2 * Copyright (c) 1982, 1986, 1988, 1990, 1993
3 * The Regents of the University of California.
4 * Copyright (c) 2004 The FreeBSD Foundation
5 * Copyright (c) 2004-2008 Robert N. M. Watson
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 4. Neither the name of the University nor the names of its contributors
17 * may be used to endorse or promote products derived from this software
18 * without specific prior written permission.
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * @(#)uipc_socket.c 8.3 (Berkeley) 4/15/94
36 * Comments on the socket life cycle:
38 * soalloc() sets of socket layer state for a socket, called only by
39 * socreate() and sonewconn(). Socket layer private.
41 * sodealloc() tears down socket layer state for a socket, called only by
42 * sofree() and sonewconn(). Socket layer private.
44 * pru_attach() associates protocol layer state with an allocated socket;
45 * called only once, may fail, aborting socket allocation. This is called
46 * from socreate() and sonewconn(). Socket layer private.
48 * pru_detach() disassociates protocol layer state from an attached socket,
49 * and will be called exactly once for sockets in which pru_attach() has
50 * been successfully called. If pru_attach() returned an error,
51 * pru_detach() will not be called. Socket layer private.
53 * pru_abort() and pru_close() notify the protocol layer that the last
54 * consumer of a socket is starting to tear down the socket, and that the
55 * protocol should terminate the connection. Historically, pru_abort() also
56 * detached protocol state from the socket state, but this is no longer the
59 * socreate() creates a socket and attaches protocol state. This is a public
60 * interface that may be used by socket layer consumers to create new
63 * sonewconn() creates a socket and attaches protocol state. This is a
64 * public interface that may be used by protocols to create new sockets when
65 * a new connection is received and will be available for accept() on a
68 * soclose() destroys a socket after possibly waiting for it to disconnect.
69 * This is a public interface that socket consumers should use to close and
70 * release a socket when done with it.
72 * soabort() destroys a socket without waiting for it to disconnect (used
73 * only for incoming connections that are already partially or fully
74 * connected). This is used internally by the socket layer when clearing
75 * listen socket queues (due to overflow or close on the listen socket), but
76 * is also a public interface protocols may use to abort connections in
77 * their incomplete listen queues should they no longer be required. Sockets
78 * placed in completed connection listen queues should not be aborted for
79 * reasons described in the comment above the soclose() implementation. This
80 * is not a general purpose close routine, and except in the specific
81 * circumstances described here, should not be used.
83 * sofree() will free a socket and its protocol state if all references on
84 * the socket have been released, and is the public interface to attempt to
85 * free a socket when a reference is removed. This is a socket layer private
88 * NOTE: In addition to socreate() and soclose(), which provide a single
89 * socket reference to the consumer to be managed as required, there are two
90 * calls to explicitly manage socket references, soref(), and sorele().
91 * Currently, these are generally required only when transitioning a socket
92 * from a listen queue to a file descriptor, in order to prevent garbage
93 * collection of the socket at an untimely moment. For a number of reasons,
94 * these interfaces are not preferred, and should be avoided.
96 * NOTE: With regard to VNETs the general rule is that callers do not set
97 * curvnet. Exceptions to this rule include soabort(), sodisconnect(),
98 * sofree() (and with that sorele(), sotryfree()), as well as sonewconn()
99 * and sorflush(), which are usually called from a pre-set VNET context.
100 * sopoll() currently does not need a VNET context to be set.
103 #include <sys/cdefs.h>
104 __FBSDID("$FreeBSD$");
106 #include "opt_inet.h"
107 #include "opt_inet6.h"
108 #include "opt_compat.h"
110 #include <sys/param.h>
111 #include <sys/systm.h>
112 #include <sys/fcntl.h>
113 #include <sys/limits.h>
114 #include <sys/lock.h>
116 #include <sys/malloc.h>
117 #include <sys/mbuf.h>
118 #include <sys/mutex.h>
119 #include <sys/domain.h>
120 #include <sys/file.h> /* for struct knote */
121 #include <sys/kernel.h>
122 #include <sys/event.h>
123 #include <sys/eventhandler.h>
124 #include <sys/poll.h>
125 #include <sys/proc.h>
126 #include <sys/protosw.h>
127 #include <sys/socket.h>
128 #include <sys/socketvar.h>
129 #include <sys/resourcevar.h>
130 #include <net/route.h>
131 #include <sys/signalvar.h>
132 #include <sys/stat.h>
134 #include <sys/sysctl.h>
136 #include <sys/jail.h>
137 #include <sys/syslog.h>
138 #include <netinet/in.h>
140 #include <net/vnet.h>
142 #include <security/mac/mac_framework.h>
146 #ifdef COMPAT_FREEBSD32
147 #include <sys/mount.h>
148 #include <sys/sysent.h>
149 #include <compat/freebsd32/freebsd32.h>
152 static int soreceive_rcvoob(struct socket *so, struct uio *uio,
155 static void filt_sordetach(struct knote *kn);
156 static int filt_soread(struct knote *kn, long hint);
157 static void filt_sowdetach(struct knote *kn);
158 static int filt_sowrite(struct knote *kn, long hint);
159 static int filt_solisten(struct knote *kn, long hint);
161 static struct filterops solisten_filtops = {
163 .f_detach = filt_sordetach,
164 .f_event = filt_solisten,
166 static struct filterops soread_filtops = {
168 .f_detach = filt_sordetach,
169 .f_event = filt_soread,
171 static struct filterops sowrite_filtops = {
173 .f_detach = filt_sowdetach,
174 .f_event = filt_sowrite,
177 so_gen_t so_gencnt; /* generation count for sockets */
179 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
180 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
182 #define VNET_SO_ASSERT(so) \
183 VNET_ASSERT(curvnet != NULL, \
184 ("%s:%d curvnet is NULL, so=%p", __func__, __LINE__, (so)));
187 * Limit on the number of connections in the listen queue waiting
189 * NB: The original sysctl somaxconn is still available but hidden
190 * to prevent confusion about the actual purpose of this number.
192 static int somaxconn = SOMAXCONN;
195 sysctl_somaxconn(SYSCTL_HANDLER_ARGS)
201 error = sysctl_handle_int(oidp, &val, 0, req);
202 if (error || !req->newptr )
205 if (val < 1 || val > USHRT_MAX)
211 SYSCTL_PROC(_kern_ipc, OID_AUTO, soacceptqueue, CTLTYPE_UINT | CTLFLAG_RW,
212 0, sizeof(int), sysctl_somaxconn, "I",
213 "Maximum listen socket pending connection accept queue size");
214 SYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn,
215 CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_SKIP,
216 0, sizeof(int), sysctl_somaxconn, "I",
217 "Maximum listen socket pending connection accept queue size (compat)");
219 static int numopensockets;
220 SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD,
221 &numopensockets, 0, "Number of open sockets");
224 * accept_mtx locks down per-socket fields relating to accept queues. See
225 * socketvar.h for an annotation of the protected fields of struct socket.
227 struct mtx accept_mtx;
228 MTX_SYSINIT(accept_mtx, &accept_mtx, "accept", MTX_DEF);
231 * so_global_mtx protects so_gencnt, numopensockets, and the per-socket
234 static struct mtx so_global_mtx;
235 MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF);
238 * General IPC sysctl name space, used by sockets and a variety of other IPC
241 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC");
244 * Initialize the socket subsystem and set up the socket
247 static uma_zone_t socket_zone;
251 socket_zone_change(void *tag)
254 maxsockets = uma_zone_set_max(socket_zone, maxsockets);
258 socket_init(void *tag)
261 socket_zone = uma_zcreate("socket", sizeof(struct socket), NULL, NULL,
262 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
263 maxsockets = uma_zone_set_max(socket_zone, maxsockets);
264 uma_zone_set_warning(socket_zone, "kern.ipc.maxsockets limit reached");
265 EVENTHANDLER_REGISTER(maxsockets_change, socket_zone_change, NULL,
266 EVENTHANDLER_PRI_FIRST);
268 SYSINIT(socket, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, socket_init, NULL);
271 * Initialise maxsockets. This SYSINIT must be run after
275 init_maxsockets(void *ignored)
278 TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
279 maxsockets = imax(maxsockets, maxfiles);
281 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL);
284 * Sysctl to get and set the maximum global sockets limit. Notify protocols
285 * of the change so that they can update their dependent limits as required.
288 sysctl_maxsockets(SYSCTL_HANDLER_ARGS)
290 int error, newmaxsockets;
292 newmaxsockets = maxsockets;
293 error = sysctl_handle_int(oidp, &newmaxsockets, 0, req);
294 if (error == 0 && req->newptr) {
295 if (newmaxsockets > maxsockets &&
296 newmaxsockets <= maxfiles) {
297 maxsockets = newmaxsockets;
298 EVENTHANDLER_INVOKE(maxsockets_change);
304 SYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets, CTLTYPE_INT|CTLFLAG_RW,
305 &maxsockets, 0, sysctl_maxsockets, "IU",
306 "Maximum number of sockets avaliable");
309 * Socket operation routines. These routines are called by the routines in
310 * sys_socket.c or from a system process, and implement the semantics of
311 * socket operations by switching out to the protocol specific routines.
