40
41#include "opt_ddb.h"
42#include "opt_inet.h"
43#include "opt_inet6.h"
44#include "opt_tcpdebug.h"
45
46#include <sys/param.h>
47#include <sys/systm.h>
48#include <sys/limits.h>
49#include <sys/malloc.h>
50#include <sys/kernel.h>
51#include <sys/sysctl.h>
52#include <sys/mbuf.h>
53#ifdef INET6
54#include <sys/domain.h>
55#endif /* INET6 */
56#include <sys/socket.h>
57#include <sys/socketvar.h>
58#include <sys/protosw.h>
59#include <sys/proc.h>
60#include <sys/jail.h>
61
62#ifdef DDB
63#include <ddb/ddb.h>
64#endif
65
66#include <net/if.h>
67#include <net/if_var.h>
68#include <net/route.h>
69#include <net/vnet.h>
70
71#include <netinet/cc.h>
72#include <netinet/in.h>
73#include <netinet/in_pcb.h>
74#include <netinet/in_systm.h>
75#include <netinet/in_var.h>
76#include <netinet/ip_var.h>
77#ifdef INET6
78#include <netinet/ip6.h>
79#include <netinet6/in6_pcb.h>
80#include <netinet6/ip6_var.h>
81#include <netinet6/scope6_var.h>
82#endif
83#include <netinet/tcp_fsm.h>
84#include <netinet/tcp_seq.h>
85#include <netinet/tcp_timer.h>
86#include <netinet/tcp_var.h>
87#include <netinet/tcpip.h>
88#ifdef TCPDEBUG
89#include <netinet/tcp_debug.h>
90#endif
91#ifdef TCP_OFFLOAD
92#include <netinet/tcp_offload.h>
93#endif
94
95/*
96 * TCP protocol interface to socket abstraction.
97 */
98static int tcp_attach(struct socket *);
99#ifdef INET
100static int tcp_connect(struct tcpcb *, struct sockaddr *,
101 struct thread *td);
102#endif /* INET */
103#ifdef INET6
104static int tcp6_connect(struct tcpcb *, struct sockaddr *,
105 struct thread *td);
106#endif /* INET6 */
107static void tcp_disconnect(struct tcpcb *);
108static void tcp_usrclosed(struct tcpcb *);
109static void tcp_fill_info(struct tcpcb *, struct tcp_info *);
110
111#ifdef TCPDEBUG
112#define TCPDEBUG0 int ostate = 0
113#define TCPDEBUG1() ostate = tp ? tp->t_state : 0
114#define TCPDEBUG2(req) if (tp && (so->so_options & SO_DEBUG)) \
115 tcp_trace(TA_USER, ostate, tp, 0, 0, req)
116#else
117#define TCPDEBUG0
118#define TCPDEBUG1()
119#define TCPDEBUG2(req)
120#endif
121
122/*
123 * TCP attaches to socket via pru_attach(), reserving space,
124 * and an internet control block.
125 */
126static int
127tcp_usr_attach(struct socket *so, int proto, struct thread *td)
128{
129 struct inpcb *inp;
130 struct tcpcb *tp = NULL;
131 int error;
132 TCPDEBUG0;
133
134 inp = sotoinpcb(so);
135 KASSERT(inp == NULL, ("tcp_usr_attach: inp != NULL"));
136 TCPDEBUG1();
137
138 error = tcp_attach(so);
139 if (error)
140 goto out;
141
142 if ((so->so_options & SO_LINGER) && so->so_linger == 0)
143 so->so_linger = TCP_LINGERTIME;
144
145 inp = sotoinpcb(so);
146 tp = intotcpcb(inp);
147out:
148 TCPDEBUG2(PRU_ATTACH);
149 return error;
150}
151
152/*
153 * tcp_detach is called when the socket layer loses its final reference
154 * to the socket, be it a file descriptor reference, a reference from TCP,
155 * etc. At this point, there is only one case in which we will keep around
156 * inpcb state: time wait.
157 *
158 * This function can probably be re-absorbed back into tcp_usr_detach() now
159 * that there is a single detach path.
160 */
161static void
162tcp_detach(struct socket *so, struct inpcb *inp)
163{
164 struct tcpcb *tp;
165
166 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
167 INP_WLOCK_ASSERT(inp);
168
169 KASSERT(so->so_pcb == inp, ("tcp_detach: so_pcb != inp"));
170 KASSERT(inp->inp_socket == so, ("tcp_detach: inp_socket != so"));
171
172 tp = intotcpcb(inp);
173
174 if (inp->inp_flags & INP_TIMEWAIT) {
175 /*
176 * There are two cases to handle: one in which the time wait
177 * state is being discarded (INP_DROPPED), and one in which
178 * this connection will remain in timewait. In the former,
179 * it is time to discard all state (except tcptw, which has
180 * already been discarded by the timewait close code, which
181 * should be further up the call stack somewhere). In the
182 * latter case, we detach from the socket, but leave the pcb
183 * present until timewait ends.
184 *
185 * XXXRW: Would it be cleaner to free the tcptw here?
186 *
187 * Astute question indeed, from twtcp perspective there are
188 * three cases to consider:
189 *
190 * #1 tcp_detach is called at tcptw creation time by
191 * tcp_twstart, then do not discard the newly created tcptw
192 * and leave inpcb present until timewait ends
193 * #2 tcp_detach is called at timewait end (or reuse) by
194 * tcp_twclose, then the tcptw has already been discarded
195 * and inpcb is freed here
196 * #3 tcp_detach is called() after timewait ends (or reuse)
197 * (e.g. by soclose), then tcptw has already been discarded
198 * and inpcb is freed here
199 *
200 * In all three cases the tcptw should not be freed here.
201 */
202 if (inp->inp_flags & INP_DROPPED) {
203 KASSERT(tp == NULL, ("tcp_detach: INP_TIMEWAIT && "
204 "INP_DROPPED && tp != NULL"));
205 in_pcbdetach(inp);
206 in_pcbfree(inp);
207 } else {
208 in_pcbdetach(inp);
209 INP_WUNLOCK(inp);
210 }
211 } else {
212 /*
213 * If the connection is not in timewait, we consider two
214 * two conditions: one in which no further processing is
215 * necessary (dropped || embryonic), and one in which TCP is
216 * not yet done, but no longer requires the socket, so the
217 * pcb will persist for the time being.
218 *
219 * XXXRW: Does the second case still occur?
220 */
221 if (inp->inp_flags & INP_DROPPED ||
222 tp->t_state < TCPS_SYN_SENT) {
223 tcp_discardcb(tp);
224 in_pcbdetach(inp);
225 in_pcbfree(inp);
226 } else {
227 in_pcbdetach(inp);
228 INP_WUNLOCK(inp);
229 }
230 }
231}
232
233/*
234 * pru_detach() detaches the TCP protocol from the socket.
235 * If the protocol state is non-embryonic, then can't
236 * do this directly: have to initiate a pru_disconnect(),
237 * which may finish later; embryonic TCB's can just
238 * be discarded here.
239 */
240static void
241tcp_usr_detach(struct socket *so)
242{
243 struct inpcb *inp;
244
245 inp = sotoinpcb(so);
246 KASSERT(inp != NULL, ("tcp_usr_detach: inp == NULL"));
247 INP_INFO_WLOCK(&V_tcbinfo);
248 INP_WLOCK(inp);
249 KASSERT(inp->inp_socket != NULL,
250 ("tcp_usr_detach: inp_socket == NULL"));
251 tcp_detach(so, inp);
252 INP_INFO_WUNLOCK(&V_tcbinfo);
253}
254
255#ifdef INET
256/*
257 * Give the socket an address.
258 */
259static int
260tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
261{
262 int error = 0;
263 struct inpcb *inp;
264 struct tcpcb *tp = NULL;
265 struct sockaddr_in *sinp;
266
267 sinp = (struct sockaddr_in *)nam;
268 if (nam->sa_len != sizeof (*sinp))
269 return (EINVAL);
270 /*
271 * Must check for multicast addresses and disallow binding
272 * to them.
273 */
274 if (sinp->sin_family == AF_INET &&
275 IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
276 return (EAFNOSUPPORT);
277
278 TCPDEBUG0;
279 inp = sotoinpcb(so);
280 KASSERT(inp != NULL, ("tcp_usr_bind: inp == NULL"));
281 INP_WLOCK(inp);
282 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
283 error = EINVAL;
284 goto out;
285 }
286 tp = intotcpcb(inp);
287 TCPDEBUG1();
288 INP_HASH_WLOCK(&V_tcbinfo);
289 error = in_pcbbind(inp, nam, td->td_ucred);
290 INP_HASH_WUNLOCK(&V_tcbinfo);
291out:
292 TCPDEBUG2(PRU_BIND);
293 INP_WUNLOCK(inp);
294
295 return (error);
296}
297#endif /* INET */
298
299#ifdef INET6
300static int
301tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
302{
303 int error = 0;
304 struct inpcb *inp;
305 struct tcpcb *tp = NULL;
306 struct sockaddr_in6 *sin6p;
307
308 sin6p = (struct sockaddr_in6 *)nam;
309 if (nam->sa_len != sizeof (*sin6p))
310 return (EINVAL);
311 /*
312 * Must check for multicast addresses and disallow binding
313 * to them.
314 */
315 if (sin6p->sin6_family == AF_INET6 &&
316 IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr))
317 return (EAFNOSUPPORT);
318
319 TCPDEBUG0;
320 inp = sotoinpcb(so);
321 KASSERT(inp != NULL, ("tcp6_usr_bind: inp == NULL"));
322 INP_WLOCK(inp);
323 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
324 error = EINVAL;
325 goto out;
326 }
327 tp = intotcpcb(inp);
328 TCPDEBUG1();
329 INP_HASH_WLOCK(&V_tcbinfo);
330 inp->inp_vflag &= ~INP_IPV4;
331 inp->inp_vflag |= INP_IPV6;
332#ifdef INET
333 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
334 if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr))
335 inp->inp_vflag |= INP_IPV4;
336 else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
337 struct sockaddr_in sin;
338
339 in6_sin6_2_sin(&sin, sin6p);
340 inp->inp_vflag |= INP_IPV4;
341 inp->inp_vflag &= ~INP_IPV6;
342 error = in_pcbbind(inp, (struct sockaddr *)&sin,
343 td->td_ucred);
344 INP_HASH_WUNLOCK(&V_tcbinfo);
345 goto out;
346 }
347 }
348#endif
349 error = in6_pcbbind(inp, nam, td->td_ucred);
350 INP_HASH_WUNLOCK(&V_tcbinfo);
351out:
352 TCPDEBUG2(PRU_BIND);
353 INP_WUNLOCK(inp);
354 return (error);
355}
356#endif /* INET6 */
357
358#ifdef INET
359/*
360 * Prepare to accept connections.
361 */
362static int
363tcp_usr_listen(struct socket *so, int backlog, struct thread *td)
364{
365 int error = 0;
366 struct inpcb *inp;
367 struct tcpcb *tp = NULL;
368
369 TCPDEBUG0;
370 inp = sotoinpcb(so);
371 KASSERT(inp != NULL, ("tcp_usr_listen: inp == NULL"));
372 INP_WLOCK(inp);
373 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
374 error = EINVAL;
375 goto out;
376 }
377 tp = intotcpcb(inp);
378 TCPDEBUG1();
379 SOCK_LOCK(so);
380 error = solisten_proto_check(so);
381 INP_HASH_WLOCK(&V_tcbinfo);
382 if (error == 0 && inp->inp_lport == 0)
383 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
384 INP_HASH_WUNLOCK(&V_tcbinfo);
385 if (error == 0) {
386 tcp_state_change(tp, TCPS_LISTEN);
387 solisten_proto(so, backlog);
388#ifdef TCP_OFFLOAD
389 if ((so->so_options & SO_NO_OFFLOAD) == 0)
390 tcp_offload_listen_start(tp);
391#endif
392 }
393 SOCK_UNLOCK(so);
394
395out:
396 TCPDEBUG2(PRU_LISTEN);
397 INP_WUNLOCK(inp);
398 return (error);
399}
400#endif /* INET */
401
402#ifdef INET6
403static int
404tcp6_usr_listen(struct socket *so, int backlog, struct thread *td)
405{
406 int error = 0;
407 struct inpcb *inp;
408 struct tcpcb *tp = NULL;
409
410 TCPDEBUG0;
411 inp = sotoinpcb(so);
412 KASSERT(inp != NULL, ("tcp6_usr_listen: inp == NULL"));
413 INP_WLOCK(inp);
414 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
415 error = EINVAL;
416 goto out;
417 }
418 tp = intotcpcb(inp);
419 TCPDEBUG1();
420 SOCK_LOCK(so);
421 error = solisten_proto_check(so);
422 INP_HASH_WLOCK(&V_tcbinfo);
423 if (error == 0 && inp->inp_lport == 0) {
424 inp->inp_vflag &= ~INP_IPV4;
425 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0)
426 inp->inp_vflag |= INP_IPV4;
427 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
428 }
429 INP_HASH_WUNLOCK(&V_tcbinfo);
430 if (error == 0) {
431 tcp_state_change(tp, TCPS_LISTEN);
432 solisten_proto(so, backlog);
433#ifdef TCP_OFFLOAD
434 if ((so->so_options & SO_NO_OFFLOAD) == 0)
435 tcp_offload_listen_start(tp);
436#endif
437 }
438 SOCK_UNLOCK(so);
439
440out:
441 TCPDEBUG2(PRU_LISTEN);
442 INP_WUNLOCK(inp);
443 return (error);
444}
445#endif /* INET6 */
446
447#ifdef INET
448/*
449 * Initiate connection to peer.
450 * Create a template for use in transmissions on this connection.
451 * Enter SYN_SENT state, and mark socket as connecting.
452 * Start keep-alive timer, and seed output sequence space.
453 * Send initial segment on connection.
454 */
455static int
456tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
457{
458 int error = 0;
459 struct inpcb *inp;
460 struct tcpcb *tp = NULL;
461 struct sockaddr_in *sinp;
462
463 sinp = (struct sockaddr_in *)nam;
464 if (nam->sa_len != sizeof (*sinp))
465 return (EINVAL);
466 /*
467 * Must disallow TCP ``connections'' to multicast addresses.
468 */
469 if (sinp->sin_family == AF_INET
470 && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
471 return (EAFNOSUPPORT);
472 if ((error = prison_remote_ip4(td->td_ucred, &sinp->sin_addr)) != 0)
473 return (error);
474
475 TCPDEBUG0;
476 inp = sotoinpcb(so);
477 KASSERT(inp != NULL, ("tcp_usr_connect: inp == NULL"));
478 INP_WLOCK(inp);
479 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
480 error = EINVAL;
481 goto out;
482 }
483 tp = intotcpcb(inp);
484 TCPDEBUG1();
485 if ((error = tcp_connect(tp, nam, td)) != 0)
486 goto out;
487#ifdef TCP_OFFLOAD
488 if (registered_toedevs > 0 &&
489 (so->so_options & SO_NO_OFFLOAD) == 0 &&
490 (error = tcp_offload_connect(so, nam)) == 0)
491 goto out;
492#endif
493 tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
494 error = tcp_output(tp);
495out:
496 TCPDEBUG2(PRU_CONNECT);
497 INP_WUNLOCK(inp);
498 return (error);
499}
500#endif /* INET */
501
502#ifdef INET6
503static int
504tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
505{
506 int error = 0;
507 struct inpcb *inp;
508 struct tcpcb *tp = NULL;
509 struct sockaddr_in6 *sin6p;
510
511 TCPDEBUG0;
512
513 sin6p = (struct sockaddr_in6 *)nam;
514 if (nam->sa_len != sizeof (*sin6p))
515 return (EINVAL);
516 /*
517 * Must disallow TCP ``connections'' to multicast addresses.
518 */
519 if (sin6p->sin6_family == AF_INET6
520 && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr))
521 return (EAFNOSUPPORT);
522
523 inp = sotoinpcb(so);
524 KASSERT(inp != NULL, ("tcp6_usr_connect: inp == NULL"));
525 INP_WLOCK(inp);
526 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
527 error = EINVAL;
528 goto out;
529 }
530 tp = intotcpcb(inp);
531 TCPDEBUG1();
532#ifdef INET
533 /*
534 * XXXRW: Some confusion: V4/V6 flags relate to binding, and
535 * therefore probably require the hash lock, which isn't held here.
536 * Is this a significant problem?
537 */
538 if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
539 struct sockaddr_in sin;
540
541 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
542 error = EINVAL;
543 goto out;
544 }
545
546 in6_sin6_2_sin(&sin, sin6p);
547 inp->inp_vflag |= INP_IPV4;
548 inp->inp_vflag &= ~INP_IPV6;
549 if ((error = prison_remote_ip4(td->td_ucred,
550 &sin.sin_addr)) != 0)
551 goto out;
552 if ((error = tcp_connect(tp, (struct sockaddr *)&sin, td)) != 0)
553 goto out;
554#ifdef TCP_OFFLOAD
555 if (registered_toedevs > 0 &&
556 (so->so_options & SO_NO_OFFLOAD) == 0 &&
557 (error = tcp_offload_connect(so, nam)) == 0)
558 goto out;
559#endif
560 error = tcp_output(tp);
561 goto out;
562 }
563#endif
564 inp->inp_vflag &= ~INP_IPV4;
565 inp->inp_vflag |= INP_IPV6;
566 inp->inp_inc.inc_flags |= INC_ISIPV6;
567 if ((error = prison_remote_ip6(td->td_ucred, &sin6p->sin6_addr)) != 0)
568 goto out;
569 if ((error = tcp6_connect(tp, nam, td)) != 0)
570 goto out;
571#ifdef TCP_OFFLOAD
572 if (registered_toedevs > 0 &&
573 (so->so_options & SO_NO_OFFLOAD) == 0 &&
574 (error = tcp_offload_connect(so, nam)) == 0)
575 goto out;
576#endif
577 tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
578 error = tcp_output(tp);
579
580out:
581 TCPDEBUG2(PRU_CONNECT);
582 INP_WUNLOCK(inp);
583 return (error);
584}
585#endif /* INET6 */
586
587/*
588 * Initiate disconnect from peer.
