tp->t_flags |= TF_MORETOCOME;
839 error = tcp_output(tp);
840 if (flags & PRUS_MORETOCOME)
841 tp->t_flags &= ~TF_MORETOCOME;
842 }
984263bc 843 }
f23061d4 844 COMMON_END((flags & PRUS_OOB) ? PRU_SENDOOB :
984263bc
MD
845 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
846}
847
848/*
6cef7136
MD
849 * NOTE: (so) is referenced from soabort*() and netmsg_pru_abort()
850 * will sofree() it when we return.
984263bc 851 */
002c1265
MD
852static void
853tcp_usr_abort(netmsg_t msg)
984263bc 854{
002c1265 855 struct socket *so = msg->abort.base.nm_so;
984263bc 856 int error = 0;
1cae611f 857 struct inpcb *inp;
984263bc
MD
858 struct tcpcb *tp;
859
69891757 860 COMMON_START(so, inp, 1);
984263bc
MD
861 tp = tcp_drop(tp, ECONNABORTED);
862 COMMON_END(PRU_ABORT);
863}
864
865/*
866 * Receive out-of-band data.
867 */
002c1265
MD
868static void
869tcp_usr_rcvoob(netmsg_t msg)
984263bc 870{
002c1265
MD
871 struct socket *so = msg->rcvoob.base.nm_so;
872 struct mbuf *m = msg->rcvoob.nm_m;
873 int flags = msg->rcvoob.nm_flags;
984263bc 874 int error = 0;
1cae611f 875 struct inpcb *inp;
984263bc
MD
876 struct tcpcb *tp;
877
69891757 878 COMMON_START(so, inp, 0);
984263bc
MD
879 if ((so->so_oobmark == 0 &&
880 (so->so_state & SS_RCVATMARK) == 0) ||
881 so->so_options & SO_OOBINLINE ||
882 tp->t_oobflags & TCPOOB_HADDATA) {
883 error = EINVAL;
884 goto out;
885 }
886 if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
887 error = EWOULDBLOCK;
888 goto out;
889 }
890 m->m_len = 1;
891 *mtod(m, caddr_t) = tp->t_iobc;
892 if ((flags & MSG_PEEK) == 0)
893 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
894 COMMON_END(PRU_RCVOOB);
895}
896
897/* xxx - should be const */
898struct pr_usrreqs tcp_usrreqs = {
fa5e758c
MD
899 .pru_abort = tcp_usr_abort,
900 .pru_accept = tcp_usr_accept,
901 .pru_attach = tcp_usr_attach,
902 .pru_bind = tcp_usr_bind,
903 .pru_connect = tcp_usr_connect,
002c1265
MD
904 .pru_connect2 = pr_generic_notsupp,
905 .pru_control = in_control_dispatch,
fa5e758c
MD
906 .pru_detach = tcp_usr_detach,
907 .pru_disconnect = tcp_usr_disconnect,
908 .pru_listen = tcp_usr_listen,
002c1265 909 .pru_peeraddr = in_setpeeraddr_dispatch,
fa5e758c
MD
910 .pru_rcvd = tcp_usr_rcvd,
911 .pru_rcvoob = tcp_usr_rcvoob,
912 .pru_send = tcp_usr_send,
913 .pru_sense = pru_sense_null,
914 .pru_shutdown = tcp_usr_shutdown,
002c1265 915 .pru_sockaddr = in_setsockaddr_dispatch,
fa5e758c 916 .pru_sosend = sosend,
8b5c39bb 917 .pru_soreceive = soreceive
984263bc
MD
918};
919
920#ifdef INET6
921struct pr_usrreqs tcp6_usrreqs = {
fa5e758c
MD
922 .pru_abort = tcp_usr_abort,
923 .pru_accept = tcp6_usr_accept,
924 .pru_attach = tcp_usr_attach,
925 .pru_bind = tcp6_usr_bind,
926 .pru_connect = tcp6_usr_connect,
002c1265
MD
927 .pru_connect2 = pr_generic_notsupp,
928 .pru_control = in6_control_dispatch,
fa5e758c
MD
929 .pru_detach = tcp_usr_detach,
930 .pru_disconnect = tcp_usr_disconnect,
931 .pru_listen = tcp6_usr_listen,
002c1265 932 .pru_peeraddr = in6_mapped_peeraddr_dispatch,
fa5e758c
MD
933 .pru_rcvd = tcp_usr_rcvd,
934 .pru_rcvoob = tcp_usr_rcvoob,
935 .pru_send = tcp_usr_send,
936 .pru_sense = pru_sense_null,
937 .pru_shutdown = tcp_usr_shutdown,
002c1265 938 .pru_sockaddr = in6_mapped_sockaddr_dispatch,
fa5e758c 939 .pru_sosend = sosend,
8b5c39bb 940 .pru_soreceive = soreceive
984263bc
MD
941};
942#endif /* INET6 */
943
984263bc 944static int
48e7b118
MD
945tcp_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf *m,
946 struct sockaddr_in *sin, struct sockaddr_in *if_sin)
984263bc
MD
947{
948 struct inpcb *inp = tp->t_inpcb, *oinp;
949 struct socket *so = inp->inp_socket;
03fb1cae 950 struct route *ro = &inp->inp_route;
984263bc 951
7fe56515 952 oinp = in_pcblookup_hash(&tcbinfo[mycpu->gd_cpuid],
002c1265
MD
953 sin->sin_addr, sin->sin_port,
954 (inp->inp_laddr.s_addr != INADDR_ANY ?
