99
100#include "opt_inet.h"
101#include "opt_inet6.h"
102#include "opt_mac.h"
103#include "opt_zero.h"
104#include "opt_compat.h"
105
106#include <sys/param.h>
107#include <sys/systm.h>
108#include <sys/fcntl.h>
109#include <sys/limits.h>
110#include <sys/lock.h>
111#include <sys/mac.h>
112#include <sys/malloc.h>
113#include <sys/mbuf.h>
114#include <sys/mutex.h>
115#include <sys/domain.h>
116#include <sys/file.h> /* for struct knote */
117#include <sys/kernel.h>
118#include <sys/event.h>
119#include <sys/eventhandler.h>
120#include <sys/poll.h>
121#include <sys/proc.h>
122#include <sys/protosw.h>
123#include <sys/socket.h>
124#include <sys/socketvar.h>
125#include <sys/resourcevar.h>
126#include <net/route.h>
127#include <sys/signalvar.h>
128#include <sys/stat.h>
129#include <sys/sx.h>
130#include <sys/sysctl.h>
131#include <sys/uio.h>
132#include <sys/jail.h>
133#include <sys/vimage.h>
134
135#include <security/mac/mac_framework.h>
136
137#include <vm/uma.h>
138
139#ifdef COMPAT_IA32
140#include <sys/mount.h>
141#include <sys/sysent.h>
142#include <compat/freebsd32/freebsd32.h>
143#endif
144
145static int soreceive_rcvoob(struct socket *so, struct uio *uio,
146 int flags);
147
148static void filt_sordetach(struct knote *kn);
149static int filt_soread(struct knote *kn, long hint);
150static void filt_sowdetach(struct knote *kn);
151static int filt_sowrite(struct knote *kn, long hint);
152static int filt_solisten(struct knote *kn, long hint);
153
154static struct filterops solisten_filtops =
155 { 1, NULL, filt_sordetach, filt_solisten };
156static struct filterops soread_filtops =
157 { 1, NULL, filt_sordetach, filt_soread };
158static struct filterops sowrite_filtops =
159 { 1, NULL, filt_sowdetach, filt_sowrite };
160
161uma_zone_t socket_zone;
162so_gen_t so_gencnt; /* generation count for sockets */
163
164int maxsockets;
165
166MALLOC_DEFINE(M_SONAME, "soname", "socket name");
167MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
168
169static int somaxconn = SOMAXCONN;
170static int sysctl_somaxconn(SYSCTL_HANDLER_ARGS);
171/* XXX: we dont have SYSCTL_USHORT */
172SYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn, CTLTYPE_UINT | CTLFLAG_RW,
173 0, sizeof(int), sysctl_somaxconn, "I", "Maximum pending socket connection "
174 "queue size");
175static int numopensockets;
176SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD,
177 &numopensockets, 0, "Number of open sockets");
178#ifdef ZERO_COPY_SOCKETS
179/* These aren't static because they're used in other files. */
180int so_zero_copy_send = 1;
181int so_zero_copy_receive = 1;
182SYSCTL_NODE(_kern_ipc, OID_AUTO, zero_copy, CTLFLAG_RD, 0,
183 "Zero copy controls");
184SYSCTL_INT(_kern_ipc_zero_copy, OID_AUTO, receive, CTLFLAG_RW,
185 &so_zero_copy_receive, 0, "Enable zero copy receive");
186SYSCTL_INT(_kern_ipc_zero_copy, OID_AUTO, send, CTLFLAG_RW,
187 &so_zero_copy_send, 0, "Enable zero copy send");
188#endif /* ZERO_COPY_SOCKETS */
189
190/*
191 * accept_mtx locks down per-socket fields relating to accept queues. See
192 * socketvar.h for an annotation of the protected fields of struct socket.
193 */
194struct mtx accept_mtx;
195MTX_SYSINIT(accept_mtx, &accept_mtx, "accept", MTX_DEF);
196
197/*
198 * so_global_mtx protects so_gencnt, numopensockets, and the per-socket
199 * so_gencnt field.
200 */
201static struct mtx so_global_mtx;
202MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF);
203
204/*
205 * General IPC sysctl name space, used by sockets and a variety of other IPC
206 * types.
207 */
208SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC");
209
210/*
211 * Sysctl to get and set the maximum global sockets limit. Notify protocols
212 * of the change so that they can update their dependent limits as required.
213 */
214static int
215sysctl_maxsockets(SYSCTL_HANDLER_ARGS)
216{
217 int error, newmaxsockets;
218
219 newmaxsockets = maxsockets;
220 error = sysctl_handle_int(oidp, &newmaxsockets, 0, req);
221 if (error == 0 && req->newptr) {
222 if (newmaxsockets > maxsockets) {
223 maxsockets = newmaxsockets;
224 if (maxsockets > ((maxfiles / 4) * 3)) {
225 maxfiles = (maxsockets * 5) / 4;
226 maxfilesperproc = (maxfiles * 9) / 10;
227 }
228 EVENTHANDLER_INVOKE(maxsockets_change);
229 } else
230 error = EINVAL;
231 }
232 return (error);
233}
234
235SYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets, CTLTYPE_INT|CTLFLAG_RW,
236 &maxsockets, 0, sysctl_maxsockets, "IU",
237 "Maximum number of sockets avaliable");
238
239/*
240 * Initialise maxsockets. This SYSINIT must be run after
241 * tunable_mbinit().
242 */
243static void
244init_maxsockets(void *ignored)
245{
246
247 TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
248 maxsockets = imax(maxsockets, imax(maxfiles, nmbclusters));
249}
250SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL);
251
252/*
253 * Socket operation routines. These routines are called by the routines in
254 * sys_socket.c or from a system process, and implement the semantics of
255 * socket operations by switching out to the protocol specific routines.
256 */
257
258/*
259 * Get a socket structure from our zone, and initialize it. Note that it
260 * would probably be better to allocate socket and PCB at the same time, but
261 * I'm not convinced that all the protocols can be easily modified to do
262 * this.
263 *
264 * soalloc() returns a socket with a ref count of 0.
265 */
266static struct socket *
267soalloc(struct vnet *vnet)
268{
269 struct socket *so;
270
271 so = uma_zalloc(socket_zone, M_NOWAIT | M_ZERO);
272 if (so == NULL)
273 return (NULL);
274#ifdef MAC
275 if (mac_socket_init(so, M_NOWAIT) != 0) {
276 uma_zfree(socket_zone, so);
277 return (NULL);
278 }
279#endif
280 SOCKBUF_LOCK_INIT(&so->so_snd, "so_snd");
281 SOCKBUF_LOCK_INIT(&so->so_rcv, "so_rcv");
282 sx_init(&so->so_snd.sb_sx, "so_snd_sx");
283 sx_init(&so->so_rcv.sb_sx, "so_rcv_sx");
284 TAILQ_INIT(&so->so_aiojobq);
285 mtx_lock(&so_global_mtx);
286 so->so_gencnt = ++so_gencnt;
287 ++numopensockets;
288#ifdef VIMAGE
289 ++vnet->sockcnt; /* Locked with so_global_mtx. */
290 so->so_vnet = vnet;
291#endif
292 mtx_unlock(&so_global_mtx);
293 return (so);
294}
295
296/*
297 * Free the storage associated with a socket at the socket layer, tear down
298 * locks, labels, etc. All protocol state is assumed already to have been
299 * torn down (and possibly never set up) by the caller.
300 */
301static void
302sodealloc(struct socket *so)
303{
304
305 KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count));
306 KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL"));
307
308 mtx_lock(&so_global_mtx);
309 so->so_gencnt = ++so_gencnt;
310 --numopensockets; /* Could be below, but faster here. */
311#ifdef VIMAGE
312 --so->so_vnet->sockcnt;
313#endif
314 mtx_unlock(&so_global_mtx);
315 if (so->so_rcv.sb_hiwat)
316 (void)chgsbsize(so->so_cred->cr_uidinfo,
317 &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY);
318 if (so->so_snd.sb_hiwat)
319 (void)chgsbsize(so->so_cred->cr_uidinfo,
320 &so->so_snd.sb_hiwat, 0, RLIM_INFINITY);
321#ifdef INET
322 /* remove acccept filter if one is present. */
323 if (so->so_accf != NULL)
324 do_setopt_accept_filter(so, NULL);
325#endif
326#ifdef MAC
327 mac_socket_destroy(so);
328#endif
329 crfree(so->so_cred);
330 sx_destroy(&so->so_snd.sb_sx);
331 sx_destroy(&so->so_rcv.sb_sx);
332 SOCKBUF_LOCK_DESTROY(&so->so_snd);
333 SOCKBUF_LOCK_DESTROY(&so->so_rcv);
334 uma_zfree(socket_zone, so);
335}
336
337/*
338 * socreate returns a socket with a ref count of 1. The socket should be
339 * closed with soclose().
340 */
341int
342socreate(int dom, struct socket **aso, int type, int proto,
343 struct ucred *cred, struct thread *td)
344{
345 struct protosw *prp;
346 struct socket *so;
347 int error;
348
349 if (proto)
350 prp = pffindproto(dom, proto, type);
351 else
352 prp = pffindtype(dom, type);
353
354 if (prp == NULL || prp->pr_usrreqs->pru_attach == NULL ||
355 prp->pr_usrreqs->pru_attach == pru_attach_notsupp)
356 return (EPROTONOSUPPORT);
357
358 if (prison_check_af(cred, prp->pr_domain->dom_family) != 0)
359 return (EPROTONOSUPPORT);
360
361 if (prp->pr_type != type)
362 return (EPROTOTYPE);
363 so = soalloc(TD_TO_VNET(td));
364 if (so == NULL)
365 return (ENOBUFS);
366
367 TAILQ_INIT(&so->so_incomp);
368 TAILQ_INIT(&so->so_comp);
369 so->so_type = type;
370 so->so_cred = crhold(cred);
371 if ((prp->pr_domain->dom_family == PF_INET) ||
372 (prp->pr_domain->dom_family == PF_ROUTE))
373 so->so_fibnum = td->td_proc->p_fibnum;
374 else
375 so->so_fibnum = 0;
376 so->so_proto = prp;
377#ifdef MAC
378 mac_socket_create(cred, so);
379#endif
380 knlist_init(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv),
381 NULL, NULL, NULL);
382 knlist_init(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd),
383 NULL, NULL, NULL);
384 so->so_count = 1;
385 /*
386 * Auto-sizing of socket buffers is managed by the protocols and
387 * the appropriate flags must be set in the pru_attach function.
388 */
389 CURVNET_SET(so->so_vnet);
390 error = (*prp->pr_usrreqs->pru_attach)(so, proto, td);
391 CURVNET_RESTORE();
392 if (error) {
393 KASSERT(so->so_count == 1, ("socreate: so_count %d",
394 so->so_count));
395 so->so_count = 0;
396 sodealloc(so);
397 return (error);
398 }
399 *aso = so;
400 return (0);
401}
402
403#ifdef REGRESSION
404static int regression_sonewconn_earlytest = 1;
405SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW,
406 ®ression_sonewconn_earlytest, 0, "Perform early sonewconn limit test");
407#endif
408
409/*
410 * When an attempt at a new connection is noted on a socket which accepts
411 * connections, sonewconn is called. If the connection is possible (subject
412 * to space constraints, etc.) then we allocate a new structure, propoerly
413 * linked into the data structure of the original socket, and return this.
414 * Connstatus may be 0, or SO_ISCONFIRMING, or SO_ISCONNECTED.
415 *
416 * Note: the ref count on the socket is 0 on return.
417 */
418struct socket *
419sonewconn(struct socket *head, int connstatus)
420{
421 struct socket *so;
422 int over;
423
424 ACCEPT_LOCK();
425 over = (head->so_qlen > 3 * head->so_qlimit / 2);
426 ACCEPT_UNLOCK();
427#ifdef REGRESSION
428 if (regression_sonewconn_earlytest && over)
429#else
430 if (over)
431#endif
432 return (NULL);
433 VNET_ASSERT(head->so_vnet);
434 so = soalloc(head->so_vnet);
435 if (so == NULL)
436 return (NULL);
437 if ((head->so_options & SO_ACCEPTFILTER) != 0)
438 connstatus = 0;
439 so->so_head = head;
440 so->so_type = head->so_type;
441 so->so_options = head->so_options &~ SO_ACCEPTCONN;
442 so->so_linger = head->so_linger;
443 so->so_state = head->so_state | SS_NOFDREF;
444 so->so_proto = head->so_proto;
445 so->so_cred = crhold(head->so_cred);
446#ifdef MAC
447 SOCK_LOCK(head);
448 mac_socket_newconn(head, so);
449 SOCK_UNLOCK(head);
450#endif
451 knlist_init(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv),
452 NULL, NULL, NULL);
453 knlist_init(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd),
454 NULL, NULL, NULL);
455 if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat) ||
456 (*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
457 sodealloc(so);
458 return (NULL);
459 }
460 so->so_rcv.sb_lowat = head->so_rcv.sb_lowat;
461 so->so_snd.sb_lowat = head->so_snd.sb_lowat;
462 so->so_rcv.sb_timeo = head->so_rcv.sb_timeo;
463 so->so_snd.sb_timeo = head->so_snd.sb_timeo;
464 so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE;
465 so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE;
466 so->so_state |= connstatus;
467 ACCEPT_LOCK();
468 if (connstatus) {
469 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
470 so->so_qstate |= SQ_COMP;
471 head->so_qlen++;
472 } else {
473 /*
474 * Keep removing sockets from the head until there's room for
475 * us to insert on the tail. In pre-locking revisions, this
476 * was a simple if(), but as we could be racing with other
477 * threads and soabort() requires dropping locks, we must
478 * loop waiting for the condition to be true.
479 */
480 while (head->so_incqlen > head->so_qlimit) {
481 struct socket *sp;
482 sp = TAILQ_FIRST(&head->so_incomp);
483 TAILQ_REMOVE(&head->so_incomp, sp, so_list);
484 head->so_incqlen--;
485 sp->so_qstate &= ~SQ_INCOMP;
486 sp->so_head = NULL;
487 ACCEPT_UNLOCK();
488 soabort(sp);
489 ACCEPT_LOCK();
490 }
491 TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list);
492 so->so_qstate |= SQ_INCOMP;
493 head->so_incqlen++;
494 }
495 ACCEPT_UNLOCK();
496 if (connstatus) {
497 sorwakeup(head);
498 wakeup_one(&head->so_timeo);
499 }
500 return (so);
501}
502
503int
504sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
505{
506 int error;
507
508 CURVNET_SET(so->so_vnet);
509 error = (*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td);
510 CURVNET_RESTORE();
511 return error;
512}
513
514/*
515 * solisten() transitions a socket from a non-listening state to a listening
516 * state, but can also be used to update the listen queue depth on an
517 * existing listen socket. The protocol will call back into the sockets
518 * layer using solisten_proto_check() and solisten_proto() to check and set
519 * socket-layer listen state. Call backs are used so that the protocol can
520 * acquire both protocol and socket layer locks in whatever order is required
521 * by the protocol.
522 *
523 * Protocol implementors are advised to hold the socket lock across the
524 * socket-layer test and set to avoid races at the socket layer.
525 */
526int
527solisten(struct socket *so, int backlog, struct thread *td)
528{
529
530 return ((*so->so_proto->pr_usrreqs->pru_listen)(so, backlog, td));
531}
532
533int
534solisten_proto_check(struct socket *so)
535{
536
537 SOCK_LOCK_ASSERT(so);
538
539 if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
540 SS_ISDISCONNECTING))
541 return (EINVAL);
542 return (0);
543}
544
545void
546solisten_proto(struct socket *so, int backlog)
547{
548
549 SOCK_LOCK_ASSERT(so);
550
551 if (backlog < 0 || backlog > somaxconn)
552 backlog = somaxconn;
553 so->so_qlimit = backlog;
554 so->so_options |= SO_ACCEPTCONN;
555}
556
557/*
558 * Attempt to free a socket. This should really be sotryfree().
559 *
560 * sofree() will succeed if:
561 *
562 * - There are no outstanding file descriptor references or related consumers
563 * (so_count == 0).
564 *
565 * - The socket has been closed by user space, if ever open (SS_NOFDREF).
566 *
567 * - The protocol does not have an outstanding strong reference on the socket
568 * (SS_PROTOREF).
569 *
570 * - The socket is not in a completed connection queue, so a process has been
571 * notified that it is present. If it is removed, the user process may
572 * block in accept() despite select() saying the socket was ready.
573 *
574 * Otherwise, it will quietly abort so that a future call to sofree(), when
575 * conditions are right, can succeed.
576 */
577void
578sofree(struct socket *so)
579{
580 struct protosw *pr = so->so_proto;
581 struct socket *head;
582
583 ACCEPT_LOCK_ASSERT();
584 SOCK_LOCK_ASSERT(so);
585
586 if ((so->so_state & SS_NOFDREF) == 0 || so->so_count != 0 ||
587 (so->so_state & SS_PROTOREF) || (so->so_qstate & SQ_COMP)) {
588 SOCK_UNLOCK(so);
589 ACCEPT_UNLOCK();
590 return;
591 }
592
593 head = so->so_head;
594 if (head != NULL) {
595 KASSERT((so->so_qstate & SQ_COMP) != 0 ||
596 (so->so_qstate & SQ_INCOMP) != 0,
597 ("sofree: so_head != NULL, but neither SQ_COMP nor "
598 "SQ_INCOMP"));
599 KASSERT((so->so_qstate & SQ_COMP) == 0 ||
600 (so->so_qstate & SQ_INCOMP) == 0,
601 ("sofree: so->so_qstate is SQ_COMP and also SQ_INCOMP"));
602 TAILQ_REMOVE(&head->so_incomp, so, so_list);
603 head->so_incqlen--;
604 so->so_qstate &= ~SQ_INCOMP;
605 so->so_head = NULL;
606 }
607 KASSERT((so->so_qstate & SQ_COMP) == 0 &&
608 (so->so_qstate & SQ_INCOMP) == 0,
609 ("sofree: so_head == NULL, but still SQ_COMP(%d) or SQ_INCOMP(%d)",
610 so->so_qstate & SQ_COMP, so->so_qstate & SQ_INCOMP));
611 if (so->so_options & SO_ACCEPTCONN) {
612 KASSERT((TAILQ_EMPTY(&so->so_comp)), ("sofree: so_comp populated"));
613 KASSERT((TAILQ_EMPTY(&so->so_incomp)), ("sofree: so_comp populated"));
614 }
615 SOCK_UNLOCK(so);
616 ACCEPT_UNLOCK();
617
618 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
619 (*pr->pr_domain->dom_dispose)(so->so_rcv.sb_mb);
620 if (pr->pr_usrreqs->pru_detach != NULL)
621 (*pr->pr_usrreqs->pru_detach)(so);
622
623 /*
624 * From this point on, we assume that no other references to this
625 * socket exist anywhere else in the stack. Therefore, no locks need
626 * to be acquired or held.
627 *
628 * We used to do a lot of socket buffer and socket locking here, as
629 * well as invoke sorflush() and perform wakeups. The direct call to
630 * dom_dispose() and sbrelease_internal() are an inlining of what was
631 * necessary from sorflush().
632 *
633 * Notice that the socket buffer and kqueue state are torn down
634 * before calling pru_detach. This means that protocols shold not
635 * assume they can perform socket wakeups, etc, in their detach code.
636 */
637 sbdestroy(&so->so_snd, so);
638 sbdestroy(&so->so_rcv, so);
639 knlist_destroy(&so->so_rcv.sb_sel.si_note);
640 knlist_destroy(&so->so_snd.sb_sel.si_note);
641 sodealloc(so);
642}
643
644/*
645 * Close a socket on last file table reference removal. Initiate disconnect
646 * if connected. Free socket when disconnect complete.
647 *
648 * This function will sorele() the socket. Note that soclose() may be called
649 * prior to the ref count reaching zero. The actual socket structure will
650 * not be freed until the ref count reaches zero.
651 */
652int
653soclose(struct socket *so)
654{
655 int error = 0;
656
657 KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter"));
658
659 CURVNET_SET(so->so_vnet);
660 funsetown(&so->so_sigio);
661 if (so->so_state & SS_ISCONNECTED) {
662 if ((so->so_state & SS_ISDISCONNECTING) == 0) {
663 error = sodisconnect(so);
664 if (error)
665 goto drop;
666 }
667 if (so->so_options & SO_LINGER) {
668 if ((so->so_state & SS_ISDISCONNECTING) &&
669 (so->so_state & SS_NBIO))
670 goto drop;
671 while (so->so_state & SS_ISCONNECTED) {
672 error = tsleep(&so->so_timeo,
673 PSOCK | PCATCH, "soclos", so->so_linger * hz);
674 if (error)
675 break;
676 }
677 }
678 }
679
680drop:
681 if (so->so_proto->pr_usrreqs->pru_close != NULL)
682 (*so->so_proto->pr_usrreqs->pru_close)(so);
683 if (so->so_options & SO_ACCEPTCONN) {
684 struct socket *sp;
685 ACCEPT_LOCK();
686 while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
687 TAILQ_REMOVE(&so->so_incomp, sp, so_list);
688 so->so_incqlen--;
689 sp->so_qstate &= ~SQ_INCOMP;
690 sp->so_head = NULL;
691 ACCEPT_UNLOCK();
692 soabort(sp);
693 ACCEPT_LOCK();
694 }
695 while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
696 TAILQ_REMOVE(&so->so_comp, sp, so_list);
697 so->so_qlen--;
698 sp->so_qstate &= ~SQ_COMP;
699 sp->so_head = NULL;
700 ACCEPT_UNLOCK();
701 soabort(sp);
702 ACCEPT_LOCK();
703 }
704 ACCEPT_UNLOCK();
705 }
706 ACCEPT_LOCK();
707 SOCK_LOCK(so);
708 KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF"));
709 so->so_state |= SS_NOFDREF;
710 sorele(so);
711 CURVNET_RESTORE();
712 return (error);
713}
714
715/*
716 * soabort() is used to abruptly tear down a connection, such as when a
717 * resource limit is reached (listen queue depth exceeded), or if a listen
718 * socket is closed while there are sockets waiting to be accepted.
719 *
720 * This interface is tricky, because it is called on an unreferenced socket,
721 * and must be called only by a thread that has actually removed the socket
722 * from the listen queue it was on, or races with other threads are risked.
723 *
724 * This interface will call into the protocol code, so must not be called
725 * with any socket locks held. Protocols do call it while holding their own
726 * recursible protocol mutexes, but this is something that should be subject
727 * to review in the future.
728 */
729void
730soabort(struct socket *so)
731{
732
733 /*
734 * In as much as is possible, assert that no references to this
735 * socket are held. This is not quite the same as asserting that the
736 * current thread is responsible for arranging for no references, but
737 * is as close as we can get for now.
738 */
739 KASSERT(so->so_count == 0, ("soabort: so_count"));
740 KASSERT((so->so_state & SS_PROTOREF) == 0, ("soabort: SS_PROTOREF"));
741 KASSERT(so->so_state & SS_NOFDREF, ("soabort: !SS_NOFDREF"));
742 KASSERT((so->so_state & SQ_COMP) == 0, ("soabort: SQ_COMP"));
743 KASSERT((so->so_state & SQ_INCOMP) == 0, ("soabort: SQ_INCOMP"));
744
745 if (so->so_proto->pr_usrreqs->pru_abort != NULL)
746 (*so->so_proto->pr_usrreqs->pru_abort)(so);
747 ACCEPT_LOCK();
748 SOCK_LOCK(so);
749 sofree(so);
750}
751
752int
753soaccept(struct socket *so, struct sockaddr **nam)
754{
755 int error;
756
757 SOCK_LOCK(so);
758 KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF"));
759 so->so_state &= ~SS_NOFDREF;
760 SOCK_UNLOCK(so);
761 error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam);
762 return (error);
763}
764
765int
766soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
767{
768 int error;
769
770 if (so->so_options & SO_ACCEPTCONN)
771 return (EOPNOTSUPP);
772 /*
773 * If protocol is connection-based, can only connect once.
774 * Otherwise, if connected, try to disconnect first. This allows
775 * user to disconnect by connecting to, e.g., a null address.
776 */
777 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
778 ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
779 (error = sodisconnect(so)))) {
780 error = EISCONN;
781 } else {
782 /*
783 * Prevent accumulated error from previous connection from
784 * biting us.
785 */
786 so->so_error = 0;
787 CURVNET_SET(so->so_vnet);
788 error = (*so->so_proto->pr_usrreqs->pru_connect)(so, nam, td);
789 CURVNET_RESTORE();
790 }
791
792 return (error);
793}
794
795int
796soconnect2(struct socket *so1, struct socket *so2)
797{
798
799 return ((*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2));
800}
801
802int
803sodisconnect(struct socket *so)
804{
805 int error;
806
807 if ((so->so_state & SS_ISCONNECTED) == 0)
808 return (ENOTCONN);
809 if (so->so_state & SS_ISDISCONNECTING)
810 return (EALREADY);
811 error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so);
812 return (error);
813}
814
815#ifdef ZERO_COPY_SOCKETS
816struct so_zerocopy_stats{
817 int size_ok;
818 int align_ok;
819 int found_ifp;
820};
821struct so_zerocopy_stats so_zerocp_stats = {0,0,0};
822#include <netinet/in.h>
823#include <net/route.h>
824#include <netinet/in_pcb.h>
825#include <vm/vm.h>
826#include <vm/vm_page.h>
827#include <vm/vm_object.h>
828
829/*
830 * sosend_copyin() is only used if zero copy sockets are enabled. Otherwise
831 * sosend_dgram() and sosend_generic() use m_uiotombuf().