315 * Get a socket structure from our zone, and initialize it. Note that it
316 * would probably be better to allocate socket and PCB at the same time, but
317 * I'm not convinced that all the protocols can be easily modified to do
320 * soalloc() returns a socket with a ref count of 0.
322 static struct socket *
323 soalloc(struct vnet *vnet)
327 so = uma_zalloc(socket_zone, M_NOWAIT | M_ZERO);
331 if (mac_socket_init(so, M_NOWAIT) != 0) {
332 uma_zfree(socket_zone, so);
336 SOCKBUF_LOCK_INIT(&so->so_snd, "so_snd");
337 SOCKBUF_LOCK_INIT(&so->so_rcv, "so_rcv");
338 sx_init(&so->so_snd.sb_sx, "so_snd_sx");
339 sx_init(&so->so_rcv.sb_sx, "so_rcv_sx");
340 TAILQ_INIT(&so->so_aiojobq);
341 mtx_lock(&so_global_mtx);
342 so->so_gencnt = ++so_gencnt;
345 VNET_ASSERT(vnet != NULL, ("%s:%d vnet is NULL, so=%p",
346 __func__, __LINE__, so));
347 vnet->vnet_sockcnt++;
350 mtx_unlock(&so_global_mtx);
355 * Free the storage associated with a socket at the socket layer, tear down
356 * locks, labels, etc. All protocol state is assumed already to have been
357 * torn down (and possibly never set up) by the caller.
360 sodealloc(struct socket *so)
363 KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count));
364 KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL"));
366 mtx_lock(&so_global_mtx);
367 so->so_gencnt = ++so_gencnt;
368 --numopensockets; /* Could be below, but faster here. */
370 VNET_ASSERT(so->so_vnet != NULL, ("%s:%d so_vnet is NULL, so=%p",
371 __func__, __LINE__, so));
372 so->so_vnet->vnet_sockcnt--;
374 mtx_unlock(&so_global_mtx);
375 if (so->so_rcv.sb_hiwat)
376 (void)chgsbsize(so->so_cred->cr_uidinfo,
377 &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY);
378 if (so->so_snd.sb_hiwat)
379 (void)chgsbsize(so->so_cred->cr_uidinfo,
380 &so->so_snd.sb_hiwat, 0, RLIM_INFINITY);
382 /* remove acccept filter if one is present. */
383 if (so->so_accf != NULL)
384 do_setopt_accept_filter(so, NULL);
387 mac_socket_destroy(so);
390 sx_destroy(&so->so_snd.sb_sx);
391 sx_destroy(&so->so_rcv.sb_sx);
392 SOCKBUF_LOCK_DESTROY(&so->so_snd);
393 SOCKBUF_LOCK_DESTROY(&so->so_rcv);
394 uma_zfree(socket_zone, so);
398 * socreate returns a socket with a ref count of 1. The socket should be
399 * closed with soclose().
402 socreate(int dom, struct socket **aso, int type, int proto,
403 struct ucred *cred, struct thread *td)
410 prp = pffindproto(dom, proto, type);
412 prp = pffindtype(dom, type);
415 /* No support for domain. */
416 if (pffinddomain(dom) == NULL)
417 return (EAFNOSUPPORT);
418 /* No support for socket type. */
419 if (proto == 0 && type != 0)
421 return (EPROTONOSUPPORT);
423 if (prp->pr_usrreqs->pru_attach == NULL ||
424 prp->pr_usrreqs->pru_attach == pru_attach_notsupp)
425 return (EPROTONOSUPPORT);
427 if (prison_check_af(cred, prp->pr_domain->dom_family) != 0)
428 return (EPROTONOSUPPORT);
430 if (prp->pr_type != type)
432 so = soalloc(CRED_TO_VNET(cred));
436 TAILQ_INIT(&so->so_incomp);
437 TAILQ_INIT(&so->so_comp);
439 so->so_cred = crhold(cred);
440 if ((prp->pr_domain->dom_family == PF_INET) ||
441 (prp->pr_domain->dom_family == PF_INET6) ||
442 (prp->pr_domain->dom_family == PF_ROUTE))
443 so->so_fibnum = td->td_proc->p_fibnum;
448 mac_socket_create(cred, so);
450 knlist_init_mtx(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv));
451 knlist_init_mtx(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd));
454 * Auto-sizing of socket buffers is managed by the protocols and
455 * the appropriate flags must be set in the pru_attach function.
457 CURVNET_SET(so->so_vnet);
458 error = (*prp->pr_usrreqs->pru_attach)(so, proto, td);
461 KASSERT(so->so_count == 1, ("socreate: so_count %d",
472 static int regression_sonewconn_earlytest = 1;
473 SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW,
474 ®ression_sonewconn_earlytest, 0, "Perform early sonewconn limit test");
478 * When an attempt at a new connection is noted on a socket which accepts
479 * connections, sonewconn is called. If the connection is possible (subject
480 * to space constraints, etc.) then we allocate a new structure, propoerly
481 * linked into the data structure of the original socket, and return this.
482 * Connstatus may be 0, or SO_ISCONFIRMING, or SO_ISCONNECTED.
484 * Note: the ref count on the socket is 0 on return.
487 sonewconn(struct socket *head, int connstatus)
489 static struct timeval lastover;
490 static struct timeval overinterval = { 60, 0 };
491 static int overcount;
497 over = (head->so_qlen > 3 * head->so_qlimit / 2);
500 if (regression_sonewconn_earlytest && over) {
506 if (ratecheck(&lastover, &overinterval)) {
507 log(LOG_DEBUG, "%s: pcb %p: Listen queue overflow: "
508 "%i already in queue awaiting acceptance "
509 "(%d occurrences)\n",
510 __func__, head->so_pcb, head->so_qlen, overcount);
517 VNET_ASSERT(head->so_vnet != NULL, ("%s:%d so_vnet is NULL, head=%p",
518 __func__, __LINE__, head));
519 so = soalloc(head->so_vnet);
521 log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: "
522 "limit reached or out of memory\n",
523 __func__, head->so_pcb);
526 if ((head->so_options & SO_ACCEPTFILTER) != 0)
529 so->so_type = head->so_type;
530 so->so_options = head->so_options &~ SO_ACCEPTCONN;
531 so->so_linger = head->so_linger;
532 so->so_state = head->so_state | SS_NOFDREF;
533 so->so_fibnum = head->so_fibnum;
534 so->so_proto = head->so_proto;
535 so->so_cred = crhold(head->so_cred);
537 mac_socket_newconn(head, so);
539 knlist_init_mtx(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv));
540 knlist_init_mtx(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd));
541 VNET_SO_ASSERT(head);
542 if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat)) {
544 log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n",
545 __func__, head->so_pcb);
548 if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
550 log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n",
551 __func__, head->so_pcb);
554 so->so_rcv.sb_lowat = head->so_rcv.sb_lowat;
555 so->so_snd.sb_lowat = head->so_snd.sb_lowat;
556 so->so_rcv.sb_timeo = head->so_rcv.sb_timeo;
557 so->so_snd.sb_timeo = head->so_snd.sb_timeo;
558 so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE;
559 so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE;
560 so->so_state |= connstatus;
563 * The accept socket may be tearing down but we just
564 * won a race on the ACCEPT_LOCK.
565 * However, if sctp_peeloff() is called on a 1-to-many
566 * style socket, the SO_ACCEPTCONN doesn't need to be set.
568 if (!(head->so_options & SO_ACCEPTCONN) &&
569 ((head->so_proto->pr_protocol != IPPROTO_SCTP) ||
570 (head->so_type != SOCK_SEQPACKET))) {
573 sofree(so); /* NB: returns ACCEPT_UNLOCK'ed. */
577 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
578 so->so_qstate |= SQ_COMP;
582 * Keep removing sockets from the head until there's room for
583 * us to insert on the tail. In pre-locking revisions, this
584 * was a simple if(), but as we could be racing with other
585 * threads and soabort() requires dropping locks, we must
586 * loop waiting for the condition to be true.
588 while (head->so_incqlen > head->so_qlimit) {
590 sp = TAILQ_FIRST(&head->so_incomp);
591 TAILQ_REMOVE(&head->so_incomp, sp, so_list);
593 sp->so_qstate &= ~SQ_INCOMP;
599 TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list);
600 so->so_qstate |= SQ_INCOMP;
606 wakeup_one(&head->so_timeo);
612 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
616 CURVNET_SET(so->so_vnet);
617 error = (*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td);
623 sobindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
627 CURVNET_SET(so->so_vnet);
628 error = (*so->so_proto->pr_usrreqs->pru_bindat)(fd, so, nam, td);
634 * solisten() transitions a socket from a non-listening state to a listening
635 * state, but can also be used to update the listen queue depth on an
636 * existing listen socket. The protocol will call back into the sockets
637 * layer using solisten_proto_check() and solisten_proto() to check and set
638 * socket-layer listen state. Call backs are used so that the protocol can
639 * acquire both protocol and socket layer locks in whatever order is required
642 * Protocol implementors are advised to hold the socket lock across the
643 * socket-layer test and set to avoid races at the socket layer.