589 * If connection never passed embryonic stage, just drop;
590 * else if don't need to let data drain, then can just drop anyways,
591 * else have to begin TCP shutdown process: mark socket disconnecting,
592 * drain unread data, state switch to reflect user close, and
593 * send segment (e.g. FIN) to peer. Socket will be really disconnected
594 * when peer sends FIN and acks ours.
595 *
596 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
597 */
598static int
599tcp_usr_disconnect(struct socket *so)
600{
601 struct inpcb *inp;
602 struct tcpcb *tp = NULL;
603 int error = 0;
604
605 TCPDEBUG0;
606 INP_INFO_WLOCK(&V_tcbinfo);
607 inp = sotoinpcb(so);
608 KASSERT(inp != NULL, ("tcp_usr_disconnect: inp == NULL"));
609 INP_WLOCK(inp);
610 if (inp->inp_flags & INP_TIMEWAIT)
611 goto out;
612 if (inp->inp_flags & INP_DROPPED) {
613 error = ECONNRESET;
614 goto out;
615 }
616 tp = intotcpcb(inp);
617 TCPDEBUG1();
618 tcp_disconnect(tp);
619out:
620 TCPDEBUG2(PRU_DISCONNECT);
621 INP_WUNLOCK(inp);
622 INP_INFO_WUNLOCK(&V_tcbinfo);
623 return (error);
624}
625
626#ifdef INET
627/*
628 * Accept a connection. Essentially all the work is done at higher levels;
629 * just return the address of the peer, storing through addr.
630 */
631static int
632tcp_usr_accept(struct socket *so, struct sockaddr **nam)
633{
634 int error = 0;
635 struct inpcb *inp = NULL;
636 struct tcpcb *tp = NULL;
637 struct in_addr addr;
638 in_port_t port = 0;
639 TCPDEBUG0;
640
641 if (so->so_state & SS_ISDISCONNECTED)
642 return (ECONNABORTED);
643
644 inp = sotoinpcb(so);
645 KASSERT(inp != NULL, ("tcp_usr_accept: inp == NULL"));
646 INP_WLOCK(inp);
647 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
648 error = ECONNABORTED;
649 goto out;
650 }
651 tp = intotcpcb(inp);
652 TCPDEBUG1();
653
654 /*
655 * We inline in_getpeeraddr and COMMON_END here, so that we can
656 * copy the data of interest and defer the malloc until after we
657 * release the lock.
658 */
659 port = inp->inp_fport;
660 addr = inp->inp_faddr;
661
662out:
663 TCPDEBUG2(PRU_ACCEPT);
664 INP_WUNLOCK(inp);
665 if (error == 0)
666 *nam = in_sockaddr(port, &addr);
667 return error;
668}
669#endif /* INET */
670
671#ifdef INET6
672static int
673tcp6_usr_accept(struct socket *so, struct sockaddr **nam)
674{
675 struct inpcb *inp = NULL;
676 int error = 0;
677 struct tcpcb *tp = NULL;
678 struct in_addr addr;
679 struct in6_addr addr6;
680 in_port_t port = 0;
681 int v4 = 0;
682 TCPDEBUG0;
683
684 if (so->so_state & SS_ISDISCONNECTED)
685 return (ECONNABORTED);
686
687 inp = sotoinpcb(so);
688 KASSERT(inp != NULL, ("tcp6_usr_accept: inp == NULL"));
689 INP_INFO_RLOCK(&V_tcbinfo);
690 INP_WLOCK(inp);
691 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
692 error = ECONNABORTED;
693 goto out;
694 }
695 tp = intotcpcb(inp);
696 TCPDEBUG1();
697
698 /*
699 * We inline in6_mapped_peeraddr and COMMON_END here, so that we can
700 * copy the data of interest and defer the malloc until after we
701 * release the lock.
702 */
703 if (inp->inp_vflag & INP_IPV4) {
704 v4 = 1;
705 port = inp->inp_fport;
706 addr = inp->inp_faddr;
707 } else {
708 port = inp->inp_fport;
709 addr6 = inp->in6p_faddr;
710 }
711
712out:
713 TCPDEBUG2(PRU_ACCEPT);
714 INP_WUNLOCK(inp);
715 INP_INFO_RUNLOCK(&V_tcbinfo);
716 if (error == 0) {
717 if (v4)
718 *nam = in6_v4mapsin6_sockaddr(port, &addr);
719 else
720 *nam = in6_sockaddr(port, &addr6);
721 }
722 return error;
723}
724#endif /* INET6 */
725
726/*
727 * Mark the connection as being incapable of further output.
728 */
729static int
730tcp_usr_shutdown(struct socket *so)
731{
732 int error = 0;
733 struct inpcb *inp;
734 struct tcpcb *tp = NULL;
735
736 TCPDEBUG0;
737 INP_INFO_WLOCK(&V_tcbinfo);
738 inp = sotoinpcb(so);
739 KASSERT(inp != NULL, ("inp == NULL"));
740 INP_WLOCK(inp);
741 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
742 error = ECONNRESET;
743 goto out;
744 }
745 tp = intotcpcb(inp);
746 TCPDEBUG1();
747 socantsendmore(so);
748 tcp_usrclosed(tp);
749 if (!(inp->inp_flags & INP_DROPPED))
750 error = tcp_output(tp);
751
752out:
753 TCPDEBUG2(PRU_SHUTDOWN);
754 INP_WUNLOCK(inp);
755 INP_INFO_WUNLOCK(&V_tcbinfo);
756
757 return (error);
758}
759
760/*
761 * After a receive, possibly send window update to peer.
762 */
763static int
764tcp_usr_rcvd(struct socket *so, int flags)
765{
766 struct inpcb *inp;
767 struct tcpcb *tp = NULL;
768 int error = 0;
769
770 TCPDEBUG0;
771 inp = sotoinpcb(so);
772 KASSERT(inp != NULL, ("tcp_usr_rcvd: inp == NULL"));
773 INP_WLOCK(inp);
774 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
775 error = ECONNRESET;
776 goto out;
777 }
778 tp = intotcpcb(inp);
779 TCPDEBUG1();
780#ifdef TCP_OFFLOAD
781 if (tp->t_flags & TF_TOE)
782 tcp_offload_rcvd(tp);
783 else
784#endif
785 tcp_output(tp);
786
787out:
788 TCPDEBUG2(PRU_RCVD);
789 INP_WUNLOCK(inp);
790 return (error);
791}
792
793/*
794 * Do a send by putting data in output queue and updating urgent
795 * marker if URG set. Possibly send more data. Unlike the other
796 * pru_*() routines, the mbuf chains are our responsibility. We
797 * must either enqueue them or free them. The other pru_* routines
798 * generally are caller-frees.
799 */
800static int
801tcp_usr_send(struct socket *so, int flags, struct mbuf *m,
802 struct sockaddr *nam, struct mbuf *control, struct thread *td)
803{
804 int error = 0;
805 struct inpcb *inp;
806 struct tcpcb *tp = NULL;
807#ifdef INET6
808 int isipv6;
809#endif
810 TCPDEBUG0;
811
812 /*
813 * We require the pcbinfo lock if we will close the socket as part of
814 * this call.
815 */
816 if (flags & PRUS_EOF)
817 INP_INFO_WLOCK(&V_tcbinfo);
818 inp = sotoinpcb(so);
819 KASSERT(inp != NULL, ("tcp_usr_send: inp == NULL"));
820 INP_WLOCK(inp);
821 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
822 if (control)
823 m_freem(control);
824 if (m)
825 m_freem(m);
826 error = ECONNRESET;
827 goto out;
828 }
829#ifdef INET6
830 isipv6 = nam && nam->sa_family == AF_INET6;
831#endif /* INET6 */
832 tp = intotcpcb(inp);
833 TCPDEBUG1();
834 if (control) {
835 /* TCP doesn't do control messages (rights, creds, etc) */
836 if (control->m_len) {
837 m_freem(control);
838 if (m)
839 m_freem(m);
840 error = EINVAL;
841 goto out;
842 }
843 m_freem(control); /* empty control, just free it */
844 }
845 if (!(flags & PRUS_OOB)) {
| 40
41#include "opt_ddb.h"
42#include "opt_inet.h"
43#include "opt_inet6.h"
44#include "opt_tcpdebug.h"
45
46#include <sys/param.h>
47#include <sys/systm.h>
48#include <sys/limits.h>
49#include <sys/malloc.h>
50#include <sys/kernel.h>
51#include <sys/sysctl.h>
52#include <sys/mbuf.h>
53#ifdef INET6
54#include <sys/domain.h>
55#endif /* INET6 */
56#include <sys/socket.h>
57#include <sys/socketvar.h>
58#include <sys/protosw.h>
59#include <sys/proc.h>
60#include <sys/jail.h>
61
62#ifdef DDB
63#include <ddb/ddb.h>
64#endif
65
66#include <net/if.h>
67#include <net/if_var.h>
68#include <net/route.h>
69#include <net/vnet.h>
70
71#include <netinet/cc.h>
72#include <netinet/in.h>
73#include <netinet/in_pcb.h>
74#include <netinet/in_systm.h>
75#include <netinet/in_var.h>
76#include <netinet/ip_var.h>
77#ifdef INET6
78#include <netinet/ip6.h>
79#include <netinet6/in6_pcb.h>
80#include <netinet6/ip6_var.h>
81#include <netinet6/scope6_var.h>
82#endif
83#include <netinet/tcp_fsm.h>
84#include <netinet/tcp_seq.h>
85#include <netinet/tcp_timer.h>
86#include <netinet/tcp_var.h>
87#include <netinet/tcpip.h>
88#ifdef TCPDEBUG
89#include <netinet/tcp_debug.h>
90#endif
91#ifdef TCP_OFFLOAD
92#include <netinet/tcp_offload.h>
93#endif
94
95/*
96 * TCP protocol interface to socket abstraction.
97 */
98static int tcp_attach(struct socket *);
99#ifdef INET
100static int tcp_connect(struct tcpcb *, struct sockaddr *,
101 struct thread *td);
102#endif /* INET */
103#ifdef INET6
104static int tcp6_connect(struct tcpcb *, struct sockaddr *,
105 struct thread *td);
106#endif /* INET6 */
107static void tcp_disconnect(struct tcpcb *);
108static void tcp_usrclosed(struct tcpcb *);
109static void tcp_fill_info(struct tcpcb *, struct tcp_info *);
110
111#ifdef TCPDEBUG
112#define TCPDEBUG0 int ostate = 0
113#define TCPDEBUG1() ostate = tp ? tp->t_state : 0
114#define TCPDEBUG2(req) if (tp && (so->so_options & SO_DEBUG)) \
115 tcp_trace(TA_USER, ostate, tp, 0, 0, req)
116#else
117#define TCPDEBUG0
118#define TCPDEBUG1()
119#define TCPDEBUG2(req)
120#endif
121
122/*
123 * TCP attaches to socket via pru_attach(), reserving space,
124 * and an internet control block.
125 */
126static int
127tcp_usr_attach(struct socket *so, int proto, struct thread *td)
128{
129 struct inpcb *inp;
130 struct tcpcb *tp = NULL;
131 int error;
132 TCPDEBUG0;
133
134 inp = sotoinpcb(so);
135 KASSERT(inp == NULL, ("tcp_usr_attach: inp != NULL"));
136 TCPDEBUG1();
137
138 error = tcp_attach(so);
139 if (error)
140 goto out;
141
142 if ((so->so_options & SO_LINGER) && so->so_linger == 0)
143 so->so_linger = TCP_LINGERTIME;
144
145 inp = sotoinpcb(so);
146 tp = intotcpcb(inp);
147out:
148 TCPDEBUG2(PRU_ATTACH);
149 return error;
150}
151
152/*
153 * tcp_detach is called when the socket layer loses its final reference
154 * to the socket, be it a file descriptor reference, a reference from TCP,
155 * etc. At this point, there is only one case in which we will keep around
156 * inpcb state: time wait.
157 *
158 * This function can probably be re-absorbed back into tcp_usr_detach() now
159 * that there is a single detach path.
160 */
161static void
162tcp_detach(struct socket *so, struct inpcb *inp)
163{
164 struct tcpcb *tp;
165
166 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
167 INP_WLOCK_ASSERT(inp);
168
169 KASSERT(so->so_pcb == inp, ("tcp_detach: so_pcb != inp"));
170 KASSERT(inp->inp_socket == so, ("tcp_detach: inp_socket != so"));
171
172 tp = intotcpcb(inp);
173
174 if (inp->inp_flags & INP_TIMEWAIT) {
175 /*
176 * There are two cases to handle: one in which the time wait
177 * state is being discarded (INP_DROPPED), and one in which
178 * this connection will remain in timewait. In the former,
179 * it is time to discard all state (except tcptw, which has
180 * already been discarded by the timewait close code, which
181 * should be further up the call stack somewhere). In the
182 * latter case, we detach from the socket, but leave the pcb
183 * present until timewait ends.
184 *
185 * XXXRW: Would it be cleaner to free the tcptw here?
186 *
187 * Astute question indeed, from twtcp perspective there are
188 * three cases to consider:
189 *
190 * #1 tcp_detach is called at tcptw creation time by
191 * tcp_twstart, then do not discard the newly created tcptw
192 * and leave inpcb present until timewait ends
193 * #2 tcp_detach is called at timewait end (or reuse) by
194 * tcp_twclose, then the tcptw has already been discarded
195 * and inpcb is freed here
196 * #3 tcp_detach is called() after timewait ends (or reuse)
197 * (e.g. by soclose), then tcptw has already been discarded
198 * and inpcb is freed here
199 *
200 * In all three cases the tcptw should not be freed here.
201 */
202 if (inp->inp_flags & INP_DROPPED) {
203 KASSERT(tp == NULL, ("tcp_detach: INP_TIMEWAIT && "
204 "INP_DROPPED && tp != NULL"));
205 in_pcbdetach(inp);
206 in_pcbfree(inp);
207 } else {
208 in_pcbdetach(inp);
209 INP_WUNLOCK(inp);
210 }
211 } else {
212 /*
213 * If the connection is not in timewait, we consider two
214 * two conditions: one in which no further processing is
215 * necessary (dropped || embryonic), and one in which TCP is
216 * not yet done, but no longer requires the socket, so the
217 * pcb will persist for the time being.
218 *
219 * XXXRW: Does the second case still occur?
220 */
221 if (inp->inp_flags & INP_DROPPED ||
222 tp->t_state < TCPS_SYN_SENT) {
223 tcp_discardcb(tp);
224 in_pcbdetach(inp);
225 in_pcbfree(inp);
226 } else {
227 in_pcbdetach(inp);
228 INP_WUNLOCK(inp);
229 }
230 }
231}
232
233/*
234 * pru_detach() detaches the TCP protocol from the socket.
235 * If the protocol state is non-embryonic, then can't
236 * do this directly: have to initiate a pru_disconnect(),
237 * which may finish later; embryonic TCB's can just
238 * be discarded here.
239 */
240static void
241tcp_usr_detach(struct socket *so)
242{
243 struct inpcb *inp;
244
245 inp = sotoinpcb(so);
246 KASSERT(inp != NULL, ("tcp_usr_detach: inp == NULL"));
247 INP_INFO_WLOCK(&V_tcbinfo);
248 INP_WLOCK(inp);
249 KASSERT(inp->inp_socket != NULL,
250 ("tcp_usr_detach: inp_socket == NULL"));
251 tcp_detach(so, inp);
252 INP_INFO_WUNLOCK(&V_tcbinfo);
253}
254
255#ifdef INET
256/*
257 * Give the socket an address.
258 */
259static int
260tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
261{
262 int error = 0;
263 struct inpcb *inp;
264 struct tcpcb *tp = NULL;
265 struct sockaddr_in *sinp;
266
267 sinp = (struct sockaddr_in *)nam;
268 if (nam->sa_len != sizeof (*sinp))
269 return (EINVAL);
270 /*
271 * Must check for multicast addresses and disallow binding
272 * to them.
273 */
274 if (sinp->sin_family == AF_INET &&
275 IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
276 return (EAFNOSUPPORT);
277
278 TCPDEBUG0;
279 inp = sotoinpcb(so);
280 KASSERT(inp != NULL, ("tcp_usr_bind: inp == NULL"));
281 INP_WLOCK(inp);
282 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
283 error = EINVAL;
284 goto out;
285 }
286 tp = intotcpcb(inp);
287 TCPDEBUG1();
288 INP_HASH_WLOCK(&V_tcbinfo);
289 error = in_pcbbind(inp, nam, td->td_ucred);
290 INP_HASH_WUNLOCK(&V_tcbinfo);
291out:
292 TCPDEBUG2(PRU_BIND);
293 INP_WUNLOCK(inp);
294
295 return (error);
296}
297#endif /* INET */
298
299#ifdef INET6
300static int
301tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
302{
303 int error = 0;
304 struct inpcb *inp;
305 struct tcpcb *tp = NULL;
306 struct sockaddr_in6 *sin6p;
307
308 sin6p = (struct sockaddr_in6 *)nam;
309 if (nam->sa_len != sizeof (*sin6p))
310 return (EINVAL);
311 /*
312 * Must check for multicast addresses and disallow binding
313 * to them.