832 *
833 * sosend_copyin() accepts a uio and prepares an mbuf chain holding part or
834 * all of the data referenced by the uio. If desired, it uses zero-copy.
835 * *space will be updated to reflect data copied in.
836 *
837 * NB: If atomic I/O is requested, the caller must already have checked that
838 * space can hold resid bytes.
839 *
840 * NB: In the event of an error, the caller may need to free the partial
841 * chain pointed to by *mpp. The contents of both *uio and *space may be
842 * modified even in the case of an error.
843 */
844static int
845sosend_copyin(struct uio *uio, struct mbuf **retmp, int atomic, long *space,
846 int flags)
847{
848 struct mbuf *m, **mp, *top;
849 long len, resid;
850 int error;
851#ifdef ZERO_COPY_SOCKETS
852 int cow_send;
853#endif
854
855 *retmp = top = NULL;
856 mp = ⊤
857 len = 0;
858 resid = uio->uio_resid;
859 error = 0;
860 do {
861#ifdef ZERO_COPY_SOCKETS
862 cow_send = 0;
863#endif /* ZERO_COPY_SOCKETS */
864 if (resid >= MINCLSIZE) {
865#ifdef ZERO_COPY_SOCKETS
866 if (top == NULL) {
867 m = m_gethdr(M_WAITOK, MT_DATA);
868 m->m_pkthdr.len = 0;
869 m->m_pkthdr.rcvif = NULL;
870 } else
871 m = m_get(M_WAITOK, MT_DATA);
872 if (so_zero_copy_send &&
873 resid>=PAGE_SIZE &&
874 *space>=PAGE_SIZE &&
875 uio->uio_iov->iov_len>=PAGE_SIZE) {
876 so_zerocp_stats.size_ok++;
877 so_zerocp_stats.align_ok++;
878 cow_send = socow_setup(m, uio);
879 len = cow_send;
880 }
881 if (!cow_send) {
882 m_clget(m, M_WAITOK);
883 len = min(min(MCLBYTES, resid), *space);
884 }
885#else /* ZERO_COPY_SOCKETS */
886 if (top == NULL) {
887 m = m_getcl(M_WAIT, MT_DATA, M_PKTHDR);
888 m->m_pkthdr.len = 0;
889 m->m_pkthdr.rcvif = NULL;
890 } else
891 m = m_getcl(M_WAIT, MT_DATA, 0);
892 len = min(min(MCLBYTES, resid), *space);
893#endif /* ZERO_COPY_SOCKETS */
894 } else {
895 if (top == NULL) {
896 m = m_gethdr(M_WAIT, MT_DATA);
897 m->m_pkthdr.len = 0;
898 m->m_pkthdr.rcvif = NULL;
899
900 len = min(min(MHLEN, resid), *space);
901 /*
902 * For datagram protocols, leave room
903 * for protocol headers in first mbuf.
904 */
905 if (atomic && m && len < MHLEN)
906 MH_ALIGN(m, len);
907 } else {
908 m = m_get(M_WAIT, MT_DATA);
909 len = min(min(MLEN, resid), *space);
910 }
911 }
912 if (m == NULL) {
913 error = ENOBUFS;
914 goto out;
915 }
916
917 *space -= len;
918#ifdef ZERO_COPY_SOCKETS
919 if (cow_send)
920 error = 0;
921 else
922#endif /* ZERO_COPY_SOCKETS */
923 error = uiomove(mtod(m, void *), (int)len, uio);
924 resid = uio->uio_resid;
925 m->m_len = len;
926 *mp = m;
927 top->m_pkthdr.len += len;
928 if (error)
929 goto out;
930 mp = &m->m_next;
931 if (resid <= 0) {
932 if (flags & MSG_EOR)
933 top->m_flags |= M_EOR;
934 break;
935 }
936 } while (*space > 0 && atomic);
937out:
938 *retmp = top;
939 return (error);
940}
941#endif /*ZERO_COPY_SOCKETS*/
942
943#define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? 0 : SBL_WAIT)
944
945int
946sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio,
947 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
948{
949 long space, resid;
950 int clen = 0, error, dontroute;
951#ifdef ZERO_COPY_SOCKETS
952 int atomic = sosendallatonce(so) || top;
953#endif
954
955 KASSERT(so->so_type == SOCK_DGRAM, ("sodgram_send: !SOCK_DGRAM"));
956 KASSERT(so->so_proto->pr_flags & PR_ATOMIC,
957 ("sodgram_send: !PR_ATOMIC"));
958
959 if (uio != NULL)
960 resid = uio->uio_resid;
961 else
962 resid = top->m_pkthdr.len;
963 /*
964 * In theory resid should be unsigned. However, space must be
965 * signed, as it might be less than 0 if we over-committed, and we
966 * must use a signed comparison of space and resid. On the other
967 * hand, a negative resid causes us to loop sending 0-length
968 * segments to the protocol.
969 *
970 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
971 * type sockets since that's an error.
972 */
973 if (resid < 0) {
974 error = EINVAL;
975 goto out;
976 }
977
978 dontroute =
979 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0;
980 if (td != NULL)
981 td->td_ru.ru_msgsnd++;
982 if (control != NULL)
983 clen = control->m_len;
984
985 SOCKBUF_LOCK(&so->so_snd);
986 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
987 SOCKBUF_UNLOCK(&so->so_snd);
988 error = EPIPE;
989 goto out;
990 }
991 if (so->so_error) {
992 error = so->so_error;
993 so->so_error = 0;
994 SOCKBUF_UNLOCK(&so->so_snd);
995 goto out;
996 }
997 if ((so->so_state & SS_ISCONNECTED) == 0) {
998 /*
999 * `sendto' and `sendmsg' is allowed on a connection-based
1000 * socket if it supports implied connect. Return ENOTCONN if
1001 * not connected and no address is supplied.
1002 */
1003 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1004 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1005 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1006 !(resid == 0 && clen != 0)) {
1007 SOCKBUF_UNLOCK(&so->so_snd);
1008 error = ENOTCONN;
1009 goto out;
1010 }
1011 } else if (addr == NULL) {
1012 if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1013 error = ENOTCONN;
1014 else
1015 error = EDESTADDRREQ;
1016 SOCKBUF_UNLOCK(&so->so_snd);
1017 goto out;
1018 }
1019 }
1020
1021 /*
1022 * Do we need MSG_OOB support in SOCK_DGRAM? Signs here may be a
1023 * problem and need fixing.
1024 */
1025 space = sbspace(&so->so_snd);
1026 if (flags & MSG_OOB)
1027 space += 1024;
1028 space -= clen;
1029 SOCKBUF_UNLOCK(&so->so_snd);
1030 if (resid > space) {
1031 error = EMSGSIZE;
1032 goto out;
1033 }
1034 if (uio == NULL) {
1035 resid = 0;
1036 if (flags & MSG_EOR)
1037 top->m_flags |= M_EOR;
1038 } else {
1039#ifdef ZERO_COPY_SOCKETS
1040 error = sosend_copyin(uio, &top, atomic, &space, flags);
1041 if (error)
1042 goto out;
1043#else
1044 /*
1045 * Copy the data from userland into a mbuf chain.
1046 * If no data is to be copied in, a single empty mbuf
1047 * is returned.
1048 */
1049 top = m_uiotombuf(uio, M_WAITOK, space, max_hdr,
1050 (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0)));
1051 if (top == NULL) {
1052 error = EFAULT; /* only possible error */
1053 goto out;
1054 }
1055 space -= resid - uio->uio_resid;
1056#endif
1057 resid = uio->uio_resid;
1058 }
1059 KASSERT(resid == 0, ("sosend_dgram: resid != 0"));
1060 /*
1061 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock
1062 * than with.
1063 */
1064 if (dontroute) {
1065 SOCK_LOCK(so);
1066 so->so_options |= SO_DONTROUTE;
1067 SOCK_UNLOCK(so);
1068 }
1069 /*
1070 * XXX all the SBS_CANTSENDMORE checks previously done could be out
1071 * of date. We could have recieved a reset packet in an interrupt or
1072 * maybe we slept while doing page faults in uiomove() etc. We could
1073 * probably recheck again inside the locking protection here, but
1074 * there are probably other places that this also happens. We must
1075 * rethink this.
1076 */
1077 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1078 (flags & MSG_OOB) ? PRUS_OOB :
1079 /*
1080 * If the user set MSG_EOF, the protocol understands this flag and
1081 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND.
1082 */
1083 ((flags & MSG_EOF) &&
1084 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1085 (resid <= 0)) ?
1086 PRUS_EOF :
1087 /* If there is more to send set PRUS_MORETOCOME */
1088 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1089 top, addr, control, td);
1090 if (dontroute) {
1091 SOCK_LOCK(so);
1092 so->so_options &= ~SO_DONTROUTE;
1093 SOCK_UNLOCK(so);
1094 }
1095 clen = 0;
1096 control = NULL;
1097 top = NULL;
1098out:
1099 if (top != NULL)
1100 m_freem(top);
1101 if (control != NULL)
1102 m_freem(control);
1103 return (error);
1104}
1105
1106/*
1107 * Send on a socket. If send must go all at once and message is larger than
1108 * send buffering, then hard error. Lock against other senders. If must go
1109 * all at once and not enough room now, then inform user that this would
1110 * block and do nothing. Otherwise, if nonblocking, send as much as
1111 * possible. The data to be sent is described by "uio" if nonzero, otherwise
1112 * by the mbuf chain "top" (which must be null if uio is not). Data provided
1113 * in mbuf chain must be small enough to send all at once.
1114 *
1115 * Returns nonzero on error, timeout or signal; callers must check for short
1116 * counts if EINTR/ERESTART are returned. Data and control buffers are freed
1117 * on return.
1118 */
1119int
1120sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio,
1121 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1122{
1123 long space, resid;
1124 int clen = 0, error, dontroute;
1125 int atomic = sosendallatonce(so) || top;
1126
1127 if (uio != NULL)
1128 resid = uio->uio_resid;
1129 else
1130 resid = top->m_pkthdr.len;
1131 /*
1132 * In theory resid should be unsigned. However, space must be
1133 * signed, as it might be less than 0 if we over-committed, and we
1134 * must use a signed comparison of space and resid. On the other
1135 * hand, a negative resid causes us to loop sending 0-length
1136 * segments to the protocol.
1137 *
1138 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
1139 * type sockets since that's an error.
1140 */
1141 if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
1142 error = EINVAL;
1143 goto out;
1144 }
1145
1146 dontroute =
1147 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
1148 (so->so_proto->pr_flags & PR_ATOMIC);
1149 if (td != NULL)
1150 td->td_ru.ru_msgsnd++;
1151 if (control != NULL)
1152 clen = control->m_len;
1153
1154 error = sblock(&so->so_snd, SBLOCKWAIT(flags));
1155 if (error)
1156 goto out;
1157
1158restart:
1159 do {
1160 SOCKBUF_LOCK(&so->so_snd);
1161 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1162 SOCKBUF_UNLOCK(&so->so_snd);
1163 error = EPIPE;
1164 goto release;
1165 }
1166 if (so->so_error) {
1167 error = so->so_error;
1168 so->so_error = 0;
1169 SOCKBUF_UNLOCK(&so->so_snd);
1170 goto release;
1171 }
1172 if ((so->so_state & SS_ISCONNECTED) == 0) {
1173 /*
1174 * `sendto' and `sendmsg' is allowed on a connection-
1175 * based socket if it supports implied connect.
1176 * Return ENOTCONN if not connected and no address is
1177 * supplied.
1178 */
1179 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1180 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1181 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1182 !(resid == 0 && clen != 0)) {
1183 SOCKBUF_UNLOCK(&so->so_snd);
1184 error = ENOTCONN;
1185 goto release;
1186 }
1187 } else if (addr == NULL) {
1188 SOCKBUF_UNLOCK(&so->so_snd);
1189 if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1190 error = ENOTCONN;
1191 else
1192 error = EDESTADDRREQ;
1193 goto release;
1194 }
1195 }
1196 space = sbspace(&so->so_snd);
1197 if (flags & MSG_OOB)
1198 space += 1024;
1199 if ((atomic && resid > so->so_snd.sb_hiwat) ||
1200 clen > so->so_snd.sb_hiwat) {
1201 SOCKBUF_UNLOCK(&so->so_snd);
1202 error = EMSGSIZE;
1203 goto release;
1204 }
1205 if (space < resid + clen &&
1206 (atomic || space < so->so_snd.sb_lowat || space < clen)) {
1207 if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO)) {
1208 SOCKBUF_UNLOCK(&so->so_snd);
1209 error = EWOULDBLOCK;
1210 goto release;
1211 }
1212 error = sbwait(&so->so_snd);
1213 SOCKBUF_UNLOCK(&so->so_snd);
1214 if (error)
1215 goto release;
1216 goto restart;
1217 }
1218 SOCKBUF_UNLOCK(&so->so_snd);
1219 space -= clen;
1220 do {
1221 if (uio == NULL) {
1222 resid = 0;
1223 if (flags & MSG_EOR)
1224 top->m_flags |= M_EOR;
1225 } else {
1226#ifdef ZERO_COPY_SOCKETS
1227 error = sosend_copyin(uio, &top, atomic,
1228 &space, flags);
1229 if (error != 0)
1230 goto release;
1231#else
1232 /*
1233 * Copy the data from userland into a mbuf
1234 * chain. If no data is to be copied in,
1235 * a single empty mbuf is returned.
1236 */
1237 top = m_uiotombuf(uio, M_WAITOK, space,
1238 (atomic ? max_hdr : 0),
1239 (atomic ? M_PKTHDR : 0) |
1240 ((flags & MSG_EOR) ? M_EOR : 0));
1241 if (top == NULL) {
1242 error = EFAULT; /* only possible error */
1243 goto release;
1244 }
1245 space -= resid - uio->uio_resid;
1246#endif
1247 resid = uio->uio_resid;
1248 }
1249 if (dontroute) {
1250 SOCK_LOCK(so);
1251 so->so_options |= SO_DONTROUTE;
1252 SOCK_UNLOCK(so);
1253 }
1254 /*
1255 * XXX all the SBS_CANTSENDMORE checks previously
1256 * done could be out of date. We could have recieved
1257 * a reset packet in an interrupt or maybe we slept
1258 * while doing page faults in uiomove() etc. We
1259 * could probably recheck again inside the locking
1260 * protection here, but there are probably other
1261 * places that this also happens. We must rethink
1262 * this.
1263 */
1264 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1265 (flags & MSG_OOB) ? PRUS_OOB :
1266 /*
1267 * If the user set MSG_EOF, the protocol understands
1268 * this flag and nothing left to send then use
1269 * PRU_SEND_EOF instead of PRU_SEND.
1270 */
1271 ((flags & MSG_EOF) &&
1272 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1273 (resid <= 0)) ?
1274 PRUS_EOF :
1275 /* If there is more to send set PRUS_MORETOCOME. */
1276 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1277 top, addr, control, td);
1278 if (dontroute) {
1279 SOCK_LOCK(so);
1280 so->so_options &= ~SO_DONTROUTE;
1281 SOCK_UNLOCK(so);
1282 }
1283 clen = 0;
1284 control = NULL;
1285 top = NULL;
1286 if (error)
1287 goto release;
1288 } while (resid && space > 0);
1289 } while (resid);
1290
1291release:
1292 sbunlock(&so->so_snd);
1293out:
1294 if (top != NULL)
1295 m_freem(top);
1296 if (control != NULL)
1297 m_freem(control);
1298 return (error);
1299}
1300
1301int
1302sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
1303 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1304{
1305 int error;
1306
1307 CURVNET_SET(so->so_vnet);
1308 error = so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio, top,
1309 control, flags, td);
1310 CURVNET_RESTORE();
1311 return (error);
1312}
1313
1314/*
1315 * The part of soreceive() that implements reading non-inline out-of-band
1316 * data from a socket. For more complete comments, see soreceive(), from
1317 * which this code originated.
1318 *
1319 * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
1320 * unable to return an mbuf chain to the caller.
1321 */
1322static int
1323soreceive_rcvoob(struct socket *so, struct uio *uio, int flags)
1324{
1325 struct protosw *pr = so->so_proto;
1326 struct mbuf *m;
1327 int error;
1328
1329 KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
1330
1331 m = m_get(M_WAIT, MT_DATA);
1332 error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK);
1333 if (error)
1334 goto bad;
1335 do {
1336#ifdef ZERO_COPY_SOCKETS
1337 if (so_zero_copy_receive) {
1338 int disposable;
1339
1340 if ((m->m_flags & M_EXT)
1341 && (m->m_ext.ext_type == EXT_DISPOSABLE))
1342 disposable = 1;
1343 else
1344 disposable = 0;
1345
1346 error = uiomoveco(mtod(m, void *),
1347 min(uio->uio_resid, m->m_len),
1348 uio, disposable);
1349 } else
1350#endif /* ZERO_COPY_SOCKETS */
1351 error = uiomove(mtod(m, void *),
1352 (int) min(uio->uio_resid, m->m_len), uio);
1353 m = m_free(m);
1354 } while (uio->uio_resid && error == 0 && m);
1355bad:
1356 if (m != NULL)
1357 m_freem(m);
1358 return (error);
1359}
1360
1361/*
1362 * Following replacement or removal of the first mbuf on the first mbuf chain
1363 * of a socket buffer, push necessary state changes back into the socket
1364 * buffer so that other consumers see the values consistently. 'nextrecord'
1365 * is the callers locally stored value of the original value of
1366 * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes.
1367 * NOTE: 'nextrecord' may be NULL.
1368 */
1369static __inline void
1370sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord)
1371{
1372
1373 SOCKBUF_LOCK_ASSERT(sb);
1374 /*
1375 * First, update for the new value of nextrecord. If necessary, make
1376 * it the first record.
1377 */
1378 if (sb->sb_mb != NULL)
1379 sb->sb_mb->m_nextpkt = nextrecord;
1380 else
1381 sb->sb_mb = nextrecord;
1382
1383 /*
1384 * Now update any dependent socket buffer fields to reflect the new
1385 * state. This is an expanded inline of SB_EMPTY_FIXUP(), with the
1386 * addition of a second clause that takes care of the case where
1387 * sb_mb has been updated, but remains the last record.
1388 */
1389 if (sb->sb_mb == NULL) {
1390 sb->sb_mbtail = NULL;
1391 sb->sb_lastrecord = NULL;
1392 } else if (sb->sb_mb->m_nextpkt == NULL)
1393 sb->sb_lastrecord = sb->sb_mb;
1394}
1395
1396
1397/*
1398 * Implement receive operations on a socket. We depend on the way that
1399 * records are added to the sockbuf by sbappend. In particular, each record
1400 * (mbufs linked through m_next) must begin with an address if the protocol
1401 * so specifies, followed by an optional mbuf or mbufs containing ancillary
1402 * data, and then zero or more mbufs of data. In order to allow parallelism
1403 * between network receive and copying to user space, as well as avoid
1404 * sleeping with a mutex held, we release the socket buffer mutex during the
1405 * user space copy. Although the sockbuf is locked, new data may still be
1406 * appended, and thus we must maintain consistency of the sockbuf during that
1407 * time.
1408 *
1409 * The caller may receive the data as a single mbuf chain by supplying an
1410 * mbuf **mp0 for use in returning the chain. The uio is then used only for
1411 * the count in uio_resid.
1412 */
1413int
1414soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio,
1415 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1416{
1417 struct mbuf *m, **mp;
1418 int flags, len, error, offset;
1419 struct protosw *pr = so->so_proto;
1420 struct mbuf *nextrecord;
1421 int moff, type = 0;
1422 int orig_resid = uio->uio_resid;
1423
1424 mp = mp0;
1425 if (psa != NULL)
1426 *psa = NULL;
1427 if (controlp != NULL)
1428 *controlp = NULL;
1429 if (flagsp != NULL)
1430 flags = *flagsp &~ MSG_EOR;
1431 else
1432 flags = 0;
1433 if (flags & MSG_OOB)
1434 return (soreceive_rcvoob(so, uio, flags));
1435 if (mp != NULL)
1436 *mp = NULL;
1437 if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING)
1438 && uio->uio_resid)
1439 (*pr->pr_usrreqs->pru_rcvd)(so, 0);
1440
1441 error = sblock(&so->so_rcv, SBLOCKWAIT(flags));
1442 if (error)
1443 return (error);
1444
1445restart:
1446 SOCKBUF_LOCK(&so->so_rcv);
1447 m = so->so_rcv.sb_mb;
1448 /*
1449 * If we have less data than requested, block awaiting more (subject
1450 * to any timeout) if:
1451 * 1. the current count is less than the low water mark, or
1452 * 2. MSG_WAITALL is set, and it is possible to do the entire
1453 * receive operation at once if we block (resid <= hiwat).
1454 * 3. MSG_DONTWAIT is not set
1455 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1456 * we have to do the receive in sections, and thus risk returning a
1457 * short count if a timeout or signal occurs after we start.
1458 */
1459 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1460 so->so_rcv.sb_cc < uio->uio_resid) &&
1461 (so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
1462 ((flags & MSG_WAITALL) && uio->uio_resid <= so->so_rcv.sb_hiwat)) &&
1463 m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
1464 KASSERT(m != NULL || !so->so_rcv.sb_cc,
1465 ("receive: m == %p so->so_rcv.sb_cc == %u",
1466 m, so->so_rcv.sb_cc));
1467 if (so->so_error) {
1468 if (m != NULL)
1469 goto dontblock;
1470 error = so->so_error;
1471 if ((flags & MSG_PEEK) == 0)
1472 so->so_error = 0;
1473 SOCKBUF_UNLOCK(&so->so_rcv);
1474 goto release;
1475 }
1476 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1477 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1478 if (m == NULL) {
1479 SOCKBUF_UNLOCK(&so->so_rcv);
1480 goto release;
1481 } else
1482 goto dontblock;
1483 }
1484 for (; m != NULL; m = m->m_next)
1485 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) {
1486 m = so->so_rcv.sb_mb;
1487 goto dontblock;
1488 }
1489 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1490 (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1491 SOCKBUF_UNLOCK(&so->so_rcv);
1492 error = ENOTCONN;
1493 goto release;
1494 }
1495 if (uio->uio_resid == 0) {
1496 SOCKBUF_UNLOCK(&so->so_rcv);
1497 goto release;
1498 }
1499 if ((so->so_state & SS_NBIO) ||
1500 (flags & (MSG_DONTWAIT|MSG_NBIO))) {
1501 SOCKBUF_UNLOCK(&so->so_rcv);
1502 error = EWOULDBLOCK;
1503 goto release;
1504 }
1505 SBLASTRECORDCHK(&so->so_rcv);
1506 SBLASTMBUFCHK(&so->so_rcv);
1507 error = sbwait(&so->so_rcv);
1508 SOCKBUF_UNLOCK(&so->so_rcv);
1509 if (error)
1510 goto release;
1511 goto restart;
1512 }
1513dontblock:
1514 /*
1515 * From this point onward, we maintain 'nextrecord' as a cache of the
1516 * pointer to the next record in the socket buffer. We must keep the
1517 * various socket buffer pointers and local stack versions of the
1518 * pointers in sync, pushing out modifications before dropping the
1519 * socket buffer mutex, and re-reading them when picking it up.
1520 *
1521 * Otherwise, we will race with the network stack appending new data
1522 * or records onto the socket buffer by using inconsistent/stale
1523 * versions of the field, possibly resulting in socket buffer
1524 * corruption.
1525 *
1526 * By holding the high-level sblock(), we prevent simultaneous
1527 * readers from pulling off the front of the socket buffer.
1528 */
1529 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1530 if (uio->uio_td)
1531 uio->uio_td->td_ru.ru_msgrcv++;
1532 KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb"));
1533 SBLASTRECORDCHK(&so->so_rcv);
1534 SBLASTMBUFCHK(&so->so_rcv);
1535 nextrecord = m->m_nextpkt;
1536 if (pr->pr_flags & PR_ADDR) {
1537 KASSERT(m->m_type == MT_SONAME,
1538 ("m->m_type == %d", m->m_type));
1539 orig_resid = 0;
1540 if (psa != NULL)
1541 *psa = sodupsockaddr(mtod(m, struct sockaddr *),
1542 M_NOWAIT);
1543 if (flags & MSG_PEEK) {
1544 m = m->m_next;
1545 } else {
1546 sbfree(&so->so_rcv, m);
1547 so->so_rcv.sb_mb = m_free(m);
1548 m = so->so_rcv.sb_mb;
1549 sockbuf_pushsync(&so->so_rcv, nextrecord);
1550 }
1551 }
1552
1553 /*
1554 * Process one or more MT_CONTROL mbufs present before any data mbufs
1555 * in the first mbuf chain on the socket buffer. If MSG_PEEK, we
1556 * just copy the data; if !MSG_PEEK, we call into the protocol to
1557 * perform externalization (or freeing if controlp == NULL).