955 inp->inp_laddr : if_sin->sin_addr),
956 inp->inp_lport, 0, NULL);
13d8907a 957 if (oinp != NULL) {
27b8aee3
AE
958 m_freem(m);
959 return (EADDRINUSE);
984263bc
MD
960 }
961 if (inp->inp_laddr.s_addr == INADDR_ANY)
13d8907a 962 inp->inp_laddr = if_sin->sin_addr;
984263bc
MD
963 inp->inp_faddr = sin->sin_addr;
964 inp->inp_fport = sin->sin_port;
7fe56515 965 inp->inp_cpcbinfo = &tcbinfo[mycpu->gd_cpuid];
13d8907a 966 in_pcbinsconnhash(inp);
984263bc 967
03fb1cae
SZ
968 /*
969 * We are now on the inpcb's owner CPU, if the cached route was
970 * freed because the rtentry's owner CPU is not the current CPU
971 * (e.g. in tcp_connect()), then we try to reallocate it here with
972 * the hope that a rtentry may be cloned from a RTF_PRCLONING
973 * rtentry.
974 */
975 if (!(inp->inp_socket->so_options & SO_DONTROUTE) && /*XXX*/
976 ro->ro_rt == NULL) {
977 bzero(&ro->ro_dst, sizeof(struct sockaddr_in));
978 ro->ro_dst.sa_family = AF_INET;
979 ro->ro_dst.sa_len = sizeof(struct sockaddr_in);
980 ((struct sockaddr_in *)&ro->ro_dst)->sin_addr =
981 sin->sin_addr;
982 rtalloc(ro);
983 }
984
3db1c8a3 985 /*
48e7b118
MD
986 * Now that no more errors can occur, change the protocol processing
987 * port to the current thread (which is the correct thread).
988 *
3db1c8a3
SZ
989 * Create TCP timer message now; we are on the tcpcb's owner
990 * CPU/thread.
991 */
992 tcp_create_timermsg(tp, &curthread->td_msgport);
0f758523 993
46e92930
MD
994 /*
995 * Compute window scaling to request. Use a larger scaling then
996 * needed for the initial receive buffer in case the receive buffer
997 * gets expanded.
998 */
999 if (tp->request_r_scale < TCP_MIN_WINSHIFT)
1000 tp->request_r_scale = TCP_MIN_WINSHIFT;
984263bc 1001 while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
46e92930
MD
1002 (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat
1003 ) {
984263bc 1004 tp->request_r_scale++;
46e92930 1005 }
984263bc
MD
1006
1007 soisconnecting(so);
1008 tcpstat.tcps_connattempt++;
1009 tp->t_state = TCPS_SYN_SENT;
2ce132be 1010 tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
984263bc 1011 tp->iss = tcp_new_isn(tp);
984263bc 1012 tcp_sendseqinit(tp);
48e7b118
MD
1013 if (m) {
1014 ssb_appendstream(&so->so_snd, m);
1015 m = NULL;
1016 if (flags & PRUS_OOB)
1017 tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
1018 }
984263bc
MD
1019
1020 /*
48e7b118
MD
1021 * Close the send side of the connection after
1022 * the data is sent if flagged.
1023 */
1024 if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
1025 socantsendmore(so);
1026 tp = tcp_usrclosed(tp);
1027 }
1028 return (tcp_output(tp));
984263bc
MD
1029}
1030
7fe56515
JH
1031/*
1032 * Common subroutine to open a TCP connection to remote host specified
1033 * by struct sockaddr_in in mbuf *nam. Call in_pcbbind to assign a local
1034 * port number if needed. Call in_pcbladdr to do the routing and to choose
27b8aee3 1035 * a local host address (interface).