314 */
315 if (sin6p->sin6_family == AF_INET6 &&
316 IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr))
317 return (EAFNOSUPPORT);
318
319 TCPDEBUG0;
320 inp = sotoinpcb(so);
321 KASSERT(inp != NULL, ("tcp6_usr_bind: inp == NULL"));
322 INP_WLOCK(inp);
323 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
324 error = EINVAL;
325 goto out;
326 }
327 tp = intotcpcb(inp);
328 TCPDEBUG1();
329 INP_HASH_WLOCK(&V_tcbinfo);
330 inp->inp_vflag &= ~INP_IPV4;
331 inp->inp_vflag |= INP_IPV6;
332#ifdef INET
333 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
334 if (IN6_IS_ADDR_UNSPECIFIED(&sin6p->sin6_addr))
335 inp->inp_vflag |= INP_IPV4;
336 else if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
337 struct sockaddr_in sin;
338
339 in6_sin6_2_sin(&sin, sin6p);
340 inp->inp_vflag |= INP_IPV4;
341 inp->inp_vflag &= ~INP_IPV6;
342 error = in_pcbbind(inp, (struct sockaddr *)&sin,
343 td->td_ucred);
344 INP_HASH_WUNLOCK(&V_tcbinfo);
345 goto out;
346 }
347 }
348#endif
349 error = in6_pcbbind(inp, nam, td->td_ucred);
350 INP_HASH_WUNLOCK(&V_tcbinfo);
351out:
352 TCPDEBUG2(PRU_BIND);
353 INP_WUNLOCK(inp);
354 return (error);
355}
356#endif /* INET6 */
357
358#ifdef INET
359/*
360 * Prepare to accept connections.
361 */
362static int
363tcp_usr_listen(struct socket *so, int backlog, struct thread *td)
364{
365 int error = 0;
366 struct inpcb *inp;
367 struct tcpcb *tp = NULL;
368
369 TCPDEBUG0;
370 inp = sotoinpcb(so);
371 KASSERT(inp != NULL, ("tcp_usr_listen: inp == NULL"));
372 INP_WLOCK(inp);
373 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
374 error = EINVAL;
375 goto out;
376 }
377 tp = intotcpcb(inp);
378 TCPDEBUG1();
379 SOCK_LOCK(so);
380 error = solisten_proto_check(so);
381 INP_HASH_WLOCK(&V_tcbinfo);
382 if (error == 0 && inp->inp_lport == 0)
383 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
384 INP_HASH_WUNLOCK(&V_tcbinfo);
385 if (error == 0) {
386 tcp_state_change(tp, TCPS_LISTEN);
387 solisten_proto(so, backlog);
388#ifdef TCP_OFFLOAD
389 if ((so->so_options & SO_NO_OFFLOAD) == 0)
390 tcp_offload_listen_start(tp);
391#endif
392 }
393 SOCK_UNLOCK(so);
394
395out:
396 TCPDEBUG2(PRU_LISTEN);
397 INP_WUNLOCK(inp);
398 return (error);
399}
400#endif /* INET */
401
402#ifdef INET6
403static int
404tcp6_usr_listen(struct socket *so, int backlog, struct thread *td)
405{
406 int error = 0;
407 struct inpcb *inp;
408 struct tcpcb *tp = NULL;
409
410 TCPDEBUG0;
411 inp = sotoinpcb(so);
412 KASSERT(inp != NULL, ("tcp6_usr_listen: inp == NULL"));
413 INP_WLOCK(inp);
414 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
415 error = EINVAL;
416 goto out;
417 }
418 tp = intotcpcb(inp);
419 TCPDEBUG1();
420 SOCK_LOCK(so);
421 error = solisten_proto_check(so);
422 INP_HASH_WLOCK(&V_tcbinfo);
423 if (error == 0 && inp->inp_lport == 0) {
424 inp->inp_vflag &= ~INP_IPV4;
425 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0)
426 inp->inp_vflag |= INP_IPV4;
427 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
428 }
429 INP_HASH_WUNLOCK(&V_tcbinfo);
430 if (error == 0) {
431 tcp_state_change(tp, TCPS_LISTEN);
432 solisten_proto(so, backlog);
433#ifdef TCP_OFFLOAD
434 if ((so->so_options & SO_NO_OFFLOAD) == 0)
435 tcp_offload_listen_start(tp);
436#endif
437 }
438 SOCK_UNLOCK(so);
439
440out:
441 TCPDEBUG2(PRU_LISTEN);
442 INP_WUNLOCK(inp);
443 return (error);
444}
445#endif /* INET6 */
446
447#ifdef INET
448/*
449 * Initiate connection to peer.
450 * Create a template for use in transmissions on this connection.
451 * Enter SYN_SENT state, and mark socket as connecting.
452 * Start keep-alive timer, and seed output sequence space.
453 * Send initial segment on connection.
454 */
455static int
456tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
457{
458 int error = 0;
459 struct inpcb *inp;
460 struct tcpcb *tp = NULL;
461 struct sockaddr_in *sinp;
462
463 sinp = (struct sockaddr_in *)nam;
464 if (nam->sa_len != sizeof (*sinp))
465 return (EINVAL);
466 /*
467 * Must disallow TCP ``connections'' to multicast addresses.
468 */
469 if (sinp->sin_family == AF_INET
470 && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
471 return (EAFNOSUPPORT);
472 if ((error = prison_remote_ip4(td->td_ucred, &sinp->sin_addr)) != 0)
473 return (error);
474
475 TCPDEBUG0;
476 inp = sotoinpcb(so);
477 KASSERT(inp != NULL, ("tcp_usr_connect: inp == NULL"));
478 INP_WLOCK(inp);
479 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
480 error = EINVAL;
481 goto out;
482 }
483 tp = intotcpcb(inp);
484 TCPDEBUG1();
485 if ((error = tcp_connect(tp, nam, td)) != 0)
486 goto out;
487#ifdef TCP_OFFLOAD
488 if (registered_toedevs > 0 &&
489 (so->so_options & SO_NO_OFFLOAD) == 0 &&
490 (error = tcp_offload_connect(so, nam)) == 0)
491 goto out;
492#endif
493 tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
494 error = tcp_output(tp);
495out:
496 TCPDEBUG2(PRU_CONNECT);
497 INP_WUNLOCK(inp);
498 return (error);
499}
500#endif /* INET */
501
502#ifdef INET6
503static int
504tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
505{
506 int error = 0;
507 struct inpcb *inp;
508 struct tcpcb *tp = NULL;
509 struct sockaddr_in6 *sin6p;
510
511 TCPDEBUG0;
512
513 sin6p = (struct sockaddr_in6 *)nam;
514 if (nam->sa_len != sizeof (*sin6p))
515 return (EINVAL);
516 /*
517 * Must disallow TCP ``connections'' to multicast addresses.
518 */
519 if (sin6p->sin6_family == AF_INET6
520 && IN6_IS_ADDR_MULTICAST(&sin6p->sin6_addr))
521 return (EAFNOSUPPORT);
522
523 inp = sotoinpcb(so);
524 KASSERT(inp != NULL, ("tcp6_usr_connect: inp == NULL"));
525 INP_WLOCK(inp);
526 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
527 error = EINVAL;
528 goto out;
529 }
530 tp = intotcpcb(inp);
531 TCPDEBUG1();
532#ifdef INET
533 /*
534 * XXXRW: Some confusion: V4/V6 flags relate to binding, and
535 * therefore probably require the hash lock, which isn't held here.
536 * Is this a significant problem?
537 */
538 if (IN6_IS_ADDR_V4MAPPED(&sin6p->sin6_addr)) {
539 struct sockaddr_in sin;
540
541 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
542 error = EINVAL;
543 goto out;
544 }
545
546 in6_sin6_2_sin(&sin, sin6p);
547 inp->inp_vflag |= INP_IPV4;
548 inp->inp_vflag &= ~INP_IPV6;
549 if ((error = prison_remote_ip4(td->td_ucred,
550 &sin.sin_addr)) != 0)
551 goto out;
552 if ((error = tcp_connect(tp, (struct sockaddr *)&sin, td)) != 0)
553 goto out;
554#ifdef TCP_OFFLOAD
555 if (registered_toedevs > 0 &&
556 (so->so_options & SO_NO_OFFLOAD) == 0 &&
557 (error = tcp_offload_connect(so, nam)) == 0)
558 goto out;
559#endif
560 error = tcp_output(tp);
561 goto out;
562 }
563#endif
564 inp->inp_vflag &= ~INP_IPV4;
565 inp->inp_vflag |= INP_IPV6;
566 inp->inp_inc.inc_flags |= INC_ISIPV6;
567 if ((error = prison_remote_ip6(td->td_ucred, &sin6p->sin6_addr)) != 0)
568 goto out;
569 if ((error = tcp6_connect(tp, nam, td)) != 0)
570 goto out;
571#ifdef TCP_OFFLOAD
572 if (registered_toedevs > 0 &&
573 (so->so_options & SO_NO_OFFLOAD) == 0 &&
574 (error = tcp_offload_connect(so, nam)) == 0)
575 goto out;
576#endif
577 tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
578 error = tcp_output(tp);
579
580out:
581 TCPDEBUG2(PRU_CONNECT);
582 INP_WUNLOCK(inp);
583 return (error);
584}
585#endif /* INET6 */
586
587/*
588 * Initiate disconnect from peer.
589 * If connection never passed embryonic stage, just drop;
590 * else if don't need to let data drain, then can just drop anyways,
591 * else have to begin TCP shutdown process: mark socket disconnecting,
592 * drain unread data, state switch to reflect user close, and
593 * send segment (e.g. FIN) to peer. Socket will be really disconnected
594 * when peer sends FIN and acks ours.
595 *
596 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
597 */
598static int
599tcp_usr_disconnect(struct socket *so)
600{
601 struct inpcb *inp;
602 struct tcpcb *tp = NULL;
603 int error = 0;
604
605 TCPDEBUG0;
606 INP_INFO_WLOCK(&V_tcbinfo);
607 inp = sotoinpcb(so);
608 KASSERT(inp != NULL, ("tcp_usr_disconnect: inp == NULL"));
609 INP_WLOCK(inp);
610 if (inp->inp_flags & INP_TIMEWAIT)
611 goto out;
612 if (inp->inp_flags & INP_DROPPED) {
613 error = ECONNRESET;
614 goto out;
615 }
616 tp = intotcpcb(inp);
617 TCPDEBUG1();
618 tcp_disconnect(tp);
619out:
620 TCPDEBUG2(PRU_DISCONNECT);
621 INP_WUNLOCK(inp);
622 INP_INFO_WUNLOCK(&V_tcbinfo);
623 return (error);
624}
625
626#ifdef INET
627/*
628 * Accept a connection. Essentially all the work is done at higher levels;
629 * just return the address of the peer, storing through addr.
630 */
631static int
632tcp_usr_accept(struct socket *so, struct sockaddr **nam)
633{
634 int error = 0;
635 struct inpcb *inp = NULL;
636 struct tcpcb *tp = NULL;
637 struct in_addr addr;
638 in_port_t port = 0;
639 TCPDEBUG0;
640
641 if (so->so_state & SS_ISDISCONNECTED)
642 return (ECONNABORTED);
643
644 inp = sotoinpcb(so);
645 KASSERT(inp != NULL, ("tcp_usr_accept: inp == NULL"));
646 INP_WLOCK(inp);
647 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
648 error = ECONNABORTED;
649 goto out;
650 }
651 tp = intotcpcb(inp);
652 TCPDEBUG1();
653
654 /*
655 * We inline in_getpeeraddr and COMMON_END here, so that we can
656 * copy the data of interest and defer the malloc until after we
657 * release the lock.
658 */
659 port = inp->inp_fport;
660 addr = inp->inp_faddr;
661
662out:
663 TCPDEBUG2(PRU_ACCEPT);
664 INP_WUNLOCK(inp);
665 if (error == 0)
666 *nam = in_sockaddr(port, &addr);
667 return error;
668}
669#endif /* INET */
670
671#ifdef INET6
672static int
673tcp6_usr_accept(struct socket *so, struct sockaddr **nam)
674{
675 struct inpcb *inp = NULL;
676 int error = 0;
677 struct tcpcb *tp = NULL;
678 struct in_addr addr;
679 struct in6_addr addr6;
680 in_port_t port = 0;
681 int v4 = 0;
682 TCPDEBUG0;
683
684 if (so->so_state & SS_ISDISCONNECTED)
685 return (ECONNABORTED);
686
687 inp = sotoinpcb(so);
688 KASSERT(inp != NULL, ("tcp6_usr_accept: inp == NULL"));
689 INP_INFO_RLOCK(&V_tcbinfo);
690 INP_WLOCK(inp);
691 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
692 error = ECONNABORTED;
693 goto out;
694 }
695 tp = intotcpcb(inp);
696 TCPDEBUG1();
697
698 /*
699 * We inline in6_mapped_peeraddr and COMMON_END here, so that we can
700 * copy the data of interest and defer the malloc until after we
701 * release the lock.
702 */
703 if (inp->inp_vflag & INP_IPV4) {
704 v4 = 1;
705 port = inp->inp_fport;
706 addr = inp->inp_faddr;
707 } else {
708 port = inp->inp_fport;
709 addr6 = inp->in6p_faddr;
710 }
711
712out:
713 TCPDEBUG2(PRU_ACCEPT);
714 INP_WUNLOCK(inp);
715 INP_INFO_RUNLOCK(&V_tcbinfo);
716 if (error == 0) {
717 if (v4)
718 *nam = in6_v4mapsin6_sockaddr(port, &addr);
719 else
720 *nam = in6_sockaddr(port, &addr6);
721 }
722 return error;
723}
724#endif /* INET6 */
725
726/*
727 * Mark the connection as being incapable of further output.
728 */
729static int
730tcp_usr_shutdown(struct socket *so)
731{
732 int error = 0;
733 struct inpcb *inp;
734 struct tcpcb *tp = NULL;
735
736 TCPDEBUG0;
737 INP_INFO_WLOCK(&V_tcbinfo);
738 inp = sotoinpcb(so);
739 KASSERT(inp != NULL, ("inp == NULL"));
740 INP_WLOCK(inp);
741 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
742 error = ECONNRESET;
743 goto out;
744 }
745 tp = intotcpcb(inp);
746 TCPDEBUG1();
747 socantsendmore(so);
748 tcp_usrclosed(tp);
749 if (!(inp->inp_flags & INP_DROPPED))
750 error = tcp_output(tp);
751
752out:
753 TCPDEBUG2(PRU_SHUTDOWN);
754 INP_WUNLOCK(inp);
755 INP_INFO_WUNLOCK(&V_tcbinfo);
756
757 return (error);
758}
759
760/*
761 * After a receive, possibly send window update to peer.
762 */
763static int
764tcp_usr_rcvd(struct socket *so, int flags)
765{
766 struct inpcb *inp;
767 struct tcpcb *tp = NULL;
768 int error = 0;
769
770 TCPDEBUG0;
771 inp = sotoinpcb(so);
772 KASSERT(inp != NULL, ("tcp_usr_rcvd: inp == NULL"));
773 INP_WLOCK(inp);
774 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
775 error = ECONNRESET;
776 goto out;
777 }
778 tp = intotcpcb(inp);
779 TCPDEBUG1();
780#ifdef TCP_OFFLOAD
781 if (tp->t_flags & TF_TOE)
782 tcp_offload_rcvd(tp);
783 else
784#endif
785 tcp_output(tp);
786
787out:
788 TCPDEBUG2(PRU_RCVD);
789 INP_WUNLOCK(inp);
790 return (error);
791}
792
793/*
794 * Do a send by putting data in output queue and updating urgent
795 * marker if URG set. Possibly send more data. Unlike the other
796 * pru_*() routines, the mbuf chains are our responsibility. We
797 * must either enqueue them or free them. The other pru_* routines
798 * generally are caller-frees.
799 */
800static int
801tcp_usr_send(struct socket *so, int flags, struct mbuf *m,
802 struct sockaddr *nam, struct mbuf *control, struct thread *td)
803{
804 int error = 0;
805 struct inpcb *inp;
806 struct tcpcb *tp = NULL;
807#ifdef INET6
808 int isipv6;
809#endif
810 TCPDEBUG0;
811
812 /*
813 * We require the pcbinfo lock if we will close the socket as part of
814 * this call.
815 */
816 if (flags & PRUS_EOF)
817 INP_INFO_WLOCK(&V_tcbinfo);
818 inp = sotoinpcb(so);
819 KASSERT(inp != NULL, ("tcp_usr_send: inp == NULL"));
820 INP_WLOCK(inp);
821 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
822 if (control)
823 m_freem(control);
824 if (m)
825 m_freem(m);
826 error = ECONNRESET;
827 goto out;
828 }
829#ifdef INET6
830 isipv6 = nam && nam->sa_family == AF_INET6;
831#endif /* INET6 */
832 tp = intotcpcb(inp);
833 TCPDEBUG1();
834 if (control) {
835 /* TCP doesn't do control messages (rights, creds, etc) */
836 if (control->m_len) {
837 m_freem(control);
838 if (m)
839 m_freem(m);
840 error = EINVAL;
841 goto out;
842 }
843 m_freem(control); /* empty control, just free it */
844 }
845 if (!(flags & PRUS_OOB)) {
|
905 SOCKBUF_UNLOCK(&so->so_snd);
906 if (nam && tp->t_state < TCPS_SYN_SENT) {
907 /*
908 * Do implied connect if not yet connected,
909 * initialize window to default value, and
910 * initialize maxseg/maxopd using peer's cached
911 * MSS.