1558 */
1559 if (m != NULL && m->m_type == MT_CONTROL) {
1560 struct mbuf *cm = NULL, *cmn;
1561 struct mbuf **cme = &cm;
1562
1563 do {
1564 if (flags & MSG_PEEK) {
1565 if (controlp != NULL) {
1566 *controlp = m_copy(m, 0, m->m_len);
1567 controlp = &(*controlp)->m_next;
1568 }
1569 m = m->m_next;
1570 } else {
1571 sbfree(&so->so_rcv, m);
1572 so->so_rcv.sb_mb = m->m_next;
1573 m->m_next = NULL;
1574 *cme = m;
1575 cme = &(*cme)->m_next;
1576 m = so->so_rcv.sb_mb;
1577 }
1578 } while (m != NULL && m->m_type == MT_CONTROL);
1579 if ((flags & MSG_PEEK) == 0)
1580 sockbuf_pushsync(&so->so_rcv, nextrecord);
1581 while (cm != NULL) {
1582 cmn = cm->m_next;
1583 cm->m_next = NULL;
1584 if (pr->pr_domain->dom_externalize != NULL) {
1585 SOCKBUF_UNLOCK(&so->so_rcv);
1586 error = (*pr->pr_domain->dom_externalize)
1587 (cm, controlp);
1588 SOCKBUF_LOCK(&so->so_rcv);
1589 } else if (controlp != NULL)
1590 *controlp = cm;
1591 else
1592 m_freem(cm);
1593 if (controlp != NULL) {
1594 orig_resid = 0;
1595 while (*controlp != NULL)
1596 controlp = &(*controlp)->m_next;
1597 }
1598 cm = cmn;
1599 }
1600 if (m != NULL)
1601 nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1602 else
1603 nextrecord = so->so_rcv.sb_mb;
1604 orig_resid = 0;
1605 }
1606 if (m != NULL) {
1607 if ((flags & MSG_PEEK) == 0) {
1608 KASSERT(m->m_nextpkt == nextrecord,
1609 ("soreceive: post-control, nextrecord !sync"));
1610 if (nextrecord == NULL) {
1611 KASSERT(so->so_rcv.sb_mb == m,
1612 ("soreceive: post-control, sb_mb!=m"));
1613 KASSERT(so->so_rcv.sb_lastrecord == m,
1614 ("soreceive: post-control, lastrecord!=m"));
1615 }
1616 }
1617 type = m->m_type;
1618 if (type == MT_OOBDATA)
1619 flags |= MSG_OOB;
1620 } else {
1621 if ((flags & MSG_PEEK) == 0) {
1622 KASSERT(so->so_rcv.sb_mb == nextrecord,
1623 ("soreceive: sb_mb != nextrecord"));
1624 if (so->so_rcv.sb_mb == NULL) {
1625 KASSERT(so->so_rcv.sb_lastrecord == NULL,
1626 ("soreceive: sb_lastercord != NULL"));
1627 }
1628 }
1629 }
1630 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1631 SBLASTRECORDCHK(&so->so_rcv);
1632 SBLASTMBUFCHK(&so->so_rcv);
1633
1634 /*
1635 * Now continue to read any data mbufs off of the head of the socket
1636 * buffer until the read request is satisfied. Note that 'type' is
1637 * used to store the type of any mbuf reads that have happened so far
1638 * such that soreceive() can stop reading if the type changes, which
1639 * causes soreceive() to return only one of regular data and inline
1640 * out-of-band data in a single socket receive operation.
1641 */
1642 moff = 0;
1643 offset = 0;
1644 while (m != NULL && uio->uio_resid > 0 && error == 0) {
1645 /*
1646 * If the type of mbuf has changed since the last mbuf
1647 * examined ('type'), end the receive operation.
1648 */
1649 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1650 if (m->m_type == MT_OOBDATA) {
1651 if (type != MT_OOBDATA)
1652 break;
1653 } else if (type == MT_OOBDATA)
1654 break;
1655 else
1656 KASSERT(m->m_type == MT_DATA,
1657 ("m->m_type == %d", m->m_type));
1658 so->so_rcv.sb_state &= ~SBS_RCVATMARK;
1659 len = uio->uio_resid;
1660 if (so->so_oobmark && len > so->so_oobmark - offset)
1661 len = so->so_oobmark - offset;
1662 if (len > m->m_len - moff)
1663 len = m->m_len - moff;
1664 /*
1665 * If mp is set, just pass back the mbufs. Otherwise copy
1666 * them out via the uio, then free. Sockbuf must be
1667 * consistent here (points to current mbuf, it points to next
1668 * record) when we drop priority; we must note any additions
1669 * to the sockbuf when we block interrupts again.
1670 */
1671 if (mp == NULL) {
1672 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1673 SBLASTRECORDCHK(&so->so_rcv);
1674 SBLASTMBUFCHK(&so->so_rcv);
1675 SOCKBUF_UNLOCK(&so->so_rcv);
1676#ifdef ZERO_COPY_SOCKETS
1677 if (so_zero_copy_receive) {
1678 int disposable;
1679
1680 if ((m->m_flags & M_EXT)
1681 && (m->m_ext.ext_type == EXT_DISPOSABLE))
1682 disposable = 1;
1683 else
1684 disposable = 0;
1685
1686 error = uiomoveco(mtod(m, char *) + moff,
1687 (int)len, uio,
1688 disposable);
1689 } else
1690#endif /* ZERO_COPY_SOCKETS */
1691 error = uiomove(mtod(m, char *) + moff, (int)len, uio);
1692 SOCKBUF_LOCK(&so->so_rcv);
1693 if (error) {
1694 /*
1695 * The MT_SONAME mbuf has already been removed
1696 * from the record, so it is necessary to
1697 * remove the data mbufs, if any, to preserve
1698 * the invariant in the case of PR_ADDR that
1699 * requires MT_SONAME mbufs at the head of
1700 * each record.
1701 */
1702 if (m && pr->pr_flags & PR_ATOMIC &&
1703 ((flags & MSG_PEEK) == 0))
1704 (void)sbdroprecord_locked(&so->so_rcv);
1705 SOCKBUF_UNLOCK(&so->so_rcv);
1706 goto release;
1707 }
1708 } else
1709 uio->uio_resid -= len;
1710 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1711 if (len == m->m_len - moff) {
1712 if (m->m_flags & M_EOR)
1713 flags |= MSG_EOR;
1714 if (flags & MSG_PEEK) {
1715 m = m->m_next;
1716 moff = 0;
1717 } else {
1718 nextrecord = m->m_nextpkt;
1719 sbfree(&so->so_rcv, m);
1720 if (mp != NULL) {
1721 *mp = m;
1722 mp = &m->m_next;
1723 so->so_rcv.sb_mb = m = m->m_next;
1724 *mp = NULL;
1725 } else {
1726 so->so_rcv.sb_mb = m_free(m);
1727 m = so->so_rcv.sb_mb;
1728 }
1729 sockbuf_pushsync(&so->so_rcv, nextrecord);
1730 SBLASTRECORDCHK(&so->so_rcv);
1731 SBLASTMBUFCHK(&so->so_rcv);
1732 }
1733 } else {
1734 if (flags & MSG_PEEK)
1735 moff += len;
1736 else {
1737 if (mp != NULL) {
1738 int copy_flag;
1739
1740 if (flags & MSG_DONTWAIT)
1741 copy_flag = M_DONTWAIT;
1742 else
1743 copy_flag = M_WAIT;
1744 if (copy_flag == M_WAIT)
1745 SOCKBUF_UNLOCK(&so->so_rcv);
1746 *mp = m_copym(m, 0, len, copy_flag);
1747 if (copy_flag == M_WAIT)
1748 SOCKBUF_LOCK(&so->so_rcv);
1749 if (*mp == NULL) {
1750 /*
1751 * m_copym() couldn't
1752 * allocate an mbuf. Adjust
1753 * uio_resid back (it was
1754 * adjusted down by len
1755 * bytes, which we didn't end
1756 * up "copying" over).
1757 */
1758 uio->uio_resid += len;
1759 break;
1760 }
1761 }
1762 m->m_data += len;
1763 m->m_len -= len;
1764 so->so_rcv.sb_cc -= len;
1765 }
1766 }
1767 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1768 if (so->so_oobmark) {
1769 if ((flags & MSG_PEEK) == 0) {
1770 so->so_oobmark -= len;
1771 if (so->so_oobmark == 0) {
1772 so->so_rcv.sb_state |= SBS_RCVATMARK;
1773 break;
1774 }
1775 } else {
1776 offset += len;
1777 if (offset == so->so_oobmark)
1778 break;
1779 }
1780 }
1781 if (flags & MSG_EOR)
1782 break;
1783 /*
1784 * If the MSG_WAITALL flag is set (for non-atomic socket), we
1785 * must not quit until "uio->uio_resid == 0" or an error
1786 * termination. If a signal/timeout occurs, return with a
1787 * short count but without error. Keep sockbuf locked
1788 * against other readers.
1789 */
1790 while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1791 !sosendallatonce(so) && nextrecord == NULL) {
1792 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1793 if (so->so_error || so->so_rcv.sb_state & SBS_CANTRCVMORE)
1794 break;
1795 /*
1796 * Notify the protocol that some data has been
1797 * drained before blocking.
1798 */
1799 if (pr->pr_flags & PR_WANTRCVD) {
1800 SOCKBUF_UNLOCK(&so->so_rcv);
1801 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
1802 SOCKBUF_LOCK(&so->so_rcv);
1803 }
1804 SBLASTRECORDCHK(&so->so_rcv);
1805 SBLASTMBUFCHK(&so->so_rcv);
1806 error = sbwait(&so->so_rcv);
1807 if (error) {
1808 SOCKBUF_UNLOCK(&so->so_rcv);
1809 goto release;
1810 }
1811 m = so->so_rcv.sb_mb;
1812 if (m != NULL)
1813 nextrecord = m->m_nextpkt;
1814 }
1815 }
1816
1817 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1818 if (m != NULL && pr->pr_flags & PR_ATOMIC) {
1819 flags |= MSG_TRUNC;
1820 if ((flags & MSG_PEEK) == 0)
1821 (void) sbdroprecord_locked(&so->so_rcv);
1822 }
1823 if ((flags & MSG_PEEK) == 0) {
1824 if (m == NULL) {
1825 /*
1826 * First part is an inline SB_EMPTY_FIXUP(). Second
1827 * part makes sure sb_lastrecord is up-to-date if
1828 * there is still data in the socket buffer.
1829 */
1830 so->so_rcv.sb_mb = nextrecord;
1831 if (so->so_rcv.sb_mb == NULL) {
1832 so->so_rcv.sb_mbtail = NULL;
1833 so->so_rcv.sb_lastrecord = NULL;
1834 } else if (nextrecord->m_nextpkt == NULL)
1835 so->so_rcv.sb_lastrecord = nextrecord;
1836 }
1837 SBLASTRECORDCHK(&so->so_rcv);
1838 SBLASTMBUFCHK(&so->so_rcv);
1839 /*
1840 * If soreceive() is being done from the socket callback,
1841 * then don't need to generate ACK to peer to update window,
1842 * since ACK will be generated on return to TCP.
1843 */
1844 if (!(flags & MSG_SOCALLBCK) &&
1845 (pr->pr_flags & PR_WANTRCVD)) {
1846 SOCKBUF_UNLOCK(&so->so_rcv);
1847 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
1848 SOCKBUF_LOCK(&so->so_rcv);
1849 }
1850 }
1851 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1852 if (orig_resid == uio->uio_resid && orig_resid &&
1853 (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
1854 SOCKBUF_UNLOCK(&so->so_rcv);
1855 goto restart;
1856 }
1857 SOCKBUF_UNLOCK(&so->so_rcv);
1858
1859 if (flagsp != NULL)
1860 *flagsp |= flags;
1861release:
1862 sbunlock(&so->so_rcv);
1863 return (error);
1864}
1865
1866/*
1867 * Optimized version of soreceive() for simple datagram cases from userspace.
1868 * Unlike in the stream case, we're able to drop a datagram if copyout()
1869 * fails, and because we handle datagrams atomically, we don't need to use a
1870 * sleep lock to prevent I/O interlacing.
1871 */
1872int
1873soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio,
1874 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1875{
1876 struct mbuf *m, *m2;
1877 int flags, len, error;
1878 struct protosw *pr = so->so_proto;
1879 struct mbuf *nextrecord;
1880
1881 if (psa != NULL)
1882 *psa = NULL;
1883 if (controlp != NULL)
1884 *controlp = NULL;
1885 if (flagsp != NULL)
1886 flags = *flagsp &~ MSG_EOR;
1887 else
1888 flags = 0;
1889
1890 /*
1891 * For any complicated cases, fall back to the full
1892 * soreceive_generic().
1893 */
1894 if (mp0 != NULL || (flags & MSG_PEEK) || (flags & MSG_OOB))
1895 return (soreceive_generic(so, psa, uio, mp0, controlp,
1896 flagsp));
1897
1898 /*
1899 * Enforce restrictions on use.
1900 */
1901 KASSERT((pr->pr_flags & PR_WANTRCVD) == 0,
1902 ("soreceive_dgram: wantrcvd"));
1903 KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic"));
1904 KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0,
1905 ("soreceive_dgram: SBS_RCVATMARK"));
1906 KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0,
1907 ("soreceive_dgram: P_CONNREQUIRED"));
1908
1909 /*
1910 * Loop blocking while waiting for a datagram.
1911 */
1912 SOCKBUF_LOCK(&so->so_rcv);
1913 while ((m = so->so_rcv.sb_mb) == NULL) {
1914 KASSERT(so->so_rcv.sb_cc == 0,
1915 ("soreceive_dgram: sb_mb NULL but sb_cc %u",
1916 so->so_rcv.sb_cc));
1917 if (so->so_error) {
1918 error = so->so_error;
1919 so->so_error = 0;
1920 SOCKBUF_UNLOCK(&so->so_rcv);
1921 return (error);
1922 }
1923 if (so->so_rcv.sb_state & SBS_CANTRCVMORE ||
1924 uio->uio_resid == 0) {
1925 SOCKBUF_UNLOCK(&so->so_rcv);
1926 return (0);
1927 }
1928 if ((so->so_state & SS_NBIO) ||
1929 (flags & (MSG_DONTWAIT|MSG_NBIO))) {
1930 SOCKBUF_UNLOCK(&so->so_rcv);
1931 return (EWOULDBLOCK);
1932 }
1933 SBLASTRECORDCHK(&so->so_rcv);
1934 SBLASTMBUFCHK(&so->so_rcv);
1935 error = sbwait(&so->so_rcv);
1936 if (error) {
1937 SOCKBUF_UNLOCK(&so->so_rcv);
1938 return (error);
1939 }
1940 }
1941 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1942
1943 if (uio->uio_td)
1944 uio->uio_td->td_ru.ru_msgrcv++;
1945 SBLASTRECORDCHK(&so->so_rcv);
1946 SBLASTMBUFCHK(&so->so_rcv);
1947 nextrecord = m->m_nextpkt;
1948 if (nextrecord == NULL) {
1949 KASSERT(so->so_rcv.sb_lastrecord == m,
1950 ("soreceive_dgram: lastrecord != m"));
1951 }
1952
1953 KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord,
1954 ("soreceive_dgram: m_nextpkt != nextrecord"));
1955
1956 /*
1957 * Pull 'm' and its chain off the front of the packet queue.
1958 */
1959 so->so_rcv.sb_mb = NULL;
1960 sockbuf_pushsync(&so->so_rcv, nextrecord);
1961
1962 /*
1963 * Walk 'm's chain and free that many bytes from the socket buffer.
1964 */
1965 for (m2 = m; m2 != NULL; m2 = m2->m_next)
1966 sbfree(&so->so_rcv, m2);
1967
1968 /*
1969 * Do a few last checks before we let go of the lock.
1970 */
1971 SBLASTRECORDCHK(&so->so_rcv);
1972 SBLASTMBUFCHK(&so->so_rcv);
1973 SOCKBUF_UNLOCK(&so->so_rcv);
1974
1975 if (pr->pr_flags & PR_ADDR) {
1976 KASSERT(m->m_type == MT_SONAME,
1977 ("m->m_type == %d", m->m_type));
1978 if (psa != NULL)
1979 *psa = sodupsockaddr(mtod(m, struct sockaddr *),
1980 M_NOWAIT);
1981 m = m_free(m);
1982 }
1983 if (m == NULL) {
1984 /* XXXRW: Can this happen? */
1985 return (0);
1986 }
1987
1988 /*
1989 * Packet to copyout() is now in 'm' and it is disconnected from the
1990 * queue.
1991 *
1992 * Process one or more MT_CONTROL mbufs present before any data mbufs
1993 * in the first mbuf chain on the socket buffer. We call into the
1994 * protocol to perform externalization (or freeing if controlp ==
1995 * NULL).
1996 */
1997 if (m->m_type == MT_CONTROL) {
1998 struct mbuf *cm = NULL, *cmn;
1999 struct mbuf **cme = &cm;
2000
2001 do {
2002 m2 = m->m_next;
2003 m->m_next = NULL;
2004 *cme = m;
2005 cme = &(*cme)->m_next;
2006 m = m2;
2007 } while (m != NULL && m->m_type == MT_CONTROL);
2008 while (cm != NULL) {
2009 cmn = cm->m_next;
2010 cm->m_next = NULL;
2011 if (pr->pr_domain->dom_externalize != NULL) {
2012 error = (*pr->pr_domain->dom_externalize)
2013 (cm, controlp);
2014 } else if (controlp != NULL)
2015 *controlp = cm;
2016 else
2017 m_freem(cm);
2018 if (controlp != NULL) {
2019 while (*controlp != NULL)
2020 controlp = &(*controlp)->m_next;
2021 }
2022 cm = cmn;
2023 }
2024 }
2025 KASSERT(m->m_type == MT_DATA, ("soreceive_dgram: !data"));
2026
2027 while (m != NULL && uio->uio_resid > 0) {
2028 len = uio->uio_resid;
2029 if (len > m->m_len)
2030 len = m->m_len;
2031 error = uiomove(mtod(m, char *), (int)len, uio);
2032 if (error) {
2033 m_freem(m);
2034 return (error);
2035 }
2036 m = m_free(m);
2037 }
2038 if (m != NULL)
2039 flags |= MSG_TRUNC;
2040 m_freem(m);
2041 if (flagsp != NULL)
2042 *flagsp |= flags;
2043 return (0);
2044}
2045
2046int
2047soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
2048 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2049{
2050
2051 return (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio, mp0,
2052 controlp, flagsp));
2053}
2054
2055int
2056soshutdown(struct socket *so, int how)
2057{
2058 struct protosw *pr = so->so_proto;
2059 int error;
2060
2061 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
2062 return (EINVAL);
2063 if (pr->pr_usrreqs->pru_flush != NULL) {
2064 (*pr->pr_usrreqs->pru_flush)(so, how);
2065 }
2066 if (how != SHUT_WR)
2067 sorflush(so);
2068 if (how != SHUT_RD) {
2069 CURVNET_SET(so->so_vnet);
2070 error = (*pr->pr_usrreqs->pru_shutdown)(so);
2071 CURVNET_RESTORE();
2072 return (error);
2073 }
2074 return (0);
2075}
2076
2077void
2078sorflush(struct socket *so)
2079{
2080 struct sockbuf *sb = &so->so_rcv;
2081 struct protosw *pr = so->so_proto;
2082 struct sockbuf asb;
2083
2084 /*
2085 * In order to avoid calling dom_dispose with the socket buffer mutex
2086 * held, and in order to generally avoid holding the lock for a long
2087 * time, we make a copy of the socket buffer and clear the original
2088 * (except locks, state). The new socket buffer copy won't have
2089 * initialized locks so we can only call routines that won't use or
2090 * assert those locks.
2091 *
2092 * Dislodge threads currently blocked in receive and wait to acquire
2093 * a lock against other simultaneous readers before clearing the
2094 * socket buffer. Don't let our acquire be interrupted by a signal
2095 * despite any existing socket disposition on interruptable waiting.
2096 */
2097 CURVNET_SET(so->so_vnet);
2098 socantrcvmore(so);
2099 (void) sblock(sb, SBL_WAIT | SBL_NOINTR);
2100
2101 /*
2102 * Invalidate/clear most of the sockbuf structure, but leave selinfo
2103 * and mutex data unchanged.
2104 */
2105 SOCKBUF_LOCK(sb);
2106 bzero(&asb, offsetof(struct sockbuf, sb_startzero));
2107 bcopy(&sb->sb_startzero, &asb.sb_startzero,
2108 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2109 bzero(&sb->sb_startzero,
2110 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2111 SOCKBUF_UNLOCK(sb);
2112 sbunlock(sb);
2113
2114 /*
2115 * Dispose of special rights and flush the socket buffer. Don't call
2116 * any unsafe routines (that rely on locks being initialized) on asb.
2117 */
2118 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
2119 (*pr->pr_domain->dom_dispose)(asb.sb_mb);
2120 sbrelease_internal(&asb, so);
2121 CURVNET_RESTORE();
2122}
2123
2124/*
2125 * Perhaps this routine, and sooptcopyout(), below, ought to come in an
2126 * additional variant to handle the case where the option value needs to be
2127 * some kind of integer, but not a specific size. In addition to their use
2128 * here, these functions are also called by the protocol-level pr_ctloutput()
2129 * routines.
2130 */
2131int
2132sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2133{
2134 size_t valsize;
2135
2136 /*
2137 * If the user gives us more than we wanted, we ignore it, but if we
2138 * don't get the minimum length the caller wants, we return EINVAL.
2139 * On success, sopt->sopt_valsize is set to however much we actually
2140 * retrieved.
2141 */
2142 if ((valsize = sopt->sopt_valsize) < minlen)
2143 return EINVAL;
2144 if (valsize > len)
2145 sopt->sopt_valsize = valsize = len;
2146
2147 if (sopt->sopt_td != NULL)
2148 return (copyin(sopt->sopt_val, buf, valsize));
2149
2150 bcopy(sopt->sopt_val, buf, valsize);
2151 return (0);
2152}
2153
2154/*
2155 * Kernel version of setsockopt(2).
2156 *
2157 * XXX: optlen is size_t, not socklen_t
2158 */
2159int
2160so_setsockopt(struct socket *so, int level, int optname, void *optval,
2161 size_t optlen)
2162{
2163 struct sockopt sopt;
2164
2165 sopt.sopt_level = level;
2166 sopt.sopt_name = optname;
2167 sopt.sopt_dir = SOPT_SET;
2168 sopt.sopt_val = optval;
2169 sopt.sopt_valsize = optlen;
2170 sopt.sopt_td = NULL;
2171 return (sosetopt(so, &sopt));
2172}
2173
2174int
2175sosetopt(struct socket *so, struct sockopt *sopt)
2176{
2177 int error, optval;
2178 struct linger l;
2179 struct timeval tv;
2180 u_long val;
2181#ifdef MAC
2182 struct mac extmac;
2183#endif
2184
2185 error = 0;
2186 if (sopt->sopt_level != SOL_SOCKET) {
2187 if (so->so_proto && so->so_proto->pr_ctloutput)
2188 return ((*so->so_proto->pr_ctloutput)
2189 (so, sopt));
2190 error = ENOPROTOOPT;
2191 } else {
2192 switch (sopt->sopt_name) {
2193#ifdef INET
2194 case SO_ACCEPTFILTER:
2195 error = do_setopt_accept_filter(so, sopt);
2196 if (error)
2197 goto bad;
2198 break;
2199#endif
2200 case SO_LINGER:
2201 error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
2202 if (error)
2203 goto bad;
2204
2205 SOCK_LOCK(so);
2206 so->so_linger = l.l_linger;
2207 if (l.l_onoff)
2208 so->so_options |= SO_LINGER;
2209 else
2210 so->so_options &= ~SO_LINGER;
2211 SOCK_UNLOCK(so);
2212 break;
2213
2214 case SO_DEBUG:
2215 case SO_KEEPALIVE:
2216 case SO_DONTROUTE:
2217 case SO_USELOOPBACK:
2218 case SO_BROADCAST:
2219 case SO_REUSEADDR:
2220 case SO_REUSEPORT:
2221 case SO_OOBINLINE:
2222 case SO_TIMESTAMP:
2223 case SO_BINTIME:
2224 case SO_NOSIGPIPE:
2225 case SO_NO_DDP:
2226 case SO_NO_OFFLOAD:
2227 error = sooptcopyin(sopt, &optval, sizeof optval,
2228 sizeof optval);
2229 if (error)
2230 goto bad;
2231 SOCK_LOCK(so);
2232 if (optval)
2233 so->so_options |= sopt->sopt_name;
2234 else
2235 so->so_options &= ~sopt->sopt_name;
2236 SOCK_UNLOCK(so);
2237 break;
2238
2239 case SO_SETFIB:
2240 error = sooptcopyin(sopt, &optval, sizeof optval,
2241 sizeof optval);
2242 if (optval < 1 || optval > rt_numfibs) {
2243 error = EINVAL;
2244 goto bad;
2245 }
2246 if ((so->so_proto->pr_domain->dom_family == PF_INET) ||
2247 (so->so_proto->pr_domain->dom_family == PF_ROUTE)) {
2248 so->so_fibnum = optval;
2249 /* Note: ignore error */
2250 if (so->so_proto && so->so_proto->pr_ctloutput)
2251 (*so->so_proto->pr_ctloutput)(so, sopt);
2252 } else {
2253 so->so_fibnum = 0;
2254 }
2255 break;
2256 case SO_SNDBUF:
2257 case SO_RCVBUF:
2258 case SO_SNDLOWAT:
2259 case SO_RCVLOWAT:
2260 error = sooptcopyin(sopt, &optval, sizeof optval,
2261 sizeof optval);
2262 if (error)
2263 goto bad;
2264
2265 /*
2266 * Values < 1 make no sense for any of these options,
2267 * so disallow them.