7fe56515
JH
1036 * Initialize connection parameters and enter SYN-SENT state.
1037 */
002c1265
MD
1038static void
1039tcp_connect(netmsg_t msg)
7fe56515 1040{
002c1265
MD
1041 struct socket *so = msg->connect.base.nm_so;
1042 struct sockaddr *nam = msg->connect.nm_nam;
1043 struct thread *td = msg->connect.nm_td;
7fe56515
JH
1044 struct sockaddr_in *sin = (struct sockaddr_in *)nam;
1045 struct sockaddr_in *if_sin;
002c1265
MD
1046 struct inpcb *inp;
1047 struct tcpcb *tp;
4fd40da5 1048 int error, calc_laddr = 1;
540a0bf0 1049#ifdef SMP
7fe56515
JH
1050 lwkt_port_t port;
1051#endif
1052
002c1265
MD
1053 COMMON_START(so, inp, 0);
1054
1055 /*
1056 * Reconnect our pcb if we have to
1057 */
1058 if (msg->connect.nm_reconnect & NMSG_RECONNECT_RECONNECT) {
1059 msg->connect.nm_reconnect &= ~NMSG_RECONNECT_RECONNECT;
1060 in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1061 }
1062
9c7b7c43
SZ
1063 /*
1064 * Bind if we have to
1065 */
1066 if (inp->inp_lport == 0) {
1067 if (tcp_lport_extension) {
db26865d
SZ
1068 KKASSERT(inp->inp_laddr.s_addr == INADDR_ANY);
1069
1070 error = in_pcbladdr(inp, nam, &if_sin, td);
1071 if (error)
1072 goto out;
1073 inp->inp_laddr.s_addr = if_sin->sin_addr.s_addr;
1074
1075 error = in_pcbconn_bind(inp, nam, td);
1076 if (error)
1077 goto out;
4fd40da5
SZ
1078
1079 calc_laddr = 0;
9c7b7c43 1080 } else {
db26865d
SZ
1081 error = in_pcbbind(inp, NULL, td);
1082 if (error)
1083 goto out;
1084 }
4fd40da5 1085 }
db26865d 1086
4fd40da5 1087 if (calc_laddr) {
db26865d
SZ
1088 /*
1089 * Calculate the correct protocol processing thread. The
1090 * connect operation must run there. Set the forwarding
1091 * port before we forward the message or it will get bounced
1092 * right back to us.
1093 */
1094 error = in_pcbladdr(inp, nam, &if_sin, td);
002c1265
MD
1095 if (error)
1096 goto out;
7fe56515 1097 }
b25d27f4
SZ
1098 KKASSERT(inp->inp_socket == so);
1099
540a0bf0 1100#ifdef SMP
7fe56515 1101 port = tcp_addrport(sin->sin_addr.s_addr, sin->sin_port,
002c1265
MD
1102 (inp->inp_laddr.s_addr ?
1103 inp->inp_laddr.s_addr : if_sin->sin_addr.s_addr),
1104 inp->inp_lport);
7fe56515 1105
fb0f29c4 1106 if (port != &curthread->td_msgport) {
a46456f3
SZ
1107 struct route *ro = &inp->inp_route;
1108
1109 /*
1110 * in_pcbladdr() may have allocated a route entry for us
1111 * on the current CPU, but we need a route entry on the
03fb1cae 1112 * inpcb's owner CPU, so free it here.
a46456f3
SZ
1113 */
1114 if (ro->ro_rt != NULL)
1115 RTFREE(ro->ro_rt);
1116 bzero(ro, sizeof(*ro));
540a0bf0 1117
48e7b118 1118 /*
0ce0603e
MD
1119 * We are moving the protocol processing port the socket
1120 * is on, we have to unlink here and re-link on the
1121 * target cpu.