912 */
913#ifdef INET6
914 if (isipv6)
915 error = tcp6_connect(tp, nam, td);
916#endif /* INET6 */
917#if defined(INET6) && defined(INET)
918 else
919#endif
920#ifdef INET
921 error = tcp_connect(tp, nam, td);
922#endif
923 if (error)
924 goto out;
925 tp->snd_wnd = TTCP_CLIENT_SND_WND;
926 tcp_mss(tp, -1);
927 }
928 tp->snd_up = tp->snd_una + sbavail(&so->so_snd);
929 tp->t_flags |= TF_FORCEDATA;
930 error = tcp_output(tp);
931 tp->t_flags &= ~TF_FORCEDATA;
932 }
933out:
934 TCPDEBUG2((flags & PRUS_OOB) ? PRU_SENDOOB :
935 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
936 INP_WUNLOCK(inp);
937 if (flags & PRUS_EOF)
938 INP_INFO_WUNLOCK(&V_tcbinfo);
939 return (error);
940}
941
942/*
943 * Abort the TCP. Drop the connection abruptly.
944 */
945static void
946tcp_usr_abort(struct socket *so)
947{
948 struct inpcb *inp;
949 struct tcpcb *tp = NULL;
950 TCPDEBUG0;
951
952 inp = sotoinpcb(so);
953 KASSERT(inp != NULL, ("tcp_usr_abort: inp == NULL"));
954
955 INP_INFO_WLOCK(&V_tcbinfo);
956 INP_WLOCK(inp);
957 KASSERT(inp->inp_socket != NULL,
958 ("tcp_usr_abort: inp_socket == NULL"));
959
960 /*
961 * If we still have full TCP state, and we're not dropped, drop.
962 */
963 if (!(inp->inp_flags & INP_TIMEWAIT) &&
964 !(inp->inp_flags & INP_DROPPED)) {
965 tp = intotcpcb(inp);
966 TCPDEBUG1();
967 tcp_drop(tp, ECONNABORTED);
968 TCPDEBUG2(PRU_ABORT);
969 }
970 if (!(inp->inp_flags & INP_DROPPED)) {
971 SOCK_LOCK(so);
972 so->so_state |= SS_PROTOREF;
973 SOCK_UNLOCK(so);
974 inp->inp_flags |= INP_SOCKREF;
975 }
976 INP_WUNLOCK(inp);
977 INP_INFO_WUNLOCK(&V_tcbinfo);
978}
979
980/*
981 * TCP socket is closed. Start friendly disconnect.
982 */
983static void
984tcp_usr_close(struct socket *so)
985{
986 struct inpcb *inp;
987 struct tcpcb *tp = NULL;
988 TCPDEBUG0;
989
990 inp = sotoinpcb(so);
991 KASSERT(inp != NULL, ("tcp_usr_close: inp == NULL"));
992
993 INP_INFO_WLOCK(&V_tcbinfo);
994 INP_WLOCK(inp);
995 KASSERT(inp->inp_socket != NULL,
996 ("tcp_usr_close: inp_socket == NULL"));
997
998 /*
999 * If we still have full TCP state, and we're not dropped, initiate
1000 * a disconnect.
1001 */
1002 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1003 !(inp->inp_flags & INP_DROPPED)) {
1004 tp = intotcpcb(inp);
1005 TCPDEBUG1();
1006 tcp_disconnect(tp);
1007 TCPDEBUG2(PRU_CLOSE);
1008 }
1009 if (!(inp->inp_flags & INP_DROPPED)) {
1010 SOCK_LOCK(so);
1011 so->so_state |= SS_PROTOREF;
1012 SOCK_UNLOCK(so);
1013 inp->inp_flags |= INP_SOCKREF;
1014 }
1015 INP_WUNLOCK(inp);
1016 INP_INFO_WUNLOCK(&V_tcbinfo);
1017}
1018
1019/*
1020 * Receive out-of-band data.
1021 */
1022static int
1023tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags)
1024{
1025 int error = 0;
1026 struct inpcb *inp;
1027 struct tcpcb *tp = NULL;
1028
1029 TCPDEBUG0;
1030 inp = sotoinpcb(so);
1031 KASSERT(inp != NULL, ("tcp_usr_rcvoob: inp == NULL"));
1032 INP_WLOCK(inp);
1033 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1034 error = ECONNRESET;
1035 goto out;
1036 }
1037 tp = intotcpcb(inp);
1038 TCPDEBUG1();
1039 if ((so->so_oobmark == 0 &&
1040 (so->so_rcv.sb_state & SBS_RCVATMARK) == 0) ||
1041 so->so_options & SO_OOBINLINE ||
1042 tp->t_oobflags & TCPOOB_HADDATA) {
1043 error = EINVAL;
1044 goto out;
1045 }
1046 if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
1047 error = EWOULDBLOCK;
1048 goto out;
1049 }
1050 m->m_len = 1;
1051 *mtod(m, caddr_t) = tp->t_iobc;
1052 if ((flags & MSG_PEEK) == 0)
1053 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
1054
1055out:
1056 TCPDEBUG2(PRU_RCVOOB);
1057 INP_WUNLOCK(inp);
1058 return (error);
1059}
1060
1061#ifdef INET
1062struct pr_usrreqs tcp_usrreqs = {
1063 .pru_abort = tcp_usr_abort,
1064 .pru_accept = tcp_usr_accept,
1065 .pru_attach = tcp_usr_attach,
1066 .pru_bind = tcp_usr_bind,
1067 .pru_connect = tcp_usr_connect,
1068 .pru_control = in_control,
1069 .pru_detach = tcp_usr_detach,
1070 .pru_disconnect = tcp_usr_disconnect,
1071 .pru_listen = tcp_usr_listen,
1072 .pru_peeraddr = in_getpeeraddr,
1073 .pru_rcvd = tcp_usr_rcvd,
1074 .pru_rcvoob = tcp_usr_rcvoob,
1075 .pru_send = tcp_usr_send,
1076 .pru_shutdown = tcp_usr_shutdown,
1077 .pru_sockaddr = in_getsockaddr,
1078 .pru_sosetlabel = in_pcbsosetlabel,
1079 .pru_close = tcp_usr_close,
1080};
1081#endif /* INET */
1082
1083#ifdef INET6
1084struct pr_usrreqs tcp6_usrreqs = {
1085 .pru_abort = tcp_usr_abort,
1086 .pru_accept = tcp6_usr_accept,
1087 .pru_attach = tcp_usr_attach,
1088 .pru_bind = tcp6_usr_bind,
1089 .pru_connect = tcp6_usr_connect,
1090 .pru_control = in6_control,
1091 .pru_detach = tcp_usr_detach,
1092 .pru_disconnect = tcp_usr_disconnect,
1093 .pru_listen = tcp6_usr_listen,
1094 .pru_peeraddr = in6_mapped_peeraddr,
1095 .pru_rcvd = tcp_usr_rcvd,
1096 .pru_rcvoob = tcp_usr_rcvoob,
1097 .pru_send = tcp_usr_send,
1098 .pru_shutdown = tcp_usr_shutdown,
1099 .pru_sockaddr = in6_mapped_sockaddr,
1100 .pru_sosetlabel = in_pcbsosetlabel,
1101 .pru_close = tcp_usr_close,
1102};
1103#endif /* INET6 */
1104
1105#ifdef INET
1106/*
1107 * Common subroutine to open a TCP connection to remote host specified
1108 * by struct sockaddr_in in mbuf *nam. Call in_pcbbind to assign a local
1109 * port number if needed. Call in_pcbconnect_setup to do the routing and
1110 * to choose a local host address (interface). If there is an existing
1111 * incarnation of the same connection in TIME-WAIT state and if the remote
1112 * host was sending CC options and if the connection duration was < MSL, then
1113 * truncate the previous TIME-WAIT state and proceed.
1114 * Initialize connection parameters and enter SYN-SENT state.
1115 */
1116static int
1117tcp_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
1118{
1119 struct inpcb *inp = tp->t_inpcb, *oinp;
1120 struct socket *so = inp->inp_socket;
1121 struct in_addr laddr;
1122 u_short lport;
1123 int error;
1124
1125 INP_WLOCK_ASSERT(inp);
1126 INP_HASH_WLOCK(&V_tcbinfo);
1127
1128 if (inp->inp_lport == 0) {
1129 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1130 if (error)
1131 goto out;
1132 }
1133
1134 /*
1135 * Cannot simply call in_pcbconnect, because there might be an
1136 * earlier incarnation of this same connection still in
1137 * TIME_WAIT state, creating an ADDRINUSE error.
1138 */
1139 laddr = inp->inp_laddr;
1140 lport = inp->inp_lport;
1141 error = in_pcbconnect_setup(inp, nam, &laddr.s_addr, &lport,
1142 &inp->inp_faddr.s_addr, &inp->inp_fport, &oinp, td->td_ucred);
1143 if (error && oinp == NULL)
1144 goto out;
1145 if (oinp) {
1146 error = EADDRINUSE;
1147 goto out;
1148 }
1149 inp->inp_laddr = laddr;
1150 in_pcbrehash(inp);
1151 INP_HASH_WUNLOCK(&V_tcbinfo);
1152
1153 /*
1154 * Compute window scaling to request:
1155 * Scale to fit into sweet spot. See tcp_syncache.c.
1156 * XXX: This should move to tcp_output().
1157 */
1158 while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1159 (TCP_MAXWIN << tp->request_r_scale) < sb_max)
1160 tp->request_r_scale++;
1161
1162 soisconnecting(so);
1163 TCPSTAT_INC(tcps_connattempt);
1164 tcp_state_change(tp, TCPS_SYN_SENT);
1165 tp->iss = tcp_new_isn(tp);
1166 tcp_sendseqinit(tp);
1167
1168 return 0;
1169
1170out:
1171 INP_HASH_WUNLOCK(&V_tcbinfo);
1172 return (error);
1173}
1174#endif /* INET */
1175
1176#ifdef INET6
1177static int
1178tcp6_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
1179{
1180 struct inpcb *inp = tp->t_inpcb;
1181 int error;
1182
1183 INP_WLOCK_ASSERT(inp);
1184 INP_HASH_WLOCK(&V_tcbinfo);
1185
1186 if (inp->inp_lport == 0) {
1187 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1188 if (error)
1189 goto out;
1190 }
1191 error = in6_pcbconnect(inp, nam, td->td_ucred);
1192 if (error != 0)
1193 goto out;
1194 INP_HASH_WUNLOCK(&V_tcbinfo);
1195
1196 /* Compute window scaling to request. */
1197 while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1198 (TCP_MAXWIN << tp->request_r_scale) < sb_max)
1199 tp->request_r_scale++;
1200
1201 soisconnecting(inp->inp_socket);
1202 TCPSTAT_INC(tcps_connattempt);
1203 tcp_state_change(tp, TCPS_SYN_SENT);
1204 tp->iss = tcp_new_isn(tp);
1205 tcp_sendseqinit(tp);
1206
1207 return 0;
1208
1209out:
1210 INP_HASH_WUNLOCK(&V_tcbinfo);
1211 return error;
1212}
1213#endif /* INET6 */
1214
1215/*
1216 * Export TCP internal state information via a struct tcp_info, based on the
1217 * Linux 2.6 API. Not ABI compatible as our constants are mapped differently
1218 * (TCP state machine, etc). We export all information using FreeBSD-native
1219 * constants -- for example, the numeric values for tcpi_state will differ
1220 * from Linux.
1221 */
1222static void
1223tcp_fill_info(struct tcpcb *tp, struct tcp_info *ti)
1224{
1225
1226 INP_WLOCK_ASSERT(tp->t_inpcb);
1227 bzero(ti, sizeof(*ti));
1228
1229 ti->tcpi_state = tp->t_state;
1230 if ((tp->t_flags & TF_REQ_TSTMP) && (tp->t_flags & TF_RCVD_TSTMP))
1231 ti->tcpi_options |= TCPI_OPT_TIMESTAMPS;
1232 if (tp->t_flags & TF_SACK_PERMIT)
1233 ti->tcpi_options |= TCPI_OPT_SACK;
1234 if ((tp->t_flags & TF_REQ_SCALE) && (tp->t_flags & TF_RCVD_SCALE)) {
1235 ti->tcpi_options |= TCPI_OPT_WSCALE;
1236 ti->tcpi_snd_wscale = tp->snd_scale;
1237 ti->tcpi_rcv_wscale = tp->rcv_scale;
1238 }
1239
1240 ti->tcpi_rto = tp->t_rxtcur * tick;
1241 ti->tcpi_last_data_recv = (long)(ticks - (int)tp->t_rcvtime) * tick;
1242 ti->tcpi_rtt = ((u_int64_t)tp->t_srtt * tick) >> TCP_RTT_SHIFT;
1243 ti->tcpi_rttvar = ((u_int64_t)tp->t_rttvar * tick) >> TCP_RTTVAR_SHIFT;
1244
1245 ti->tcpi_snd_ssthresh = tp->snd_ssthresh;
1246 ti->tcpi_snd_cwnd = tp->snd_cwnd;
1247
1248 /*
1249 * FreeBSD-specific extension fields for tcp_info.
1250 */
1251 ti->tcpi_rcv_space = tp->rcv_wnd;
1252 ti->tcpi_rcv_nxt = tp->rcv_nxt;
1253 ti->tcpi_snd_wnd = tp->snd_wnd;
1254 ti->tcpi_snd_bwnd = 0; /* Unused, kept for compat. */
1255 ti->tcpi_snd_nxt = tp->snd_nxt;
1256 ti->tcpi_snd_mss = tp->t_maxseg;
1257 ti->tcpi_rcv_mss = tp->t_maxseg;
1258 if (tp->t_flags & TF_TOE)
1259 ti->tcpi_options |= TCPI_OPT_TOE;
1260 ti->tcpi_snd_rexmitpack = tp->t_sndrexmitpack;
1261 ti->tcpi_rcv_ooopack = tp->t_rcvoopack;
1262 ti->tcpi_snd_zerowin = tp->t_sndzerowin;
1263}
1264
1265/*
1266 * tcp_ctloutput() must drop the inpcb lock before performing copyin on
1267 * socket option arguments. When it re-acquires the lock after the copy, it
1268 * has to revalidate that the connection is still valid for the socket
1269 * option.
1270 */
1271#define INP_WLOCK_RECHECK(inp) do { \
1272 INP_WLOCK(inp); \
1273 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { \
1274 INP_WUNLOCK(inp); \
1275 return (ECONNRESET); \
1276 } \
1277 tp = intotcpcb(inp); \
1278} while(0)
1279
1280int
1281tcp_ctloutput(struct socket *so, struct sockopt *sopt)
1282{
1283 int error, opt, optval;
1284 u_int ui;
1285 struct inpcb *inp;
1286 struct tcpcb *tp;
1287 struct tcp_info ti;
1288 char buf[TCP_CA_NAME_MAX];
1289 struct cc_algo *algo;
1290
1291 error = 0;
1292 inp = sotoinpcb(so);
1293 KASSERT(inp != NULL, ("tcp_ctloutput: inp == NULL"));
1294 INP_WLOCK(inp);
1295 if (sopt->sopt_level != IPPROTO_TCP) {
1296#ifdef INET6
1297 if (inp->inp_vflag & INP_IPV6PROTO) {
1298 INP_WUNLOCK(inp);
1299 error = ip6_ctloutput(so, sopt);
1300 }
1301#endif /* INET6 */
1302#if defined(INET6) && defined(INET)
1303 else
1304#endif
1305#ifdef INET
1306 {
1307 INP_WUNLOCK(inp);
1308 error = ip_ctloutput(so, sopt);
1309 }
1310#endif
1311 return (error);
1312 }
1313 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1314 INP_WUNLOCK(inp);
1315 return (ECONNRESET);
1316 }
1317
1318 switch (sopt->sopt_dir) {
1319 case SOPT_SET:
1320 switch (sopt->sopt_name) {
1321#ifdef TCP_SIGNATURE
1322 case TCP_MD5SIG:
1323 INP_WUNLOCK(inp);
1324 error = sooptcopyin(sopt, &optval, sizeof optval,
1325 sizeof optval);
1326 if (error)
1327 return (error);
1328
1329 INP_WLOCK_RECHECK(inp);
1330 if (optval > 0)
1331 tp->t_flags |= TF_SIGNATURE;
1332 else
1333 tp->t_flags &= ~TF_SIGNATURE;
1334 goto unlock_and_done;
1335#endif /* TCP_SIGNATURE */
1336
1337 case TCP_NODELAY:
1338 case TCP_NOOPT:
1339 INP_WUNLOCK(inp);
1340 error = sooptcopyin(sopt, &optval, sizeof optval,
1341 sizeof optval);
1342 if (error)
1343 return (error);
1344
1345 INP_WLOCK_RECHECK(inp);
1346 switch (sopt->sopt_name) {
1347 case TCP_NODELAY:
1348 opt = TF_NODELAY;
1349 break;
1350 case TCP_NOOPT:
1351 opt = TF_NOOPT;
1352 break;
1353 default:
1354 opt = 0; /* dead code to fool gcc */
1355 break;
1356 }
1357
1358 if (optval)
1359 tp->t_flags |= opt;
1360 else
1361 tp->t_flags &= ~opt;
1362unlock_and_done:
1363#ifdef TCP_OFFLOAD
1364 if (tp->t_flags & TF_TOE) {
1365 tcp_offload_ctloutput(tp, sopt->sopt_dir,
1366 sopt->sopt_name);
1367 }
1368#endif
1369 INP_WUNLOCK(inp);
1370 break;
1371
1372 case TCP_NOPUSH:
1373 INP_WUNLOCK(inp);
1374 error = sooptcopyin(sopt, &optval, sizeof optval,
1375 sizeof optval);
1376 if (error)
1377 return (error);
1378
1379 INP_WLOCK_RECHECK(inp);
1380 if (optval)
1381 tp->t_flags |= TF_NOPUSH;
1382 else if (tp->t_flags & TF_NOPUSH) {
1383 tp->t_flags &= ~TF_NOPUSH;
1384 if (TCPS_HAVEESTABLISHED(tp->t_state))
1385 error = tcp_output(tp);
1386 }
1387 goto unlock_and_done;
1388
1389 case TCP_MAXSEG:
1390 INP_WUNLOCK(inp);
1391 error = sooptcopyin(sopt, &optval, sizeof optval,
1392 sizeof optval);
1393 if (error)
1394 return (error);
1395
1396 INP_WLOCK_RECHECK(inp);
1397 if (optval > 0 && optval <= tp->t_maxseg &&
1398 optval + 40 >= V_tcp_minmss)
1399 tp->t_maxseg = optval;
1400 else
1401 error = EINVAL;
1402 goto unlock_and_done;
1403
1404 case TCP_INFO:
1405 INP_WUNLOCK(inp);
1406 error = EINVAL;
1407 break;
1408
1409 case TCP_CONGESTION:
1410 INP_WUNLOCK(inp);
1411 bzero(buf, sizeof(buf));
1412 error = sooptcopyin(sopt, &buf, sizeof(buf), 1);
1413 if (error)
1414 break;
1415 INP_WLOCK_RECHECK(inp);
1416 /*
1417 * Return EINVAL if we can't find the requested cc algo.