646 solisten(struct socket *so, int backlog, struct thread *td)
650 CURVNET_SET(so->so_vnet);
651 error = (*so->so_proto->pr_usrreqs->pru_listen)(so, backlog, td);
657 solisten_proto_check(struct socket *so)
660 SOCK_LOCK_ASSERT(so);
662 if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
669 solisten_proto(struct socket *so, int backlog)
672 SOCK_LOCK_ASSERT(so);
674 if (backlog < 0 || backlog > somaxconn)
676 so->so_qlimit = backlog;
677 so->so_options |= SO_ACCEPTCONN;
681 * Evaluate the reference count and named references on a socket; if no
682 * references remain, free it. This should be called whenever a reference is
683 * released, such as in sorele(), but also when named reference flags are
684 * cleared in socket or protocol code.
686 * sofree() will free the socket if:
688 * - There are no outstanding file descriptor references or related consumers
691 * - The socket has been closed by user space, if ever open (SS_NOFDREF).
693 * - The protocol does not have an outstanding strong reference on the socket
696 * - The socket is not in a completed connection queue, so a process has been
697 * notified that it is present. If it is removed, the user process may
698 * block in accept() despite select() saying the socket was ready.
701 sofree(struct socket *so)
703 struct protosw *pr = so->so_proto;
706 ACCEPT_LOCK_ASSERT();
707 SOCK_LOCK_ASSERT(so);
709 if ((so->so_state & SS_NOFDREF) == 0 || so->so_count != 0 ||
710 (so->so_state & SS_PROTOREF) || (so->so_qstate & SQ_COMP)) {
718 KASSERT((so->so_qstate & SQ_COMP) != 0 ||
719 (so->so_qstate & SQ_INCOMP) != 0,
720 ("sofree: so_head != NULL, but neither SQ_COMP nor "
722 KASSERT((so->so_qstate & SQ_COMP) == 0 ||
723 (so->so_qstate & SQ_INCOMP) == 0,
724 ("sofree: so->so_qstate is SQ_COMP and also SQ_INCOMP"));
725 TAILQ_REMOVE(&head->so_incomp, so, so_list);
727 so->so_qstate &= ~SQ_INCOMP;
730 KASSERT((so->so_qstate & SQ_COMP) == 0 &&
731 (so->so_qstate & SQ_INCOMP) == 0,
732 ("sofree: so_head == NULL, but still SQ_COMP(%d) or SQ_INCOMP(%d)",
733 so->so_qstate & SQ_COMP, so->so_qstate & SQ_INCOMP));
734 if (so->so_options & SO_ACCEPTCONN) {
735 KASSERT((TAILQ_EMPTY(&so->so_comp)),
736 ("sofree: so_comp populated"));
737 KASSERT((TAILQ_EMPTY(&so->so_incomp)),
738 ("sofree: so_incomp populated"));
744 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL) {
745 if (pr->pr_domain->dom_family == AF_LOCAL)
748 (*pr->pr_domain->dom_dispose)(so->so_rcv.sb_mb);
750 if (pr->pr_usrreqs->pru_detach != NULL)
751 (*pr->pr_usrreqs->pru_detach)(so);
754 * From this point on, we assume that no other references to this
755 * socket exist anywhere else in the stack. Therefore, no locks need
756 * to be acquired or held.
758 * We used to do a lot of socket buffer and socket locking here, as
759 * well as invoke sorflush() and perform wakeups. The direct call to
760 * dom_dispose() and sbrelease_internal() are an inlining of what was
761 * necessary from sorflush().
763 * Notice that the socket buffer and kqueue state are torn down
764 * before calling pru_detach. This means that protocols shold not
765 * assume they can perform socket wakeups, etc, in their detach code.
767 sbdestroy(&so->so_snd, so);
768 sbdestroy(&so->so_rcv, so);
769 seldrain(&so->so_snd.sb_sel);
770 seldrain(&so->so_rcv.sb_sel);
771 knlist_destroy(&so->so_rcv.sb_sel.si_note);
772 knlist_destroy(&so->so_snd.sb_sel.si_note);
777 * Close a socket on last file table reference removal. Initiate disconnect
778 * if connected. Free socket when disconnect complete.
780 * This function will sorele() the socket. Note that soclose() may be called
781 * prior to the ref count reaching zero. The actual socket structure will
782 * not be freed until the ref count reaches zero.
785 soclose(struct socket *so)
789 KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter"));
791 CURVNET_SET(so->so_vnet);
792 funsetown(&so->so_sigio);
793 if (so->so_state & SS_ISCONNECTED) {
794 if ((so->so_state & SS_ISDISCONNECTING) == 0) {
795 error = sodisconnect(so);
797 if (error == ENOTCONN)
802 if (so->so_options & SO_LINGER) {
803 if ((so->so_state & SS_ISDISCONNECTING) &&
804 (so->so_state & SS_NBIO))
806 while (so->so_state & SS_ISCONNECTED) {
807 error = tsleep(&so->so_timeo,
808 PSOCK | PCATCH, "soclos",
817 if (so->so_proto->pr_usrreqs->pru_close != NULL)
818 (*so->so_proto->pr_usrreqs->pru_close)(so);
820 if (so->so_options & SO_ACCEPTCONN) {
823 * Prevent new additions to the accept queues due
824 * to ACCEPT_LOCK races while we are draining them.
826 so->so_options &= ~SO_ACCEPTCONN;
827 while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
828 TAILQ_REMOVE(&so->so_incomp, sp, so_list);
830 sp->so_qstate &= ~SQ_INCOMP;
836 while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
837 TAILQ_REMOVE(&so->so_comp, sp, so_list);
839 sp->so_qstate &= ~SQ_COMP;
845 KASSERT((TAILQ_EMPTY(&so->so_comp)),
846 ("%s: so_comp populated", __func__));
847 KASSERT((TAILQ_EMPTY(&so->so_incomp)),
848 ("%s: so_incomp populated", __func__));
851 KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF"));
852 so->so_state |= SS_NOFDREF;
853 sorele(so); /* NB: Returns with ACCEPT_UNLOCK(). */
859 * soabort() is used to abruptly tear down a connection, such as when a
860 * resource limit is reached (listen queue depth exceeded), or if a listen
861 * socket is closed while there are sockets waiting to be accepted.
863 * This interface is tricky, because it is called on an unreferenced socket,
864 * and must be called only by a thread that has actually removed the socket
865 * from the listen queue it was on, or races with other threads are risked.
867 * This interface will call into the protocol code, so must not be called
868 * with any socket locks held. Protocols do call it while holding their own
869 * recursible protocol mutexes, but this is something that should be subject
870 * to review in the future.
873 soabort(struct socket *so)
877 * In as much as is possible, assert that no references to this
878 * socket are held. This is not quite the same as asserting that the
879 * current thread is responsible for arranging for no references, but
880 * is as close as we can get for now.
882 KASSERT(so->so_count == 0, ("soabort: so_count"));
883 KASSERT((so->so_state & SS_PROTOREF) == 0, ("soabort: SS_PROTOREF"));
884 KASSERT(so->so_state & SS_NOFDREF, ("soabort: !SS_NOFDREF"));
885 KASSERT((so->so_state & SQ_COMP) == 0, ("soabort: SQ_COMP"));
886 KASSERT((so->so_state & SQ_INCOMP) == 0, ("soabort: SQ_INCOMP"));
889 if (so->so_proto->pr_usrreqs->pru_abort != NULL)
890 (*so->so_proto->pr_usrreqs->pru_abort)(so);
897 soaccept(struct socket *so, struct sockaddr **nam)
902 KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF"));
903 so->so_state &= ~SS_NOFDREF;
906 CURVNET_SET(so->so_vnet);
907 error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam);
913 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
916 return (soconnectat(AT_FDCWD, so, nam, td));
920 soconnectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
924 if (so->so_options & SO_ACCEPTCONN)
927 CURVNET_SET(so->so_vnet);
929 * If protocol is connection-based, can only connect once.
930 * Otherwise, if connected, try to disconnect first. This allows
931 * user to disconnect by connecting to, e.g., a null address.
933 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
934 ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
935 (error = sodisconnect(so)))) {
939 * Prevent accumulated error from previous connection from
943 if (fd == AT_FDCWD) {
944 error = (*so->so_proto->pr_usrreqs->pru_connect)(so,
947 error = (*so->so_proto->pr_usrreqs->pru_connectat)(fd,
957 soconnect2(struct socket *so1, struct socket *so2)
961 CURVNET_SET(so1->so_vnet);
962 error = (*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2);
968 sodisconnect(struct socket *so)
972 if ((so->so_state & SS_ISCONNECTED) == 0)
974 if (so->so_state & SS_ISDISCONNECTING)
977 error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so);
981 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? 0 : SBL_WAIT)
984 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio,
985 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
989 int clen = 0, error, dontroute;
991 KASSERT(so->so_type == SOCK_DGRAM, ("sosend_dgram: !SOCK_DGRAM"));
992 KASSERT(so->so_proto->pr_flags & PR_ATOMIC,
993 ("sosend_dgram: !PR_ATOMIC"));
996 resid = uio->uio_resid;
998 resid = top->m_pkthdr.len;
1000 * In theory resid should be unsigned. However, space must be
1001 * signed, as it might be less than 0 if we over-committed, and we
1002 * must use a signed comparison of space and resid. On the other
1003 * hand, a negative resid causes us to loop sending 0-length
1004 * segments to the protocol.