48e7b118 1122 */
0ce0603e
MD
1123 in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1124 sosetport(so, port);
002c1265
MD
1125 msg->connect.nm_reconnect |= NMSG_RECONNECT_RECONNECT;
1126 msg->connect.base.nm_dispatch = tcp_connect;
0ce0603e 1127
002c1265
MD
1128 lwkt_forwardmsg(port, &msg->connect.base.lmsg);
1129 /* msg invalid now */
1130 return;
48e7b118
MD
1131 }
1132#else
0ce0603e 1133 KKASSERT(so->so_port == &curthread->td_msgport);
7fe56515 1134#endif
002c1265
MD
1135 error = tcp_connect_oncpu(tp, msg->connect.nm_flags,
1136 msg->connect.nm_m, sin, if_sin);
1137 msg->connect.nm_m = NULL;
1138out:
1139 if (msg->connect.nm_m) {
1140 m_freem(msg->connect.nm_m);
1141 msg->connect.nm_m = NULL;
1142 }
1143 if (msg->connect.nm_reconnect & NMSG_RECONNECT_NAMALLOC) {
1144 kfree(msg->connect.nm_nam, M_LWKTMSG);
1145 msg->connect.nm_nam = NULL;
1146 }
1147 lwkt_replymsg(&msg->connect.base.lmsg, error);
1148 /* msg invalid now */
7fe56515
JH
1149}
1150
984263bc 1151#ifdef INET6
65f3e756 1152
002c1265
MD
1153static void
1154tcp6_connect(netmsg_t msg)
984263bc 1155{
002c1265
MD
1156 struct tcpcb *tp;
1157 struct socket *so = msg->connect.base.nm_so;
1158 struct sockaddr *nam = msg->connect.nm_nam;
1159 struct thread *td = msg->connect.nm_td;
1160 struct inpcb *inp;
984263bc
MD
1161 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)nam;
1162 struct in6_addr *addr6;
f890e67a 1163#ifdef SMP
65f3e756 1164 lwkt_port_t port;
f890e67a 1165#endif
984263bc
MD
1166 int error;
1167
002c1265
MD
1168 COMMON_START(so, inp, 0);
1169
1170 /*
1171 * Reconnect our pcb if we have to
1172 */
1173 if (msg->connect.nm_reconnect & NMSG_RECONNECT_RECONNECT) {
1174 msg->connect.nm_reconnect &= ~NMSG_RECONNECT_RECONNECT;
1175 in_pcblink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1176 }
1177
1178 /*
1179 * Bind if we have to
1180 */
984263bc 1181 if (inp->inp_lport == 0) {
60233e58 1182 error = in6_pcbbind(inp, NULL, td);
002c1265
MD
1183 if (error)
1184 goto out;
984263bc
MD
1185 }
1186
1187 /*
1188 * Cannot simply call in_pcbconnect, because there might be an
1189 * earlier incarnation of this same connection still in
1190 * TIME_WAIT state, creating an ADDRINUSE error.
1191 */
3e4150ef 1192 error = in6_pcbladdr(inp, nam, &addr6, td);
002c1265
MD
1193 if (error)
1194 goto out;
65f3e756
MD
1195
1196#ifdef SMP
1197 port = tcp6_addrport(); /* XXX hack for now, always cpu0 */
1198
1199 if (port != &curthread->td_msgport) {
65f3e756
MD
1200 struct route *ro = &inp->inp_route;
1201
1202 /*
1203 * in_pcbladdr() may have allocated a route entry for us
1204 * on the current CPU, but we need a route entry on the
1205 * inpcb's owner CPU, so free it here.
1206 */
1207 if (ro->ro_rt != NULL)
1208 RTFREE(ro->ro_rt);
1209 bzero(ro, sizeof(*ro));
1210
002c1265
MD
1211 in_pcbunlink(so->so_pcb, &tcbinfo[mycpu->gd_cpuid]);
1212 sosetport(so, port);
1213 msg->connect.nm_reconnect |= NMSG_RECONNECT_RECONNECT;
1214 msg->connect.base.nm_dispatch = tcp6_connect;
1215
1216 lwkt_forwardmsg(port, &msg->connect.base.lmsg);
1217 /* msg invalid now */
1218 return;
48e7b118 1219 }
65f3e756 1220#endif
002c1265
MD
1221 error = tcp6_connect_oncpu(tp, msg->connect.nm_flags,
1222 &msg->connect.nm_m, sin6, addr6);
1223 /* nm_m may still be intact */
1224out:
1225 if (error && (msg->connect.nm_reconnect & NMSG_RECONNECT_FALLBACK)) {
1226 tcp_connect(msg);
1227 /* msg invalid now */
1228 } else {
1229 if (msg->connect.nm_m) {
1230 m_freem(msg->connect.nm_m);
1231 msg->connect.nm_m = NULL;
1232 }
1233 if (msg->connect.nm_reconnect & NMSG_RECONNECT_NAMALLOC) {
1234 kfree(msg->connect.nm_nam, M_LWKTMSG);
1235 msg->connect.nm_nam = NULL;
1236 }
1237 lwkt_replymsg(&msg->connect.base.lmsg, error);
1238 /* msg invalid now */
1239 }
65f3e756
MD
1240}
1241
1242static int
002c1265 1243tcp6_connect_oncpu(struct tcpcb *tp, int flags, struct mbuf **mp,
48e7b118 1244 struct sockaddr_in6 *sin6, struct in6_addr *addr6)
65f3e756 1245{
002c1265 1246 struct mbuf *m = *mp;
65f3e756
MD
1247 struct inpcb *inp = tp->t_inpcb;
1248 struct socket *so = inp->inp_socket;
1249 struct inpcb *oinp;
65f3e756 1250
48e7b118
MD
1251 /*
1252 * Cannot simply call in_pcbconnect, because there might be an
1253 * earlier incarnation of this same connection still in
1254 * TIME_WAIT state, creating an ADDRINUSE error.