1418 */
1419 error = EINVAL;
1420 CC_LIST_RLOCK();
1421 STAILQ_FOREACH(algo, &cc_list, entries) {
1422 if (strncmp(buf, algo->name, TCP_CA_NAME_MAX)
1423 == 0) {
1424 /* We've found the requested algo. */
1425 error = 0;
1426 /*
1427 * We hold a write lock over the tcb
1428 * so it's safe to do these things
1429 * without ordering concerns.
1430 */
1431 if (CC_ALGO(tp)->cb_destroy != NULL)
1432 CC_ALGO(tp)->cb_destroy(tp->ccv);
1433 CC_ALGO(tp) = algo;
1434 /*
1435 * If something goes pear shaped
1436 * initialising the new algo,
1437 * fall back to newreno (which
1438 * does not require initialisation).
1439 */
1440 if (algo->cb_init != NULL)
1441 if (algo->cb_init(tp->ccv) > 0) {
1442 CC_ALGO(tp) = &newreno_cc_algo;
1443 /*
1444 * The only reason init
1445 * should fail is
1446 * because of malloc.
1447 */
1448 error = ENOMEM;
1449 }
1450 break; /* Break the STAILQ_FOREACH. */
1451 }
1452 }
1453 CC_LIST_RUNLOCK();
1454 goto unlock_and_done;
1455
1456 case TCP_KEEPIDLE:
1457 case TCP_KEEPINTVL:
1458 case TCP_KEEPINIT:
1459 INP_WUNLOCK(inp);
1460 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
1461 if (error)
1462 return (error);
1463
1464 if (ui > (UINT_MAX / hz)) {
1465 error = EINVAL;
1466 break;
1467 }
1468 ui *= hz;
1469
1470 INP_WLOCK_RECHECK(inp);
1471 switch (sopt->sopt_name) {
1472 case TCP_KEEPIDLE:
1473 tp->t_keepidle = ui;
1474 /*
1475 * XXX: better check current remaining
1476 * timeout and "merge" it with new value.
1477 */
1478 if ((tp->t_state > TCPS_LISTEN) &&
1479 (tp->t_state <= TCPS_CLOSING))
1480 tcp_timer_activate(tp, TT_KEEP,
1481 TP_KEEPIDLE(tp));
1482 break;
1483 case TCP_KEEPINTVL:
1484 tp->t_keepintvl = ui;
1485 if ((tp->t_state == TCPS_FIN_WAIT_2) &&
1486 (TP_MAXIDLE(tp) > 0))
1487 tcp_timer_activate(tp, TT_2MSL,
1488 TP_MAXIDLE(tp));
1489 break;
1490 case TCP_KEEPINIT:
1491 tp->t_keepinit = ui;
1492 if (tp->t_state == TCPS_SYN_RECEIVED ||
1493 tp->t_state == TCPS_SYN_SENT)
1494 tcp_timer_activate(tp, TT_KEEP,
1495 TP_KEEPINIT(tp));
1496 break;
1497 }
1498 goto unlock_and_done;
1499
1500 case TCP_KEEPCNT:
1501 INP_WUNLOCK(inp);
1502 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
1503 if (error)
1504 return (error);
1505
1506 INP_WLOCK_RECHECK(inp);
1507 tp->t_keepcnt = ui;
1508 if ((tp->t_state == TCPS_FIN_WAIT_2) &&
1509 (TP_MAXIDLE(tp) > 0))
1510 tcp_timer_activate(tp, TT_2MSL,
1511 TP_MAXIDLE(tp));
1512 goto unlock_and_done;
1513
1514 default:
1515 INP_WUNLOCK(inp);
1516 error = ENOPROTOOPT;
1517 break;
1518 }
1519 break;
1520
1521 case SOPT_GET:
1522 tp = intotcpcb(inp);
1523 switch (sopt->sopt_name) {
1524#ifdef TCP_SIGNATURE
1525 case TCP_MD5SIG:
1526 optval = (tp->t_flags & TF_SIGNATURE) ? 1 : 0;
1527 INP_WUNLOCK(inp);
1528 error = sooptcopyout(sopt, &optval, sizeof optval);
1529 break;
1530#endif
1531
1532 case TCP_NODELAY:
1533 optval = tp->t_flags & TF_NODELAY;
1534 INP_WUNLOCK(inp);
1535 error = sooptcopyout(sopt, &optval, sizeof optval);
1536 break;
1537 case TCP_MAXSEG:
1538 optval = tp->t_maxseg;
1539 INP_WUNLOCK(inp);
1540 error = sooptcopyout(sopt, &optval, sizeof optval);
1541 break;
1542 case TCP_NOOPT:
1543 optval = tp->t_flags & TF_NOOPT;
1544 INP_WUNLOCK(inp);
1545 error = sooptcopyout(sopt, &optval, sizeof optval);
1546 break;
1547 case TCP_NOPUSH:
1548 optval = tp->t_flags & TF_NOPUSH;
1549 INP_WUNLOCK(inp);
1550 error = sooptcopyout(sopt, &optval, sizeof optval);
1551 break;
1552 case TCP_INFO:
1553 tcp_fill_info(tp, &ti);
1554 INP_WUNLOCK(inp);
1555 error = sooptcopyout(sopt, &ti, sizeof ti);
1556 break;
1557 case TCP_CONGESTION:
1558 bzero(buf, sizeof(buf));
1559 strlcpy(buf, CC_ALGO(tp)->name, TCP_CA_NAME_MAX);
1560 INP_WUNLOCK(inp);
1561 error = sooptcopyout(sopt, buf, TCP_CA_NAME_MAX);
1562 break;
1563 case TCP_KEEPIDLE:
1564 case TCP_KEEPINTVL:
1565 case TCP_KEEPINIT:
1566 case TCP_KEEPCNT:
1567 switch (sopt->sopt_name) {
1568 case TCP_KEEPIDLE:
1569 ui = tp->t_keepidle / hz;
1570 break;
1571 case TCP_KEEPINTVL:
1572 ui = tp->t_keepintvl / hz;
1573 break;
1574 case TCP_KEEPINIT:
1575 ui = tp->t_keepinit / hz;
1576 break;
1577 case TCP_KEEPCNT:
1578 ui = tp->t_keepcnt;
1579 break;
1580 }
1581 INP_WUNLOCK(inp);
1582 error = sooptcopyout(sopt, &ui, sizeof(ui));
1583 break;
1584 default:
1585 INP_WUNLOCK(inp);
1586 error = ENOPROTOOPT;
1587 break;
1588 }
1589 break;
1590 }
1591 return (error);
1592}
1593#undef INP_WLOCK_RECHECK
1594
1595/*
1596 * Attach TCP protocol to socket, allocating
1597 * internet protocol control block, tcp control block,
1598 * bufer space, and entering LISTEN state if to accept connections.
1599 */
1600static int
1601tcp_attach(struct socket *so)
1602{
1603 struct tcpcb *tp;
1604 struct inpcb *inp;
1605 int error;
1606
1607 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
1608 error = soreserve(so, V_tcp_sendspace, V_tcp_recvspace);
1609 if (error)
1610 return (error);
1611 }
1612 so->so_rcv.sb_flags |= SB_AUTOSIZE;
1613 so->so_snd.sb_flags |= SB_AUTOSIZE;
1614 INP_INFO_WLOCK(&V_tcbinfo);
1615 error = in_pcballoc(so, &V_tcbinfo);
1616 if (error) {
1617 INP_INFO_WUNLOCK(&V_tcbinfo);
1618 return (error);
1619 }
1620 inp = sotoinpcb(so);
1621#ifdef INET6
1622 if (inp->inp_vflag & INP_IPV6PROTO) {
1623 inp->inp_vflag |= INP_IPV6;
1624 inp->in6p_hops = -1; /* use kernel default */
1625 }
1626 else
1627#endif
1628 inp->inp_vflag |= INP_IPV4;
1629 tp = tcp_newtcpcb(inp);
1630 if (tp == NULL) {
1631 in_pcbdetach(inp);
1632 in_pcbfree(inp);
1633 INP_INFO_WUNLOCK(&V_tcbinfo);
1634 return (ENOBUFS);
1635 }
1636 tp->t_state = TCPS_CLOSED;
1637 INP_WUNLOCK(inp);
1638 INP_INFO_WUNLOCK(&V_tcbinfo);
1639 return (0);
1640}
1641
1642/*
1643 * Initiate (or continue) disconnect.
1644 * If embryonic state, just send reset (once).
1645 * If in ``let data drain'' option and linger null, just drop.
1646 * Otherwise (hard), mark socket disconnecting and drop
1647 * current input data; switch states based on user close, and
1648 * send segment to peer (with FIN).
1649 */
1650static void
1651tcp_disconnect(struct tcpcb *tp)
1652{
1653 struct inpcb *inp = tp->t_inpcb;
1654 struct socket *so = inp->inp_socket;
1655
1656 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1657 INP_WLOCK_ASSERT(inp);
1658
1659 /*
1660 * Neither tcp_close() nor tcp_drop() should return NULL, as the
1661 * socket is still open.
1662 */
1663 if (tp->t_state < TCPS_ESTABLISHED) {
1664 tp = tcp_close(tp);
1665 KASSERT(tp != NULL,
1666 ("tcp_disconnect: tcp_close() returned NULL"));
1667 } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
1668 tp = tcp_drop(tp, 0);
1669 KASSERT(tp != NULL,
1670 ("tcp_disconnect: tcp_drop() returned NULL"));
1671 } else {
1672 soisdisconnecting(so);
1673 sbflush(&so->so_rcv);
1674 tcp_usrclosed(tp);
1675 if (!(inp->inp_flags & INP_DROPPED))
1676 tcp_output(tp);
1677 }
1678}
1679
1680/*
1681 * User issued close, and wish to trail through shutdown states:
1682 * if never received SYN, just forget it. If got a SYN from peer,
1683 * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
1684 * If already got a FIN from peer, then almost done; go to LAST_ACK
1685 * state. In all other cases, have already sent FIN to peer (e.g.
1686 * after PRU_SHUTDOWN), and just have to play tedious game waiting
1687 * for peer to send FIN or not respond to keep-alives, etc.
1688 * We can let the user exit from the close as soon as the FIN is acked.
1689 */
1690static void
1691tcp_usrclosed(struct tcpcb *tp)
1692{
1693
1694 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1695 INP_WLOCK_ASSERT(tp->t_inpcb);
1696
1697 switch (tp->t_state) {
1698 case TCPS_LISTEN:
1699#ifdef TCP_OFFLOAD
1700 tcp_offload_listen_stop(tp);
1701#endif
1702 /* FALLTHROUGH */
1703 case TCPS_CLOSED:
1704 tcp_state_change(tp, TCPS_CLOSED);
1705 tp = tcp_close(tp);
1706 /*
1707 * tcp_close() should never return NULL here as the socket is
1708 * still open.
1709 */
1710 KASSERT(tp != NULL,
1711 ("tcp_usrclosed: tcp_close() returned NULL"));
1712 break;
1713
1714 case TCPS_SYN_SENT:
1715 case TCPS_SYN_RECEIVED:
1716 tp->t_flags |= TF_NEEDFIN;
1717 break;
1718
1719 case TCPS_ESTABLISHED:
1720 tcp_state_change(tp, TCPS_FIN_WAIT_1);
1721 break;
1722
1723 case TCPS_CLOSE_WAIT:
1724 tcp_state_change(tp, TCPS_LAST_ACK);
1725 break;
1726 }
1727 if (tp->t_state >= TCPS_FIN_WAIT_2) {
1728 soisdisconnected(tp->t_inpcb->inp_socket);
1729 /* Prevent the connection hanging in FIN_WAIT_2 forever. */
1730 if (tp->t_state == TCPS_FIN_WAIT_2) {
1731 int timeout;
1732
1733 timeout = (tcp_fast_finwait2_recycle) ?