2268 */
2269 if (optval < 1) {
2270 error = EINVAL;
2271 goto bad;
2272 }
2273
2274 switch (sopt->sopt_name) {
2275 case SO_SNDBUF:
2276 case SO_RCVBUF:
2277 if (sbreserve(sopt->sopt_name == SO_SNDBUF ?
2278 &so->so_snd : &so->so_rcv, (u_long)optval,
2279 so, curthread) == 0) {
2280 error = ENOBUFS;
2281 goto bad;
2282 }
2283 (sopt->sopt_name == SO_SNDBUF ? &so->so_snd :
2284 &so->so_rcv)->sb_flags &= ~SB_AUTOSIZE;
2285 break;
2286
2287 /*
2288 * Make sure the low-water is never greater than the
2289 * high-water.
2290 */
2291 case SO_SNDLOWAT:
2292 SOCKBUF_LOCK(&so->so_snd);
2293 so->so_snd.sb_lowat =
2294 (optval > so->so_snd.sb_hiwat) ?
2295 so->so_snd.sb_hiwat : optval;
2296 SOCKBUF_UNLOCK(&so->so_snd);
2297 break;
2298 case SO_RCVLOWAT:
2299 SOCKBUF_LOCK(&so->so_rcv);
2300 so->so_rcv.sb_lowat =
2301 (optval > so->so_rcv.sb_hiwat) ?
2302 so->so_rcv.sb_hiwat : optval;
2303 SOCKBUF_UNLOCK(&so->so_rcv);
2304 break;
2305 }
2306 break;
2307
2308 case SO_SNDTIMEO:
2309 case SO_RCVTIMEO:
2310#ifdef COMPAT_IA32
2311 if (SV_CURPROC_FLAG(SV_ILP32)) {
2312 struct timeval32 tv32;
2313
2314 error = sooptcopyin(sopt, &tv32, sizeof tv32,
2315 sizeof tv32);
2316 CP(tv32, tv, tv_sec);
2317 CP(tv32, tv, tv_usec);
2318 } else
2319#endif
2320 error = sooptcopyin(sopt, &tv, sizeof tv,
2321 sizeof tv);
2322 if (error)
2323 goto bad;
2324
2325 /* assert(hz > 0); */
2326 if (tv.tv_sec < 0 || tv.tv_sec > INT_MAX / hz ||
2327 tv.tv_usec < 0 || tv.tv_usec >= 1000000) {
2328 error = EDOM;
2329 goto bad;
2330 }
2331 /* assert(tick > 0); */
2332 /* assert(ULONG_MAX - INT_MAX >= 1000000); */
2333 val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / tick;
2334 if (val > INT_MAX) {
2335 error = EDOM;
2336 goto bad;
2337 }
2338 if (val == 0 && tv.tv_usec != 0)
2339 val = 1;
2340
2341 switch (sopt->sopt_name) {
2342 case SO_SNDTIMEO:
2343 so->so_snd.sb_timeo = val;
2344 break;
2345 case SO_RCVTIMEO:
2346 so->so_rcv.sb_timeo = val;
2347 break;
2348 }
2349 break;
2350
2351 case SO_LABEL:
2352#ifdef MAC
2353 error = sooptcopyin(sopt, &extmac, sizeof extmac,
2354 sizeof extmac);
2355 if (error)
2356 goto bad;
2357 error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
2358 so, &extmac);
2359#else
2360 error = EOPNOTSUPP;
2361#endif
2362 break;
2363
2364 default:
2365 error = ENOPROTOOPT;
2366 break;
2367 }
2368 if (error == 0 && so->so_proto != NULL &&
2369 so->so_proto->pr_ctloutput != NULL) {
2370 (void) ((*so->so_proto->pr_ctloutput)
2371 (so, sopt));
2372 }
2373 }
2374bad:
2375 return (error);
2376}
2377
2378/*
2379 * Helper routine for getsockopt.
2380 */
2381int
2382sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
2383{
2384 int error;
2385 size_t valsize;
2386
2387 error = 0;
2388
2389 /*
2390 * Documented get behavior is that we always return a value, possibly
2391 * truncated to fit in the user's buffer. Traditional behavior is
2392 * that we always tell the user precisely how much we copied, rather
2393 * than something useful like the total amount we had available for
2394 * her. Note that this interface is not idempotent; the entire
2395 * answer must generated ahead of time.
2396 */
2397 valsize = min(len, sopt->sopt_valsize);
2398 sopt->sopt_valsize = valsize;
2399 if (sopt->sopt_val != NULL) {
2400 if (sopt->sopt_td != NULL)
2401 error = copyout(buf, sopt->sopt_val, valsize);
2402 else
2403 bcopy(buf, sopt->sopt_val, valsize);
2404 }
2405 return (error);
2406}
2407
2408int
2409sogetopt(struct socket *so, struct sockopt *sopt)
2410{
2411 int error, optval;
2412 struct linger l;
2413 struct timeval tv;
2414#ifdef MAC
2415 struct mac extmac;
2416#endif
2417
2418 error = 0;
2419 if (sopt->sopt_level != SOL_SOCKET) {
2420 if (so->so_proto && so->so_proto->pr_ctloutput) {
2421 return ((*so->so_proto->pr_ctloutput)
2422 (so, sopt));
2423 } else
2424 return (ENOPROTOOPT);
2425 } else {
2426 switch (sopt->sopt_name) {
2427#ifdef INET
2428 case SO_ACCEPTFILTER:
2429 error = do_getopt_accept_filter(so, sopt);
2430 break;
2431#endif
2432 case SO_LINGER:
2433 SOCK_LOCK(so);
2434 l.l_onoff = so->so_options & SO_LINGER;
2435 l.l_linger = so->so_linger;
2436 SOCK_UNLOCK(so);
2437 error = sooptcopyout(sopt, &l, sizeof l);
2438 break;
2439
2440 case SO_USELOOPBACK:
2441 case SO_DONTROUTE:
2442 case SO_DEBUG:
2443 case SO_KEEPALIVE:
2444 case SO_REUSEADDR:
2445 case SO_REUSEPORT:
2446 case SO_BROADCAST:
2447 case SO_OOBINLINE:
2448 case SO_ACCEPTCONN:
2449 case SO_TIMESTAMP:
2450 case SO_BINTIME:
2451 case SO_NOSIGPIPE:
2452 optval = so->so_options & sopt->sopt_name;
2453integer:
2454 error = sooptcopyout(sopt, &optval, sizeof optval);
2455 break;
2456
2457 case SO_TYPE:
2458 optval = so->so_type;
2459 goto integer;
2460
2461 case SO_ERROR:
2462 SOCK_LOCK(so);
2463 optval = so->so_error;
2464 so->so_error = 0;
2465 SOCK_UNLOCK(so);
2466 goto integer;
2467
2468 case SO_SNDBUF:
2469 optval = so->so_snd.sb_hiwat;
2470 goto integer;
2471
2472 case SO_RCVBUF:
2473 optval = so->so_rcv.sb_hiwat;
2474 goto integer;
2475
2476 case SO_SNDLOWAT:
2477 optval = so->so_snd.sb_lowat;
2478 goto integer;
2479
2480 case SO_RCVLOWAT:
2481 optval = so->so_rcv.sb_lowat;
2482 goto integer;
2483
2484 case SO_SNDTIMEO:
2485 case SO_RCVTIMEO:
2486 optval = (sopt->sopt_name == SO_SNDTIMEO ?
2487 so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
2488
2489 tv.tv_sec = optval / hz;
2490 tv.tv_usec = (optval % hz) * tick;
2491#ifdef COMPAT_IA32
2492 if (SV_CURPROC_FLAG(SV_ILP32)) {
2493 struct timeval32 tv32;
2494
2495 CP(tv, tv32, tv_sec);
2496 CP(tv, tv32, tv_usec);
2497 error = sooptcopyout(sopt, &tv32, sizeof tv32);
2498 } else
2499#endif
2500 error = sooptcopyout(sopt, &tv, sizeof tv);
2501 break;
2502
2503 case SO_LABEL:
2504#ifdef MAC
2505 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
2506 sizeof(extmac));
2507 if (error)
2508 return (error);
2509 error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
2510 so, &extmac);
2511 if (error)
2512 return (error);
2513 error = sooptcopyout(sopt, &extmac, sizeof extmac);
2514#else
2515 error = EOPNOTSUPP;
2516#endif
2517 break;
2518
2519 case SO_PEERLABEL:
2520#ifdef MAC
2521 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
2522 sizeof(extmac));
2523 if (error)
2524 return (error);
2525 error = mac_getsockopt_peerlabel(
2526 sopt->sopt_td->td_ucred, so, &extmac);
2527 if (error)
2528 return (error);
2529 error = sooptcopyout(sopt, &extmac, sizeof extmac);
2530#else
2531 error = EOPNOTSUPP;
2532#endif
2533 break;
2534
2535 case SO_LISTENQLIMIT:
2536 optval = so->so_qlimit;
2537 goto integer;
2538
2539 case SO_LISTENQLEN:
2540 optval = so->so_qlen;
2541 goto integer;
2542
2543 case SO_LISTENINCQLEN:
2544 optval = so->so_incqlen;
2545 goto integer;
2546
2547 default:
2548 error = ENOPROTOOPT;
2549 break;
2550 }
2551 return (error);
2552 }
2553}
2554
2555/* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */
2556int
2557soopt_getm(struct sockopt *sopt, struct mbuf **mp)
2558{
2559 struct mbuf *m, *m_prev;
2560 int sopt_size = sopt->sopt_valsize;
2561
2562 MGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT, MT_DATA);
2563 if (m == NULL)
2564 return ENOBUFS;
2565 if (sopt_size > MLEN) {
2566 MCLGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT);
2567 if ((m->m_flags & M_EXT) == 0) {
2568 m_free(m);
2569 return ENOBUFS;
2570 }
2571 m->m_len = min(MCLBYTES, sopt_size);
2572 } else {
2573 m->m_len = min(MLEN, sopt_size);
2574 }
2575 sopt_size -= m->m_len;
2576 *mp = m;
2577 m_prev = m;
2578
2579 while (sopt_size) {
2580 MGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT, MT_DATA);
2581 if (m == NULL) {
2582 m_freem(*mp);
2583 return ENOBUFS;
2584 }
2585 if (sopt_size > MLEN) {
2586 MCLGET(m, sopt->sopt_td != NULL ? M_WAIT :
2587 M_DONTWAIT);
2588 if ((m->m_flags & M_EXT) == 0) {
2589 m_freem(m);
2590 m_freem(*mp);
2591 return ENOBUFS;
2592 }
2593 m->m_len = min(MCLBYTES, sopt_size);
2594 } else {
2595 m->m_len = min(MLEN, sopt_size);
2596 }
2597 sopt_size -= m->m_len;
2598 m_prev->m_next = m;
2599 m_prev = m;
2600 }
2601 return (0);
2602}
2603
2604/* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */
2605int
2606soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
2607{
2608 struct mbuf *m0 = m;
2609
2610 if (sopt->sopt_val == NULL)
2611 return (0);
2612 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
2613 if (sopt->sopt_td != NULL) {
2614 int error;
2615
2616 error = copyin(sopt->sopt_val, mtod(m, char *),
2617 m->m_len);
2618 if (error != 0) {
2619 m_freem(m0);
2620 return(error);
2621 }
2622 } else
2623 bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
2624 sopt->sopt_valsize -= m->m_len;
2625 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
2626 m = m->m_next;
2627 }
2628 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
2629 panic("ip6_sooptmcopyin");
2630 return (0);
2631}
2632
2633/* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */
2634int
2635soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
2636{
2637 struct mbuf *m0 = m;
2638 size_t valsize = 0;
2639
2640 if (sopt->sopt_val == NULL)
2641 return (0);
2642 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
2643 if (sopt->sopt_td != NULL) {
2644 int error;
2645
2646 error = copyout(mtod(m, char *), sopt->sopt_val,
2647 m->m_len);
2648 if (error != 0) {
2649 m_freem(m0);
2650 return(error);
2651 }
2652 } else
2653 bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
2654 sopt->sopt_valsize -= m->m_len;
2655 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
2656 valsize += m->m_len;
2657 m = m->m_next;
2658 }
2659 if (m != NULL) {
2660 /* enough soopt buffer should be given from user-land */
2661 m_freem(m0);
2662 return(EINVAL);
2663 }
2664 sopt->sopt_valsize = valsize;
2665 return (0);
2666}
2667
2668/*
2669 * sohasoutofband(): protocol notifies socket layer of the arrival of new
2670 * out-of-band data, which will then notify socket consumers.
2671 */
2672void
2673sohasoutofband(struct socket *so)
2674{
2675
2676 if (so->so_sigio != NULL)
2677 pgsigio(&so->so_sigio, SIGURG, 0);
2678 selwakeuppri(&so->so_rcv.sb_sel, PSOCK);
2679}
2680
2681int
2682sopoll(struct socket *so, int events, struct ucred *active_cred,
2683 struct thread *td)
2684{
2685
2686 return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred,
2687 td));
2688}
2689
2690int
2691sopoll_generic(struct socket *so, int events, struct ucred *active_cred,
2692 struct thread *td)
2693{
2694 int revents = 0;
2695
2696 SOCKBUF_LOCK(&so->so_snd);
2697 SOCKBUF_LOCK(&so->so_rcv);
2698 if (events & (POLLIN | POLLRDNORM))
2699 if (soreadable(so))
2700 revents |= events & (POLLIN | POLLRDNORM);
2701
2702 if (events & POLLINIGNEOF)
2703 if (so->so_rcv.sb_cc >= so->so_rcv.sb_lowat ||
2704 !TAILQ_EMPTY(&so->so_comp) || so->so_error)
2705 revents |= POLLINIGNEOF;
2706
2707 if (events & (POLLOUT | POLLWRNORM))
2708 if (sowriteable(so))
2709 revents |= events & (POLLOUT | POLLWRNORM);
2710
2711 if (events & (POLLPRI | POLLRDBAND))
2712 if (so->so_oobmark || (so->so_rcv.sb_state & SBS_RCVATMARK))
2713 revents |= events & (POLLPRI | POLLRDBAND);
2714
2715 if (revents == 0) {
2716 if (events &
2717 (POLLIN | POLLINIGNEOF | POLLPRI | POLLRDNORM |
2718 POLLRDBAND)) {
2719 selrecord(td, &so->so_rcv.sb_sel);
2720 so->so_rcv.sb_flags |= SB_SEL;
2721 }
2722
2723 if (events & (POLLOUT | POLLWRNORM)) {
2724 selrecord(td, &so->so_snd.sb_sel);
2725 so->so_snd.sb_flags |= SB_SEL;
2726 }
2727 }
2728
2729 SOCKBUF_UNLOCK(&so->so_rcv);
2730 SOCKBUF_UNLOCK(&so->so_snd);
2731 return (revents);
2732}
2733
2734int
2735soo_kqfilter(struct file *fp, struct knote *kn)
2736{
2737 struct socket *so = kn->kn_fp->f_data;
2738 struct sockbuf *sb;
2739
2740 switch (kn->kn_filter) {
2741 case EVFILT_READ:
2742 if (so->so_options & SO_ACCEPTCONN)
2743 kn->kn_fop = &solisten_filtops;
2744 else
2745 kn->kn_fop = &soread_filtops;
2746 sb = &so->so_rcv;
2747 break;
2748 case EVFILT_WRITE:
2749 kn->kn_fop = &sowrite_filtops;
2750 sb = &so->so_snd;
2751 break;
2752 default:
2753 return (EINVAL);
2754 }
2755
2756 SOCKBUF_LOCK(sb);
2757 knlist_add(&sb->sb_sel.si_note, kn, 1);
2758 sb->sb_flags |= SB_KNOTE;
2759 SOCKBUF_UNLOCK(sb);
2760 return (0);
2761}
2762
2763/*
2764 * Some routines that return EOPNOTSUPP for entry points that are not
2765 * supported by a protocol. Fill in as needed.
2766 */
2767int
2768pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
2769{
2770
2771 return EOPNOTSUPP;
2772}
2773
2774int
2775pru_attach_notsupp(struct socket *so, int proto, struct thread *td)
2776{
2777
2778 return EOPNOTSUPP;
2779}
2780
2781int
2782pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
2783{
2784
2785 return EOPNOTSUPP;
2786}
2787
2788int
2789pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
2790{
2791
2792 return EOPNOTSUPP;
2793}
2794
2795int
2796pru_connect2_notsupp(struct socket *so1, struct socket *so2)
2797{
2798
2799 return EOPNOTSUPP;
2800}
2801
2802int
2803pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
2804 struct ifnet *ifp, struct thread *td)
2805{
2806
2807 return EOPNOTSUPP;
2808}
2809
2810int
2811pru_disconnect_notsupp(struct socket *so)
2812{
2813
2814 return EOPNOTSUPP;
2815}
2816
2817int
2818pru_listen_notsupp(struct socket *so, int backlog, struct thread *td)
2819{
2820
2821 return EOPNOTSUPP;
2822}
2823
2824int
2825pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam)
2826{
2827
2828 return EOPNOTSUPP;
2829}
2830
2831int
2832pru_rcvd_notsupp(struct socket *so, int flags)
2833{
2834
2835 return EOPNOTSUPP;
2836}
2837
2838int
2839pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
2840{
2841
2842 return EOPNOTSUPP;
2843}
2844
2845int
2846pru_send_notsupp(struct socket *so, int flags, struct mbuf *m,
2847 struct sockaddr *addr, struct mbuf *control, struct thread *td)
2848{
2849
2850 return EOPNOTSUPP;
2851}
2852
2853/*
2854 * This isn't really a ``null'' operation, but it's the default one and
2855 * doesn't do anything destructive.
2856 */
2857int
2858pru_sense_null(struct socket *so, struct stat *sb)
2859{
2860
2861 sb->st_blksize = so->so_snd.sb_hiwat;
2862 return 0;
2863}
2864
2865int
2866pru_shutdown_notsupp(struct socket *so)
2867{
2868
2869 return EOPNOTSUPP;
2870}
2871
2872int
2873pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam)
2874{
2875
2876 return EOPNOTSUPP;
2877}
2878
2879int
2880pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio,
2881 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
2882{
2883
2884 return EOPNOTSUPP;
2885}
2886
2887int
2888pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr,
2889 struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2890{
2891
2892 return EOPNOTSUPP;
2893}
2894
2895int
2896pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred,
2897 struct thread *td)
2898{
2899
2900 return EOPNOTSUPP;
2901}
2902
2903static void
2904filt_sordetach(struct knote *kn)
2905{
2906 struct socket *so = kn->kn_fp->f_data;
2907
2908 SOCKBUF_LOCK(&so->so_rcv);
2909 knlist_remove(&so->so_rcv.sb_sel.si_note, kn, 1);
2910 if (knlist_empty(&so->so_rcv.sb_sel.si_note))
2911 so->so_rcv.sb_flags &= ~SB_KNOTE;
2912 SOCKBUF_UNLOCK(&so->so_rcv);
2913}
2914
2915/*ARGSUSED*/
2916static int
2917filt_soread(struct knote *kn, long hint)
2918{
2919 struct socket *so;
2920
2921 so = kn->kn_fp->f_data;
2922 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2923
2924 kn->kn_data = so->so_rcv.sb_cc - so->so_rcv.sb_ctl;
2925 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2926 kn->kn_flags |= EV_EOF;
2927 kn->kn_fflags = so->so_error;
2928 return (1);
2929 } else if (so->so_error) /* temporary udp error */
2930 return (1);
2931 else if (kn->kn_sfflags & NOTE_LOWAT)
2932 return (kn->kn_data >= kn->kn_sdata);
2933 else
2934 return (so->so_rcv.sb_cc >= so->so_rcv.sb_lowat);
2935}
2936
2937static void
2938filt_sowdetach(struct knote *kn)
2939{
2940 struct socket *so = kn->kn_fp->f_data;
2941
2942 SOCKBUF_LOCK(&so->so_snd);
2943 knlist_remove(&so->so_snd.sb_sel.si_note, kn, 1);
2944 if (knlist_empty(&so->so_snd.sb_sel.si_note))
2945 so->so_snd.sb_flags &= ~SB_KNOTE;
2946 SOCKBUF_UNLOCK(&so->so_snd);
2947}
2948
2949/*ARGSUSED*/
2950static int
2951filt_sowrite(struct knote *kn, long hint)
2952{
2953 struct socket *so;
2954
2955 so = kn->kn_fp->f_data;
2956 SOCKBUF_LOCK_ASSERT(&so->so_snd);
2957 kn->kn_data = sbspace(&so->so_snd);
2958 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
2959 kn->kn_flags |= EV_EOF;
2960 kn->kn_fflags = so->so_error;
2961 return (1);
2962 } else if (so->so_error) /* temporary udp error */
2963 return (1);
2964 else if (((so->so_state & SS_ISCONNECTED) == 0) &&
2965 (so->so_proto->pr_flags & PR_CONNREQUIRED))
2966 return (0);
2967 else if (kn->kn_sfflags & NOTE_LOWAT)
2968 return (kn->kn_data >= kn->kn_sdata);
2969 else
2970 return (kn->kn_data >= so->so_snd.sb_lowat);
2971}
2972
2973/*ARGSUSED*/
2974static int
2975filt_solisten(struct knote *kn, long hint)
2976{
2977 struct socket *so = kn->kn_fp->f_data;
2978
2979 kn->kn_data = so->so_qlen;
2980 return (! TAILQ_EMPTY(&so->so_comp));
2981}
2982
2983int
2984socheckuid(struct socket *so, uid_t uid)
2985{
2986
2987 if (so == NULL)
2988 return (EPERM);
2989 if (so->so_cred->cr_uid != uid)
2990 return (EPERM);
2991 return (0);
2992}
2993
2994static int
2995sysctl_somaxconn(SYSCTL_HANDLER_ARGS)
2996{
2997 int error;
2998 int val;
2999
3000 val = somaxconn;
3001 error = sysctl_handle_int(oidp, &val, 0, req);
3002 if (error || !req->newptr )
3003 return (error);
3004
3005 if (val < 1 || val > USHRT_MAX)
3006 return (EINVAL);
3007
3008 somaxconn = val;
3009 return (0);
3010}
3011
3012/*
3013 * These functions are used by protocols to notify the socket layer (and its
3014 * consumers) of state changes in the sockets driven by protocol-side events.
3015 */
3016
3017/*
3018 * Procedures to manipulate state flags of socket and do appropriate wakeups.
3019 *
3020 * Normal sequence from the active (originating) side is that
3021 * soisconnecting() is called during processing of connect() call, resulting
3022 * in an eventual call to soisconnected() if/when the connection is
3023 * established. When the connection is torn down soisdisconnecting() is
3024 * called during processing of disconnect() call, and soisdisconnected() is
3025 * called when the connection to the peer is totally severed. The semantics
3026 * of these routines are such that connectionless protocols can call
3027 * soisconnected() and soisdisconnected() only, bypassing the in-progress
3028 * calls when setting up a ``connection'' takes no time.
3029 *
3030 * From the passive side, a socket is created with two queues of sockets:
3031 * so_incomp for connections in progress and so_comp for connections already
3032 * made and awaiting user acceptance. As a protocol is preparing incoming
3033 * connections, it creates a socket structure queued on so_incomp by calling
3034 * sonewconn(). When the connection is established, soisconnected() is
3035 * called, and transfers the socket structure to so_comp, making it available
3036 * to accept().
3037 *
3038 * If a socket is closed with sockets on either so_incomp or so_comp, these
3039 * sockets are dropped.
3040 *
3041 * If higher-level protocols are implemented in the kernel, the wakeups done
3042 * here will sometimes cause software-interrupt process scheduling.