1255 */
7fe56515 1256 oinp = in6_pcblookup_hash(inp->inp_cpcbinfo,
984263bc 1257 &sin6->sin6_addr, sin6->sin6_port,
002c1265
MD
1258 (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr) ?
1259 addr6 : &inp->in6p_laddr),
984263bc 1260 inp->inp_lport, 0, NULL);
002c1265 1261 if (oinp)
27b8aee3 1262 return (EADDRINUSE);
002c1265 1263
984263bc
MD
1264 if (IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_laddr))
1265 inp->in6p_laddr = *addr6;
1266 inp->in6p_faddr = sin6->sin6_addr;
1267 inp->inp_fport = sin6->sin6_port;
3641b7ca 1268 if ((sin6->sin6_flowinfo & IPV6_FLOWINFO_MASK) != 0)
984263bc 1269 inp->in6p_flowinfo = sin6->sin6_flowinfo;
13d8907a 1270 in_pcbinsconnhash(inp);
984263bc 1271
48e7b118
MD
1272 /*
1273 * Now that no more errors can occur, change the protocol processing
1274 * port to the current thread (which is the correct thread).
1275 *
1276 * Create TCP timer message now; we are on the tcpcb's owner
1277 * CPU/thread.
1278 */
3db1c8a3
SZ
1279 tcp_create_timermsg(tp, &curthread->td_msgport);
1280
984263bc 1281 /* Compute window scaling to request. */
46e92930
MD
1282 if (tp->request_r_scale < TCP_MIN_WINSHIFT)
1283 tp->request_r_scale = TCP_MIN_WINSHIFT;
984263bc 1284 while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
46e92930 1285 (TCP_MAXWIN << tp->request_r_scale) < so->so_rcv.ssb_hiwat) {
984263bc 1286 tp->request_r_scale++;
46e92930 1287 }
984263bc
MD
1288
1289 soisconnecting(so);
1290 tcpstat.tcps_connattempt++;
1291 tp->t_state = TCPS_SYN_SENT;
2ce132be 1292 tcp_callout_reset(tp, tp->tt_keep, tp->t_keepinit, tcp_timer_keep);
984263bc 1293 tp->iss = tcp_new_isn(tp);
984263bc 1294 tcp_sendseqinit(tp);
48e7b118
MD
1295 if (m) {
1296 ssb_appendstream(&so->so_snd, m);
002c1265 1297 *mp = NULL;
48e7b118
MD
1298 if (flags & PRUS_OOB)
1299 tp->snd_up = tp->snd_una + so->so_snd.ssb_cc;
1300 }
984263bc
MD
1301
1302 /*
48e7b118
MD
1303 * Close the send side of the connection after
1304 * the data is sent if flagged.
1305 */
1306 if ((flags & (PRUS_OOB|PRUS_EOF)) == PRUS_EOF) {
1307 socantsendmore(so);
1308 tp = tcp_usrclosed(tp);
1309 }
1310 return (tcp_output(tp));
984263bc 1311}
65f3e756 1312
984263bc
MD
1313#endif /* INET6 */
1314
1315/*
1316 * The new sockopt interface makes it possible for us to block in the
1cae611f
MD
1317 * copyin/out step (if we take a page fault). Taking a page fault while
1318 * in a critical section is probably a Bad Thing. (Since sockets and pcbs
1319 * both now use TSM, there probably isn't any need for this function to
1320 * run in a critical section any more. This needs more examination.)