1734 tcp_finwait2_timeout : TP_MAXIDLE(tp);
1735 tcp_timer_activate(tp, TT_2MSL, timeout);
1736 }
1737 }
1738}
1739
1740#ifdef DDB
1741static void
1742db_print_indent(int indent)
1743{
1744 int i;
1745
1746 for (i = 0; i < indent; i++)
1747 db_printf(" ");
1748}
1749
1750static void
1751db_print_tstate(int t_state)
1752{
1753
1754 switch (t_state) {
1755 case TCPS_CLOSED:
1756 db_printf("TCPS_CLOSED");
1757 return;
1758
1759 case TCPS_LISTEN:
1760 db_printf("TCPS_LISTEN");
1761 return;
1762
1763 case TCPS_SYN_SENT:
1764 db_printf("TCPS_SYN_SENT");
1765 return;
1766
1767 case TCPS_SYN_RECEIVED:
1768 db_printf("TCPS_SYN_RECEIVED");
1769 return;
1770
1771 case TCPS_ESTABLISHED:
1772 db_printf("TCPS_ESTABLISHED");
1773 return;
1774
1775 case TCPS_CLOSE_WAIT:
1776 db_printf("TCPS_CLOSE_WAIT");
1777 return;
1778
1779 case TCPS_FIN_WAIT_1:
1780 db_printf("TCPS_FIN_WAIT_1");
1781 return;
1782
1783 case TCPS_CLOSING:
1784 db_printf("TCPS_CLOSING");
1785 return;
1786
1787 case TCPS_LAST_ACK:
1788 db_printf("TCPS_LAST_ACK");
1789 return;
1790
1791 case TCPS_FIN_WAIT_2:
1792 db_printf("TCPS_FIN_WAIT_2");
1793 return;
1794
1795 case TCPS_TIME_WAIT:
1796 db_printf("TCPS_TIME_WAIT");
1797 return;
1798
1799 default:
1800 db_printf("unknown");
1801 return;
1802 }
1803}
1804
1805static void
1806db_print_tflags(u_int t_flags)
1807{
1808 int comma;
1809
1810 comma = 0;
1811 if (t_flags & TF_ACKNOW) {
1812 db_printf("%sTF_ACKNOW", comma ? ", " : "");
1813 comma = 1;
1814 }
1815 if (t_flags & TF_DELACK) {
1816 db_printf("%sTF_DELACK", comma ? ", " : "");
1817 comma = 1;
1818 }
1819 if (t_flags & TF_NODELAY) {
1820 db_printf("%sTF_NODELAY", comma ? ", " : "");
1821 comma = 1;
1822 }
1823 if (t_flags & TF_NOOPT) {
1824 db_printf("%sTF_NOOPT", comma ? ", " : "");
1825 comma = 1;
1826 }
1827 if (t_flags & TF_SENTFIN) {
1828 db_printf("%sTF_SENTFIN", comma ? ", " : "");
1829 comma = 1;
1830 }
1831 if (t_flags & TF_REQ_SCALE) {
1832 db_printf("%sTF_REQ_SCALE", comma ? ", " : "");
1833 comma = 1;
1834 }
1835 if (t_flags & TF_RCVD_SCALE) {
1836 db_printf("%sTF_RECVD_SCALE", comma ? ", " : "");
1837 comma = 1;
1838 }
1839 if (t_flags & TF_REQ_TSTMP) {
1840 db_printf("%sTF_REQ_TSTMP", comma ? ", " : "");
1841 comma = 1;
1842 }
1843 if (t_flags & TF_RCVD_TSTMP) {
1844 db_printf("%sTF_RCVD_TSTMP", comma ? ", " : "");
1845 comma = 1;
1846 }
1847 if (t_flags & TF_SACK_PERMIT) {
1848 db_printf("%sTF_SACK_PERMIT", comma ? ", " : "");
1849 comma = 1;
1850 }
1851 if (t_flags & TF_NEEDSYN) {
1852 db_printf("%sTF_NEEDSYN", comma ? ", " : "");
1853 comma = 1;
1854 }
1855 if (t_flags & TF_NEEDFIN) {
1856 db_printf("%sTF_NEEDFIN", comma ? ", " : "");
1857 comma = 1;
1858 }
1859 if (t_flags & TF_NOPUSH) {
1860 db_printf("%sTF_NOPUSH", comma ? ", " : "");
1861 comma = 1;
1862 }
1863 if (t_flags & TF_MORETOCOME) {
1864 db_printf("%sTF_MORETOCOME", comma ? ", " : "");
1865 comma = 1;
1866 }
1867 if (t_flags & TF_LQ_OVERFLOW) {
1868 db_printf("%sTF_LQ_OVERFLOW", comma ? ", " : "");
1869 comma = 1;
1870 }
1871 if (t_flags & TF_LASTIDLE) {
1872 db_printf("%sTF_LASTIDLE", comma ? ", " : "");
1873 comma = 1;
1874 }
1875 if (t_flags & TF_RXWIN0SENT) {
1876 db_printf("%sTF_RXWIN0SENT", comma ? ", " : "");
1877 comma = 1;
1878 }
1879 if (t_flags & TF_FASTRECOVERY) {
1880 db_printf("%sTF_FASTRECOVERY", comma ? ", " : "");
1881 comma = 1;
1882 }
1883 if (t_flags & TF_CONGRECOVERY) {
1884 db_printf("%sTF_CONGRECOVERY", comma ? ", " : "");
1885 comma = 1;
1886 }
1887 if (t_flags & TF_WASFRECOVERY) {
1888 db_printf("%sTF_WASFRECOVERY", comma ? ", " : "");
1889 comma = 1;
1890 }
1891 if (t_flags & TF_SIGNATURE) {
1892 db_printf("%sTF_SIGNATURE", comma ? ", " : "");
1893 comma = 1;
1894 }
1895 if (t_flags & TF_FORCEDATA) {
1896 db_printf("%sTF_FORCEDATA", comma ? ", " : "");
1897 comma = 1;
1898 }
1899 if (t_flags & TF_TSO) {
1900 db_printf("%sTF_TSO", comma ? ", " : "");
1901 comma = 1;
1902 }
1903 if (t_flags & TF_ECN_PERMIT) {
1904 db_printf("%sTF_ECN_PERMIT", comma ? ", " : "");
1905 comma = 1;
1906 }
1907}
1908
1909static void
1910db_print_toobflags(char t_oobflags)
1911{
1912 int comma;
1913
1914 comma = 0;
1915 if (t_oobflags & TCPOOB_HAVEDATA) {
1916 db_printf("%sTCPOOB_HAVEDATA", comma ? ", " : "");
1917 comma = 1;
1918 }
1919 if (t_oobflags & TCPOOB_HADDATA) {
1920 db_printf("%sTCPOOB_HADDATA", comma ? ", " : "");
1921 comma = 1;
1922 }
1923}
1924
1925static void
1926db_print_tcpcb(struct tcpcb *tp, const char *name, int indent)
1927{
1928
1929 db_print_indent(indent);
1930 db_printf("%s at %p\n", name, tp);
1931
1932 indent += 2;
1933
1934 db_print_indent(indent);
1935 db_printf("t_segq first: %p t_segqlen: %d t_dupacks: %d\n",
1936 tp->t_segq, tp->t_segqlen, tp->t_dupacks);
1937
1938 db_print_indent(indent);
1939 db_printf("tt_rexmt: %p tt_persist: %p tt_keep: %p\n",
1940 &tp->t_timers->tt_rexmt, &tp->t_timers->tt_persist, &tp->t_timers->tt_keep);
1941
1942 db_print_indent(indent);
1943 db_printf("tt_2msl: %p tt_delack: %p t_inpcb: %p\n", &tp->t_timers->tt_2msl,
1944 &tp->t_timers->tt_delack, tp->t_inpcb);
1945
1946 db_print_indent(indent);
1947 db_printf("t_state: %d (", tp->t_state);
1948 db_print_tstate(tp->t_state);
1949 db_printf(")\n");
1950
1951 db_print_indent(indent);
1952 db_printf("t_flags: 0x%x (", tp->t_flags);
1953 db_print_tflags(tp->t_flags);
1954 db_printf(")\n");
1955
1956 db_print_indent(indent);
1957 db_printf("snd_una: 0x%08x snd_max: 0x%08x snd_nxt: x0%08x\n",
1958 tp->snd_una, tp->snd_max, tp->snd_nxt);
1959
1960 db_print_indent(indent);
1961 db_printf("snd_up: 0x%08x snd_wl1: 0x%08x snd_wl2: 0x%08x\n",
1962 tp->snd_up, tp->snd_wl1, tp->snd_wl2);
1963
1964 db_print_indent(indent);
1965 db_printf("iss: 0x%08x irs: 0x%08x rcv_nxt: 0x%08x\n",
1966 tp->iss, tp->irs, tp->rcv_nxt);
1967
1968 db_print_indent(indent);
1969 db_printf("rcv_adv: 0x%08x rcv_wnd: %lu rcv_up: 0x%08x\n",
1970 tp->rcv_adv, tp->rcv_wnd, tp->rcv_up);
1971
1972 db_print_indent(indent);
1973 db_printf("snd_wnd: %lu snd_cwnd: %lu\n",
1974 tp->snd_wnd, tp->snd_cwnd);
1975
1976 db_print_indent(indent);
1977 db_printf("snd_ssthresh: %lu snd_recover: "
1978 "0x%08x\n", tp->snd_ssthresh, tp->snd_recover);
1979
1980 db_print_indent(indent);
1981 db_printf("t_maxopd: %u t_rcvtime: %u t_startime: %u\n",
1982 tp->t_maxopd, tp->t_rcvtime, tp->t_starttime);
1983
1984 db_print_indent(indent);
1985 db_printf("t_rttime: %u t_rtsq: 0x%08x\n",
1986 tp->t_rtttime, tp->t_rtseq);
1987
1988 db_print_indent(indent);
1989 db_printf("t_rxtcur: %d t_maxseg: %u t_srtt: %d\n",
1990 tp->t_rxtcur, tp->t_maxseg, tp->t_srtt);
1991
1992 db_print_indent(indent);
1993 db_printf("t_rttvar: %d t_rxtshift: %d t_rttmin: %u "
1994 "t_rttbest: %u\n", tp->t_rttvar, tp->t_rxtshift, tp->t_rttmin,
1995 tp->t_rttbest);
1996
1997 db_print_indent(indent);
1998 db_printf("t_rttupdated: %lu max_sndwnd: %lu t_softerror: %d\n",
1999 tp->t_rttupdated, tp->max_sndwnd, tp->t_softerror);
2000
2001 db_print_indent(indent);
2002 db_printf("t_oobflags: 0x%x (", tp->t_oobflags);
2003 db_print_toobflags(tp->t_oobflags);
2004 db_printf(") t_iobc: 0x%02x\n", tp->t_iobc);
2005
2006 db_print_indent(indent);
2007 db_printf("snd_scale: %u rcv_scale: %u request_r_scale: %u\n",
2008 tp->snd_scale, tp->rcv_scale, tp->request_r_scale);
2009
2010 db_print_indent(indent);
2011 db_printf("ts_recent: %u ts_recent_age: %u\n",
2012 tp->ts_recent, tp->ts_recent_age);
2013
2014 db_print_indent(indent);
2015 db_printf("ts_offset: %u last_ack_sent: 0x%08x snd_cwnd_prev: "
2016 "%lu\n", tp->ts_offset, tp->last_ack_sent, tp->snd_cwnd_prev);
2017
2018 db_print_indent(indent);
2019 db_printf("snd_ssthresh_prev: %lu snd_recover_prev: 0x%08x "
2020 "t_badrxtwin: %u\n", tp->snd_ssthresh_prev,
2021 tp->snd_recover_prev, tp->t_badrxtwin);
2022
2023 db_print_indent(indent);
2024 db_printf("snd_numholes: %d snd_holes first: %p\n",
2025 tp->snd_numholes, TAILQ_FIRST(&tp->snd_holes));
2026
2027 db_print_indent(indent);
2028 db_printf("snd_fack: 0x%08x rcv_numsacks: %d sack_newdata: "
2029 "0x%08x\n", tp->snd_fack, tp->rcv_numsacks, tp->sack_newdata);
2030
2031 /* Skip sackblks, sackhint. */
2032
2033 db_print_indent(indent);
2034 db_printf("t_rttlow: %d rfbuf_ts: %u rfbuf_cnt: %d\n",
2035 tp->t_rttlow, tp->rfbuf_ts, tp->rfbuf_cnt);
2036}
2037
2038DB_SHOW_COMMAND(tcpcb, db_show_tcpcb)
2039{
2040 struct tcpcb *tp;
2041
2042 if (!have_addr) {
2043 db_printf("usage: show tcpcb <addr>\n");
2044 return;
2045 }
2046 tp = (struct tcpcb *)addr;
2047
2048 db_print_tcpcb(tp, "tcpcb", 0);
2049}
2050#endif
| 905 SOCKBUF_UNLOCK(&so->so_snd);
906 if (nam && tp->t_state < TCPS_SYN_SENT) {
907 /*
908 * Do implied connect if not yet connected,
909 * initialize window to default value, and
910 * initialize maxseg/maxopd using peer's cached
911 * MSS.
912 */
913#ifdef INET6
914 if (isipv6)
915 error = tcp6_connect(tp, nam, td);
916#endif /* INET6 */
917#if defined(INET6) && defined(INET)
918 else
919#endif
920#ifdef INET
921 error = tcp_connect(tp, nam, td);
922#endif
923 if (error)
924 goto out;
925 tp->snd_wnd = TTCP_CLIENT_SND_WND;
926 tcp_mss(tp, -1);
927 }
928 tp->snd_up = tp->snd_una + sbavail(&so->so_snd);
929 tp->t_flags |= TF_FORCEDATA;
930 error = tcp_output(tp);
931 tp->t_flags &= ~TF_FORCEDATA;
932 }
933out:
934 TCPDEBUG2((flags & PRUS_OOB) ? PRU_SENDOOB :
935 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
936 INP_WUNLOCK(inp);
937 if (flags & PRUS_EOF)
938 INP_INFO_WUNLOCK(&V_tcbinfo);
939 return (error);
940}
941
942/*
943 * Abort the TCP. Drop the connection abruptly.
944 */
945static void
946tcp_usr_abort(struct socket *so)
947{
948 struct inpcb *inp;
949 struct tcpcb *tp = NULL;
950 TCPDEBUG0;
951
952 inp = sotoinpcb(so);
953 KASSERT(inp != NULL, ("tcp_usr_abort: inp == NULL"));
954
955 INP_INFO_WLOCK(&V_tcbinfo);
956 INP_WLOCK(inp);
957 KASSERT(inp->inp_socket != NULL,
958 ("tcp_usr_abort: inp_socket == NULL"));
959
960 /*
961 * If we still have full TCP state, and we're not dropped, drop.
962 */
963 if (!(inp->inp_flags & INP_TIMEWAIT) &&
964 !(inp->inp_flags & INP_DROPPED)) {
965 tp = intotcpcb(inp);
966 TCPDEBUG1();
967 tcp_drop(tp, ECONNABORTED);
968 TCPDEBUG2(PRU_ABORT);
969 }
970 if (!(inp->inp_flags & INP_DROPPED)) {
971 SOCK_LOCK(so);
972 so->so_state |= SS_PROTOREF;
973 SOCK_UNLOCK(so);
974 inp->inp_flags |= INP_SOCKREF;
975 }
976 INP_WUNLOCK(inp);
977 INP_INFO_WUNLOCK(&V_tcbinfo);
978}
979
980/*
981 * TCP socket is closed. Start friendly disconnect.
982 */
983static void
984tcp_usr_close(struct socket *so)
985{
986 struct inpcb *inp;
987 struct tcpcb *tp = NULL;
988 TCPDEBUG0;
989
990 inp = sotoinpcb(so);
991 KASSERT(inp != NULL, ("tcp_usr_close: inp == NULL"));
992
993 INP_INFO_WLOCK(&V_tcbinfo);
994 INP_WLOCK(inp);
995 KASSERT(inp->inp_socket != NULL,
996 ("tcp_usr_close: inp_socket == NULL"));
997
998 /*
999 * If we still have full TCP state, and we're not dropped, initiate
1000 * a disconnect.
1001 */
1002 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1003 !(inp->inp_flags & INP_DROPPED)) {
1004 tp = intotcpcb(inp);
1005 TCPDEBUG1();
1006 tcp_disconnect(tp);
1007 TCPDEBUG2(PRU_CLOSE);
1008 }
1009 if (!(inp->inp_flags & INP_DROPPED)) {
1010 SOCK_LOCK(so);
1011 so->so_state |= SS_PROTOREF;
1012 SOCK_UNLOCK(so);
1013 inp->inp_flags |= INP_SOCKREF;
1014 }
1015 INP_WUNLOCK(inp);
1016 INP_INFO_WUNLOCK(&V_tcbinfo);
1017}
1018
1019/*
1020 * Receive out-of-band data.
1021 */
1022static int
1023tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags)
1024{
1025 int error = 0;
1026 struct inpcb *inp;
1027 struct tcpcb *tp = NULL;
1028
1029 TCPDEBUG0;
1030 inp = sotoinpcb(so);
1031 KASSERT(inp != NULL, ("tcp_usr_rcvoob: inp == NULL"));
1032 INP_WLOCK(inp);
1033 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1034 error = ECONNRESET;
1035 goto out;
1036 }
1037 tp = intotcpcb(inp);
1038 TCPDEBUG1();
1039 if ((so->so_oobmark == 0 &&
1040 (so->so_rcv.sb_state & SBS_RCVATMARK) == 0) ||
1041 so->so_options & SO_OOBINLINE ||
1042 tp->t_oobflags & TCPOOB_HADDATA) {
1043 error = EINVAL;
1044 goto out;
1045 }
1046 if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
1047 error = EWOULDBLOCK;
1048 goto out;
1049 }
1050 m->m_len = 1;
1051 *mtod(m, caddr_t) = tp->t_iobc;
1052 if ((flags & MSG_PEEK) == 0)
1053 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
1054
1055out:
1056 TCPDEBUG2(PRU_RCVOOB);
1057 INP_WUNLOCK(inp);
1058 return (error);
1059}
1060
1061#ifdef INET
1062struct pr_usrreqs tcp_usrreqs = {
1063 .pru_abort = tcp_usr_abort,
1064 .pru_accept = tcp_usr_accept,
1065 .pru_attach = tcp_usr_attach,
1066 .pru_bind = tcp_usr_bind,
1067 .pru_connect = tcp_usr_connect,
1068 .pru_control = in_control,
1069 .pru_detach = tcp_usr_detach,
1070 .pru_disconnect = tcp_usr_disconnect,
1071 .pru_listen = tcp_usr_listen,
1072 .pru_peeraddr = in_getpeeraddr,
1073 .pru_rcvd = tcp_usr_rcvd,
1074 .pru_rcvoob = tcp_usr_rcvoob,
1075 .pru_send = tcp_usr_send,
1076 .pru_shutdown = tcp_usr_shutdown,
1077 .pru_sockaddr = in_getsockaddr,
1078 .pru_sosetlabel = in_pcbsosetlabel,
1079 .pru_close = tcp_usr_close,
1080};
1081#endif /* INET */
1082
1083#ifdef INET6
1084struct pr_usrreqs tcp6_usrreqs = {
1085 .pru_abort = tcp_usr_abort,
1086 .pru_accept = tcp6_usr_accept,
1087 .pru_attach = tcp_usr_attach,
1088 .pru_bind = tcp6_usr_bind,
1089 .pru_connect = tcp6_usr_connect,
1090 .pru_control = in6_control,
1091 .pru_detach = tcp_usr_detach,
1092 .pru_disconnect = tcp_usr_disconnect,
1093 .pru_listen = tcp6_usr_listen,
1094 .pru_peeraddr = in6_mapped_peeraddr,
1095 .pru_rcvd = tcp_usr_rcvd,
1096 .pru_rcvoob = tcp_usr_rcvoob,
1097 .pru_send = tcp_usr_send,
1098 .pru_shutdown = tcp_usr_shutdown,
1099 .pru_sockaddr = in6_mapped_sockaddr,
1100 .pru_sosetlabel = in_pcbsosetlabel,
1101 .pru_close = tcp_usr_close,
1102};
1103#endif /* INET6 */
1104
1105#ifdef INET
1106/*
1107 * Common subroutine to open a TCP connection to remote host specified
1108 * by struct sockaddr_in in mbuf *nam. Call in_pcbbind to assign a local
1109 * port number if needed. Call in_pcbconnect_setup to do the routing and
1110 * to choose a local host address (interface). If there is an existing
1111 * incarnation of the same connection in TIME-WAIT state and if the remote
1112 * host was sending CC options and if the connection duration was < MSL, then
1113 * truncate the previous TIME-WAIT state and proceed.
1114 * Initialize connection parameters and enter SYN-SENT state.
1115 */
1116static int
1117tcp_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
1118{
1119 struct inpcb *inp = tp->t_inpcb, *oinp;
1120 struct socket *so = inp->inp_socket;
1121 struct in_addr laddr;
1122 u_short lport;
1123 int error;
1124
1125 INP_WLOCK_ASSERT(inp);
1126 INP_HASH_WLOCK(&V_tcbinfo);
1127
1128 if (inp->inp_lport == 0) {
1129 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1130 if (error)
1131 goto out;
1132 }
1133
1134 /*
1135 * Cannot simply call in_pcbconnect, because there might be an
1136 * earlier incarnation of this same connection still in
1137 * TIME_WAIT state, creating an ADDRINUSE error.