1012 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0;
1014 td->td_ru.ru_msgsnd++;
1015 if (control != NULL)
1016 clen = control->m_len;
1018 SOCKBUF_LOCK(&so->so_snd);
1019 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1020 SOCKBUF_UNLOCK(&so->so_snd);
1025 error = so->so_error;
1027 SOCKBUF_UNLOCK(&so->so_snd);
1030 if ((so->so_state & SS_ISCONNECTED) == 0) {
1032 * `sendto' and `sendmsg' is allowed on a connection-based
1033 * socket if it supports implied connect. Return ENOTCONN if
1034 * not connected and no address is supplied.
1036 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1037 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1038 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1039 !(resid == 0 && clen != 0)) {
1040 SOCKBUF_UNLOCK(&so->so_snd);
1044 } else if (addr == NULL) {
1045 if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1048 error = EDESTADDRREQ;
1049 SOCKBUF_UNLOCK(&so->so_snd);
1055 * Do we need MSG_OOB support in SOCK_DGRAM? Signs here may be a
1056 * problem and need fixing.
1058 space = sbspace(&so->so_snd);
1059 if (flags & MSG_OOB)
1062 SOCKBUF_UNLOCK(&so->so_snd);
1063 if (resid > space) {
1069 if (flags & MSG_EOR)
1070 top->m_flags |= M_EOR;
1073 * Copy the data from userland into a mbuf chain.
1074 * If no data is to be copied in, a single empty mbuf
1077 top = m_uiotombuf(uio, M_WAITOK, space, max_hdr,
1078 (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0)));
1080 error = EFAULT; /* only possible error */
1083 space -= resid - uio->uio_resid;
1084 resid = uio->uio_resid;
1086 KASSERT(resid == 0, ("sosend_dgram: resid != 0"));
1088 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock
1093 so->so_options |= SO_DONTROUTE;
1097 * XXX all the SBS_CANTSENDMORE checks previously done could be out
1098 * of date. We could have received a reset packet in an interrupt or
1099 * maybe we slept while doing page faults in uiomove() etc. We could
1100 * probably recheck again inside the locking protection here, but
1101 * there are probably other places that this also happens. We must
1105 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1106 (flags & MSG_OOB) ? PRUS_OOB :
1108 * If the user set MSG_EOF, the protocol understands this flag and
1109 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND.
1111 ((flags & MSG_EOF) &&
1112 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1115 /* If there is more to send set PRUS_MORETOCOME */
1116 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1117 top, addr, control, td);
1120 so->so_options &= ~SO_DONTROUTE;
1129 if (control != NULL)
1135 * Send on a socket. If send must go all at once and message is larger than
1136 * send buffering, then hard error. Lock against other senders. If must go
1137 * all at once and not enough room now, then inform user that this would
1138 * block and do nothing. Otherwise, if nonblocking, send as much as
1139 * possible. The data to be sent is described by "uio" if nonzero, otherwise
1140 * by the mbuf chain "top" (which must be null if uio is not). Data provided
1141 * in mbuf chain must be small enough to send all at once.
1143 * Returns nonzero on error, timeout or signal; callers must check for short
1144 * counts if EINTR/ERESTART are returned. Data and control buffers are freed
1148 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio,
1149 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1153 int clen = 0, error, dontroute;
1154 int atomic = sosendallatonce(so) || top;
1157 resid = uio->uio_resid;
1159 resid = top->m_pkthdr.len;
1161 * In theory resid should be unsigned. However, space must be
1162 * signed, as it might be less than 0 if we over-committed, and we
1163 * must use a signed comparison of space and resid. On the other
1164 * hand, a negative resid causes us to loop sending 0-length
1165 * segments to the protocol.
1167 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
1168 * type sockets since that's an error.
1170 if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
1176 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
1177 (so->so_proto->pr_flags & PR_ATOMIC);
1179 td->td_ru.ru_msgsnd++;
1180 if (control != NULL)
1181 clen = control->m_len;
1183 error = sblock(&so->so_snd, SBLOCKWAIT(flags));
1189 SOCKBUF_LOCK(&so->so_snd);
1190 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1191 SOCKBUF_UNLOCK(&so->so_snd);
1196 error = so->so_error;
1198 SOCKBUF_UNLOCK(&so->so_snd);
1201 if ((so->so_state & SS_ISCONNECTED) == 0) {
1203 * `sendto' and `sendmsg' is allowed on a connection-
1204 * based socket if it supports implied connect.
1205 * Return ENOTCONN if not connected and no address is
1208 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1209 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1210 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1211 !(resid == 0 && clen != 0)) {
1212 SOCKBUF_UNLOCK(&so->so_snd);
1216 } else if (addr == NULL) {
1217 SOCKBUF_UNLOCK(&so->so_snd);
1218 if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1221 error = EDESTADDRREQ;
1225 space = sbspace(&so->so_snd);
1226 if (flags & MSG_OOB)
1228 if ((atomic && resid > so->so_snd.sb_hiwat) ||
1229 clen > so->so_snd.sb_hiwat) {
1230 SOCKBUF_UNLOCK(&so->so_snd);
1234 if (space < resid + clen &&
1235 (atomic || space < so->so_snd.sb_lowat || space < clen)) {
1236 if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO)) {
1237 SOCKBUF_UNLOCK(&so->so_snd);
1238 error = EWOULDBLOCK;
1241 error = sbwait(&so->so_snd);
1242 SOCKBUF_UNLOCK(&so->so_snd);
1247 SOCKBUF_UNLOCK(&so->so_snd);
1252 if (flags & MSG_EOR)
1253 top->m_flags |= M_EOR;
1256 * Copy the data from userland into a mbuf
1257 * chain. If no data is to be copied in,
1258 * a single empty mbuf is returned.
1260 top = m_uiotombuf(uio, M_WAITOK, space,
1261 (atomic ? max_hdr : 0),
1262 (atomic ? M_PKTHDR : 0) |
1263 ((flags & MSG_EOR) ? M_EOR : 0));
1265 error = EFAULT; /* only possible error */
1268 space -= resid - uio->uio_resid;
1269 resid = uio->uio_resid;
1273 so->so_options |= SO_DONTROUTE;
1277 * XXX all the SBS_CANTSENDMORE checks previously
1278 * done could be out of date. We could have received
1279 * a reset packet in an interrupt or maybe we slept
1280 * while doing page faults in uiomove() etc. We
1281 * could probably recheck again inside the locking
1282 * protection here, but there are probably other
1283 * places that this also happens. We must rethink
1287 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1288 (flags & MSG_OOB) ? PRUS_OOB :
1290 * If the user set MSG_EOF, the protocol understands
1291 * this flag and nothing left to send then use
1292 * PRU_SEND_EOF instead of PRU_SEND.
1294 ((flags & MSG_EOF) &&
1295 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1298 /* If there is more to send set PRUS_MORETOCOME. */
1299 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1300 top, addr, control, td);
1303 so->so_options &= ~SO_DONTROUTE;
1311 } while (resid && space > 0);
1315 sbunlock(&so->so_snd);
1319 if (control != NULL)
1325 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
1326 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1330 CURVNET_SET(so->so_vnet);
1331 error = so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio, top,
1332 control, flags, td);
1338 * The part of soreceive() that implements reading non-inline out-of-band
1339 * data from a socket. For more complete comments, see soreceive(), from
1340 * which this code originated.
1342 * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
1343 * unable to return an mbuf chain to the caller.
1346 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags)
1348 struct protosw *pr = so->so_proto;
1352 KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
1355 m = m_get(M_WAITOK, MT_DATA);
1356 error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK);
1360 error = uiomove(mtod(m, void *),
1361 (int) min(uio->uio_resid, m->m_len), uio);
1363 } while (uio->uio_resid && error == 0 && m);
1371 * Following replacement or removal of the first mbuf on the first mbuf chain
1372 * of a socket buffer, push necessary state changes back into the socket
1373 * buffer so that other consumers see the values consistently. 'nextrecord'
1374 * is the callers locally stored value of the original value of
1375 * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes.
1376 * NOTE: 'nextrecord' may be NULL.
1378 static __inline void
1379 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord)
1382 SOCKBUF_LOCK_ASSERT(sb);
1384 * First, update for the new value of nextrecord. If necessary, make
1385 * it the first record.
1387 if (sb->sb_mb != NULL)
1388 sb->sb_mb->m_nextpkt = nextrecord;
1390 sb->sb_mb = nextrecord;
1393 * Now update any dependent socket buffer fields to reflect the new
1394 * state. This is an expanded inline of SB_EMPTY_FIXUP(), with the
1395 * addition of a second clause that takes care of the case where
1396 * sb_mb has been updated, but remains the last record.