3043 */
3044void
3045soisconnecting(struct socket *so)
3046{
3047
3048 SOCK_LOCK(so);
3049 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
3050 so->so_state |= SS_ISCONNECTING;
3051 SOCK_UNLOCK(so);
3052}
3053
3054void
3055soisconnected(struct socket *so)
3056{
| 99
100#include "opt_inet.h"
101#include "opt_inet6.h"
102#include "opt_mac.h"
103#include "opt_zero.h"
104#include "opt_compat.h"
105
106#include <sys/param.h>
107#include <sys/systm.h>
108#include <sys/fcntl.h>
109#include <sys/limits.h>
110#include <sys/lock.h>
111#include <sys/mac.h>
112#include <sys/malloc.h>
113#include <sys/mbuf.h>
114#include <sys/mutex.h>
115#include <sys/domain.h>
116#include <sys/file.h> /* for struct knote */
117#include <sys/kernel.h>
118#include <sys/event.h>
119#include <sys/eventhandler.h>
120#include <sys/poll.h>
121#include <sys/proc.h>
122#include <sys/protosw.h>
123#include <sys/socket.h>
124#include <sys/socketvar.h>
125#include <sys/resourcevar.h>
126#include <net/route.h>
127#include <sys/signalvar.h>
128#include <sys/stat.h>
129#include <sys/sx.h>
130#include <sys/sysctl.h>
131#include <sys/uio.h>
132#include <sys/jail.h>
133#include <sys/vimage.h>
134
135#include <security/mac/mac_framework.h>
136
137#include <vm/uma.h>
138
139#ifdef COMPAT_IA32
140#include <sys/mount.h>
141#include <sys/sysent.h>
142#include <compat/freebsd32/freebsd32.h>
143#endif
144
145static int soreceive_rcvoob(struct socket *so, struct uio *uio,
146 int flags);
147
148static void filt_sordetach(struct knote *kn);
149static int filt_soread(struct knote *kn, long hint);
150static void filt_sowdetach(struct knote *kn);
151static int filt_sowrite(struct knote *kn, long hint);
152static int filt_solisten(struct knote *kn, long hint);
153
154static struct filterops solisten_filtops =
155 { 1, NULL, filt_sordetach, filt_solisten };
156static struct filterops soread_filtops =
157 { 1, NULL, filt_sordetach, filt_soread };
158static struct filterops sowrite_filtops =
159 { 1, NULL, filt_sowdetach, filt_sowrite };
160
161uma_zone_t socket_zone;
162so_gen_t so_gencnt; /* generation count for sockets */
163
164int maxsockets;
165
166MALLOC_DEFINE(M_SONAME, "soname", "socket name");
167MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
168
169static int somaxconn = SOMAXCONN;
170static int sysctl_somaxconn(SYSCTL_HANDLER_ARGS);
171/* XXX: we dont have SYSCTL_USHORT */
172SYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn, CTLTYPE_UINT | CTLFLAG_RW,
173 0, sizeof(int), sysctl_somaxconn, "I", "Maximum pending socket connection "
174 "queue size");
175static int numopensockets;
176SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD,
177 &numopensockets, 0, "Number of open sockets");
178#ifdef ZERO_COPY_SOCKETS
179/* These aren't static because they're used in other files. */
180int so_zero_copy_send = 1;
181int so_zero_copy_receive = 1;
182SYSCTL_NODE(_kern_ipc, OID_AUTO, zero_copy, CTLFLAG_RD, 0,
183 "Zero copy controls");
184SYSCTL_INT(_kern_ipc_zero_copy, OID_AUTO, receive, CTLFLAG_RW,
185 &so_zero_copy_receive, 0, "Enable zero copy receive");
186SYSCTL_INT(_kern_ipc_zero_copy, OID_AUTO, send, CTLFLAG_RW,
187 &so_zero_copy_send, 0, "Enable zero copy send");
188#endif /* ZERO_COPY_SOCKETS */
189
190/*
191 * accept_mtx locks down per-socket fields relating to accept queues. See
192 * socketvar.h for an annotation of the protected fields of struct socket.
193 */
194struct mtx accept_mtx;
195MTX_SYSINIT(accept_mtx, &accept_mtx, "accept", MTX_DEF);
196
197/*
198 * so_global_mtx protects so_gencnt, numopensockets, and the per-socket
199 * so_gencnt field.
200 */
201static struct mtx so_global_mtx;
202MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF);
203
204/*
205 * General IPC sysctl name space, used by sockets and a variety of other IPC
206 * types.
207 */
208SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC");
209
210/*
211 * Sysctl to get and set the maximum global sockets limit. Notify protocols
212 * of the change so that they can update their dependent limits as required.
213 */
214static int
215sysctl_maxsockets(SYSCTL_HANDLER_ARGS)
216{
217 int error, newmaxsockets;
218
219 newmaxsockets = maxsockets;
220 error = sysctl_handle_int(oidp, &newmaxsockets, 0, req);
221 if (error == 0 && req->newptr) {
222 if (newmaxsockets > maxsockets) {
223 maxsockets = newmaxsockets;
224 if (maxsockets > ((maxfiles / 4) * 3)) {
225 maxfiles = (maxsockets * 5) / 4;
226 maxfilesperproc = (maxfiles * 9) / 10;
227 }
228 EVENTHANDLER_INVOKE(maxsockets_change);
229 } else
230 error = EINVAL;
231 }
232 return (error);
233}
234
235SYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets, CTLTYPE_INT|CTLFLAG_RW,
236 &maxsockets, 0, sysctl_maxsockets, "IU",
237 "Maximum number of sockets avaliable");
238
239/*
240 * Initialise maxsockets. This SYSINIT must be run after
241 * tunable_mbinit().
242 */
243static void
244init_maxsockets(void *ignored)
245{
246
247 TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
248 maxsockets = imax(maxsockets, imax(maxfiles, nmbclusters));
249}
250SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL);
251
252/*
253 * Socket operation routines. These routines are called by the routines in
254 * sys_socket.c or from a system process, and implement the semantics of
255 * socket operations by switching out to the protocol specific routines.
256 */
257
258/*
259 * Get a socket structure from our zone, and initialize it. Note that it
260 * would probably be better to allocate socket and PCB at the same time, but
261 * I'm not convinced that all the protocols can be easily modified to do
262 * this.
263 *
264 * soalloc() returns a socket with a ref count of 0.
265 */
266static struct socket *
267soalloc(struct vnet *vnet)
268{
269 struct socket *so;
270
271 so = uma_zalloc(socket_zone, M_NOWAIT | M_ZERO);
272 if (so == NULL)
273 return (NULL);
274#ifdef MAC
275 if (mac_socket_init(so, M_NOWAIT) != 0) {
276 uma_zfree(socket_zone, so);
277 return (NULL);
278 }
279#endif
280 SOCKBUF_LOCK_INIT(&so->so_snd, "so_snd");
281 SOCKBUF_LOCK_INIT(&so->so_rcv, "so_rcv");
282 sx_init(&so->so_snd.sb_sx, "so_snd_sx");
283 sx_init(&so->so_rcv.sb_sx, "so_rcv_sx");
284 TAILQ_INIT(&so->so_aiojobq);
285 mtx_lock(&so_global_mtx);
286 so->so_gencnt = ++so_gencnt;
287 ++numopensockets;
288#ifdef VIMAGE
289 ++vnet->sockcnt; /* Locked with so_global_mtx. */
290 so->so_vnet = vnet;
291#endif
292 mtx_unlock(&so_global_mtx);
293 return (so);
294}
295
296/*
297 * Free the storage associated with a socket at the socket layer, tear down
298 * locks, labels, etc. All protocol state is assumed already to have been
299 * torn down (and possibly never set up) by the caller.
300 */
301static void
302sodealloc(struct socket *so)
303{
304
305 KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count));
306 KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL"));
307
308 mtx_lock(&so_global_mtx);
309 so->so_gencnt = ++so_gencnt;
310 --numopensockets; /* Could be below, but faster here. */
311#ifdef VIMAGE
312 --so->so_vnet->sockcnt;
313#endif
314 mtx_unlock(&so_global_mtx);
315 if (so->so_rcv.sb_hiwat)
316 (void)chgsbsize(so->so_cred->cr_uidinfo,
317 &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY);
318 if (so->so_snd.sb_hiwat)
319 (void)chgsbsize(so->so_cred->cr_uidinfo,
320 &so->so_snd.sb_hiwat, 0, RLIM_INFINITY);
321#ifdef INET
322 /* remove acccept filter if one is present. */
323 if (so->so_accf != NULL)
324 do_setopt_accept_filter(so, NULL);
325#endif
326#ifdef MAC
327 mac_socket_destroy(so);
328#endif
329 crfree(so->so_cred);
330 sx_destroy(&so->so_snd.sb_sx);
331 sx_destroy(&so->so_rcv.sb_sx);
332 SOCKBUF_LOCK_DESTROY(&so->so_snd);
333 SOCKBUF_LOCK_DESTROY(&so->so_rcv);
334 uma_zfree(socket_zone, so);
335}
336
337/*
338 * socreate returns a socket with a ref count of 1. The socket should be
339 * closed with soclose().
340 */
341int
342socreate(int dom, struct socket **aso, int type, int proto,
343 struct ucred *cred, struct thread *td)
344{
345 struct protosw *prp;
346 struct socket *so;
347 int error;
348
349 if (proto)
350 prp = pffindproto(dom, proto, type);
351 else
352 prp = pffindtype(dom, type);
353
354 if (prp == NULL || prp->pr_usrreqs->pru_attach == NULL ||
355 prp->pr_usrreqs->pru_attach == pru_attach_notsupp)
356 return (EPROTONOSUPPORT);
357
358 if (prison_check_af(cred, prp->pr_domain->dom_family) != 0)
359 return (EPROTONOSUPPORT);
360
361 if (prp->pr_type != type)
362 return (EPROTOTYPE);
363 so = soalloc(TD_TO_VNET(td));
364 if (so == NULL)
365 return (ENOBUFS);
366
367 TAILQ_INIT(&so->so_incomp);
368 TAILQ_INIT(&so->so_comp);
369 so->so_type = type;
370 so->so_cred = crhold(cred);
371 if ((prp->pr_domain->dom_family == PF_INET) ||
372 (prp->pr_domain->dom_family == PF_ROUTE))
373 so->so_fibnum = td->td_proc->p_fibnum;
374 else
375 so->so_fibnum = 0;
376 so->so_proto = prp;
377#ifdef MAC
378 mac_socket_create(cred, so);
379#endif
380 knlist_init(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv),
381 NULL, NULL, NULL);
382 knlist_init(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd),
383 NULL, NULL, NULL);
384 so->so_count = 1;
385 /*
386 * Auto-sizing of socket buffers is managed by the protocols and
387 * the appropriate flags must be set in the pru_attach function.
388 */
389 CURVNET_SET(so->so_vnet);
390 error = (*prp->pr_usrreqs->pru_attach)(so, proto, td);
391 CURVNET_RESTORE();
392 if (error) {
393 KASSERT(so->so_count == 1, ("socreate: so_count %d",
394 so->so_count));
395 so->so_count = 0;
396 sodealloc(so);
397 return (error);
398 }
399 *aso = so;
400 return (0);
401}
402
403#ifdef REGRESSION
404static int regression_sonewconn_earlytest = 1;
405SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW,
406 ®ression_sonewconn_earlytest, 0, "Perform early sonewconn limit test");
407#endif
408
409/*
410 * When an attempt at a new connection is noted on a socket which accepts
411 * connections, sonewconn is called. If the connection is possible (subject
412 * to space constraints, etc.) then we allocate a new structure, propoerly
413 * linked into the data structure of the original socket, and return this.
414 * Connstatus may be 0, or SO_ISCONFIRMING, or SO_ISCONNECTED.
415 *
416 * Note: the ref count on the socket is 0 on return.
417 */
418struct socket *
419sonewconn(struct socket *head, int connstatus)
420{
421 struct socket *so;
422 int over;
423
424 ACCEPT_LOCK();
425 over = (head->so_qlen > 3 * head->so_qlimit / 2);
426 ACCEPT_UNLOCK();
427#ifdef REGRESSION
428 if (regression_sonewconn_earlytest && over)
429#else
430 if (over)
431#endif
432 return (NULL);
433 VNET_ASSERT(head->so_vnet);
434 so = soalloc(head->so_vnet);
435 if (so == NULL)
436 return (NULL);
437 if ((head->so_options & SO_ACCEPTFILTER) != 0)
438 connstatus = 0;
439 so->so_head = head;
440 so->so_type = head->so_type;
441 so->so_options = head->so_options &~ SO_ACCEPTCONN;
442 so->so_linger = head->so_linger;
443 so->so_state = head->so_state | SS_NOFDREF;
444 so->so_proto = head->so_proto;
445 so->so_cred = crhold(head->so_cred);
446#ifdef MAC
447 SOCK_LOCK(head);
448 mac_socket_newconn(head, so);
449 SOCK_UNLOCK(head);
450#endif
451 knlist_init(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv),
452 NULL, NULL, NULL);
453 knlist_init(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd),
454 NULL, NULL, NULL);
455 if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat) ||
456 (*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
457 sodealloc(so);
458 return (NULL);
459 }
460 so->so_rcv.sb_lowat = head->so_rcv.sb_lowat;
461 so->so_snd.sb_lowat = head->so_snd.sb_lowat;
462 so->so_rcv.sb_timeo = head->so_rcv.sb_timeo;
463 so->so_snd.sb_timeo = head->so_snd.sb_timeo;
464 so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE;
465 so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE;
466 so->so_state |= connstatus;
467 ACCEPT_LOCK();
468 if (connstatus) {
469 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
470 so->so_qstate |= SQ_COMP;
471 head->so_qlen++;
472 } else {
473 /*
474 * Keep removing sockets from the head until there's room for
475 * us to insert on the tail. In pre-locking revisions, this
476 * was a simple if(), but as we could be racing with other
477 * threads and soabort() requires dropping locks, we must
478 * loop waiting for the condition to be true.
479 */
480 while (head->so_incqlen > head->so_qlimit) {
481 struct socket *sp;
482 sp = TAILQ_FIRST(&head->so_incomp);
483 TAILQ_REMOVE(&head->so_incomp, sp, so_list);
484 head->so_incqlen--;
485 sp->so_qstate &= ~SQ_INCOMP;
486 sp->so_head = NULL;
487 ACCEPT_UNLOCK();
488 soabort(sp);
489 ACCEPT_LOCK();
490 }
491 TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list);
492 so->so_qstate |= SQ_INCOMP;
493 head->so_incqlen++;
494 }
495 ACCEPT_UNLOCK();
496 if (connstatus) {
497 sorwakeup(head);
498 wakeup_one(&head->so_timeo);
499 }
500 return (so);
501}
502
503int
504sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
505{
506 int error;
507
508 CURVNET_SET(so->so_vnet);
509 error = (*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td);
510 CURVNET_RESTORE();
511 return error;
512}
513
514/*
515 * solisten() transitions a socket from a non-listening state to a listening
516 * state, but can also be used to update the listen queue depth on an
517 * existing listen socket. The protocol will call back into the sockets
518 * layer using solisten_proto_check() and solisten_proto() to check and set
519 * socket-layer listen state. Call backs are used so that the protocol can
520 * acquire both protocol and socket layer locks in whatever order is required
521 * by the protocol.
522 *
523 * Protocol implementors are advised to hold the socket lock across the
524 * socket-layer test and set to avoid races at the socket layer.
525 */
526int
527solisten(struct socket *so, int backlog, struct thread *td)
528{
529
530 return ((*so->so_proto->pr_usrreqs->pru_listen)(so, backlog, td));
531}
532
533int
534solisten_proto_check(struct socket *so)
535{
536
537 SOCK_LOCK_ASSERT(so);
538
539 if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
540 SS_ISDISCONNECTING))
541 return (EINVAL);
542 return (0);
543}
544
545void
546solisten_proto(struct socket *so, int backlog)
547{
548
549 SOCK_LOCK_ASSERT(so);
550
551 if (backlog < 0 || backlog > somaxconn)
552 backlog = somaxconn;
553 so->so_qlimit = backlog;
554 so->so_options |= SO_ACCEPTCONN;
555}
556
557/*
558 * Attempt to free a socket. This should really be sotryfree().
559 *
560 * sofree() will succeed if:
561 *
562 * - There are no outstanding file descriptor references or related consumers
563 * (so_count == 0).
564 *
565 * - The socket has been closed by user space, if ever open (SS_NOFDREF).
566 *
567 * - The protocol does not have an outstanding strong reference on the socket
568 * (SS_PROTOREF).
569 *
570 * - The socket is not in a completed connection queue, so a process has been
571 * notified that it is present. If it is removed, the user process may
572 * block in accept() despite select() saying the socket was ready.
573 *
574 * Otherwise, it will quietly abort so that a future call to sofree(), when
575 * conditions are right, can succeed.
576 */
577void
578sofree(struct socket *so)
579{
580 struct protosw *pr = so->so_proto;
581 struct socket *head;
582
583 ACCEPT_LOCK_ASSERT();
584 SOCK_LOCK_ASSERT(so);
585
586 if ((so->so_state & SS_NOFDREF) == 0 || so->so_count != 0 ||
587 (so->so_state & SS_PROTOREF) || (so->so_qstate & SQ_COMP)) {
588 SOCK_UNLOCK(so);
589 ACCEPT_UNLOCK();
590 return;
591 }
592
593 head = so->so_head;
594 if (head != NULL) {
595 KASSERT((so->so_qstate & SQ_COMP) != 0 ||
596 (so->so_qstate & SQ_INCOMP) != 0,
597 ("sofree: so_head != NULL, but neither SQ_COMP nor "
598 "SQ_INCOMP"));
599 KASSERT((so->so_qstate & SQ_COMP) == 0 ||
600 (so->so_qstate & SQ_INCOMP) == 0,
601 ("sofree: so->so_qstate is SQ_COMP and also SQ_INCOMP"));
602 TAILQ_REMOVE(&head->so_incomp, so, so_list);
603 head->so_incqlen--;
604 so->so_qstate &= ~SQ_INCOMP;
605 so->so_head = NULL;
606 }
607 KASSERT((so->so_qstate & SQ_COMP) == 0 &&
608 (so->so_qstate & SQ_INCOMP) == 0,
609 ("sofree: so_head == NULL, but still SQ_COMP(%d) or SQ_INCOMP(%d)",
610 so->so_qstate & SQ_COMP, so->so_qstate & SQ_INCOMP));
611 if (so->so_options & SO_ACCEPTCONN) {
612 KASSERT((TAILQ_EMPTY(&so->so_comp)), ("sofree: so_comp populated"));
613 KASSERT((TAILQ_EMPTY(&so->so_incomp)), ("sofree: so_comp populated"));
614 }
615 SOCK_UNLOCK(so);
616 ACCEPT_UNLOCK();
617
618 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
619 (*pr->pr_domain->dom_dispose)(so->so_rcv.sb_mb);
620 if (pr->pr_usrreqs->pru_detach != NULL)
621 (*pr->pr_usrreqs->pru_detach)(so);
622
623 /*
624 * From this point on, we assume that no other references to this
625 * socket exist anywhere else in the stack. Therefore, no locks need
626 * to be acquired or held.
627 *
628 * We used to do a lot of socket buffer and socket locking here, as
629 * well as invoke sorflush() and perform wakeups. The direct call to
630 * dom_dispose() and sbrelease_internal() are an inlining of what was
631 * necessary from sorflush().
632 *
633 * Notice that the socket buffer and kqueue state are torn down
634 * before calling pru_detach. This means that protocols shold not
635 * assume they can perform socket wakeups, etc, in their detach code.
636 */
637 sbdestroy(&so->so_snd, so);
638 sbdestroy(&so->so_rcv, so);
639 knlist_destroy(&so->so_rcv.sb_sel.si_note);
640 knlist_destroy(&so->so_snd.sb_sel.si_note);
641 sodealloc(so);
642}
643
644/*
645 * Close a socket on last file table reference removal. Initiate disconnect
646 * if connected. Free socket when disconnect complete.
647 *
648 * This function will sorele() the socket. Note that soclose() may be called
649 * prior to the ref count reaching zero. The actual socket structure will
650 * not be freed until the ref count reaches zero.
651 */
652int
653soclose(struct socket *so)
654{
655 int error = 0;
656
657 KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter"));
658
659 CURVNET_SET(so->so_vnet);
660 funsetown(&so->so_sigio);
661 if (so->so_state & SS_ISCONNECTED) {
662 if ((so->so_state & SS_ISDISCONNECTING) == 0) {
663 error = sodisconnect(so);
664 if (error)
665 goto drop;
666 }
667 if (so->so_options & SO_LINGER) {
668 if ((so->so_state & SS_ISDISCONNECTING) &&
669 (so->so_state & SS_NBIO))
670 goto drop;
671 while (so->so_state & SS_ISCONNECTED) {
672 error = tsleep(&so->so_timeo,
673 PSOCK | PCATCH, "soclos", so->so_linger * hz);
674 if (error)
675 break;
676 }
677 }
678 }
679
680drop:
681 if (so->so_proto->pr_usrreqs->pru_close != NULL)
682 (*so->so_proto->pr_usrreqs->pru_close)(so);
683 if (so->so_options & SO_ACCEPTCONN) {
684 struct socket *sp;
685 ACCEPT_LOCK();
686 while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
687 TAILQ_REMOVE(&so->so_incomp, sp, so_list);
688 so->so_incqlen--;
689 sp->so_qstate &= ~SQ_INCOMP;
690 sp->so_head = NULL;
691 ACCEPT_UNLOCK();
692 soabort(sp);
693 ACCEPT_LOCK();
694 }
695 while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
696 TAILQ_REMOVE(&so->so_comp, sp, so_list);
697 so->so_qlen--;
698 sp->so_qstate &= ~SQ_COMP;
699 sp->so_head = NULL;
700 ACCEPT_UNLOCK();
701 soabort(sp);
702 ACCEPT_LOCK();
703 }
704 ACCEPT_UNLOCK();
705 }
706 ACCEPT_LOCK();
707 SOCK_LOCK(so);
708 KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF"));
709 so->so_state |= SS_NOFDREF;
710 sorele(so);
711 CURVNET_RESTORE();
712 return (error);
713}
714
715/*
716 * soabort() is used to abruptly tear down a connection, such as when a
717 * resource limit is reached (listen queue depth exceeded), or if a listen
718 * socket is closed while there are sockets waiting to be accepted.
719 *
720 * This interface is tricky, because it is called on an unreferenced socket,
721 * and must be called only by a thread that has actually removed the socket
722 * from the listen queue it was on, or races with other threads are risked.
723 *
724 * This interface will call into the protocol code, so must not be called
725 * with any socket locks held. Protocols do call it while holding their own
726 * recursible protocol mutexes, but this is something that should be subject
727 * to review in the future.
728 */
729void
730soabort(struct socket *so)
731{
732
733 /*
734 * In as much as is possible, assert that no references to this
735 * socket are held. This is not quite the same as asserting that the
736 * current thread is responsible for arranging for no references, but
737 * is as close as we can get for now.
738 */
739 KASSERT(so->so_count == 0, ("soabort: so_count"));
740 KASSERT((so->so_state & SS_PROTOREF) == 0, ("soabort: SS_PROTOREF"));
741 KASSERT(so->so_state & SS_NOFDREF, ("soabort: !SS_NOFDREF"));
742 KASSERT((so->so_state & SQ_COMP) == 0, ("soabort: SQ_COMP"));
743 KASSERT((so->so_state & SQ_INCOMP) == 0, ("soabort: SQ_INCOMP"));
744
745 if (so->so_proto->pr_usrreqs->pru_abort != NULL)
746 (*so->so_proto->pr_usrreqs->pru_abort)(so);
747 ACCEPT_LOCK();
748 SOCK_LOCK(so);
749 sofree(so);
750}
751
752int
753soaccept(struct socket *so, struct sockaddr **nam)
754{
755 int error;
756
757 SOCK_LOCK(so);
758 KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF"));
759 so->so_state &= ~SS_NOFDREF;
760 SOCK_UNLOCK(so);
761 error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam);
762 return (error);
763}
764
765int
766soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
767{
768 int error;
769
770 if (so->so_options & SO_ACCEPTCONN)
771 return (EOPNOTSUPP);
772 /*
773 * If protocol is connection-based, can only connect once.
774 * Otherwise, if connected, try to disconnect first. This allows
775 * user to disconnect by connecting to, e.g., a null address.
776 */
777 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
778 ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
779 (error = sodisconnect(so)))) {
780 error = EISCONN;
781 } else {
782 /*
783 * Prevent accumulated error from previous connection from
784 * biting us.
785 */
786 so->so_error = 0;
787 CURVNET_SET(so->so_vnet);
788 error = (*so->so_proto->pr_usrreqs->pru_connect)(so, nam, td);
789 CURVNET_RESTORE();
790 }
791
792 return (error);
793}
794
795int
796soconnect2(struct socket *so1, struct socket *so2)
797{
798
799 return ((*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2));
800}
801
802int
803sodisconnect(struct socket *so)
804{
805 int error;
806
807 if ((so->so_state & SS_ISCONNECTED) == 0)
808 return (ENOTCONN);
809 if (so->so_state & SS_ISDISCONNECTING)
810 return (EALREADY);
811 error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so);
812 return (error);
813}
814
815#ifdef ZERO_COPY_SOCKETS
816struct so_zerocopy_stats{
817 int size_ok;
818 int align_ok;
819 int found_ifp;
820};
821struct so_zerocopy_stats so_zerocp_stats = {0,0,0};
822#include <netinet/in.h>
823#include <net/route.h>
824#include <netinet/in_pcb.h>
825#include <vm/vm.h>
826#include <vm/vm_page.h>
827#include <vm/vm_object.h>
828
829/*
830 * sosend_copyin() is only used if zero copy sockets are enabled. Otherwise
831 * sosend_dgram() and sosend_generic() use m_uiotombuf().