1138 */
1139 laddr = inp->inp_laddr;
1140 lport = inp->inp_lport;
1141 error = in_pcbconnect_setup(inp, nam, &laddr.s_addr, &lport,
1142 &inp->inp_faddr.s_addr, &inp->inp_fport, &oinp, td->td_ucred);
1143 if (error && oinp == NULL)
1144 goto out;
1145 if (oinp) {
1146 error = EADDRINUSE;
1147 goto out;
1148 }
1149 inp->inp_laddr = laddr;
1150 in_pcbrehash(inp);
1151 INP_HASH_WUNLOCK(&V_tcbinfo);
1152
1153 /*
1154 * Compute window scaling to request:
1155 * Scale to fit into sweet spot. See tcp_syncache.c.
1156 * XXX: This should move to tcp_output().
1157 */
1158 while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1159 (TCP_MAXWIN << tp->request_r_scale) < sb_max)
1160 tp->request_r_scale++;
1161
1162 soisconnecting(so);
1163 TCPSTAT_INC(tcps_connattempt);
1164 tcp_state_change(tp, TCPS_SYN_SENT);
1165 tp->iss = tcp_new_isn(tp);
1166 tcp_sendseqinit(tp);
1167
1168 return 0;
1169
1170out:
1171 INP_HASH_WUNLOCK(&V_tcbinfo);
1172 return (error);
1173}
1174#endif /* INET */
1175
1176#ifdef INET6
1177static int
1178tcp6_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
1179{
1180 struct inpcb *inp = tp->t_inpcb;
1181 int error;
1182
1183 INP_WLOCK_ASSERT(inp);
1184 INP_HASH_WLOCK(&V_tcbinfo);
1185
1186 if (inp->inp_lport == 0) {
1187 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1188 if (error)
1189 goto out;
1190 }
1191 error = in6_pcbconnect(inp, nam, td->td_ucred);
1192 if (error != 0)
1193 goto out;
1194 INP_HASH_WUNLOCK(&V_tcbinfo);
1195
1196 /* Compute window scaling to request. */
1197 while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1198 (TCP_MAXWIN << tp->request_r_scale) < sb_max)
1199 tp->request_r_scale++;
1200
1201 soisconnecting(inp->inp_socket);
1202 TCPSTAT_INC(tcps_connattempt);
1203 tcp_state_change(tp, TCPS_SYN_SENT);
1204 tp->iss = tcp_new_isn(tp);
1205 tcp_sendseqinit(tp);
1206
1207 return 0;
1208
1209out:
1210 INP_HASH_WUNLOCK(&V_tcbinfo);
1211 return error;
1212}
1213#endif /* INET6 */
1214
1215/*
1216 * Export TCP internal state information via a struct tcp_info, based on the
1217 * Linux 2.6 API. Not ABI compatible as our constants are mapped differently
1218 * (TCP state machine, etc). We export all information using FreeBSD-native
1219 * constants -- for example, the numeric values for tcpi_state will differ
1220 * from Linux.
1221 */
1222static void
1223tcp_fill_info(struct tcpcb *tp, struct tcp_info *ti)
1224{
1225
1226 INP_WLOCK_ASSERT(tp->t_inpcb);
1227 bzero(ti, sizeof(*ti));
1228
1229 ti->tcpi_state = tp->t_state;
1230 if ((tp->t_flags & TF_REQ_TSTMP) && (tp->t_flags & TF_RCVD_TSTMP))
1231 ti->tcpi_options |= TCPI_OPT_TIMESTAMPS;
1232 if (tp->t_flags & TF_SACK_PERMIT)
1233 ti->tcpi_options |= TCPI_OPT_SACK;
1234 if ((tp->t_flags & TF_REQ_SCALE) && (tp->t_flags & TF_RCVD_SCALE)) {
1235 ti->tcpi_options |= TCPI_OPT_WSCALE;
1236 ti->tcpi_snd_wscale = tp->snd_scale;
1237 ti->tcpi_rcv_wscale = tp->rcv_scale;
1238 }
1239
1240 ti->tcpi_rto = tp->t_rxtcur * tick;
1241 ti->tcpi_last_data_recv = (long)(ticks - (int)tp->t_rcvtime) * tick;
1242 ti->tcpi_rtt = ((u_int64_t)tp->t_srtt * tick) >> TCP_RTT_SHIFT;
1243 ti->tcpi_rttvar = ((u_int64_t)tp->t_rttvar * tick) >> TCP_RTTVAR_SHIFT;
1244
1245 ti->tcpi_snd_ssthresh = tp->snd_ssthresh;
1246 ti->tcpi_snd_cwnd = tp->snd_cwnd;
1247
1248 /*
1249 * FreeBSD-specific extension fields for tcp_info.
1250 */
1251 ti->tcpi_rcv_space = tp->rcv_wnd;
1252 ti->tcpi_rcv_nxt = tp->rcv_nxt;
1253 ti->tcpi_snd_wnd = tp->snd_wnd;
1254 ti->tcpi_snd_bwnd = 0; /* Unused, kept for compat. */
1255 ti->tcpi_snd_nxt = tp->snd_nxt;
1256 ti->tcpi_snd_mss = tp->t_maxseg;
1257 ti->tcpi_rcv_mss = tp->t_maxseg;
1258 if (tp->t_flags & TF_TOE)
1259 ti->tcpi_options |= TCPI_OPT_TOE;
1260 ti->tcpi_snd_rexmitpack = tp->t_sndrexmitpack;
1261 ti->tcpi_rcv_ooopack = tp->t_rcvoopack;
1262 ti->tcpi_snd_zerowin = tp->t_sndzerowin;
1263}
1264
1265/*
1266 * tcp_ctloutput() must drop the inpcb lock before performing copyin on
1267 * socket option arguments. When it re-acquires the lock after the copy, it
1268 * has to revalidate that the connection is still valid for the socket
1269 * option.
1270 */
1271#define INP_WLOCK_RECHECK(inp) do { \
1272 INP_WLOCK(inp); \
1273 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { \
1274 INP_WUNLOCK(inp); \
1275 return (ECONNRESET); \
1276 } \
1277 tp = intotcpcb(inp); \
1278} while(0)
1279
1280int
1281tcp_ctloutput(struct socket *so, struct sockopt *sopt)
1282{
1283 int error, opt, optval;
1284 u_int ui;
1285 struct inpcb *inp;
1286 struct tcpcb *tp;
1287 struct tcp_info ti;
1288 char buf[TCP_CA_NAME_MAX];
1289 struct cc_algo *algo;
1290
1291 error = 0;
1292 inp = sotoinpcb(so);
1293 KASSERT(inp != NULL, ("tcp_ctloutput: inp == NULL"));
1294 INP_WLOCK(inp);
1295 if (sopt->sopt_level != IPPROTO_TCP) {
1296#ifdef INET6
1297 if (inp->inp_vflag & INP_IPV6PROTO) {
1298 INP_WUNLOCK(inp);
1299 error = ip6_ctloutput(so, sopt);
1300 }
1301#endif /* INET6 */
1302#if defined(INET6) && defined(INET)
1303 else
1304#endif
1305#ifdef INET
1306 {
1307 INP_WUNLOCK(inp);
1308 error = ip_ctloutput(so, sopt);
1309 }
1310#endif
1311 return (error);
1312 }
1313 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1314 INP_WUNLOCK(inp);
1315 return (ECONNRESET);
1316 }
1317
1318 switch (sopt->sopt_dir) {
1319 case SOPT_SET:
1320 switch (sopt->sopt_name) {
1321#ifdef TCP_SIGNATURE
1322 case TCP_MD5SIG:
1323 INP_WUNLOCK(inp);
1324 error = sooptcopyin(sopt, &optval, sizeof optval,
1325 sizeof optval);
1326 if (error)
1327 return (error);
1328
1329 INP_WLOCK_RECHECK(inp);
1330 if (optval > 0)
1331 tp->t_flags |= TF_SIGNATURE;
1332 else
1333 tp->t_flags &= ~TF_SIGNATURE;
1334 goto unlock_and_done;
1335#endif /* TCP_SIGNATURE */
1336
1337 case TCP_NODELAY:
1338 case TCP_NOOPT:
1339 INP_WUNLOCK(inp);
1340 error = sooptcopyin(sopt, &optval, sizeof optval,
1341 sizeof optval);
1342 if (error)
1343 return (error);
1344
1345 INP_WLOCK_RECHECK(inp);
1346 switch (sopt->sopt_name) {
1347 case TCP_NODELAY:
1348 opt = TF_NODELAY;
1349 break;
1350 case TCP_NOOPT:
1351 opt = TF_NOOPT;
1352 break;
1353 default:
1354 opt = 0; /* dead code to fool gcc */
1355 break;
1356 }
1357
1358 if (optval)
1359 tp->t_flags |= opt;
1360 else
1361 tp->t_flags &= ~opt;
1362unlock_and_done:
1363#ifdef TCP_OFFLOAD
1364 if (tp->t_flags & TF_TOE) {
1365 tcp_offload_ctloutput(tp, sopt->sopt_dir,
1366 sopt->sopt_name);
1367 }
1368#endif
1369 INP_WUNLOCK(inp);
1370 break;
1371
1372 case TCP_NOPUSH:
1373 INP_WUNLOCK(inp);
1374 error = sooptcopyin(sopt, &optval, sizeof optval,
1375 sizeof optval);
1376 if (error)
1377 return (error);
1378
1379 INP_WLOCK_RECHECK(inp);
1380 if (optval)
1381 tp->t_flags |= TF_NOPUSH;
1382 else if (tp->t_flags & TF_NOPUSH) {
1383 tp->t_flags &= ~TF_NOPUSH;
1384 if (TCPS_HAVEESTABLISHED(tp->t_state))
1385 error = tcp_output(tp);
1386 }
1387 goto unlock_and_done;
1388
1389 case TCP_MAXSEG:
1390 INP_WUNLOCK(inp);
1391 error = sooptcopyin(sopt, &optval, sizeof optval,
1392 sizeof optval);
1393 if (error)
1394 return (error);
1395
1396 INP_WLOCK_RECHECK(inp);
1397 if (optval > 0 && optval <= tp->t_maxseg &&
1398 optval + 40 >= V_tcp_minmss)
1399 tp->t_maxseg = optval;
1400 else
1401 error = EINVAL;
1402 goto unlock_and_done;
1403
1404 case TCP_INFO:
1405 INP_WUNLOCK(inp);
1406 error = EINVAL;
1407 break;
1408
1409 case TCP_CONGESTION:
1410 INP_WUNLOCK(inp);
1411 bzero(buf, sizeof(buf));
1412 error = sooptcopyin(sopt, &buf, sizeof(buf), 1);
1413 if (error)
1414 break;
1415 INP_WLOCK_RECHECK(inp);
1416 /*
1417 * Return EINVAL if we can't find the requested cc algo.
1418 */
1419 error = EINVAL;
1420 CC_LIST_RLOCK();
1421 STAILQ_FOREACH(algo, &cc_list, entries) {
1422 if (strncmp(buf, algo->name, TCP_CA_NAME_MAX)
1423 == 0) {
1424 /* We've found the requested algo. */
1425 error = 0;
1426 /*
1427 * We hold a write lock over the tcb
1428 * so it's safe to do these things
1429 * without ordering concerns.
1430 */
1431 if (CC_ALGO(tp)->cb_destroy != NULL)
1432 CC_ALGO(tp)->cb_destroy(tp->ccv);
1433 CC_ALGO(tp) = algo;
1434 /*
1435 * If something goes pear shaped
1436 * initialising the new algo,
1437 * fall back to newreno (which
1438 * does not require initialisation).
1439 */
1440 if (algo->cb_init != NULL)
1441 if (algo->cb_init(tp->ccv) > 0) {
1442 CC_ALGO(tp) = &newreno_cc_algo;
1443 /*
1444 * The only reason init
1445 * should fail is
1446 * because of malloc.
1447 */
1448 error = ENOMEM;
1449 }
1450 break; /* Break the STAILQ_FOREACH. */
1451 }
1452 }
1453 CC_LIST_RUNLOCK();
1454 goto unlock_and_done;
1455
1456 case TCP_KEEPIDLE:
1457 case TCP_KEEPINTVL:
1458 case TCP_KEEPINIT:
1459 INP_WUNLOCK(inp);
1460 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
1461 if (error)
1462 return (error);
1463
1464 if (ui > (UINT_MAX / hz)) {
1465 error = EINVAL;
1466 break;
1467 }
1468 ui *= hz;
1469
1470 INP_WLOCK_RECHECK(inp);
1471 switch (sopt->sopt_name) {
1472 case TCP_KEEPIDLE:
1473 tp->t_keepidle = ui;
1474 /*
1475 * XXX: better check current remaining
1476 * timeout and "merge" it with new value.
1477 */
1478 if ((tp->t_state > TCPS_LISTEN) &&
1479 (tp->t_state <= TCPS_CLOSING))
1480 tcp_timer_activate(tp, TT_KEEP,
1481 TP_KEEPIDLE(tp));
1482 break;
1483 case TCP_KEEPINTVL:
1484 tp->t_keepintvl = ui;
1485 if ((tp->t_state == TCPS_FIN_WAIT_2) &&
1486 (TP_MAXIDLE(tp) > 0))
1487 tcp_timer_activate(tp, TT_2MSL,
1488 TP_MAXIDLE(tp));
1489 break;
1490 case TCP_KEEPINIT:
1491 tp->t_keepinit = ui;
1492 if (tp->t_state == TCPS_SYN_RECEIVED ||
1493 tp->t_state == TCPS_SYN_SENT)
1494 tcp_timer_activate(tp, TT_KEEP,
1495 TP_KEEPINIT(tp));
1496 break;
1497 }
1498 goto unlock_and_done;
1499
1500 case TCP_KEEPCNT:
1501 INP_WUNLOCK(inp);
1502 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
1503 if (error)
1504 return (error);
1505
1506 INP_WLOCK_RECHECK(inp);
1507 tp->t_keepcnt = ui;
1508 if ((tp->t_state == TCPS_FIN_WAIT_2) &&
1509 (TP_MAXIDLE(tp) > 0))
1510 tcp_timer_activate(tp, TT_2MSL,
1511 TP_MAXIDLE(tp));
1512 goto unlock_and_done;
1513
1514 default:
1515 INP_WUNLOCK(inp);
1516 error = ENOPROTOOPT;
1517 break;
1518 }
1519 break;
1520
1521 case SOPT_GET:
1522 tp = intotcpcb(inp);
1523 switch (sopt->sopt_name) {
1524#ifdef TCP_SIGNATURE
1525 case TCP_MD5SIG:
1526 optval = (tp->t_flags & TF_SIGNATURE) ? 1 : 0;
1527 INP_WUNLOCK(inp);
1528 error = sooptcopyout(sopt, &optval, sizeof optval);
1529 break;
1530#endif
1531
1532 case TCP_NODELAY:
1533 optval = tp->t_flags & TF_NODELAY;
1534 INP_WUNLOCK(inp);
1535 error = sooptcopyout(sopt, &optval, sizeof optval);
1536 break;
1537 case TCP_MAXSEG:
1538 optval = tp->t_maxseg;
1539 INP_WUNLOCK(inp);
1540 error = sooptcopyout(sopt, &optval, sizeof optval);
1541 break;
1542 case TCP_NOOPT:
1543 optval = tp->t_flags & TF_NOOPT;
1544 INP_WUNLOCK(inp);
1545 error = sooptcopyout(sopt, &optval, sizeof optval);
1546 break;
1547 case TCP_NOPUSH:
1548 optval = tp->t_flags & TF_NOPUSH;
1549 INP_WUNLOCK(inp);
1550 error = sooptcopyout(sopt, &optval, sizeof optval);
1551 break;
1552 case TCP_INFO:
1553 tcp_fill_info(tp, &ti);
1554 INP_WUNLOCK(inp);
1555 error = sooptcopyout(sopt, &ti, sizeof ti);
1556 break;
1557 case TCP_CONGESTION:
1558 bzero(buf, sizeof(buf));
1559 strlcpy(buf, CC_ALGO(tp)->name, TCP_CA_NAME_MAX);
1560 INP_WUNLOCK(inp);
1561 error = sooptcopyout(sopt, buf, TCP_CA_NAME_MAX);
1562 break;
1563 case TCP_KEEPIDLE:
1564 case TCP_KEEPINTVL:
1565 case TCP_KEEPINIT:
1566 case TCP_KEEPCNT:
1567 switch (sopt->sopt_name) {
1568 case TCP_KEEPIDLE:
1569 ui = tp->t_keepidle / hz;
1570 break;
1571 case TCP_KEEPINTVL:
1572 ui = tp->t_keepintvl / hz;
1573 break;
1574 case TCP_KEEPINIT:
1575 ui = tp->t_keepinit / hz;
1576 break;
1577 case TCP_KEEPCNT:
1578 ui = tp->t_keepcnt;
1579 break;
1580 }
1581 INP_WUNLOCK(inp);
1582 error = sooptcopyout(sopt, &ui, sizeof(ui));
1583 break;
1584 default:
1585 INP_WUNLOCK(inp);
1586 error = ENOPROTOOPT;
1587 break;
1588 }
1589 break;
1590 }
1591 return (error);
1592}
1593#undef INP_WLOCK_RECHECK
1594
1595/*
1596 * Attach TCP protocol to socket, allocating
1597 * internet protocol control block, tcp control block,
1598 * bufer space, and entering LISTEN state if to accept connections.