1398 if (sb->sb_mb == NULL) {
1399 sb->sb_mbtail = NULL;
1400 sb->sb_lastrecord = NULL;
1401 } else if (sb->sb_mb->m_nextpkt == NULL)
1402 sb->sb_lastrecord = sb->sb_mb;
1406 * Implement receive operations on a socket. We depend on the way that
1407 * records are added to the sockbuf by sbappend. In particular, each record
1408 * (mbufs linked through m_next) must begin with an address if the protocol
1409 * so specifies, followed by an optional mbuf or mbufs containing ancillary
1410 * data, and then zero or more mbufs of data. In order to allow parallelism
1411 * between network receive and copying to user space, as well as avoid
1412 * sleeping with a mutex held, we release the socket buffer mutex during the
1413 * user space copy. Although the sockbuf is locked, new data may still be
1414 * appended, and thus we must maintain consistency of the sockbuf during that
1417 * The caller may receive the data as a single mbuf chain by supplying an
1418 * mbuf **mp0 for use in returning the chain. The uio is then used only for
1419 * the count in uio_resid.
1422 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio,
1423 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1425 struct mbuf *m, **mp;
1426 int flags, error, offset;
1428 struct protosw *pr = so->so_proto;
1429 struct mbuf *nextrecord;
1431 ssize_t orig_resid = uio->uio_resid;
1436 if (controlp != NULL)
1439 flags = *flagsp &~ MSG_EOR;
1442 if (flags & MSG_OOB)
1443 return (soreceive_rcvoob(so, uio, flags));
1446 if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING)
1447 && uio->uio_resid) {
1449 (*pr->pr_usrreqs->pru_rcvd)(so, 0);
1452 error = sblock(&so->so_rcv, SBLOCKWAIT(flags));
1457 SOCKBUF_LOCK(&so->so_rcv);
1458 m = so->so_rcv.sb_mb;
1460 * If we have less data than requested, block awaiting more (subject
1461 * to any timeout) if:
1462 * 1. the current count is less than the low water mark, or
1463 * 2. MSG_DONTWAIT is not set
1465 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1466 so->so_rcv.sb_cc < uio->uio_resid) &&
1467 so->so_rcv.sb_cc < so->so_rcv.sb_lowat &&
1468 m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
1469 KASSERT(m != NULL || !so->so_rcv.sb_cc,
1470 ("receive: m == %p so->so_rcv.sb_cc == %u",
1471 m, so->so_rcv.sb_cc));
1475 error = so->so_error;
1476 if ((flags & MSG_PEEK) == 0)
1478 SOCKBUF_UNLOCK(&so->so_rcv);
1481 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1482 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1484 SOCKBUF_UNLOCK(&so->so_rcv);
1489 for (; m != NULL; m = m->m_next)
1490 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) {
1491 m = so->so_rcv.sb_mb;
1494 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1495 (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1496 SOCKBUF_UNLOCK(&so->so_rcv);
1500 if (uio->uio_resid == 0) {
1501 SOCKBUF_UNLOCK(&so->so_rcv);
1504 if ((so->so_state & SS_NBIO) ||
1505 (flags & (MSG_DONTWAIT|MSG_NBIO))) {
1506 SOCKBUF_UNLOCK(&so->so_rcv);
1507 error = EWOULDBLOCK;
1510 SBLASTRECORDCHK(&so->so_rcv);
1511 SBLASTMBUFCHK(&so->so_rcv);
1512 error = sbwait(&so->so_rcv);
1513 SOCKBUF_UNLOCK(&so->so_rcv);
1520 * From this point onward, we maintain 'nextrecord' as a cache of the
1521 * pointer to the next record in the socket buffer. We must keep the
1522 * various socket buffer pointers and local stack versions of the
1523 * pointers in sync, pushing out modifications before dropping the
1524 * socket buffer mutex, and re-reading them when picking it up.
1526 * Otherwise, we will race with the network stack appending new data
1527 * or records onto the socket buffer by using inconsistent/stale
1528 * versions of the field, possibly resulting in socket buffer
1531 * By holding the high-level sblock(), we prevent simultaneous
1532 * readers from pulling off the front of the socket buffer.
1534 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1536 uio->uio_td->td_ru.ru_msgrcv++;
1537 KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb"));
1538 SBLASTRECORDCHK(&so->so_rcv);
1539 SBLASTMBUFCHK(&so->so_rcv);
1540 nextrecord = m->m_nextpkt;
1541 if (pr->pr_flags & PR_ADDR) {
1542 KASSERT(m->m_type == MT_SONAME,
1543 ("m->m_type == %d", m->m_type));
1546 *psa = sodupsockaddr(mtod(m, struct sockaddr *),
1548 if (flags & MSG_PEEK) {
1551 sbfree(&so->so_rcv, m);
1552 so->so_rcv.sb_mb = m_free(m);
1553 m = so->so_rcv.sb_mb;
1554 sockbuf_pushsync(&so->so_rcv, nextrecord);
1559 * Process one or more MT_CONTROL mbufs present before any data mbufs
1560 * in the first mbuf chain on the socket buffer. If MSG_PEEK, we
1561 * just copy the data; if !MSG_PEEK, we call into the protocol to
1562 * perform externalization (or freeing if controlp == NULL).
1564 if (m != NULL && m->m_type == MT_CONTROL) {
1565 struct mbuf *cm = NULL, *cmn;
1566 struct mbuf **cme = &cm;
1569 if (flags & MSG_PEEK) {
1570 if (controlp != NULL) {
1571 *controlp = m_copy(m, 0, m->m_len);
1572 controlp = &(*controlp)->m_next;
1576 sbfree(&so->so_rcv, m);
1577 so->so_rcv.sb_mb = m->m_next;
1580 cme = &(*cme)->m_next;
1581 m = so->so_rcv.sb_mb;
1583 } while (m != NULL && m->m_type == MT_CONTROL);
1584 if ((flags & MSG_PEEK) == 0)
1585 sockbuf_pushsync(&so->so_rcv, nextrecord);
1586 while (cm != NULL) {
1589 if (pr->pr_domain->dom_externalize != NULL) {
1590 SOCKBUF_UNLOCK(&so->so_rcv);
1592 error = (*pr->pr_domain->dom_externalize)
1593 (cm, controlp, flags);
1594 SOCKBUF_LOCK(&so->so_rcv);
1595 } else if (controlp != NULL)
1599 if (controlp != NULL) {
1601 while (*controlp != NULL)
1602 controlp = &(*controlp)->m_next;
1607 nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1609 nextrecord = so->so_rcv.sb_mb;
1613 if ((flags & MSG_PEEK) == 0) {
1614 KASSERT(m->m_nextpkt == nextrecord,
1615 ("soreceive: post-control, nextrecord !sync"));
1616 if (nextrecord == NULL) {
1617 KASSERT(so->so_rcv.sb_mb == m,
1618 ("soreceive: post-control, sb_mb!=m"));
1619 KASSERT(so->so_rcv.sb_lastrecord == m,
1620 ("soreceive: post-control, lastrecord!=m"));
1624 if (type == MT_OOBDATA)
1627 if ((flags & MSG_PEEK) == 0) {
1628 KASSERT(so->so_rcv.sb_mb == nextrecord,
1629 ("soreceive: sb_mb != nextrecord"));
1630 if (so->so_rcv.sb_mb == NULL) {
1631 KASSERT(so->so_rcv.sb_lastrecord == NULL,
1632 ("soreceive: sb_lastercord != NULL"));
1636 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1637 SBLASTRECORDCHK(&so->so_rcv);
1638 SBLASTMBUFCHK(&so->so_rcv);
1641 * Now continue to read any data mbufs off of the head of the socket
1642 * buffer until the read request is satisfied. Note that 'type' is
1643 * used to store the type of any mbuf reads that have happened so far
1644 * such that soreceive() can stop reading if the type changes, which
1645 * causes soreceive() to return only one of regular data and inline
1646 * out-of-band data in a single socket receive operation.
1650 while (m != NULL && uio->uio_resid > 0 && error == 0) {
1652 * If the type of mbuf has changed since the last mbuf
1653 * examined ('type'), end the receive operation.
1655 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1656 if (m->m_type == MT_OOBDATA || m->m_type == MT_CONTROL) {
1657 if (type != m->m_type)
1659 } else if (type == MT_OOBDATA)
1662 KASSERT(m->m_type == MT_DATA,
1663 ("m->m_type == %d", m->m_type));
1664 so->so_rcv.sb_state &= ~SBS_RCVATMARK;
1665 len = uio->uio_resid;
1666 if (so->so_oobmark && len > so->so_oobmark - offset)
1667 len = so->so_oobmark - offset;
1668 if (len > m->m_len - moff)
1669 len = m->m_len - moff;
1671 * If mp is set, just pass back the mbufs. Otherwise copy
1672 * them out via the uio, then free. Sockbuf must be
1673 * consistent here (points to current mbuf, it points to next
1674 * record) when we drop priority; we must note any additions
1675 * to the sockbuf when we block interrupts again.