984263bc 1321 */
002c1265
MD
1322void
1323tcp_ctloutput(netmsg_t msg)
984263bc 1324{
002c1265
MD
1325 struct socket *so = msg->base.nm_so;
1326 struct sockopt *sopt = msg->ctloutput.nm_sopt;
2ce132be 1327 int error, opt, optval, opthz;
984263bc
MD
1328 struct inpcb *inp;
1329 struct tcpcb *tp;
1330
1331 error = 0;
ed894f8c 1332 inp = so->so_pcb;
002c1265
MD
1333 if (inp == NULL) {
1334 error = ECONNRESET;
1335 goto done;
1336 }
6cef7136 1337
984263bc
MD
1338 if (sopt->sopt_level != IPPROTO_TCP) {
1339#ifdef INET6
1340 if (INP_CHECK_SOCKAF(so, AF_INET6))
002c1265 1341 ip6_ctloutput_dispatch(msg);
984263bc
MD
1342 else
1343#endif /* INET6 */
002c1265
MD
1344 ip_ctloutput(msg);
1345 /* msg invalid now */
1346 return;
984263bc
MD
1347 }
1348 tp = intotcpcb(inp);
1349
1350 switch (sopt->sopt_dir) {
1351 case SOPT_SET:
e71a125f
AE
1352 error = soopt_to_kbuf(sopt, &optval, sizeof optval,
1353 sizeof optval);
1354 if (error)
1355 break;
984263bc 1356 switch (sopt->sopt_name) {
2a0c7c52 1357 case TCP_FASTKEEP:
0ecd93f9
MD
1358 if (optval > 0) {
1359 if ((tp->t_flags & TF_FASTKEEP) == 0) {
1360 tp->t_flags |= TF_FASTKEEP;
1361 tcp_timer_keep_activity(tp, 0);
1362 }
1363 } else {
2a0c7c52 1364 tp->t_flags &= ~TF_FASTKEEP;
0ecd93f9 1365 }
2a0c7c52 1366 break;
b1992928
MD
1367#ifdef TCP_SIGNATURE
1368 case TCP_SIGNATURE_ENABLE:
1369 if (optval > 0)
1370 tp->t_flags |= TF_SIGNATURE;
1371 else
1372 tp->t_flags &= ~TF_SIGNATURE;
1373 break;
1374#endif /* TCP_SIGNATURE */
984263bc
MD
1375 case TCP_NODELAY:
1376 case TCP_NOOPT:
984263bc
MD
1377 switch (sopt->sopt_name) {
1378 case TCP_NODELAY:
1379 opt = TF_NODELAY;
1380 break;
1381 case TCP_NOOPT:
1382 opt = TF_NOOPT;
1383 break;
1384 default:
1385 opt = 0; /* dead code to fool gcc */
1386 break;
1387 }
1388
1389 if (optval)
1390 tp->t_flags |= opt;
1391 else
1392 tp->t_flags &= ~opt;
1393 break;
1394
1395 case TCP_NOPUSH:
984263bc
MD
1396 if (optval)
1397 tp->t_flags |= TF_NOPUSH;
1398 else {
1399 tp->t_flags &= ~TF_NOPUSH;
1400 error = tcp_output(tp);
1401 }
1402 break;
1403
1404 case TCP_MAXSEG:
153d34ac
MD
1405 /*
1406 * Must be between 0 and maxseg. If the requested
1407 * maxseg is too small to satisfy the desired minmss,
1408 * pump it up (silently so sysctl modifications of
1409 * minmss do not create unexpected program failures).
1410 * Handle degenerate cases.
1411 */
1412 if (optval > 0 && optval <= tp->t_maxseg) {
1413 if (optval + 40 < tcp_minmss) {
1414 optval = tcp_minmss - 40;
1415 if (optval < 0)
1416 optval = 1;
1417 }
984263bc 1418 tp->t_maxseg = optval;
153d34ac 1419 } else {
984263bc 1420 error = EINVAL;
153d34ac 1421 }
984263bc
MD
1422 break;
1423
2ce132be
SZ
1424 case TCP_KEEPINIT:
1425 opthz = ((int64_t)optval * hz) / 1000;
1426 if (opthz >= 1)
1427 tp->t_keepinit = opthz;
1428 else
1429 error = EINVAL;
1430 break;
1431
7ea3a353
SZ
1432 case TCP_KEEPIDLE:
1433 opthz = ((int64_t)optval * hz) / 1000;
1434 if (opthz >= 1)
1435 tp->t_keepidle = opthz;
1436 else
1437 error = EINVAL;
1438 break;
1439
5d61ded3
SZ
1440 case TCP_KEEPINTVL:
1441 opthz = ((int64_t)optval * hz) / 1000;
1442 if (opthz >= 1) {
1443 tp->t_keepintvl = opthz;
1444 tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt;
1445 } else {
1446 error = EINVAL;
1447 }
1448 break;
1449
1450 case TCP_KEEPCNT:
1451 if (optval > 0) {
1452 tp->t_keepcnt = optval;