1599 */
1600static int
1601tcp_attach(struct socket *so)
1602{
1603 struct tcpcb *tp;
1604 struct inpcb *inp;
1605 int error;
1606
1607 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
1608 error = soreserve(so, V_tcp_sendspace, V_tcp_recvspace);
1609 if (error)
1610 return (error);
1611 }
1612 so->so_rcv.sb_flags |= SB_AUTOSIZE;
1613 so->so_snd.sb_flags |= SB_AUTOSIZE;
1614 INP_INFO_WLOCK(&V_tcbinfo);
1615 error = in_pcballoc(so, &V_tcbinfo);
1616 if (error) {
1617 INP_INFO_WUNLOCK(&V_tcbinfo);
1618 return (error);
1619 }
1620 inp = sotoinpcb(so);
1621#ifdef INET6
1622 if (inp->inp_vflag & INP_IPV6PROTO) {
1623 inp->inp_vflag |= INP_IPV6;
1624 inp->in6p_hops = -1; /* use kernel default */
1625 }
1626 else
1627#endif
1628 inp->inp_vflag |= INP_IPV4;
1629 tp = tcp_newtcpcb(inp);
1630 if (tp == NULL) {
1631 in_pcbdetach(inp);
1632 in_pcbfree(inp);
1633 INP_INFO_WUNLOCK(&V_tcbinfo);
1634 return (ENOBUFS);
1635 }
1636 tp->t_state = TCPS_CLOSED;
1637 INP_WUNLOCK(inp);
1638 INP_INFO_WUNLOCK(&V_tcbinfo);
1639 return (0);
1640}
1641
1642/*
1643 * Initiate (or continue) disconnect.
1644 * If embryonic state, just send reset (once).
1645 * If in ``let data drain'' option and linger null, just drop.
1646 * Otherwise (hard), mark socket disconnecting and drop
1647 * current input data; switch states based on user close, and
1648 * send segment to peer (with FIN).
1649 */
1650static void
1651tcp_disconnect(struct tcpcb *tp)
1652{
1653 struct inpcb *inp = tp->t_inpcb;
1654 struct socket *so = inp->inp_socket;
1655
1656 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1657 INP_WLOCK_ASSERT(inp);
1658
1659 /*
1660 * Neither tcp_close() nor tcp_drop() should return NULL, as the
1661 * socket is still open.
1662 */
1663 if (tp->t_state < TCPS_ESTABLISHED) {
1664 tp = tcp_close(tp);
1665 KASSERT(tp != NULL,
1666 ("tcp_disconnect: tcp_close() returned NULL"));
1667 } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
1668 tp = tcp_drop(tp, 0);
1669 KASSERT(tp != NULL,
1670 ("tcp_disconnect: tcp_drop() returned NULL"));
1671 } else {
1672 soisdisconnecting(so);
1673 sbflush(&so->so_rcv);
1674 tcp_usrclosed(tp);
1675 if (!(inp->inp_flags & INP_DROPPED))
1676 tcp_output(tp);
1677 }
1678}
1679
1680/*
1681 * User issued close, and wish to trail through shutdown states:
1682 * if never received SYN, just forget it. If got a SYN from peer,
1683 * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
1684 * If already got a FIN from peer, then almost done; go to LAST_ACK
1685 * state. In all other cases, have already sent FIN to peer (e.g.
1686 * after PRU_SHUTDOWN), and just have to play tedious game waiting
1687 * for peer to send FIN or not respond to keep-alives, etc.
1688 * We can let the user exit from the close as soon as the FIN is acked.
1689 */
1690static void
1691tcp_usrclosed(struct tcpcb *tp)
1692{
1693
1694 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1695 INP_WLOCK_ASSERT(tp->t_inpcb);
1696
1697 switch (tp->t_state) {
1698 case TCPS_LISTEN:
1699#ifdef TCP_OFFLOAD
1700 tcp_offload_listen_stop(tp);
1701#endif
1702 /* FALLTHROUGH */
1703 case TCPS_CLOSED:
1704 tcp_state_change(tp, TCPS_CLOSED);
1705 tp = tcp_close(tp);
1706 /*
1707 * tcp_close() should never return NULL here as the socket is
1708 * still open.
1709 */
1710 KASSERT(tp != NULL,
1711 ("tcp_usrclosed: tcp_close() returned NULL"));
1712 break;
1713
1714 case TCPS_SYN_SENT:
1715 case TCPS_SYN_RECEIVED:
1716 tp->t_flags |= TF_NEEDFIN;
1717 break;
1718
1719 case TCPS_ESTABLISHED:
1720 tcp_state_change(tp, TCPS_FIN_WAIT_1);
1721 break;
1722
1723 case TCPS_CLOSE_WAIT:
1724 tcp_state_change(tp, TCPS_LAST_ACK);
1725 break;
1726 }
1727 if (tp->t_state >= TCPS_FIN_WAIT_2) {
1728 soisdisconnected(tp->t_inpcb->inp_socket);
1729 /* Prevent the connection hanging in FIN_WAIT_2 forever. */
1730 if (tp->t_state == TCPS_FIN_WAIT_2) {
1731 int timeout;
1732
1733 timeout = (tcp_fast_finwait2_recycle) ?
1734 tcp_finwait2_timeout : TP_MAXIDLE(tp);
1735 tcp_timer_activate(tp, TT_2MSL, timeout);
1736 }
1737 }
1738}
1739
1740#ifdef DDB
1741static void
1742db_print_indent(int indent)
1743{
1744 int i;
1745
1746 for (i = 0; i < indent; i++)
1747 db_printf(" ");
1748}
1749
1750static void
1751db_print_tstate(int t_state)
1752{
1753
1754 switch (t_state) {
1755 case TCPS_CLOSED:
1756 db_printf("TCPS_CLOSED");
1757 return;
1758
1759 case TCPS_LISTEN:
1760 db_printf("TCPS_LISTEN");
1761 return;
1762
1763 case TCPS_SYN_SENT:
1764 db_printf("TCPS_SYN_SENT");
1765 return;
1766
1767 case TCPS_SYN_RECEIVED:
1768 db_printf("TCPS_SYN_RECEIVED");
1769 return;
1770
1771 case TCPS_ESTABLISHED:
1772 db_printf("TCPS_ESTABLISHED");
1773 return;
1774
1775 case TCPS_CLOSE_WAIT:
1776 db_printf("TCPS_CLOSE_WAIT");
1777 return;
1778
1779 case TCPS_FIN_WAIT_1:
1780 db_printf("TCPS_FIN_WAIT_1");
1781 return;
1782
1783 case TCPS_CLOSING:
1784 db_printf("TCPS_CLOSING");
1785 return;
1786
1787 case TCPS_LAST_ACK:
1788 db_printf("TCPS_LAST_ACK");
1789 return;
1790
1791 case TCPS_FIN_WAIT_2:
1792 db_printf("TCPS_FIN_WAIT_2");
1793 return;
1794
1795 case TCPS_TIME_WAIT:
1796 db_printf("TCPS_TIME_WAIT");
1797 return;
1798
1799 default:
1800 db_printf("unknown");
1801 return;
1802 }
1803}
1804
1805static void
1806db_print_tflags(u_int t_flags)
1807{
1808 int comma;
1809
1810 comma = 0;
1811 if (t_flags & TF_ACKNOW) {
1812 db_printf("%sTF_ACKNOW", comma ? ", " : "");
1813 comma = 1;
1814 }
1815 if (t_flags & TF_DELACK) {
1816 db_printf("%sTF_DELACK", comma ? ", " : "");
1817 comma = 1;
1818 }
1819 if (t_flags & TF_NODELAY) {
1820 db_printf("%sTF_NODELAY", comma ? ", " : "");
1821 comma = 1;
1822 }
1823 if (t_flags & TF_NOOPT) {
1824 db_printf("%sTF_NOOPT", comma ? ", " : "");
1825 comma = 1;
1826 }
1827 if (t_flags & TF_SENTFIN) {
1828 db_printf("%sTF_SENTFIN", comma ? ", " : "");
1829 comma = 1;
1830 }
1831 if (t_flags & TF_REQ_SCALE) {
1832 db_printf("%sTF_REQ_SCALE", comma ? ", " : "");
1833 comma = 1;
1834 }
1835 if (t_flags & TF_RCVD_SCALE) {
1836 db_printf("%sTF_RECVD_SCALE", comma ? ", " : "");
1837 comma = 1;
1838 }
1839 if (t_flags & TF_REQ_TSTMP) {
1840 db_printf("%sTF_REQ_TSTMP", comma ? ", " : "");
1841 comma = 1;
1842 }
1843 if (t_flags & TF_RCVD_TSTMP) {
1844 db_printf("%sTF_RCVD_TSTMP", comma ? ", " : "");
1845 comma = 1;
1846 }
1847 if (t_flags & TF_SACK_PERMIT) {
1848 db_printf("%sTF_SACK_PERMIT", comma ? ", " : "");
1849 comma = 1;
1850 }
1851 if (t_flags & TF_NEEDSYN) {
1852 db_printf("%sTF_NEEDSYN", comma ? ", " : "");
1853 comma = 1;
1854 }
1855 if (t_flags & TF_NEEDFIN) {
1856 db_printf("%sTF_NEEDFIN", comma ? ", " : "");
1857 comma = 1;
1858 }
1859 if (t_flags & TF_NOPUSH) {
1860 db_printf("%sTF_NOPUSH", comma ? ", " : "");
1861 comma = 1;
1862 }
1863 if (t_flags & TF_MORETOCOME) {
1864 db_printf("%sTF_MORETOCOME", comma ? ", " : "");
1865 comma = 1;
1866 }
1867 if (t_flags & TF_LQ_OVERFLOW) {
1868 db_printf("%sTF_LQ_OVERFLOW", comma ? ", " : "");
1869 comma = 1;
1870 }
1871 if (t_flags & TF_LASTIDLE) {
1872 db_printf("%sTF_LASTIDLE", comma ? ", " : "");
1873 comma = 1;
1874 }
1875 if (t_flags & TF_RXWIN0SENT) {
1876 db_printf("%sTF_RXWIN0SENT", comma ? ", " : "");
1877 comma = 1;
1878 }
1879 if (t_flags & TF_FASTRECOVERY) {
1880 db_printf("%sTF_FASTRECOVERY", comma ? ", " : "");
1881 comma = 1;
1882 }
1883 if (t_flags & TF_CONGRECOVERY) {
1884 db_printf("%sTF_CONGRECOVERY", comma ? ", " : "");
1885 comma = 1;
1886 }
1887 if (t_flags & TF_WASFRECOVERY) {
1888 db_printf("%sTF_WASFRECOVERY", comma ? ", " : "");
1889 comma = 1;
1890 }
1891 if (t_flags & TF_SIGNATURE) {
1892 db_printf("%sTF_SIGNATURE", comma ? ", " : "");
1893 comma = 1;
1894 }
1895 if (t_flags & TF_FORCEDATA) {
1896 db_printf("%sTF_FORCEDATA", comma ? ", " : "");
1897 comma = 1;
1898 }
1899 if (t_flags & TF_TSO) {
1900 db_printf("%sTF_TSO", comma ? ", " : "");
1901 comma = 1;
1902 }
1903 if (t_flags & TF_ECN_PERMIT) {
1904 db_printf("%sTF_ECN_PERMIT", comma ? ", " : "");
1905 comma = 1;
1906 }
1907}
1908
1909static void
1910db_print_toobflags(char t_oobflags)
1911{
1912 int comma;
1913
1914 comma = 0;
1915 if (t_oobflags & TCPOOB_HAVEDATA) {
1916 db_printf("%sTCPOOB_HAVEDATA", comma ? ", " : "");
1917 comma = 1;
1918 }
1919 if (t_oobflags & TCPOOB_HADDATA) {
1920 db_printf("%sTCPOOB_HADDATA", comma ? ", " : "");
1921 comma = 1;
1922 }
1923}
1924
1925static void
1926db_print_tcpcb(struct tcpcb *tp, const char *name, int indent)
1927{
1928
1929 db_print_indent(indent);
1930 db_printf("%s at %p\n", name, tp);
1931
1932 indent += 2;
1933
1934 db_print_indent(indent);
1935 db_printf("t_segq first: %p t_segqlen: %d t_dupacks: %d\n",
1936 tp->t_segq, tp->t_segqlen, tp->t_dupacks);
1937
1938 db_print_indent(indent);
1939 db_printf("tt_rexmt: %p tt_persist: %p tt_keep: %p\n",
1940 &tp->t_timers->tt_rexmt, &tp->t_timers->tt_persist, &tp->t_timers->tt_keep);
1941
1942 db_print_indent(indent);
1943 db_printf("tt_2msl: %p tt_delack: %p t_inpcb: %p\n", &tp->t_timers->tt_2msl,
1944 &tp->t_timers->tt_delack, tp->t_inpcb);
1945
1946 db_print_indent(indent);
1947 db_printf("t_state: %d (", tp->t_state);
1948 db_print_tstate(tp->t_state);
1949 db_printf(")\n");
1950
1951 db_print_indent(indent);
1952 db_printf("t_flags: 0x%x (", tp->t_flags);
1953 db_print_tflags(tp->t_flags);
1954 db_printf(")\n");
1955
1956 db_print_indent(indent);
1957 db_printf("snd_una: 0x%08x snd_max: 0x%08x snd_nxt: x0%08x\n",
1958 tp->snd_una, tp->snd_max, tp->snd_nxt);
1959
1960 db_print_indent(indent);
1961 db_printf("snd_up: 0x%08x snd_wl1: 0x%08x snd_wl2: 0x%08x\n",
1962 tp->snd_up, tp->snd_wl1, tp->snd_wl2);
1963
1964 db_print_indent(indent);
1965 db_printf("iss: 0x%08x irs: 0x%08x rcv_nxt: 0x%08x\n",
1966 tp->iss, tp->irs, tp->rcv_nxt);
1967
1968 db_print_indent(indent);
1969 db_printf("rcv_adv: 0x%08x rcv_wnd: %lu rcv_up: 0x%08x\n",
1970 tp->rcv_adv, tp->rcv_wnd, tp->rcv_up);
1971
1972 db_print_indent(indent);
1973 db_printf("snd_wnd: %lu snd_cwnd: %lu\n",
1974 tp->snd_wnd, tp->snd_cwnd);
1975
1976 db_print_indent(indent);
1977 db_printf("snd_ssthresh: %lu snd_recover: "
1978 "0x%08x\n", tp->snd_ssthresh, tp->snd_recover);
1979
1980 db_print_indent(indent);
1981 db_printf("t_maxopd: %u t_rcvtime: %u t_startime: %u\n",
1982 tp->t_maxopd, tp->t_rcvtime, tp->t_starttime);
1983
1984 db_print_indent(indent);
1985 db_printf("t_rttime: %u t_rtsq: 0x%08x\n",
1986 tp->t_rtttime, tp->t_rtseq);
1987
1988 db_print_indent(indent);
1989 db_printf("t_rxtcur: %d t_maxseg: %u t_srtt: %d\n",
1990 tp->t_rxtcur, tp->t_maxseg, tp->t_srtt);
1991
1992 db_print_indent(indent);
1993 db_printf("t_rttvar: %d t_rxtshift: %d t_rttmin: %u "
1994 "t_rttbest: %u\n", tp->t_rttvar, tp->t_rxtshift, tp->t_rttmin,
1995 tp->t_rttbest);
1996
1997 db_print_indent(indent);
1998 db_printf("t_rttupdated: %lu max_sndwnd: %lu t_softerror: %d\n",
1999 tp->t_rttupdated, tp->max_sndwnd, tp->t_softerror);
2000
2001 db_print_indent(indent);
2002 db_printf("t_oobflags: 0x%x (", tp->t_oobflags);
2003 db_print_toobflags(tp->t_oobflags);
2004 db_printf(") t_iobc: 0x%02x\n", tp->t_iobc);
2005
2006 db_print_indent(indent);
2007 db_printf("snd_scale: %u rcv_scale: %u request_r_scale: %u\n",
2008 tp->snd_scale, tp->rcv_scale, tp->request_r_scale);
2009
2010 db_print_indent(indent);
2011 db_printf("ts_recent: %u ts_recent_age: %u\n",
2012 tp->ts_recent, tp->ts_recent_age);
2013
2014 db_print_indent(indent);
2015 db_printf("ts_offset: %u last_ack_sent: 0x%08x snd_cwnd_prev: "
2016 "%lu\n", tp->ts_offset, tp->last_ack_sent, tp->snd_cwnd_prev);
2017
2018 db_print_indent(indent);
2019 db_printf("snd_ssthresh_prev: %lu snd_recover_prev: 0x%08x "
2020 "t_badrxtwin: %u\n", tp->snd_ssthresh_prev,
2021 tp->snd_recover_prev, tp->t_badrxtwin);
2022
2023 db_print_indent(indent);
2024 db_printf("snd_numholes: %d snd_holes first: %p\n",
2025 tp->snd_numholes, TAILQ_FIRST(&tp->snd_holes));
2026
2027 db_print_indent(indent);
2028 db_printf("snd_fack: 0x%08x rcv_numsacks: %d sack_newdata: "
2029 "0x%08x\n", tp->snd_fack, tp->rcv_numsacks, tp->sack_newdata);
2030
2031 /* Skip sackblks, sackhint. */
2032
2033 db_print_indent(indent);
2034 db_printf("t_rttlow: %d rfbuf_ts: %u rfbuf_cnt: %d\n",
2035 tp->t_rttlow, tp->rfbuf_ts, tp->rfbuf_cnt);
2036}
2037
2038DB_SHOW_COMMAND(tcpcb, db_show_tcpcb)
2039{
2040 struct tcpcb *tp;
2041
2042 if (!have_addr) {
2043 db_printf("usage: show tcpcb <addr>\n");
2044 return;
2045 }
2046 tp = (struct tcpcb *)addr;
2047
2048 db_print_tcpcb(tp, "tcpcb", 0);
2049}
2050#endif
|