1678 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1679 SBLASTRECORDCHK(&so->so_rcv);
1680 SBLASTMBUFCHK(&so->so_rcv);
1681 SOCKBUF_UNLOCK(&so->so_rcv);
1682 error = uiomove(mtod(m, char *) + moff, (int)len, uio);
1683 SOCKBUF_LOCK(&so->so_rcv);
1686 * The MT_SONAME mbuf has already been removed
1687 * from the record, so it is necessary to
1688 * remove the data mbufs, if any, to preserve
1689 * the invariant in the case of PR_ADDR that
1690 * requires MT_SONAME mbufs at the head of
1693 if (m && pr->pr_flags & PR_ATOMIC &&
1694 ((flags & MSG_PEEK) == 0))
1695 (void)sbdroprecord_locked(&so->so_rcv);
1696 SOCKBUF_UNLOCK(&so->so_rcv);
1700 uio->uio_resid -= len;
1701 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1702 if (len == m->m_len - moff) {
1703 if (m->m_flags & M_EOR)
1705 if (flags & MSG_PEEK) {
1709 nextrecord = m->m_nextpkt;
1710 sbfree(&so->so_rcv, m);
1712 m->m_nextpkt = NULL;
1715 so->so_rcv.sb_mb = m = m->m_next;
1718 so->so_rcv.sb_mb = m_free(m);
1719 m = so->so_rcv.sb_mb;
1721 sockbuf_pushsync(&so->so_rcv, nextrecord);
1722 SBLASTRECORDCHK(&so->so_rcv);
1723 SBLASTMBUFCHK(&so->so_rcv);
1726 if (flags & MSG_PEEK)
1732 if (flags & MSG_DONTWAIT)
1733 copy_flag = M_NOWAIT;
1736 if (copy_flag == M_WAITOK)
1737 SOCKBUF_UNLOCK(&so->so_rcv);
1738 *mp = m_copym(m, 0, len, copy_flag);
1739 if (copy_flag == M_WAITOK)
1740 SOCKBUF_LOCK(&so->so_rcv);
1743 * m_copym() couldn't
1744 * allocate an mbuf. Adjust
1745 * uio_resid back (it was
1746 * adjusted down by len
1747 * bytes, which we didn't end
1748 * up "copying" over).
1750 uio->uio_resid += len;
1756 so->so_rcv.sb_cc -= len;
1759 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1760 if (so->so_oobmark) {
1761 if ((flags & MSG_PEEK) == 0) {
1762 so->so_oobmark -= len;
1763 if (so->so_oobmark == 0) {
1764 so->so_rcv.sb_state |= SBS_RCVATMARK;
1769 if (offset == so->so_oobmark)
1773 if (flags & MSG_EOR)
1776 * If the MSG_WAITALL flag is set (for non-atomic socket), we
1777 * must not quit until "uio->uio_resid == 0" or an error
1778 * termination. If a signal/timeout occurs, return with a
1779 * short count but without error. Keep sockbuf locked
1780 * against other readers.
1782 while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1783 !sosendallatonce(so) && nextrecord == NULL) {
1784 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1786 so->so_rcv.sb_state & SBS_CANTRCVMORE)
1789 * Notify the protocol that some data has been
1790 * drained before blocking.
1792 if (pr->pr_flags & PR_WANTRCVD) {
1793 SOCKBUF_UNLOCK(&so->so_rcv);
1795 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
1796 SOCKBUF_LOCK(&so->so_rcv);
1798 SBLASTRECORDCHK(&so->so_rcv);
1799 SBLASTMBUFCHK(&so->so_rcv);
1801 * We could receive some data while was notifying
1802 * the protocol. Skip blocking in this case.
1804 if (so->so_rcv.sb_mb == NULL) {
1805 error = sbwait(&so->so_rcv);
1807 SOCKBUF_UNLOCK(&so->so_rcv);
1811 m = so->so_rcv.sb_mb;
1813 nextrecord = m->m_nextpkt;
1817 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1818 if (m != NULL && pr->pr_flags & PR_ATOMIC) {
1820 if ((flags & MSG_PEEK) == 0)
1821 (void) sbdroprecord_locked(&so->so_rcv);
1823 if ((flags & MSG_PEEK) == 0) {
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.
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;
1837 SBLASTRECORDCHK(&so->so_rcv);
1838 SBLASTMBUFCHK(&so->so_rcv);
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.
1844 if (!(flags & MSG_SOCALLBCK) &&
1845 (pr->pr_flags & PR_WANTRCVD)) {
1846 SOCKBUF_UNLOCK(&so->so_rcv);
1848 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
1849 SOCKBUF_LOCK(&so->so_rcv);
1852 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1853 if (orig_resid == uio->uio_resid && orig_resid &&
1854 (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
1855 SOCKBUF_UNLOCK(&so->so_rcv);
1858 SOCKBUF_UNLOCK(&so->so_rcv);
1863 sbunlock(&so->so_rcv);
1868 * Optimized version of soreceive() for stream (TCP) sockets.
1869 * XXXAO: (MSG_WAITALL | MSG_PEEK) isn't properly handled.
1872 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio,
1873 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1875 int len = 0, error = 0, flags, oresid;
1877 struct mbuf *m, *n = NULL;
1879 /* We only do stream sockets. */
1880 if (so->so_type != SOCK_STREAM)
1884 if (controlp != NULL)
1887 flags = *flagsp &~ MSG_EOR;
1890 if (flags & MSG_OOB)
1891 return (soreceive_rcvoob(so, uio, flags));
1897 /* Prevent other readers from entering the socket. */
1898 error = sblock(sb, SBLOCKWAIT(flags));
1903 /* Easy one, no space to copyout anything. */
1904 if (uio->uio_resid == 0) {
1908 oresid = uio->uio_resid;
1910 /* We will never ever get anything unless we are or were connected. */
1911 if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) {
1917 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1919 /* Abort if socket has reported problems. */
1923 if (oresid > uio->uio_resid)
1925 error = so->so_error;
1926 if (!(flags & MSG_PEEK))
1931 /* Door is closed. Deliver what is left, if any. */
1932 if (sb->sb_state & SBS_CANTRCVMORE) {
1939 /* Socket buffer is empty and we shall not block. */
1940 if (sb->sb_cc == 0 &&
1941 ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) {
1946 /* Socket buffer got some data that we shall deliver now. */
1947 if (sb->sb_cc > 0 && !(flags & MSG_WAITALL) &&
1948 ((so->so_state & SS_NBIO) ||
1949 (flags & (MSG_DONTWAIT|MSG_NBIO)) ||
1950 sb->sb_cc >= sb->sb_lowat ||
1951 sb->sb_cc >= uio->uio_resid ||
1952 sb->sb_cc >= sb->sb_hiwat) ) {
1956 /* On MSG_WAITALL we must wait until all data or error arrives. */
1957 if ((flags & MSG_WAITALL) &&
1958 (sb->sb_cc >= uio->uio_resid || sb->sb_cc >= sb->sb_hiwat))
1962 * Wait and block until (more) data comes in.
1963 * NB: Drops the sockbuf lock during wait.
1971 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1972 KASSERT(sb->sb_cc > 0, ("%s: sockbuf empty", __func__));
1973 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__));
1977 uio->uio_td->td_ru.ru_msgrcv++;
1979 /* Fill uio until full or current end of socket buffer is reached. */
1980 len = min(uio->uio_resid, sb->sb_cc);
1982 /* Dequeue as many mbufs as possible. */
1983 if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) {
1987 m_cat(*mp0, sb->sb_mb);
1989 m != NULL && m->m_len <= len;
1992 uio->uio_resid -= m->m_len;
1998 sb->sb_lastrecord = sb->sb_mb;
1999 if (sb->sb_mb == NULL)
2002 /* Copy the remainder. */
2004 KASSERT(sb->sb_mb != NULL,
2005 ("%s: len > 0 && sb->sb_mb empty", __func__));
2007 m = m_copym(sb->sb_mb, 0, len, M_NOWAIT);
2009 len = 0; /* Don't flush data from sockbuf. */
2011 uio->uio_resid -= len;
2022 /* NB: Must unlock socket buffer as uiomove may sleep. */
2024 error = m_mbuftouio(uio, sb->sb_mb, len);
2029 SBLASTRECORDCHK(sb);
2033 * Remove the delivered data from the socket buffer unless we
2034 * were only peeking.
2036 if (!(flags & MSG_PEEK)) {
2038 sbdrop_locked(sb, len);
2040 /* Notify protocol that we drained some data. */
2041 if ((so->so_proto->pr_flags & PR_WANTRCVD) &&
2042 (((flags & MSG_WAITALL) && uio->uio_resid > 0) ||
2043 !(flags & MSG_SOCALLBCK))) {
2046 (*so->so_proto->pr_usrreqs->pru_rcvd)(so, flags);
2052 * For MSG_WAITALL we may have to loop again and wait for
2053 * more data to come in.
2055 if ((flags & MSG_WAITALL) && uio->uio_resid > 0)
2058 SOCKBUF_LOCK_ASSERT(sb);
2059 SBLASTRECORDCHK(sb);
2067 * Optimized version of soreceive() for simple datagram cases from userspace.
2068 * Unlike in the stream case, we're able to drop a datagram if copyout()
2069 * fails, and because we handle datagrams atomically, we don't need to use a
2070 * sleep lock to prevent I/O interlacing.