1453 tp->t_maxidle = tp->t_keepintvl * tp->t_keepcnt;
1454 } else {
1455 error = EINVAL;
1456 }
1457 break;
1458
984263bc
MD
1459 default:
1460 error = ENOPROTOOPT;
1461 break;
1462 }
1463 break;
1464
1465 case SOPT_GET:
1466 switch (sopt->sopt_name) {
b1992928
MD
1467#ifdef TCP_SIGNATURE
1468 case TCP_SIGNATURE_ENABLE:
1469 optval = (tp->t_flags & TF_SIGNATURE) ? 1 : 0;
1470 break;
1471#endif /* TCP_SIGNATURE */
984263bc
MD
1472 case TCP_NODELAY:
1473 optval = tp->t_flags & TF_NODELAY;
1474 break;
1475 case TCP_MAXSEG:
1476 optval = tp->t_maxseg;
1477 break;
1478 case TCP_NOOPT:
1479 optval = tp->t_flags & TF_NOOPT;
1480 break;
1481 case TCP_NOPUSH:
1482 optval = tp->t_flags & TF_NOPUSH;
1483 break;
2ce132be
SZ
1484 case TCP_KEEPINIT:
1485 optval = ((int64_t)tp->t_keepinit * 1000) / hz;
1486 break;
7ea3a353
SZ
1487 case TCP_KEEPIDLE:
1488 optval = ((int64_t)tp->t_keepidle * 1000) / hz;
1489 break;
5d61ded3
SZ
1490 case TCP_KEEPINTVL:
1491 optval = ((int64_t)tp->t_keepintvl * 1000) / hz;
1492 break;
1493 case TCP_KEEPCNT:
1494 optval = tp->t_keepcnt;
1495 break;
984263bc
MD
1496 default:
1497 error = ENOPROTOOPT;
1498 break;
1499 }
1500 if (error == 0)
e71a125f 1501 soopt_from_kbuf(sopt, &optval, sizeof optval);
984263bc
MD
1502 break;
1503 }
002c1265
MD
1504done:
1505 lwkt_replymsg(&msg->lmsg, error);
984263bc
MD
1506}
1507
1508/*
1509 * tcp_sendspace and tcp_recvspace are the default send and receive window
1510 * sizes, respectively. These are obsolescent (this information should
1511 * be set by the route).
0560a274
MD
1512 *
1513 * Use a default that does not require tcp window scaling to be turned
1514 * on. Individual programs or the administrator can increase the default.
984263bc 1515 */
0560a274 1516u_long tcp_sendspace = 57344; /* largest multiple of PAGE_SIZE < 64k */
f23061d4 1517SYSCTL_INT(_net_inet_tcp, TCPCTL_SENDSPACE, sendspace, CTLFLAG_RW,
984263bc
MD
1518 &tcp_sendspace , 0, "Maximum outgoing TCP datagram size");
1519u_long tcp_recvspace = 57344; /* largest multiple of PAGE_SIZE < 64k */
f23061d4 1520SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_RW,
984263bc
MD
1521 &tcp_recvspace , 0, "Maximum incoming TCP datagram size");
1522
1523/*
0ce0603e
MD
1524 * Attach TCP protocol to socket, allocating internet protocol control
1525 * block, tcp control block, bufer space, and entering LISTEN state
1526 * if to accept connections.
984263bc
MD
1527 */
1528static int
e4700d00 1529tcp_attach(struct socket *so, struct pru_attach_info *ai)
984263bc 1530{
2256ba69 1531 struct tcpcb *tp;
984263bc
MD
1532 struct inpcb *inp;
1533 int error;
6ca1a1cd 1534 int cpu;
984263bc 1535#ifdef INET6
3641b7ca 1536 int isipv6 = INP_CHECK_SOCKAF(so, AF_INET6) != 0;
984263bc
MD
1537#endif
1538
6d49aa6f 1539 if (so->so_snd.ssb_hiwat == 0 || so->so_rcv.ssb_hiwat == 0) {
6cef7136 1540 lwkt_gettoken(&so->so_rcv.ssb_token);
e4700d00
JH
1541 error = soreserve(so, tcp_sendspace, tcp_recvspace,
1542 ai->sb_rlimit);
6cef7136 1543 lwkt_reltoken(&so->so_rcv.ssb_token);
984263bc
MD
1544 if (error)
1545 return (error);
1546 }
14343ad3
MD
1547 atomic_set_int(&so->so_rcv.ssb_flags, SSB_AUTOSIZE);
1548 atomic_set_int(&so->so_snd.ssb_flags, SSB_AUTOSIZE);
6ca1a1cd 1549 cpu = mycpu->gd_cpuid;
0ce0603e
MD
1550
1551 /*
1552 * Set the default port for protocol processing. This will likely
1553 * change when we connect.