832 *
833 * sosend_copyin() accepts a uio and prepares an mbuf chain holding part or
834 * all of the data referenced by the uio. If desired, it uses zero-copy.
835 * *space will be updated to reflect data copied in.
836 *
837 * NB: If atomic I/O is requested, the caller must already have checked that
838 * space can hold resid bytes.
839 *
840 * NB: In the event of an error, the caller may need to free the partial
841 * chain pointed to by *mpp. The contents of both *uio and *space may be
842 * modified even in the case of an error.
843 */
844static int
845sosend_copyin(struct uio *uio, struct mbuf **retmp, int atomic, long *space,
846 int flags)
847{
848 struct mbuf *m, **mp, *top;
849 long len, resid;
850 int error;
851#ifdef ZERO_COPY_SOCKETS
852 int cow_send;
853#endif
854
855 *retmp = top = NULL;
856 mp = ⊤
857 len = 0;
858 resid = uio->uio_resid;
859 error = 0;
860 do {
861#ifdef ZERO_COPY_SOCKETS
862 cow_send = 0;
863#endif /* ZERO_COPY_SOCKETS */
864 if (resid >= MINCLSIZE) {
865#ifdef ZERO_COPY_SOCKETS
866 if (top == NULL) {
867 m = m_gethdr(M_WAITOK, MT_DATA);
868 m->m_pkthdr.len = 0;
869 m->m_pkthdr.rcvif = NULL;
870 } else
871 m = m_get(M_WAITOK, MT_DATA);
872 if (so_zero_copy_send &&
873 resid>=PAGE_SIZE &&
874 *space>=PAGE_SIZE &&
875 uio->uio_iov->iov_len>=PAGE_SIZE) {
876 so_zerocp_stats.size_ok++;
877 so_zerocp_stats.align_ok++;
878 cow_send = socow_setup(m, uio);
879 len = cow_send;
880 }
881 if (!cow_send) {
882 m_clget(m, M_WAITOK);
883 len = min(min(MCLBYTES, resid), *space);
884 }
885#else /* ZERO_COPY_SOCKETS */
886 if (top == NULL) {
887 m = m_getcl(M_WAIT, MT_DATA, M_PKTHDR);
888 m->m_pkthdr.len = 0;
889 m->m_pkthdr.rcvif = NULL;
890 } else
891 m = m_getcl(M_WAIT, MT_DATA, 0);
892 len = min(min(MCLBYTES, resid), *space);
893#endif /* ZERO_COPY_SOCKETS */
894 } else {
895 if (top == NULL) {
896 m = m_gethdr(M_WAIT, MT_DATA);
897 m->m_pkthdr.len = 0;
898 m->m_pkthdr.rcvif = NULL;
899
900 len = min(min(MHLEN, resid), *space);
901 /*
902 * For datagram protocols, leave room
903 * for protocol headers in first mbuf.
904 */
905 if (atomic && m && len < MHLEN)
906 MH_ALIGN(m, len);
907 } else {
908 m = m_get(M_WAIT, MT_DATA);
909 len = min(min(MLEN, resid), *space);
910 }
911 }
912 if (m == NULL) {
913 error = ENOBUFS;
914 goto out;
915 }
916
917 *space -= len;
918#ifdef ZERO_COPY_SOCKETS
919 if (cow_send)
920 error = 0;
921 else
922#endif /* ZERO_COPY_SOCKETS */
923 error = uiomove(mtod(m, void *), (int)len, uio);
924 resid = uio->uio_resid;
925 m->m_len = len;
926 *mp = m;
927 top->m_pkthdr.len += len;
928 if (error)
929 goto out;
930 mp = &m->m_next;
931 if (resid <= 0) {
932 if (flags & MSG_EOR)
933 top->m_flags |= M_EOR;
934 break;
935 }
936 } while (*space > 0 && atomic);
937out:
938 *retmp = top;
939 return (error);
940}
941#endif /*ZERO_COPY_SOCKETS*/
942
943#define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? 0 : SBL_WAIT)
944
945int
946sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio,
947 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
948{
949 long space, resid;
950 int clen = 0, error, dontroute;
951#ifdef ZERO_COPY_SOCKETS
952 int atomic = sosendallatonce(so) || top;
953#endif
954
955 KASSERT(so->so_type == SOCK_DGRAM, ("sodgram_send: !SOCK_DGRAM"));
956 KASSERT(so->so_proto->pr_flags & PR_ATOMIC,
957 ("sodgram_send: !PR_ATOMIC"));
958
959 if (uio != NULL)
960 resid = uio->uio_resid;
961 else
962 resid = top->m_pkthdr.len;
963 /*
964 * In theory resid should be unsigned. However, space must be
965 * signed, as it might be less than 0 if we over-committed, and we
966 * must use a signed comparison of space and resid. On the other
967 * hand, a negative resid causes us to loop sending 0-length
968 * segments to the protocol.
969 *
970 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
971 * type sockets since that's an error.
972 */
973 if (resid < 0) {
974 error = EINVAL;
975 goto out;
976 }
977
978 dontroute =
979 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0;
980 if (td != NULL)
981 td->td_ru.ru_msgsnd++;
982 if (control != NULL)
983 clen = control->m_len;
984
985 SOCKBUF_LOCK(&so->so_snd);
986 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
987 SOCKBUF_UNLOCK(&so->so_snd);
988 error = EPIPE;
989 goto out;
990 }
991 if (so->so_error) {
992 error = so->so_error;
993 so->so_error = 0;
994 SOCKBUF_UNLOCK(&so->so_snd);
995 goto out;
996 }
997 if ((so->so_state & SS_ISCONNECTED) == 0) {
998 /*
999 * `sendto' and `sendmsg' is allowed on a connection-based
1000 * socket if it supports implied connect. Return ENOTCONN if
1001 * not connected and no address is supplied.
1002 */
1003 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1004 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1005 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1006 !(resid == 0 && clen != 0)) {
1007 SOCKBUF_UNLOCK(&so->so_snd);
1008 error = ENOTCONN;
1009 goto out;
1010 }
1011 } else if (addr == NULL) {
1012 if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1013 error = ENOTCONN;
1014 else
1015 error = EDESTADDRREQ;
1016 SOCKBUF_UNLOCK(&so->so_snd);
1017 goto out;
1018 }
1019 }
1020
1021 /*
1022 * Do we need MSG_OOB support in SOCK_DGRAM? Signs here may be a
1023 * problem and need fixing.
1024 */
1025 space = sbspace(&so->so_snd);
1026 if (flags & MSG_OOB)
1027 space += 1024;
1028 space -= clen;
1029 SOCKBUF_UNLOCK(&so->so_snd);
1030 if (resid > space) {
1031 error = EMSGSIZE;
1032 goto out;
1033 }
1034 if (uio == NULL) {
1035 resid = 0;
1036 if (flags & MSG_EOR)
1037 top->m_flags |= M_EOR;
1038 } else {
1039#ifdef ZERO_COPY_SOCKETS
1040 error = sosend_copyin(uio, &top, atomic, &space, flags);
1041 if (error)
1042 goto out;
1043#else
1044 /*
1045 * Copy the data from userland into a mbuf chain.
1046 * If no data is to be copied in, a single empty mbuf
1047 * is returned.
1048 */
1049 top = m_uiotombuf(uio, M_WAITOK, space, max_hdr,
1050 (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0)));
1051 if (top == NULL) {
1052 error = EFAULT; /* only possible error */
1053 goto out;
1054 }
1055 space -= resid - uio->uio_resid;
1056#endif
1057 resid = uio->uio_resid;
1058 }
1059 KASSERT(resid == 0, ("sosend_dgram: resid != 0"));
1060 /*
1061 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock
1062 * than with.
1063 */
1064 if (dontroute) {
1065 SOCK_LOCK(so);
1066 so->so_options |= SO_DONTROUTE;
1067 SOCK_UNLOCK(so);
1068 }
1069 /*
1070 * XXX all the SBS_CANTSENDMORE checks previously done could be out
1071 * of date. We could have recieved a reset packet in an interrupt or
1072 * maybe we slept while doing page faults in uiomove() etc. We could
1073 * probably recheck again inside the locking protection here, but
1074 * there are probably other places that this also happens. We must
1075 * rethink this.
1076 */
1077 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1078 (flags & MSG_OOB) ? PRUS_OOB :
1079 /*
1080 * If the user set MSG_EOF, the protocol understands this flag and
1081 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND.
1082 */
1083 ((flags & MSG_EOF) &&
1084 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1085 (resid <= 0)) ?
1086 PRUS_EOF :
1087 /* If there is more to send set PRUS_MORETOCOME */
1088 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1089 top, addr, control, td);
1090 if (dontroute) {
1091 SOCK_LOCK(so);
1092 so->so_options &= ~SO_DONTROUTE;
1093 SOCK_UNLOCK(so);
1094 }
1095 clen = 0;
1096 control = NULL;
1097 top = NULL;
1098out:
1099 if (top != NULL)
1100 m_freem(top);
1101 if (control != NULL)
1102 m_freem(control);
1103 return (error);
1104}
1105
1106/*
1107 * Send on a socket. If send must go all at once and message is larger than
1108 * send buffering, then hard error. Lock against other senders. If must go
1109 * all at once and not enough room now, then inform user that this would
1110 * block and do nothing. Otherwise, if nonblocking, send as much as
1111 * possible. The data to be sent is described by "uio" if nonzero, otherwise
1112 * by the mbuf chain "top" (which must be null if uio is not). Data provided
1113 * in mbuf chain must be small enough to send all at once.
1114 *
1115 * Returns nonzero on error, timeout or signal; callers must check for short
1116 * counts if EINTR/ERESTART are returned. Data and control buffers are freed
1117 * on return.
1118 */
1119int
1120sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio,
1121 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1122{
1123 long space, resid;
1124 int clen = 0, error, dontroute;
1125 int atomic = sosendallatonce(so) || top;
1126
1127 if (uio != NULL)
1128 resid = uio->uio_resid;
1129 else
1130 resid = top->m_pkthdr.len;
1131 /*
1132 * In theory resid should be unsigned. However, space must be
1133 * signed, as it might be less than 0 if we over-committed, and we
1134 * must use a signed comparison of space and resid. On the other
1135 * hand, a negative resid causes us to loop sending 0-length
1136 * segments to the protocol.
1137 *
1138 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
1139 * type sockets since that's an error.
1140 */
1141 if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
1142 error = EINVAL;
1143 goto out;
1144 }
1145
1146 dontroute =
1147 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
1148 (so->so_proto->pr_flags & PR_ATOMIC);
1149 if (td != NULL)
1150 td->td_ru.ru_msgsnd++;
1151 if (control != NULL)
1152 clen = control->m_len;
1153
1154 error = sblock(&so->so_snd, SBLOCKWAIT(flags));
1155 if (error)
1156 goto out;
1157
1158restart:
1159 do {
1160 SOCKBUF_LOCK(&so->so_snd);
1161 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1162 SOCKBUF_UNLOCK(&so->so_snd);
1163 error = EPIPE;
1164 goto release;
1165 }
1166 if (so->so_error) {
1167 error = so->so_error;
1168 so->so_error = 0;
1169 SOCKBUF_UNLOCK(&so->so_snd);
1170 goto release;
1171 }
1172 if ((so->so_state & SS_ISCONNECTED) == 0) {
1173 /*
1174 * `sendto' and `sendmsg' is allowed on a connection-
1175 * based socket if it supports implied connect.
1176 * Return ENOTCONN if not connected and no address is
1177 * supplied.
1178 */
1179 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1180 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1181 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1182 !(resid == 0 && clen != 0)) {
1183 SOCKBUF_UNLOCK(&so->so_snd);
1184 error = ENOTCONN;
1185 goto release;
1186 }
1187 } else if (addr == NULL) {
1188 SOCKBUF_UNLOCK(&so->so_snd);
1189 if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1190 error = ENOTCONN;
1191 else
1192 error = EDESTADDRREQ;
1193 goto release;
1194 }
1195 }
1196 space = sbspace(&so->so_snd);
1197 if (flags & MSG_OOB)
1198 space += 1024;
1199 if ((atomic && resid > so->so_snd.sb_hiwat) ||
1200 clen > so->so_snd.sb_hiwat) {
1201 SOCKBUF_UNLOCK(&so->so_snd);
1202 error = EMSGSIZE;
1203 goto release;
1204 }
1205 if (space < resid + clen &&
1206 (atomic || space < so->so_snd.sb_lowat || space < clen)) {
1207 if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO)) {
1208 SOCKBUF_UNLOCK(&so->so_snd);
1209 error = EWOULDBLOCK;
1210 goto release;
1211 }
1212 error = sbwait(&so->so_snd);
1213 SOCKBUF_UNLOCK(&so->so_snd);
1214 if (error)
1215 goto release;
1216 goto restart;
1217 }
1218 SOCKBUF_UNLOCK(&so->so_snd);
1219 space -= clen;
1220 do {
1221 if (uio == NULL) {
1222 resid = 0;
1223 if (flags & MSG_EOR)
1224 top->m_flags |= M_EOR;
1225 } else {
1226#ifdef ZERO_COPY_SOCKETS
1227 error = sosend_copyin(uio, &top, atomic,
1228 &space, flags);
1229 if (error != 0)
1230 goto release;
1231#else
1232 /*
1233 * Copy the data from userland into a mbuf
1234 * chain. If no data is to be copied in,
1235 * a single empty mbuf is returned.
1236 */
1237 top = m_uiotombuf(uio, M_WAITOK, space,
1238 (atomic ? max_hdr : 0),
1239 (atomic ? M_PKTHDR : 0) |
1240 ((flags & MSG_EOR) ? M_EOR : 0));
1241 if (top == NULL) {
1242 error = EFAULT; /* only possible error */
1243 goto release;
1244 }
1245 space -= resid - uio->uio_resid;
1246#endif
1247 resid = uio->uio_resid;
1248 }
1249 if (dontroute) {
1250 SOCK_LOCK(so);
1251 so->so_options |= SO_DONTROUTE;
1252 SOCK_UNLOCK(so);
1253 }
1254 /*
1255 * XXX all the SBS_CANTSENDMORE checks previously
1256 * done could be out of date. We could have recieved
1257 * a reset packet in an interrupt or maybe we slept
1258 * while doing page faults in uiomove() etc. We
1259 * could probably recheck again inside the locking
1260 * protection here, but there are probably other
1261 * places that this also happens. We must rethink
1262 * this.
1263 */
1264 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1265 (flags & MSG_OOB) ? PRUS_OOB :
1266 /*
1267 * If the user set MSG_EOF, the protocol understands
1268 * this flag and nothing left to send then use
1269 * PRU_SEND_EOF instead of PRU_SEND.
1270 */
1271 ((flags & MSG_EOF) &&
1272 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1273 (resid <= 0)) ?
1274 PRUS_EOF :
1275 /* If there is more to send set PRUS_MORETOCOME. */
1276 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1277 top, addr, control, td);
1278 if (dontroute) {
1279 SOCK_LOCK(so);
1280 so->so_options &= ~SO_DONTROUTE;
1281 SOCK_UNLOCK(so);
1282 }
1283 clen = 0;
1284 control = NULL;
1285 top = NULL;
1286 if (error)
1287 goto release;
1288 } while (resid && space > 0);
1289 } while (resid);
1290
1291release:
1292 sbunlock(&so->so_snd);
1293out:
1294 if (top != NULL)
1295 m_freem(top);
1296 if (control != NULL)
1297 m_freem(control);
1298 return (error);
1299}
1300
1301int
1302sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
1303 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1304{
1305 int error;
1306
1307 CURVNET_SET(so->so_vnet);
1308 error = so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio, top,
1309 control, flags, td);
1310 CURVNET_RESTORE();
1311 return (error);
1312}
1313
1314/*
1315 * The part of soreceive() that implements reading non-inline out-of-band
1316 * data from a socket. For more complete comments, see soreceive(), from
1317 * which this code originated.
1318 *
1319 * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
1320 * unable to return an mbuf chain to the caller.
1321 */
1322static int
1323soreceive_rcvoob(struct socket *so, struct uio *uio, int flags)
1324{
1325 struct protosw *pr = so->so_proto;
1326 struct mbuf *m;
1327 int error;
1328
1329 KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
1330
1331 m = m_get(M_WAIT, MT_DATA);
1332 error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK);
1333 if (error)
1334 goto bad;
1335 do {
1336#ifdef ZERO_COPY_SOCKETS
1337 if (so_zero_copy_receive) {
1338 int disposable;
1339
1340 if ((m->m_flags & M_EXT)
1341 && (m->m_ext.ext_type == EXT_DISPOSABLE))
1342 disposable = 1;
1343 else
1344 disposable = 0;
1345
1346 error = uiomoveco(mtod(m, void *),
1347 min(uio->uio_resid, m->m_len),
1348 uio, disposable);
1349 } else
1350#endif /* ZERO_COPY_SOCKETS */
1351 error = uiomove(mtod(m, void *),
1352 (int) min(uio->uio_resid, m->m_len), uio);
1353 m = m_free(m);
1354 } while (uio->uio_resid && error == 0 && m);
1355bad:
1356 if (m != NULL)
1357 m_freem(m);
1358 return (error);
1359}
1360
1361/*
1362 * Following replacement or removal of the first mbuf on the first mbuf chain
1363 * of a socket buffer, push necessary state changes back into the socket
1364 * buffer so that other consumers see the values consistently. 'nextrecord'
1365 * is the callers locally stored value of the original value of
1366 * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes.
1367 * NOTE: 'nextrecord' may be NULL.
1368 */
1369static __inline void
1370sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord)
1371{
1372
1373 SOCKBUF_LOCK_ASSERT(sb);
1374 /*
1375 * First, update for the new value of nextrecord. If necessary, make
1376 * it the first record.
1377 */
1378 if (sb->sb_mb != NULL)
1379 sb->sb_mb->m_nextpkt = nextrecord;
1380 else
1381 sb->sb_mb = nextrecord;
1382
1383 /*
1384 * Now update any dependent socket buffer fields to reflect the new
1385 * state. This is an expanded inline of SB_EMPTY_FIXUP(), with the
1386 * addition of a second clause that takes care of the case where
1387 * sb_mb has been updated, but remains the last record.
1388 */
1389 if (sb->sb_mb == NULL) {
1390 sb->sb_mbtail = NULL;
1391 sb->sb_lastrecord = NULL;
1392 } else if (sb->sb_mb->m_nextpkt == NULL)
1393 sb->sb_lastrecord = sb->sb_mb;
1394}
1395
1396
1397/*
1398 * Implement receive operations on a socket. We depend on the way that
1399 * records are added to the sockbuf by sbappend. In particular, each record
1400 * (mbufs linked through m_next) must begin with an address if the protocol
1401 * so specifies, followed by an optional mbuf or mbufs containing ancillary
1402 * data, and then zero or more mbufs of data. In order to allow parallelism
1403 * between network receive and copying to user space, as well as avoid
1404 * sleeping with a mutex held, we release the socket buffer mutex during the
1405 * user space copy. Although the sockbuf is locked, new data may still be
1406 * appended, and thus we must maintain consistency of the sockbuf during that
1407 * time.
1408 *
1409 * The caller may receive the data as a single mbuf chain by supplying an
1410 * mbuf **mp0 for use in returning the chain. The uio is then used only for
1411 * the count in uio_resid.
1412 */
1413int
1414soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio,
1415 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1416{
1417 struct mbuf *m, **mp;
1418 int flags, len, error, offset;
1419 struct protosw *pr = so->so_proto;
1420 struct mbuf *nextrecord;
1421 int moff, type = 0;
1422 int orig_resid = uio->uio_resid;
1423
1424 mp = mp0;
1425 if (psa != NULL)
1426 *psa = NULL;
1427 if (controlp != NULL)
1428 *controlp = NULL;
1429 if (flagsp != NULL)
1430 flags = *flagsp &~ MSG_EOR;
1431 else
1432 flags = 0;
1433 if (flags & MSG_OOB)
1434 return (soreceive_rcvoob(so, uio, flags));
1435 if (mp != NULL)
1436 *mp = NULL;
1437 if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING)
1438 && uio->uio_resid)
1439 (*pr->pr_usrreqs->pru_rcvd)(so, 0);
1440
1441 error = sblock(&so->so_rcv, SBLOCKWAIT(flags));
1442 if (error)
1443 return (error);
1444
1445restart:
1446 SOCKBUF_LOCK(&so->so_rcv);
1447 m = so->so_rcv.sb_mb;
1448 /*
1449 * If we have less data than requested, block awaiting more (subject
1450 * to any timeout) if:
1451 * 1. the current count is less than the low water mark, or
1452 * 2. MSG_WAITALL is set, and it is possible to do the entire
1453 * receive operation at once if we block (resid <= hiwat).
1454 * 3. MSG_DONTWAIT is not set
1455 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1456 * we have to do the receive in sections, and thus risk returning a
1457 * short count if a timeout or signal occurs after we start.
1458 */
1459 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1460 so->so_rcv.sb_cc < uio->uio_resid) &&
1461 (so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
1462 ((flags & MSG_WAITALL) && uio->uio_resid <= so->so_rcv.sb_hiwat)) &&
1463 m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
1464 KASSERT(m != NULL || !so->so_rcv.sb_cc,
1465 ("receive: m == %p so->so_rcv.sb_cc == %u",
1466 m, so->so_rcv.sb_cc));
1467 if (so->so_error) {
1468 if (m != NULL)
1469 goto dontblock;
1470 error = so->so_error;
1471 if ((flags & MSG_PEEK) == 0)
1472 so->so_error = 0;
1473 SOCKBUF_UNLOCK(&so->so_rcv);
1474 goto release;
1475 }
1476 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1477 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1478 if (m == NULL) {
1479 SOCKBUF_UNLOCK(&so->so_rcv);
1480 goto release;
1481 } else
1482 goto dontblock;
1483 }
1484 for (; m != NULL; m = m->m_next)
1485 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) {
1486 m = so->so_rcv.sb_mb;
1487 goto dontblock;
1488 }
1489 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1490 (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1491 SOCKBUF_UNLOCK(&so->so_rcv);
1492 error = ENOTCONN;
1493 goto release;
1494 }
1495 if (uio->uio_resid == 0) {
1496 SOCKBUF_UNLOCK(&so->so_rcv);
1497 goto release;
1498 }
1499 if ((so->so_state & SS_NBIO) ||
1500 (flags & (MSG_DONTWAIT|MSG_NBIO))) {
1501 SOCKBUF_UNLOCK(&so->so_rcv);
1502 error = EWOULDBLOCK;
1503 goto release;
1504 }
1505 SBLASTRECORDCHK(&so->so_rcv);
1506 SBLASTMBUFCHK(&so->so_rcv);
1507 error = sbwait(&so->so_rcv);
1508 SOCKBUF_UNLOCK(&so->so_rcv);
1509 if (error)
1510 goto release;
1511 goto restart;
1512 }
1513dontblock:
1514 /*
1515 * From this point onward, we maintain 'nextrecord' as a cache of the
1516 * pointer to the next record in the socket buffer. We must keep the
1517 * various socket buffer pointers and local stack versions of the
1518 * pointers in sync, pushing out modifications before dropping the
1519 * socket buffer mutex, and re-reading them when picking it up.
1520 *
1521 * Otherwise, we will race with the network stack appending new data
1522 * or records onto the socket buffer by using inconsistent/stale
1523 * versions of the field, possibly resulting in socket buffer
1524 * corruption.
1525 *
1526 * By holding the high-level sblock(), we prevent simultaneous
1527 * readers from pulling off the front of the socket buffer.
1528 */
1529 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1530 if (uio->uio_td)
1531 uio->uio_td->td_ru.ru_msgrcv++;
1532 KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb"));
1533 SBLASTRECORDCHK(&so->so_rcv);
1534 SBLASTMBUFCHK(&so->so_rcv);
1535 nextrecord = m->m_nextpkt;
1536 if (pr->pr_flags & PR_ADDR) {
1537 KASSERT(m->m_type == MT_SONAME,
1538 ("m->m_type == %d", m->m_type));
1539 orig_resid = 0;
1540 if (psa != NULL)
1541 *psa = sodupsockaddr(mtod(m, struct sockaddr *),
1542 M_NOWAIT);
1543 if (flags & MSG_PEEK) {
1544 m = m->m_next;
1545 } else {
1546 sbfree(&so->so_rcv, m);
1547 so->so_rcv.sb_mb = m_free(m);
1548 m = so->so_rcv.sb_mb;
1549 sockbuf_pushsync(&so->so_rcv, nextrecord);
1550 }
1551 }
1552
1553 /*
1554 * Process one or more MT_CONTROL mbufs present before any data mbufs
1555 * in the first mbuf chain on the socket buffer. If MSG_PEEK, we
1556 * just copy the data; if !MSG_PEEK, we call into the protocol to
1557 * perform externalization (or freeing if controlp == NULL).