2073 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio,
2074 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2076 struct mbuf *m, *m2;
2079 struct protosw *pr = so->so_proto;
2080 struct mbuf *nextrecord;
2084 if (controlp != NULL)
2087 flags = *flagsp &~ MSG_EOR;
2092 * For any complicated cases, fall back to the full
2093 * soreceive_generic().
2095 if (mp0 != NULL || (flags & MSG_PEEK) || (flags & MSG_OOB))
2096 return (soreceive_generic(so, psa, uio, mp0, controlp,
2100 * Enforce restrictions on use.
2102 KASSERT((pr->pr_flags & PR_WANTRCVD) == 0,
2103 ("soreceive_dgram: wantrcvd"));
2104 KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic"));
2105 KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0,
2106 ("soreceive_dgram: SBS_RCVATMARK"));
2107 KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0,
2108 ("soreceive_dgram: P_CONNREQUIRED"));
2111 * Loop blocking while waiting for a datagram.
2113 SOCKBUF_LOCK(&so->so_rcv);
2114 while ((m = so->so_rcv.sb_mb) == NULL) {
2115 KASSERT(so->so_rcv.sb_cc == 0,
2116 ("soreceive_dgram: sb_mb NULL but sb_cc %u",
2119 error = so->so_error;
2121 SOCKBUF_UNLOCK(&so->so_rcv);
2124 if (so->so_rcv.sb_state & SBS_CANTRCVMORE ||
2125 uio->uio_resid == 0) {
2126 SOCKBUF_UNLOCK(&so->so_rcv);
2129 if ((so->so_state & SS_NBIO) ||
2130 (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2131 SOCKBUF_UNLOCK(&so->so_rcv);
2132 return (EWOULDBLOCK);
2134 SBLASTRECORDCHK(&so->so_rcv);
2135 SBLASTMBUFCHK(&so->so_rcv);
2136 error = sbwait(&so->so_rcv);
2138 SOCKBUF_UNLOCK(&so->so_rcv);
2142 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2145 uio->uio_td->td_ru.ru_msgrcv++;
2146 SBLASTRECORDCHK(&so->so_rcv);
2147 SBLASTMBUFCHK(&so->so_rcv);
2148 nextrecord = m->m_nextpkt;
2149 if (nextrecord == NULL) {
2150 KASSERT(so->so_rcv.sb_lastrecord == m,
2151 ("soreceive_dgram: lastrecord != m"));
2154 KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord,
2155 ("soreceive_dgram: m_nextpkt != nextrecord"));
2158 * Pull 'm' and its chain off the front of the packet queue.
2160 so->so_rcv.sb_mb = NULL;
2161 sockbuf_pushsync(&so->so_rcv, nextrecord);
2164 * Walk 'm's chain and free that many bytes from the socket buffer.
2166 for (m2 = m; m2 != NULL; m2 = m2->m_next)
2167 sbfree(&so->so_rcv, m2);
2170 * Do a few last checks before we let go of the lock.
2172 SBLASTRECORDCHK(&so->so_rcv);
2173 SBLASTMBUFCHK(&so->so_rcv);
2174 SOCKBUF_UNLOCK(&so->so_rcv);
2176 if (pr->pr_flags & PR_ADDR) {
2177 KASSERT(m->m_type == MT_SONAME,
2178 ("m->m_type == %d", m->m_type));
2180 *psa = sodupsockaddr(mtod(m, struct sockaddr *),
2185 /* XXXRW: Can this happen? */
2190 * Packet to copyout() is now in 'm' and it is disconnected from the
2193 * Process one or more MT_CONTROL mbufs present before any data mbufs
2194 * in the first mbuf chain on the socket buffer. We call into the
2195 * protocol to perform externalization (or freeing if controlp ==
2196 * NULL). In some cases there can be only MT_CONTROL mbufs without
2199 if (m->m_type == MT_CONTROL) {
2200 struct mbuf *cm = NULL, *cmn;
2201 struct mbuf **cme = &cm;
2207 cme = &(*cme)->m_next;
2209 } while (m != NULL && m->m_type == MT_CONTROL);
2210 while (cm != NULL) {
2213 if (pr->pr_domain->dom_externalize != NULL) {
2214 error = (*pr->pr_domain->dom_externalize)
2215 (cm, controlp, flags);
2216 } else if (controlp != NULL)
2220 if (controlp != NULL) {
2221 while (*controlp != NULL)
2222 controlp = &(*controlp)->m_next;
2227 KASSERT(m == NULL || m->m_type == MT_DATA,
2228 ("soreceive_dgram: !data"));
2229 while (m != NULL && uio->uio_resid > 0) {
2230 len = uio->uio_resid;
2233 error = uiomove(mtod(m, char *), (int)len, uio);
2238 if (len == m->m_len)
2255 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
2256 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2260 CURVNET_SET(so->so_vnet);
2261 error = (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio, mp0,
2268 soshutdown(struct socket *so, int how)
2270 struct protosw *pr = so->so_proto;
2273 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
2276 CURVNET_SET(so->so_vnet);
2277 if (pr->pr_usrreqs->pru_flush != NULL)
2278 (*pr->pr_usrreqs->pru_flush)(so, how);
2281 if (how != SHUT_RD) {
2282 error = (*pr->pr_usrreqs->pru_shutdown)(so);
2283 wakeup(&so->so_timeo);
2287 wakeup(&so->so_timeo);
2293 sorflush(struct socket *so)
2295 struct sockbuf *sb = &so->so_rcv;
2296 struct protosw *pr = so->so_proto;
2302 * In order to avoid calling dom_dispose with the socket buffer mutex
2303 * held, and in order to generally avoid holding the lock for a long
2304 * time, we make a copy of the socket buffer and clear the original
2305 * (except locks, state). The new socket buffer copy won't have
2306 * initialized locks so we can only call routines that won't use or
2307 * assert those locks.
2309 * Dislodge threads currently blocked in receive and wait to acquire
2310 * a lock against other simultaneous readers before clearing the
2311 * socket buffer. Don't let our acquire be interrupted by a signal
2312 * despite any existing socket disposition on interruptable waiting.
2315 (void) sblock(sb, SBL_WAIT | SBL_NOINTR);
2318 * Invalidate/clear most of the sockbuf structure, but leave selinfo
2319 * and mutex data unchanged.
2322 bzero(&aso, sizeof(aso));
2323 aso.so_pcb = so->so_pcb;
2324 bcopy(&sb->sb_startzero, &aso.so_rcv.sb_startzero,
2325 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2326 bzero(&sb->sb_startzero,
2327 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2332 * Dispose of special rights and flush the copied socket. Don't call
2333 * any unsafe routines (that rely on locks being initialized) on aso.
2335 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL) {
2336 if (pr->pr_domain->dom_family == AF_LOCAL)
2337 unp_dispose_so(&aso);
2339 (*pr->pr_domain->dom_dispose)(aso.so_rcv.sb_mb);
2341 sbrelease_internal(&aso.so_rcv, so);
2345 * Perhaps this routine, and sooptcopyout(), below, ought to come in an
2346 * additional variant to handle the case where the option value needs to be
2347 * some kind of integer, but not a specific size. In addition to their use
2348 * here, these functions are also called by the protocol-level pr_ctloutput()
2352 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2357 * If the user gives us more than we wanted, we ignore it, but if we
2358 * don't get the minimum length the caller wants, we return EINVAL.
2359 * On success, sopt->sopt_valsize is set to however much we actually
2362 if ((valsize = sopt->sopt_valsize) < minlen)
2365 sopt->sopt_valsize = valsize = len;
2367 if (sopt->sopt_td != NULL)
2368 return (copyin(sopt->sopt_val, buf, valsize));
2370 bcopy(sopt->sopt_val, buf, valsize);
2375 * Kernel version of setsockopt(2).
2377 * XXX: optlen is size_t, not socklen_t
2380 so_setsockopt(struct socket *so, int level, int optname, void *optval,
2383 struct sockopt sopt;
2385 sopt.sopt_level = level;
2386 sopt.sopt_name = optname;
2387 sopt.sopt_dir = SOPT_SET;
2388 sopt.sopt_val = optval;
2389 sopt.sopt_valsize = optlen;
2390 sopt.sopt_td = NULL;
2391 return (sosetopt(so, &sopt));
2395 sosetopt(struct socket *so, struct sockopt *sopt)
2406 CURVNET_SET(so->so_vnet);
2408 if (sopt->sopt_level != SOL_SOCKET) {
2409 if (so->so_proto->pr_ctloutput != NULL) {
2410 error = (*so->so_proto->pr_ctloutput)(so, sopt);
2414 error = ENOPROTOOPT;
2416 switch (sopt->sopt_name) {
2418 case SO_ACCEPTFILTER:
2419 error = do_setopt_accept_filter(so, sopt);
2425 error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
2430 so->so_linger = l.l_linger;
2432 so->so_options |= SO_LINGER;
2434 so->so_options &= ~SO_LINGER;
2441 case SO_USELOOPBACK:
2451 error = sooptcopyin(sopt, &optval, sizeof optval,
2457 so->so_options |= sopt->sopt_name;
2459 so->so_options &= ~sopt->sopt_name;
2464 error = sooptcopyin(sopt, &optval, sizeof optval,
2469 if (optval < 0 || optval >= rt_numfibs) {
2473 if (((so->so_proto->pr_domain->dom_family == PF_INET) ||
2474 (so->so_proto->pr_domain->dom_family == PF_INET6) ||
2475 (so->so_proto->pr_domain->dom_family == PF_ROUTE)))
2476 so->so_fibnum = optval;
2481 case SO_USER_COOKIE:
2482 error = sooptcopyin(sopt, &val32, sizeof val32,
2486 so->so_user_cookie = val32;
2493 error = sooptcopyin(sopt, &optval, sizeof optval,
2499 * Values < 1 make no sense for any of these options,
2507 switch (sopt->sopt_name) {
2510 if (sbreserve(sopt->sopt_name == SO_SNDBUF ?