1554 */
6ca1a1cd 1555 error = in_pcballoc(so, &tcbinfo[cpu]);
984263bc
MD
1556 if (error)
1557 return (error);
ed894f8c 1558 inp = so->so_pcb;
984263bc
MD
1559#ifdef INET6
1560 if (isipv6) {
1561 inp->inp_vflag |= INP_IPV6;
1562 inp->in6p_hops = -1; /* use kernel default */
1563 }
1564 else
1565#endif
1566 inp->inp_vflag |= INP_IPV4;
1567 tp = tcp_newtcpcb(inp);
6cef7136
MD
1568 if (tp == NULL) {
1569 /*
1570 * Make sure the socket is destroyed by the pcbdetach.
1571 */
1572 soreference(so);
984263bc
MD
1573#ifdef INET6
1574 if (isipv6)
1575 in6_pcbdetach(inp);
1576 else
1577#endif
1578 in_pcbdetach(inp);
6cef7136 1579 sofree(so); /* from ref above */
984263bc
MD
1580 return (ENOBUFS);
1581 }
1582 tp->t_state = TCPS_CLOSED;
1583 return (0);
1584}
1585
1586/*
1587 * Initiate (or continue) disconnect.
1588 * If embryonic state, just send reset (once).
1589 * If in ``let data drain'' option and linger null, just drop.
1590 * Otherwise (hard), mark socket disconnecting and drop
1591 * current input data; switch states based on user close, and
1592 * send segment to peer (with FIN).
1593 */
1594static struct tcpcb *
f3f70f0d 1595tcp_disconnect(struct tcpcb *tp)
984263bc
MD
1596{
1597 struct socket *so = tp->t_inpcb->inp_socket;
1598
6cef7136 1599 if (tp->t_state < TCPS_ESTABLISHED) {
984263bc 1600 tp = tcp_close(tp);
6cef7136 1601 } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
984263bc 1602 tp = tcp_drop(tp, 0);
6cef7136
MD
1603 } else {
1604 lwkt_gettoken(&so->so_rcv.ssb_token);
984263bc 1605 soisdisconnecting(so);
6d49aa6f 1606 sbflush(&so->so_rcv.sb);
984263bc
MD
1607 tp = tcp_usrclosed(tp);
1608 if (tp)
f23061d4 1609 tcp_output(tp);
6cef7136 1610 lwkt_reltoken(&so->so_rcv.ssb_token);
984263bc
MD
1611 }
1612 return (tp);
1613}
1614
1615/*
1616 * User issued close, and wish to trail through shutdown states:
1617 * if never received SYN, just forget it. If got a SYN from peer,
1618 * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
1619 * If already got a FIN from peer, then almost done; go to LAST_ACK
1620 * state. In all other cases, have already sent FIN to peer (e.g.
1621 * after PRU_SHUTDOWN), and just have to play tedious game waiting
1622 * for peer to send FIN or not respond to keep-alives, etc.
1623 * We can let the user exit from the close as soon as the FIN is acked.
1624 */
1625static struct tcpcb *
f3f70f0d 1626tcp_usrclosed(struct tcpcb *tp)
984263bc
MD
1627{
1628
1629 switch (tp->t_state) {
1630
1631 case TCPS_CLOSED:
1632 case TCPS_LISTEN:
1633 tp->t_state = TCPS_CLOSED;
1634 tp = tcp_close(tp);
1635 break;
1636
1637 case TCPS_SYN_SENT:
1638 case TCPS_SYN_RECEIVED:
1639 tp->t_flags |= TF_NEEDFIN;
1640 break;
1641
1642 case TCPS_ESTABLISHED:
1643 tp->t_state = TCPS_FIN_WAIT_1;
1644 break;
1645
1646 case TCPS_CLOSE_WAIT:
1647 tp->t_state = TCPS_LAST_ACK;
1648 break;
1649 }
1650 if (tp && tp->t_state >= TCPS_FIN_WAIT_2) {
1651 soisdisconnected(tp->t_inpcb->inp_socket);
1652 /* To prevent the connection hanging in FIN_WAIT_2 forever. */
a48c5dd5 1653 if (tp->t_state == TCPS_FIN_WAIT_2) {
5d61ded3 1654 tcp_callout_reset(tp, tp->tt_2msl, tp->t_maxidle,
a48c5dd5
SZ
1655 tcp_timer_2msl);
1656 }
984263bc
MD
1657 }
1658 return (tp);
1659}