1558 */
1559 if (m != NULL && m->m_type == MT_CONTROL) {
1560 struct mbuf *cm = NULL, *cmn;
1561 struct mbuf **cme = &cm;
1562
1563 do {
1564 if (flags & MSG_PEEK) {
1565 if (controlp != NULL) {
1566 *controlp = m_copy(m, 0, m->m_len);
1567 controlp = &(*controlp)->m_next;
1568 }
1569 m = m->m_next;
1570 } else {
1571 sbfree(&so->so_rcv, m);
1572 so->so_rcv.sb_mb = m->m_next;
1573 m->m_next = NULL;
1574 *cme = m;
1575 cme = &(*cme)->m_next;
1576 m = so->so_rcv.sb_mb;
1577 }
1578 } while (m != NULL && m->m_type == MT_CONTROL);
1579 if ((flags & MSG_PEEK) == 0)
1580 sockbuf_pushsync(&so->so_rcv, nextrecord);
1581 while (cm != NULL) {
1582 cmn = cm->m_next;
1583 cm->m_next = NULL;
1584 if (pr->pr_domain->dom_externalize != NULL) {
1585 SOCKBUF_UNLOCK(&so->so_rcv);
1586 error = (*pr->pr_domain->dom_externalize)
1587 (cm, controlp);
1588 SOCKBUF_LOCK(&so->so_rcv);
1589 } else if (controlp != NULL)
1590 *controlp = cm;
1591 else
1592 m_freem(cm);
1593 if (controlp != NULL) {
1594 orig_resid = 0;
1595 while (*controlp != NULL)
1596 controlp = &(*controlp)->m_next;
1597 }
1598 cm = cmn;
1599 }
1600 if (m != NULL)
1601 nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1602 else
1603 nextrecord = so->so_rcv.sb_mb;
1604 orig_resid = 0;
1605 }
1606 if (m != NULL) {
1607 if ((flags & MSG_PEEK) == 0) {
1608 KASSERT(m->m_nextpkt == nextrecord,
1609 ("soreceive: post-control, nextrecord !sync"));
1610 if (nextrecord == NULL) {
1611 KASSERT(so->so_rcv.sb_mb == m,
1612 ("soreceive: post-control, sb_mb!=m"));
1613 KASSERT(so->so_rcv.sb_lastrecord == m,
1614 ("soreceive: post-control, lastrecord!=m"));
1615 }
1616 }
1617 type = m->m_type;
1618 if (type == MT_OOBDATA)
1619 flags |= MSG_OOB;
1620 } else {
1621 if ((flags & MSG_PEEK) == 0) {
1622 KASSERT(so->so_rcv.sb_mb == nextrecord,
1623 ("soreceive: sb_mb != nextrecord"));
1624 if (so->so_rcv.sb_mb == NULL) {
1625 KASSERT(so->so_rcv.sb_lastrecord == NULL,
1626 ("soreceive: sb_lastercord != NULL"));
1627 }
1628 }
1629 }
1630 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1631 SBLASTRECORDCHK(&so->so_rcv);
1632 SBLASTMBUFCHK(&so->so_rcv);
1633
1634 /*
1635 * Now continue to read any data mbufs off of the head of the socket
1636 * buffer until the read request is satisfied. Note that 'type' is
1637 * used to store the type of any mbuf reads that have happened so far
1638 * such that soreceive() can stop reading if the type changes, which
1639 * causes soreceive() to return only one of regular data and inline
1640 * out-of-band data in a single socket receive operation.
1641 */
1642 moff = 0;
1643 offset = 0;
1644 while (m != NULL && uio->uio_resid > 0 && error == 0) {
1645 /*
1646 * If the type of mbuf has changed since the last mbuf
1647 * examined ('type'), end the receive operation.
1648 */
1649 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1650 if (m->m_type == MT_OOBDATA) {
1651 if (type != MT_OOBDATA)
1652 break;
1653 } else if (type == MT_OOBDATA)
1654 break;
1655 else
1656 KASSERT(m->m_type == MT_DATA,
1657 ("m->m_type == %d", m->m_type));
1658 so->so_rcv.sb_state &= ~SBS_RCVATMARK;
1659 len = uio->uio_resid;
1660 if (so->so_oobmark && len > so->so_oobmark - offset)
1661 len = so->so_oobmark - offset;
1662 if (len > m->m_len - moff)
1663 len = m->m_len - moff;
1664 /*
1665 * If mp is set, just pass back the mbufs. Otherwise copy
1666 * them out via the uio, then free. Sockbuf must be
1667 * consistent here (points to current mbuf, it points to next
1668 * record) when we drop priority; we must note any additions
1669 * to the sockbuf when we block interrupts again.
1670 */
1671 if (mp == NULL) {
1672 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1673 SBLASTRECORDCHK(&so->so_rcv);
1674 SBLASTMBUFCHK(&so->so_rcv);
1675 SOCKBUF_UNLOCK(&so->so_rcv);
1676#ifdef ZERO_COPY_SOCKETS
1677 if (so_zero_copy_receive) {
1678 int disposable;
1679
1680 if ((m->m_flags & M_EXT)
1681 && (m->m_ext.ext_type == EXT_DISPOSABLE))
1682 disposable = 1;
1683 else
1684 disposable = 0;
1685
1686 error = uiomoveco(mtod(m, char *) + moff,
1687 (int)len, uio,
1688 disposable);
1689 } else
1690#endif /* ZERO_COPY_SOCKETS */
1691 error = uiomove(mtod(m, char *) + moff, (int)len, uio);
1692 SOCKBUF_LOCK(&so->so_rcv);
1693 if (error) {
1694 /*
1695 * The MT_SONAME mbuf has already been removed
1696 * from the record, so it is necessary to
1697 * remove the data mbufs, if any, to preserve
1698 * the invariant in the case of PR_ADDR that
1699 * requires MT_SONAME mbufs at the head of
1700 * each record.
1701 */
1702 if (m && pr->pr_flags & PR_ATOMIC &&
1703 ((flags & MSG_PEEK) == 0))
1704 (void)sbdroprecord_locked(&so->so_rcv);
1705 SOCKBUF_UNLOCK(&so->so_rcv);
1706 goto release;
1707 }
1708 } else
1709 uio->uio_resid -= len;
1710 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1711 if (len == m->m_len - moff) {
1712 if (m->m_flags & M_EOR)
1713 flags |= MSG_EOR;
1714 if (flags & MSG_PEEK) {
1715 m = m->m_next;
1716 moff = 0;
1717 } else {
1718 nextrecord = m->m_nextpkt;
1719 sbfree(&so->so_rcv, m);
1720 if (mp != NULL) {
1721 *mp = m;
1722 mp = &m->m_next;
1723 so->so_rcv.sb_mb = m = m->m_next;
1724 *mp = NULL;
1725 } else {
1726 so->so_rcv.sb_mb = m_free(m);
1727 m = so->so_rcv.sb_mb;
1728 }
1729 sockbuf_pushsync(&so->so_rcv, nextrecord);
1730 SBLASTRECORDCHK(&so->so_rcv);
1731 SBLASTMBUFCHK(&so->so_rcv);
1732 }
1733 } else {
1734 if (flags & MSG_PEEK)
1735 moff += len;
1736 else {
1737 if (mp != NULL) {
1738 int copy_flag;
1739
1740 if (flags & MSG_DONTWAIT)
1741 copy_flag = M_DONTWAIT;
1742 else
1743 copy_flag = M_WAIT;
1744 if (copy_flag == M_WAIT)
1745 SOCKBUF_UNLOCK(&so->so_rcv);
1746 *mp = m_copym(m, 0, len, copy_flag);
1747 if (copy_flag == M_WAIT)
1748 SOCKBUF_LOCK(&so->so_rcv);
1749 if (*mp == NULL) {
1750 /*
1751 * m_copym() couldn't
1752 * allocate an mbuf. Adjust
1753 * uio_resid back (it was
1754 * adjusted down by len
1755 * bytes, which we didn't end
1756 * up "copying" over).
1757 */
1758 uio->uio_resid += len;
1759 break;
1760 }
1761 }
1762 m->m_data += len;
1763 m->m_len -= len;
1764 so->so_rcv.sb_cc -= len;
1765 }
1766 }
1767 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1768 if (so->so_oobmark) {
1769 if ((flags & MSG_PEEK) == 0) {
1770 so->so_oobmark -= len;
1771 if (so->so_oobmark == 0) {
1772 so->so_rcv.sb_state |= SBS_RCVATMARK;
1773 break;
1774 }
1775 } else {
1776 offset += len;
1777 if (offset == so->so_oobmark)
1778 break;
1779 }
1780 }
1781 if (flags & MSG_EOR)
1782 break;
1783 /*
1784 * If the MSG_WAITALL flag is set (for non-atomic socket), we
1785 * must not quit until "uio->uio_resid == 0" or an error
1786 * termination. If a signal/timeout occurs, return with a
1787 * short count but without error. Keep sockbuf locked
1788 * against other readers.
1789 */
1790 while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1791 !sosendallatonce(so) && nextrecord == NULL) {
1792 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1793 if (so->so_error || so->so_rcv.sb_state & SBS_CANTRCVMORE)
1794 break;
1795 /*
1796 * Notify the protocol that some data has been
1797 * drained before blocking.
1798 */
1799 if (pr->pr_flags & PR_WANTRCVD) {
1800 SOCKBUF_UNLOCK(&so->so_rcv);
1801 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
1802 SOCKBUF_LOCK(&so->so_rcv);
1803 }
1804 SBLASTRECORDCHK(&so->so_rcv);
1805 SBLASTMBUFCHK(&so->so_rcv);
1806 error = sbwait(&so->so_rcv);
1807 if (error) {
1808 SOCKBUF_UNLOCK(&so->so_rcv);
1809 goto release;
1810 }
1811 m = so->so_rcv.sb_mb;
1812 if (m != NULL)
1813 nextrecord = m->m_nextpkt;
1814 }
1815 }
1816
1817 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1818 if (m != NULL && pr->pr_flags & PR_ATOMIC) {
1819 flags |= MSG_TRUNC;
1820 if ((flags & MSG_PEEK) == 0)
1821 (void) sbdroprecord_locked(&so->so_rcv);
1822 }
1823 if ((flags & MSG_PEEK) == 0) {
1824 if (m == NULL) {
1825 /*
1826 * First part is an inline SB_EMPTY_FIXUP(). Second
1827 * part makes sure sb_lastrecord is up-to-date if
1828 * there is still data in the socket buffer.
1829 */
1830 so->so_rcv.sb_mb = nextrecord;
1831 if (so->so_rcv.sb_mb == NULL) {
1832 so->so_rcv.sb_mbtail = NULL;
1833 so->so_rcv.sb_lastrecord = NULL;
1834 } else if (nextrecord->m_nextpkt == NULL)
1835 so->so_rcv.sb_lastrecord = nextrecord;
1836 }
1837 SBLASTRECORDCHK(&so->so_rcv);
1838 SBLASTMBUFCHK(&so->so_rcv);
1839 /*
1840 * If soreceive() is being done from the socket callback,
1841 * then don't need to generate ACK to peer to update window,
1842 * since ACK will be generated on return to TCP.
1843 */
1844 if (!(flags & MSG_SOCALLBCK) &&
1845 (pr->pr_flags & PR_WANTRCVD)) {
1846 SOCKBUF_UNLOCK(&so->so_rcv);
1847 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
1848 SOCKBUF_LOCK(&so->so_rcv);
1849 }
1850 }
1851 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1852 if (orig_resid == uio->uio_resid && orig_resid &&
1853 (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
1854 SOCKBUF_UNLOCK(&so->so_rcv);
1855 goto restart;
1856 }
1857 SOCKBUF_UNLOCK(&so->so_rcv);
1858
1859 if (flagsp != NULL)
1860 *flagsp |= flags;
1861release:
1862 sbunlock(&so->so_rcv);
1863 return (error);
1864}
1865
1866/*
1867 * Optimized version of soreceive() for simple datagram cases from userspace.
1868 * Unlike in the stream case, we're able to drop a datagram if copyout()
1869 * fails, and because we handle datagrams atomically, we don't need to use a
1870 * sleep lock to prevent I/O interlacing.
1871 */
1872int
1873soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio,
1874 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1875{
1876 struct mbuf *m, *m2;
1877 int flags, len, error;
1878 struct protosw *pr = so->so_proto;
1879 struct mbuf *nextrecord;
1880
1881 if (psa != NULL)
1882 *psa = NULL;
1883 if (controlp != NULL)
1884 *controlp = NULL;
1885 if (flagsp != NULL)
1886 flags = *flagsp &~ MSG_EOR;
1887 else
1888 flags = 0;
1889
1890 /*
1891 * For any complicated cases, fall back to the full
1892 * soreceive_generic().
1893 */
1894 if (mp0 != NULL || (flags & MSG_PEEK) || (flags & MSG_OOB))
1895 return (soreceive_generic(so, psa, uio, mp0, controlp,
1896 flagsp));
1897
1898 /*
1899 * Enforce restrictions on use.
1900 */
1901 KASSERT((pr->pr_flags & PR_WANTRCVD) == 0,
1902 ("soreceive_dgram: wantrcvd"));
1903 KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic"));
1904 KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0,
1905 ("soreceive_dgram: SBS_RCVATMARK"));
1906 KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0,
1907 ("soreceive_dgram: P_CONNREQUIRED"));
1908
1909 /*
1910 * Loop blocking while waiting for a datagram.
1911 */
1912 SOCKBUF_LOCK(&so->so_rcv);
1913 while ((m = so->so_rcv.sb_mb) == NULL) {
1914 KASSERT(so->so_rcv.sb_cc == 0,
1915 ("soreceive_dgram: sb_mb NULL but sb_cc %u",
1916 so->so_rcv.sb_cc));
1917 if (so->so_error) {
1918 error = so->so_error;
1919 so->so_error = 0;
1920 SOCKBUF_UNLOCK(&so->so_rcv);
1921 return (error);
1922 }
1923 if (so->so_rcv.sb_state & SBS_CANTRCVMORE ||
1924 uio->uio_resid == 0) {
1925 SOCKBUF_UNLOCK(&so->so_rcv);
1926 return (0);
1927 }
1928 if ((so->so_state & SS_NBIO) ||
1929 (flags & (MSG_DONTWAIT|MSG_NBIO))) {
1930 SOCKBUF_UNLOCK(&so->so_rcv);
1931 return (EWOULDBLOCK);
1932 }
1933 SBLASTRECORDCHK(&so->so_rcv);
1934 SBLASTMBUFCHK(&so->so_rcv);
1935 error = sbwait(&so->so_rcv);
1936 if (error) {
1937 SOCKBUF_UNLOCK(&so->so_rcv);
1938 return (error);
1939 }
1940 }
1941 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1942
1943 if (uio->uio_td)
1944 uio->uio_td->td_ru.ru_msgrcv++;
1945 SBLASTRECORDCHK(&so->so_rcv);
1946 SBLASTMBUFCHK(&so->so_rcv);
1947 nextrecord = m->m_nextpkt;
1948 if (nextrecord == NULL) {
1949 KASSERT(so->so_rcv.sb_lastrecord == m,
1950 ("soreceive_dgram: lastrecord != m"));
1951 }
1952
1953 KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord,
1954 ("soreceive_dgram: m_nextpkt != nextrecord"));
1955
1956 /*
1957 * Pull 'm' and its chain off the front of the packet queue.
1958 */
1959 so->so_rcv.sb_mb = NULL;
1960 sockbuf_pushsync(&so->so_rcv, nextrecord);
1961
1962 /*
1963 * Walk 'm's chain and free that many bytes from the socket buffer.
1964 */
1965 for (m2 = m; m2 != NULL; m2 = m2->m_next)
1966 sbfree(&so->so_rcv, m2);
1967
1968 /*
1969 * Do a few last checks before we let go of the lock.
1970 */
1971 SBLASTRECORDCHK(&so->so_rcv);
1972 SBLASTMBUFCHK(&so->so_rcv);
1973 SOCKBUF_UNLOCK(&so->so_rcv);
1974
1975 if (pr->pr_flags & PR_ADDR) {
1976 KASSERT(m->m_type == MT_SONAME,
1977 ("m->m_type == %d", m->m_type));
1978 if (psa != NULL)
1979 *psa = sodupsockaddr(mtod(m, struct sockaddr *),
1980 M_NOWAIT);
1981 m = m_free(m);
1982 }
1983 if (m == NULL) {
1984 /* XXXRW: Can this happen? */
1985 return (0);
1986 }
1987
1988 /*
1989 * Packet to copyout() is now in 'm' and it is disconnected from the
1990 * queue.
1991 *
1992 * Process one or more MT_CONTROL mbufs present before any data mbufs
1993 * in the first mbuf chain on the socket buffer. We call into the
1994 * protocol to perform externalization (or freeing if controlp ==
1995 * NULL).
1996 */
1997 if (m->m_type == MT_CONTROL) {
1998 struct mbuf *cm = NULL, *cmn;
1999 struct mbuf **cme = &cm;
2000
2001 do {
2002 m2 = m->m_next;
2003 m->m_next = NULL;
2004 *cme = m;
2005 cme = &(*cme)->m_next;
2006 m = m2;
2007 } while (m != NULL && m->m_type == MT_CONTROL);
2008 while (cm != NULL) {
2009 cmn = cm->m_next;
2010 cm->m_next = NULL;
2011 if (pr->pr_domain->dom_externalize != NULL) {
2012 error = (*pr->pr_domain->dom_externalize)
2013 (cm, controlp);
2014 } else if (controlp != NULL)
2015 *controlp = cm;
2016 else
2017 m_freem(cm);
2018 if (controlp != NULL) {
2019 while (*controlp != NULL)
2020 controlp = &(*controlp)->m_next;
2021 }
2022 cm = cmn;
2023 }
2024 }
2025 KASSERT(m->m_type == MT_DATA, ("soreceive_dgram: !data"));
2026
2027 while (m != NULL && uio->uio_resid > 0) {
2028 len = uio->uio_resid;
2029 if (len > m->m_len)
2030 len = m->m_len;
2031 error = uiomove(mtod(m, char *), (int)len, uio);
2032 if (error) {
2033 m_freem(m);
2034 return (error);
2035 }
2036 m = m_free(m);
2037 }
2038 if (m != NULL)
2039 flags |= MSG_TRUNC;
2040 m_freem(m);
2041 if (flagsp != NULL)
2042 *flagsp |= flags;
2043 return (0);
2044}
2045
2046int
2047soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
2048 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2049{
2050
2051 return (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio, mp0,
2052 controlp, flagsp));
2053}
2054
2055int
2056soshutdown(struct socket *so, int how)
2057{
2058 struct protosw *pr = so->so_proto;
2059 int error;
2060
2061 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
2062 return (EINVAL);
2063 if (pr->pr_usrreqs->pru_flush != NULL) {
2064 (*pr->pr_usrreqs->pru_flush)(so, how);
2065 }
2066 if (how != SHUT_WR)
2067 sorflush(so);
2068 if (how != SHUT_RD) {
2069 CURVNET_SET(so->so_vnet);
2070 error = (*pr->pr_usrreqs->pru_shutdown)(so);
2071 CURVNET_RESTORE();
2072 return (error);
2073 }
2074 return (0);
2075}
2076
2077void
2078sorflush(struct socket *so)
2079{
2080 struct sockbuf *sb = &so->so_rcv;
2081 struct protosw *pr = so->so_proto;
2082 struct sockbuf asb;
2083
2084 /*
2085 * In order to avoid calling dom_dispose with the socket buffer mutex
2086 * held, and in order to generally avoid holding the lock for a long
2087 * time, we make a copy of the socket buffer and clear the original
2088 * (except locks, state). The new socket buffer copy won't have
2089 * initialized locks so we can only call routines that won't use or
2090 * assert those locks.
2091 *
2092 * Dislodge threads currently blocked in receive and wait to acquire
2093 * a lock against other simultaneous readers before clearing the
2094 * socket buffer. Don't let our acquire be interrupted by a signal
2095 * despite any existing socket disposition on interruptable waiting.
2096 */
2097 CURVNET_SET(so->so_vnet);
2098 socantrcvmore(so);
2099 (void) sblock(sb, SBL_WAIT | SBL_NOINTR);
2100
2101 /*
2102 * Invalidate/clear most of the sockbuf structure, but leave selinfo
2103 * and mutex data unchanged.
2104 */
2105 SOCKBUF_LOCK(sb);
2106 bzero(&asb, offsetof(struct sockbuf, sb_startzero));
2107 bcopy(&sb->sb_startzero, &asb.sb_startzero,
2108 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2109 bzero(&sb->sb_startzero,
2110 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2111 SOCKBUF_UNLOCK(sb);
2112 sbunlock(sb);
2113
2114 /*
2115 * Dispose of special rights and flush the socket buffer. Don't call
2116 * any unsafe routines (that rely on locks being initialized) on asb.
2117 */
2118 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
2119 (*pr->pr_domain->dom_dispose)(asb.sb_mb);
2120 sbrelease_internal(&asb, so);
2121 CURVNET_RESTORE();
2122}
2123
2124/*
2125 * Perhaps this routine, and sooptcopyout(), below, ought to come in an
2126 * additional variant to handle the case where the option value needs to be
2127 * some kind of integer, but not a specific size. In addition to their use
2128 * here, these functions are also called by the protocol-level pr_ctloutput()
2129 * routines.
2130 */
2131int
2132sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2133{
2134 size_t valsize;
2135
2136 /*
2137 * If the user gives us more than we wanted, we ignore it, but if we
2138 * don't get the minimum length the caller wants, we return EINVAL.
2139 * On success, sopt->sopt_valsize is set to however much we actually
2140 * retrieved.
2141 */
2142 if ((valsize = sopt->sopt_valsize) < minlen)
2143 return EINVAL;
2144 if (valsize > len)
2145 sopt->sopt_valsize = valsize = len;
2146
2147 if (sopt->sopt_td != NULL)
2148 return (copyin(sopt->sopt_val, buf, valsize));
2149
2150 bcopy(sopt->sopt_val, buf, valsize);
2151 return (0);
2152}
2153
2154/*
2155 * Kernel version of setsockopt(2).
2156 *
2157 * XXX: optlen is size_t, not socklen_t
2158 */
2159int
2160so_setsockopt(struct socket *so, int level, int optname, void *optval,
2161 size_t optlen)
2162{
2163 struct sockopt sopt;
2164
2165 sopt.sopt_level = level;
2166 sopt.sopt_name = optname;
2167 sopt.sopt_dir = SOPT_SET;
2168 sopt.sopt_val = optval;
2169 sopt.sopt_valsize = optlen;
2170 sopt.sopt_td = NULL;
2171 return (sosetopt(so, &sopt));
2172}
2173
2174int
2175sosetopt(struct socket *so, struct sockopt *sopt)
2176{
2177 int error, optval;
2178 struct linger l;
2179 struct timeval tv;
2180 u_long val;
2181#ifdef MAC
2182 struct mac extmac;
2183#endif
2184
2185 error = 0;
2186 if (sopt->sopt_level != SOL_SOCKET) {
2187 if (so->so_proto && so->so_proto->pr_ctloutput)
2188 return ((*so->so_proto->pr_ctloutput)
2189 (so, sopt));
2190 error = ENOPROTOOPT;
2191 } else {
2192 switch (sopt->sopt_name) {
2193#ifdef INET
2194 case SO_ACCEPTFILTER:
2195 error = do_setopt_accept_filter(so, sopt);
2196 if (error)
2197 goto bad;
2198 break;
2199#endif
2200 case SO_LINGER:
2201 error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
2202 if (error)
2203 goto bad;
2204
2205 SOCK_LOCK(so);
2206 so->so_linger = l.l_linger;
2207 if (l.l_onoff)
2208 so->so_options |= SO_LINGER;
2209 else
2210 so->so_options &= ~SO_LINGER;
2211 SOCK_UNLOCK(so);
2212 break;
2213
2214 case SO_DEBUG:
2215 case SO_KEEPALIVE:
2216 case SO_DONTROUTE:
2217 case SO_USELOOPBACK:
2218 case SO_BROADCAST:
2219 case SO_REUSEADDR:
2220 case SO_REUSEPORT:
2221 case SO_OOBINLINE:
2222 case SO_TIMESTAMP:
2223 case SO_BINTIME:
2224 case SO_NOSIGPIPE:
2225 case SO_NO_DDP:
2226 case SO_NO_OFFLOAD:
2227 error = sooptcopyin(sopt, &optval, sizeof optval,
2228 sizeof optval);
2229 if (error)
2230 goto bad;
2231 SOCK_LOCK(so);
2232 if (optval)
2233 so->so_options |= sopt->sopt_name;
2234 else
2235 so->so_options &= ~sopt->sopt_name;
2236 SOCK_UNLOCK(so);
2237 break;
2238
2239 case SO_SETFIB:
2240 error = sooptcopyin(sopt, &optval, sizeof optval,
2241 sizeof optval);
2242 if (optval < 1 || optval > rt_numfibs) {
2243 error = EINVAL;
2244 goto bad;
2245 }
2246 if ((so->so_proto->pr_domain->dom_family == PF_INET) ||
2247 (so->so_proto->pr_domain->dom_family == PF_ROUTE)) {
2248 so->so_fibnum = optval;
2249 /* Note: ignore error */
2250 if (so->so_proto && so->so_proto->pr_ctloutput)
2251 (*so->so_proto->pr_ctloutput)(so, sopt);
2252 } else {
2253 so->so_fibnum = 0;
2254 }
2255 break;
2256 case SO_SNDBUF:
2257 case SO_RCVBUF:
2258 case SO_SNDLOWAT:
2259 case SO_RCVLOWAT:
2260 error = sooptcopyin(sopt, &optval, sizeof optval,
2261 sizeof optval);
2262 if (error)
2263 goto bad;
2264
2265 /*
2266 * Values < 1 make no sense for any of these options,
2267 * so disallow them.
2268 */
2269 if (optval < 1) {
2270 error = EINVAL;
2271 goto bad;
2272 }
2273
2274 switch (sopt->sopt_name) {
2275 case SO_SNDBUF:
2276 case SO_RCVBUF:
2277 if (sbreserve(sopt->sopt_name == SO_SNDBUF ?
2278 &so->so_snd : &so->so_rcv, (u_long)optval,
2279 so, curthread) == 0) {
2280 error = ENOBUFS;
2281 goto bad;
2282 }
2283 (sopt->sopt_name == SO_SNDBUF ? &so->so_snd :
2284 &so->so_rcv)->sb_flags &= ~SB_AUTOSIZE;
2285 break;
2286
2287 /*
2288 * Make sure the low-water is never greater than the
2289 * high-water.
2290 */
2291 case SO_SNDLOWAT:
2292 SOCKBUF_LOCK(&so->so_snd);
2293 so->so_snd.sb_lowat =
2294 (optval > so->so_snd.sb_hiwat) ?
2295 so->so_snd.sb_hiwat : optval;
2296 SOCKBUF_UNLOCK(&so->so_snd);
2297 break;
2298 case SO_RCVLOWAT:
2299 SOCKBUF_LOCK(&so->so_rcv);
2300 so->so_rcv.sb_lowat =
2301 (optval > so->so_rcv.sb_hiwat) ?
2302 so->so_rcv.sb_hiwat : optval;
2303 SOCKBUF_UNLOCK(&so->so_rcv);
2304 break;
2305 }
2306 break;
2307
2308 case SO_SNDTIMEO:
2309 case SO_RCVTIMEO:
2310#ifdef COMPAT_IA32
2311 if (SV_CURPROC_FLAG(SV_ILP32)) {
2312 struct timeval32 tv32;
2313
2314 error = sooptcopyin(sopt, &tv32, sizeof tv32,
2315 sizeof tv32);
2316 CP(tv32, tv, tv_sec);
2317 CP(tv32, tv, tv_usec);
2318 } else
2319#endif
2320 error = sooptcopyin(sopt, &tv, sizeof tv,
2321 sizeof tv);
2322 if (error)
2323 goto bad;
2324
2325 /* assert(hz > 0); */
2326 if (tv.tv_sec < 0 || tv.tv_sec > INT_MAX / hz ||
2327 tv.tv_usec < 0 || tv.tv_usec >= 1000000) {
2328 error = EDOM;
2329 goto bad;
2330 }
2331 /* assert(tick > 0); */
2332 /* assert(ULONG_MAX - INT_MAX >= 1000000); */
2333 val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / tick;
2334 if (val > INT_MAX) {
2335 error = EDOM;
2336 goto bad;
2337 }
2338 if (val == 0 && tv.tv_usec != 0)
2339 val = 1;
2340
2341 switch (sopt->sopt_name) {
2342 case SO_SNDTIMEO:
2343 so->so_snd.sb_timeo = val;
2344 break;
2345 case SO_RCVTIMEO:
2346 so->so_rcv.sb_timeo = val;
2347 break;
2348 }
2349 break;
2350
2351 case SO_LABEL:
2352#ifdef MAC
2353 error = sooptcopyin(sopt, &extmac, sizeof extmac,
2354 sizeof extmac);
2355 if (error)
2356 goto bad;
2357 error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
2358 so, &extmac);
2359#else
2360 error = EOPNOTSUPP;
2361#endif
2362 break;
2363
2364 default:
2365 error = ENOPROTOOPT;
2366 break;
2367 }
2368 if (error == 0 && so->so_proto != NULL &&
2369 so->so_proto->pr_ctloutput != NULL) {
2370 (void) ((*so->so_proto->pr_ctloutput)
2371 (so, sopt));
2372 }
2373 }
2374bad:
2375 return (error);
2376}
2377
2378/*
2379 * Helper routine for getsockopt.
2380 */
2381int
2382sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
2383{
2384 int error;
2385 size_t valsize;
2386
2387 error = 0;
2388
2389 /*
2390 * Documented get behavior is that we always return a value, possibly
2391 * truncated to fit in the user's buffer. Traditional behavior is
2392 * that we always tell the user precisely how much we copied, rather
2393 * than something useful like the total amount we had available for
2394 * her. Note that this interface is not idempotent; the entire
2395 * answer must generated ahead of time.
2396 */
2397 valsize = min(len, sopt->sopt_valsize);
2398 sopt->sopt_valsize = valsize;
2399 if (sopt->sopt_val != NULL) {
2400 if (sopt->sopt_td != NULL)
2401 error = copyout(buf, sopt->sopt_val, valsize);
2402 else
2403 bcopy(buf, sopt->sopt_val, valsize);
2404 }
2405 return (error);
2406}
2407
2408int
2409sogetopt(struct socket *so, struct sockopt *sopt)
2410{
2411 int error, optval;
2412 struct linger l;
2413 struct timeval tv;
2414#ifdef MAC
2415 struct mac extmac;
2416#endif
2417
2418 error = 0;
2419 if (sopt->sopt_level != SOL_SOCKET) {
2420 if (so->so_proto && so->so_proto->pr_ctloutput) {
2421 return ((*so->so_proto->pr_ctloutput)
2422 (so, sopt));
2423 } else
2424 return (ENOPROTOOPT);
2425 } else {
2426 switch (sopt->sopt_name) {
2427#ifdef INET
2428 case SO_ACCEPTFILTER:
2429 error = do_getopt_accept_filter(so, sopt);
2430 break;
2431#endif
2432 case SO_LINGER:
2433 SOCK_LOCK(so);
2434 l.l_onoff = so->so_options & SO_LINGER;
2435 l.l_linger = so->so_linger;
2436 SOCK_UNLOCK(so);
2437 error = sooptcopyout(sopt, &l, sizeof l);
2438 break;
2439
2440 case SO_USELOOPBACK:
2441 case SO_DONTROUTE:
2442 case SO_DEBUG:
2443 case SO_KEEPALIVE:
2444 case SO_REUSEADDR:
2445 case SO_REUSEPORT:
2446 case SO_BROADCAST:
2447 case SO_OOBINLINE:
2448 case SO_ACCEPTCONN:
2449 case SO_TIMESTAMP:
2450 case SO_BINTIME:
2451 case SO_NOSIGPIPE:
2452 optval = so->so_options & sopt->sopt_name;
2453integer:
2454 error = sooptcopyout(sopt, &optval, sizeof optval);
2455 break;
2456
2457 case SO_TYPE:
2458 optval = so->so_type;
2459 goto integer;
2460
2461 case SO_ERROR:
2462 SOCK_LOCK(so);
2463 optval = so->so_error;
2464 so->so_error = 0;
2465 SOCK_UNLOCK(so);
2466 goto integer;
2467
2468 case SO_SNDBUF:
2469 optval = so->so_snd.sb_hiwat;
2470 goto integer;
2471
2472 case SO_RCVBUF:
2473 optval = so->so_rcv.sb_hiwat;
2474 goto integer;
2475
2476 case SO_SNDLOWAT:
2477 optval = so->so_snd.sb_lowat;
2478 goto integer;
2479
2480 case SO_RCVLOWAT:
2481 optval = so->so_rcv.sb_lowat;
2482 goto integer;
2483
2484 case SO_SNDTIMEO:
2485 case SO_RCVTIMEO:
2486 optval = (sopt->sopt_name == SO_SNDTIMEO ?
2487 so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
2488
2489 tv.tv_sec = optval / hz;
2490 tv.tv_usec = (optval % hz) * tick;
2491#ifdef COMPAT_IA32
2492 if (SV_CURPROC_FLAG(SV_ILP32)) {
2493 struct timeval32 tv32;
2494
2495 CP(tv, tv32, tv_sec);
2496 CP(tv, tv32, tv_usec);
2497 error = sooptcopyout(sopt, &tv32, sizeof tv32);
2498 } else
2499#endif
2500 error = sooptcopyout(sopt, &tv, sizeof tv);
2501 break;
2502
2503 case SO_LABEL:
2504#ifdef MAC
2505 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
2506 sizeof(extmac));
2507 if (error)
2508 return (error);
2509 error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
2510 so, &extmac);
2511 if (error)
2512 return (error);
2513 error = sooptcopyout(sopt, &extmac, sizeof extmac);
2514#else
2515 error = EOPNOTSUPP;
2516#endif
2517 break;
2518
2519 case SO_PEERLABEL:
2520#ifdef MAC
2521 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
2522 sizeof(extmac));
2523 if (error)
2524 return (error);
2525 error = mac_getsockopt_peerlabel(
2526 sopt->sopt_td->td_ucred, so, &extmac);
2527 if (error)
2528 return (error);
2529 error = sooptcopyout(sopt, &extmac, sizeof extmac);
2530#else
2531 error = EOPNOTSUPP;
2532#endif
2533 break;
2534
2535 case SO_LISTENQLIMIT:
2536 optval = so->so_qlimit;
2537 goto integer;
2538
2539 case SO_LISTENQLEN:
2540 optval = so->so_qlen;
2541 goto integer;
2542
2543 case SO_LISTENINCQLEN:
2544 optval = so->so_incqlen;
2545 goto integer;
2546
2547 default:
2548 error = ENOPROTOOPT;
2549 break;
2550 }
2551 return (error);
2552 }
2553}
2554
2555/* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */
2556int
2557soopt_getm(struct sockopt *sopt, struct mbuf **mp)
2558{
2559 struct mbuf *m, *m_prev;
2560 int sopt_size = sopt->sopt_valsize;
2561
2562 MGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT, MT_DATA);
2563 if (m == NULL)
2564 return ENOBUFS;
2565 if (sopt_size > MLEN) {
2566 MCLGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT);
2567 if ((m->m_flags & M_EXT) == 0) {
2568 m_free(m);
2569 return ENOBUFS;
2570 }
2571 m->m_len = min(MCLBYTES, sopt_size);
2572 } else {
2573 m->m_len = min(MLEN, sopt_size);
2574 }
2575 sopt_size -= m->m_len;
2576 *mp = m;
2577 m_prev = m;
2578
2579 while (sopt_size) {
2580 MGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT, MT_DATA);
2581 if (m == NULL) {
2582 m_freem(*mp);
2583 return ENOBUFS;
2584 }
2585 if (sopt_size > MLEN) {
2586 MCLGET(m, sopt->sopt_td != NULL ? M_WAIT :
2587 M_DONTWAIT);
2588 if ((m->m_flags & M_EXT) == 0) {
2589 m_freem(m);
2590 m_freem(*mp);
2591 return ENOBUFS;
2592 }
2593 m->m_len = min(MCLBYTES, sopt_size);
2594 } else {
2595 m->m_len = min(MLEN, sopt_size);
2596 }
2597 sopt_size -= m->m_len;
2598 m_prev->m_next = m;
2599 m_prev = m;
2600 }
2601 return (0);
2602}
2603
2604/* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */
2605int
2606soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
2607{
2608 struct mbuf *m0 = m;
2609
2610 if (sopt->sopt_val == NULL)
2611 return (0);
2612 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
2613 if (sopt->sopt_td != NULL) {
2614 int error;
2615
2616 error = copyin(sopt->sopt_val, mtod(m, char *),
2617 m->m_len);
2618 if (error != 0) {
2619 m_freem(m0);
2620 return(error);
2621 }
2622 } else
2623 bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
2624 sopt->sopt_valsize -= m->m_len;
2625 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
2626 m = m->m_next;
2627 }
2628 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
2629 panic("ip6_sooptmcopyin");
2630 return (0);
2631}
2632
2633/* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */
2634int
2635soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
2636{
2637 struct mbuf *m0 = m;
2638 size_t valsize = 0;
2639
2640 if (sopt->sopt_val == NULL)
2641 return (0);
2642 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
2643 if (sopt->sopt_td != NULL) {
2644 int error;
2645
2646 error = copyout(mtod(m, char *), sopt->sopt_val,
2647 m->m_len);
2648 if (error != 0) {
2649 m_freem(m0);
2650 return(error);
2651 }
2652 } else
2653 bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
2654 sopt->sopt_valsize -= m->m_len;
2655 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
2656 valsize += m->m_len;
2657 m = m->m_next;
2658 }
2659 if (m != NULL) {
2660 /* enough soopt buffer should be given from user-land */
2661 m_freem(m0);
2662 return(EINVAL);
2663 }
2664 sopt->sopt_valsize = valsize;
2665 return (0);
2666}
2667
2668/*
2669 * sohasoutofband(): protocol notifies socket layer of the arrival of new
2670 * out-of-band data, which will then notify socket consumers.
2671 */
2672void
2673sohasoutofband(struct socket *so)
2674{
2675
2676 if (so->so_sigio != NULL)
2677 pgsigio(&so->so_sigio, SIGURG, 0);
2678 selwakeuppri(&so->so_rcv.sb_sel, PSOCK);
2679}
2680
2681int
2682sopoll(struct socket *so, int events, struct ucred *active_cred,
2683 struct thread *td)
2684{
2685
2686 return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred,
2687 td));
2688}
2689
2690int
2691sopoll_generic(struct socket *so, int events, struct ucred *active_cred,
2692 struct thread *td)
2693{
2694 int revents = 0;
2695
2696 SOCKBUF_LOCK(&so->so_snd);
2697 SOCKBUF_LOCK(&so->so_rcv);
2698 if (events & (POLLIN | POLLRDNORM))
2699 if (soreadable(so))
2700 revents |= events & (POLLIN | POLLRDNORM);
2701
2702 if (events & POLLINIGNEOF)
2703 if (so->so_rcv.sb_cc >= so->so_rcv.sb_lowat ||
2704 !TAILQ_EMPTY(&so->so_comp) || so->so_error)
2705 revents |= POLLINIGNEOF;
2706
2707 if (events & (POLLOUT | POLLWRNORM))
2708 if (sowriteable(so))
2709 revents |= events & (POLLOUT | POLLWRNORM);
2710
2711 if (events & (POLLPRI | POLLRDBAND))
2712 if (so->so_oobmark || (so->so_rcv.sb_state & SBS_RCVATMARK))
2713 revents |= events & (POLLPRI | POLLRDBAND);
2714
2715 if (revents == 0) {
2716 if (events &
2717 (POLLIN | POLLINIGNEOF | POLLPRI | POLLRDNORM |
2718 POLLRDBAND)) {
2719 selrecord(td, &so->so_rcv.sb_sel);
2720 so->so_rcv.sb_flags |= SB_SEL;
2721 }
2722
2723 if (events & (POLLOUT | POLLWRNORM)) {
2724 selrecord(td, &so->so_snd.sb_sel);
2725 so->so_snd.sb_flags |= SB_SEL;
2726 }
2727 }
2728
2729 SOCKBUF_UNLOCK(&so->so_rcv);
2730 SOCKBUF_UNLOCK(&so->so_snd);
2731 return (revents);
2732}
2733
2734int
2735soo_kqfilter(struct file *fp, struct knote *kn)
2736{
2737 struct socket *so = kn->kn_fp->f_data;
2738 struct sockbuf *sb;
2739
2740 switch (kn->kn_filter) {
2741 case EVFILT_READ:
2742 if (so->so_options & SO_ACCEPTCONN)
2743 kn->kn_fop = &solisten_filtops;
2744 else
2745 kn->kn_fop = &soread_filtops;
2746 sb = &so->so_rcv;
2747 break;
2748 case EVFILT_WRITE:
2749 kn->kn_fop = &sowrite_filtops;
2750 sb = &so->so_snd;
2751 break;
2752 default:
2753 return (EINVAL);
2754 }
2755
2756 SOCKBUF_LOCK(sb);
2757 knlist_add(&sb->sb_sel.si_note, kn, 1);
2758 sb->sb_flags |= SB_KNOTE;
2759 SOCKBUF_UNLOCK(sb);
2760 return (0);
2761}
2762
2763/*
2764 * Some routines that return EOPNOTSUPP for entry points that are not
2765 * supported by a protocol. Fill in as needed.
2766 */
2767int
2768pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
2769{
2770
2771 return EOPNOTSUPP;
2772}
2773
2774int
2775pru_attach_notsupp(struct socket *so, int proto, struct thread *td)
2776{
2777
2778 return EOPNOTSUPP;
2779}
2780
2781int
2782pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
2783{
2784
2785 return EOPNOTSUPP;
2786}
2787
2788int
2789pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
2790{
2791
2792 return EOPNOTSUPP;
2793}
2794
2795int
2796pru_connect2_notsupp(struct socket *so1, struct socket *so2)
2797{
2798
2799 return EOPNOTSUPP;
2800}
2801
2802int
2803pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
2804 struct ifnet *ifp, struct thread *td)
2805{
2806
2807 return EOPNOTSUPP;
2808}
2809
2810int
2811pru_disconnect_notsupp(struct socket *so)
2812{
2813
2814 return EOPNOTSUPP;
2815}
2816
2817int
2818pru_listen_notsupp(struct socket *so, int backlog, struct thread *td)
2819{
2820
2821 return EOPNOTSUPP;
2822}
2823
2824int
2825pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam)
2826{
2827
2828 return EOPNOTSUPP;
2829}
2830
2831int
2832pru_rcvd_notsupp(struct socket *so, int flags)
2833{
2834
2835 return EOPNOTSUPP;
2836}
2837
2838int
2839pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
2840{
2841
2842 return EOPNOTSUPP;
2843}
2844
2845int
2846pru_send_notsupp(struct socket *so, int flags, struct mbuf *m,
2847 struct sockaddr *addr, struct mbuf *control, struct thread *td)
2848{
2849
2850 return EOPNOTSUPP;
2851}
2852
2853/*
2854 * This isn't really a ``null'' operation, but it's the default one and
2855 * doesn't do anything destructive.
2856 */
2857int
2858pru_sense_null(struct socket *so, struct stat *sb)
2859{
2860
2861 sb->st_blksize = so->so_snd.sb_hiwat;
2862 return 0;
2863}
2864
2865int
2866pru_shutdown_notsupp(struct socket *so)
2867{
2868
2869 return EOPNOTSUPP;
2870}
2871
2872int
2873pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam)
2874{
2875
2876 return EOPNOTSUPP;
2877}
2878
2879int
2880pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio,
2881 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
2882{
2883
2884 return EOPNOTSUPP;
2885}
2886
2887int
2888pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr,
2889 struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2890{
2891
2892 return EOPNOTSUPP;
2893}
2894
2895int
2896pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred,
2897 struct thread *td)
2898{
2899
2900 return EOPNOTSUPP;
2901}
2902
2903static void
2904filt_sordetach(struct knote *kn)
2905{
2906 struct socket *so = kn->kn_fp->f_data;
2907
2908 SOCKBUF_LOCK(&so->so_rcv);
2909 knlist_remove(&so->so_rcv.sb_sel.si_note, kn, 1);
2910 if (knlist_empty(&so->so_rcv.sb_sel.si_note))
2911 so->so_rcv.sb_flags &= ~SB_KNOTE;
2912 SOCKBUF_UNLOCK(&so->so_rcv);
2913}
2914
2915/*ARGSUSED*/
2916static int
2917filt_soread(struct knote *kn, long hint)
2918{
2919 struct socket *so;
2920
2921 so = kn->kn_fp->f_data;
2922 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2923
2924 kn->kn_data = so->so_rcv.sb_cc - so->so_rcv.sb_ctl;
2925 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2926 kn->kn_flags |= EV_EOF;
2927 kn->kn_fflags = so->so_error;
2928 return (1);
2929 } else if (so->so_error) /* temporary udp error */
2930 return (1);
2931 else if (kn->kn_sfflags & NOTE_LOWAT)
2932 return (kn->kn_data >= kn->kn_sdata);
2933 else
2934 return (so->so_rcv.sb_cc >= so->so_rcv.sb_lowat);
2935}
2936
2937static void
2938filt_sowdetach(struct knote *kn)
2939{
2940 struct socket *so = kn->kn_fp->f_data;
2941
2942 SOCKBUF_LOCK(&so->so_snd);
2943 knlist_remove(&so->so_snd.sb_sel.si_note, kn, 1);
2944 if (knlist_empty(&so->so_snd.sb_sel.si_note))
2945 so->so_snd.sb_flags &= ~SB_KNOTE;
2946 SOCKBUF_UNLOCK(&so->so_snd);
2947}
2948
2949/*ARGSUSED*/
2950static int
2951filt_sowrite(struct knote *kn, long hint)
2952{
2953 struct socket *so;
2954
2955 so = kn->kn_fp->f_data;
2956 SOCKBUF_LOCK_ASSERT(&so->so_snd);
2957 kn->kn_data = sbspace(&so->so_snd);
2958 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
2959 kn->kn_flags |= EV_EOF;
2960 kn->kn_fflags = so->so_error;
2961 return (1);
2962 } else if (so->so_error) /* temporary udp error */
2963 return (1);
2964 else if (((so->so_state & SS_ISCONNECTED) == 0) &&
2965 (so->so_proto->pr_flags & PR_CONNREQUIRED))
2966 return (0);
2967 else if (kn->kn_sfflags & NOTE_LOWAT)
2968 return (kn->kn_data >= kn->kn_sdata);
2969 else
2970 return (kn->kn_data >= so->so_snd.sb_lowat);
2971}
2972
2973/*ARGSUSED*/
2974static int
2975filt_solisten(struct knote *kn, long hint)
2976{
2977 struct socket *so = kn->kn_fp->f_data;
2978
2979 kn->kn_data = so->so_qlen;
2980 return (! TAILQ_EMPTY(&so->so_comp));
2981}
2982
2983int
2984socheckuid(struct socket *so, uid_t uid)
2985{
2986
2987 if (so == NULL)
2988 return (EPERM);
2989 if (so->so_cred->cr_uid != uid)
2990 return (EPERM);
2991 return (0);
2992}
2993
2994static int
2995sysctl_somaxconn(SYSCTL_HANDLER_ARGS)
2996{
2997 int error;
2998 int val;
2999
3000 val = somaxconn;
3001 error = sysctl_handle_int(oidp, &val, 0, req);
3002 if (error || !req->newptr )
3003 return (error);
3004
3005 if (val < 1 || val > USHRT_MAX)
3006 return (EINVAL);
3007
3008 somaxconn = val;
3009 return (0);
3010}
3011
3012/*
3013 * These functions are used by protocols to notify the socket layer (and its
3014 * consumers) of state changes in the sockets driven by protocol-side events.
3015 */
3016
3017/*
3018 * Procedures to manipulate state flags of socket and do appropriate wakeups.
3019 *
3020 * Normal sequence from the active (originating) side is that
3021 * soisconnecting() is called during processing of connect() call, resulting
3022 * in an eventual call to soisconnected() if/when the connection is
3023 * established. When the connection is torn down soisdisconnecting() is
3024 * called during processing of disconnect() call, and soisdisconnected() is
3025 * called when the connection to the peer is totally severed. The semantics
3026 * of these routines are such that connectionless protocols can call
3027 * soisconnected() and soisdisconnected() only, bypassing the in-progress
3028 * calls when setting up a ``connection'' takes no time.
3029 *
3030 * From the passive side, a socket is created with two queues of sockets:
3031 * so_incomp for connections in progress and so_comp for connections already
3032 * made and awaiting user acceptance. As a protocol is preparing incoming
3033 * connections, it creates a socket structure queued on so_incomp by calling
3034 * sonewconn(). When the connection is established, soisconnected() is
3035 * called, and transfers the socket structure to so_comp, making it available
3036 * to accept().
3037 *
3038 * If a socket is closed with sockets on either so_incomp or so_comp, these
3039 * sockets are dropped.
3040 *
3041 * If higher-level protocols are implemented in the kernel, the wakeups done
3042 * here will sometimes cause software-interrupt process scheduling.
3043 */
3044void
3045soisconnecting(struct socket *so)
3046{
3047
3048 SOCK_LOCK(so);
3049 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
3050 so->so_state |= SS_ISCONNECTING;
3051 SOCK_UNLOCK(so);
3052}
3053
3054void
3055soisconnected(struct socket *so)
3056{
|