2511 &so->so_snd : &so->so_rcv, (u_long)optval,
2512 so, curthread) == 0) {
2516 (sopt->sopt_name == SO_SNDBUF ? &so->so_snd :
2517 &so->so_rcv)->sb_flags &= ~SB_AUTOSIZE;
2521 * Make sure the low-water is never greater than the
2525 SOCKBUF_LOCK(&so->so_snd);
2526 so->so_snd.sb_lowat =
2527 (optval > so->so_snd.sb_hiwat) ?
2528 so->so_snd.sb_hiwat : optval;
2529 SOCKBUF_UNLOCK(&so->so_snd);
2532 SOCKBUF_LOCK(&so->so_rcv);
2533 so->so_rcv.sb_lowat =
2534 (optval > so->so_rcv.sb_hiwat) ?
2535 so->so_rcv.sb_hiwat : optval;
2536 SOCKBUF_UNLOCK(&so->so_rcv);
2543 #ifdef COMPAT_FREEBSD32
2544 if (SV_CURPROC_FLAG(SV_ILP32)) {
2545 struct timeval32 tv32;
2547 error = sooptcopyin(sopt, &tv32, sizeof tv32,
2549 CP(tv32, tv, tv_sec);
2550 CP(tv32, tv, tv_usec);
2553 error = sooptcopyin(sopt, &tv, sizeof tv,
2557 if (tv.tv_sec < 0 || tv.tv_usec < 0 ||
2558 tv.tv_usec >= 1000000) {
2562 if (tv.tv_sec > INT32_MAX)
2566 switch (sopt->sopt_name) {
2568 so->so_snd.sb_timeo = val;
2571 so->so_rcv.sb_timeo = val;
2578 error = sooptcopyin(sopt, &extmac, sizeof extmac,
2582 error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
2590 error = ENOPROTOOPT;
2593 if (error == 0 && so->so_proto->pr_ctloutput != NULL)
2594 (void)(*so->so_proto->pr_ctloutput)(so, sopt);
2602 * Helper routine for getsockopt.
2605 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
2613 * Documented get behavior is that we always return a value, possibly
2614 * truncated to fit in the user's buffer. Traditional behavior is
2615 * that we always tell the user precisely how much we copied, rather
2616 * than something useful like the total amount we had available for
2617 * her. Note that this interface is not idempotent; the entire
2618 * answer must generated ahead of time.
2620 valsize = min(len, sopt->sopt_valsize);
2621 sopt->sopt_valsize = valsize;
2622 if (sopt->sopt_val != NULL) {
2623 if (sopt->sopt_td != NULL)
2624 error = copyout(buf, sopt->sopt_val, valsize);
2626 bcopy(buf, sopt->sopt_val, valsize);
2632 sogetopt(struct socket *so, struct sockopt *sopt)
2641 CURVNET_SET(so->so_vnet);
2643 if (sopt->sopt_level != SOL_SOCKET) {
2644 if (so->so_proto->pr_ctloutput != NULL)
2645 error = (*so->so_proto->pr_ctloutput)(so, sopt);
2647 error = ENOPROTOOPT;
2651 switch (sopt->sopt_name) {
2653 case SO_ACCEPTFILTER:
2654 error = do_getopt_accept_filter(so, sopt);
2659 l.l_onoff = so->so_options & SO_LINGER;
2660 l.l_linger = so->so_linger;
2662 error = sooptcopyout(sopt, &l, sizeof l);
2665 case SO_USELOOPBACK:
2677 optval = so->so_options & sopt->sopt_name;
2679 error = sooptcopyout(sopt, &optval, sizeof optval);
2683 optval = so->so_type;
2687 optval = so->so_proto->pr_protocol;
2692 optval = so->so_error;
2698 optval = so->so_snd.sb_hiwat;
2702 optval = so->so_rcv.sb_hiwat;
2706 optval = so->so_snd.sb_lowat;
2710 optval = so->so_rcv.sb_lowat;
2715 tv = sbttotv(sopt->sopt_name == SO_SNDTIMEO ?
2716 so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
2717 #ifdef COMPAT_FREEBSD32
2718 if (SV_CURPROC_FLAG(SV_ILP32)) {
2719 struct timeval32 tv32;
2721 CP(tv, tv32, tv_sec);
2722 CP(tv, tv32, tv_usec);
2723 error = sooptcopyout(sopt, &tv32, sizeof tv32);
2726 error = sooptcopyout(sopt, &tv, sizeof tv);
2731 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
2735 error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
2739 error = sooptcopyout(sopt, &extmac, sizeof extmac);
2747 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
2751 error = mac_getsockopt_peerlabel(
2752 sopt->sopt_td->td_ucred, so, &extmac);
2755 error = sooptcopyout(sopt, &extmac, sizeof extmac);
2761 case SO_LISTENQLIMIT:
2762 optval = so->so_qlimit;
2766 optval = so->so_qlen;
2769 case SO_LISTENINCQLEN:
2770 optval = so->so_incqlen;
2774 error = ENOPROTOOPT;
2786 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
2788 struct mbuf *m, *m_prev;
2789 int sopt_size = sopt->sopt_valsize;
2791 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
2794 if (sopt_size > MLEN) {
2795 MCLGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT);
2796 if ((m->m_flags & M_EXT) == 0) {
2800 m->m_len = min(MCLBYTES, sopt_size);
2802 m->m_len = min(MLEN, sopt_size);
2804 sopt_size -= m->m_len;
2809 MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
2814 if (sopt_size > MLEN) {
2815 MCLGET(m, sopt->sopt_td != NULL ? M_WAITOK :
2817 if ((m->m_flags & M_EXT) == 0) {
2822 m->m_len = min(MCLBYTES, sopt_size);
2824 m->m_len = min(MLEN, sopt_size);
2826 sopt_size -= m->m_len;
2834 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
2836 struct mbuf *m0 = m;
2838 if (sopt->sopt_val == NULL)
2840 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
2841 if (sopt->sopt_td != NULL) {
2844 error = copyin(sopt->sopt_val, mtod(m, char *),
2851 bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
2852 sopt->sopt_valsize -= m->m_len;
2853 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
2856 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
2857 panic("ip6_sooptmcopyin");
2862 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
2864 struct mbuf *m0 = m;
2867 if (sopt->sopt_val == NULL)
2869 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
2870 if (sopt->sopt_td != NULL) {
2873 error = copyout(mtod(m, char *), sopt->sopt_val,
2880 bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
2881 sopt->sopt_valsize -= m->m_len;
2882 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
2883 valsize += m->m_len;
2887 /* enough soopt buffer should be given from user-land */
2891 sopt->sopt_valsize = valsize;
2896 * sohasoutofband(): protocol notifies socket layer of the arrival of new
2897 * out-of-band data, which will then notify socket consumers.
2900 sohasoutofband(struct socket *so)
2903 if (so->so_sigio != NULL)
2904 pgsigio(&so->so_sigio, SIGURG, 0);
2905 selwakeuppri(&so->so_rcv.sb_sel, PSOCK);
2909 sopoll(struct socket *so, int events, struct ucred *active_cred,
2914 * We do not need to set or assert curvnet as long as everyone uses
2917 return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred,
2922 sopoll_generic(struct socket *so, int events, struct ucred *active_cred,
2927 SOCKBUF_LOCK(&so->so_snd);
2928 SOCKBUF_LOCK(&so->so_rcv);
2929 if (events & (POLLIN | POLLRDNORM))
2930 if (soreadabledata(so))
2931 revents |= events & (POLLIN | POLLRDNORM);
2933 if (events & (POLLOUT | POLLWRNORM))
2934 if (sowriteable(so))
2935 revents |= events & (POLLOUT | POLLWRNORM);
2937 if (events & (POLLPRI | POLLRDBAND))
2938 if (so->so_oobmark || (so->so_rcv.sb_state & SBS_RCVATMARK))
2939 revents |= events & (POLLPRI | POLLRDBAND);
2941 if ((events & POLLINIGNEOF) == 0) {
2942 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2943 revents |= events & (POLLIN | POLLRDNORM);
2944 if (so->so_snd.sb_state & SBS_CANTSENDMORE)
2950 if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
2951 selrecord(td, &so->so_rcv.sb_sel);
2952 so->so_rcv.sb_flags |= SB_SEL;
2955 if (events & (POLLOUT | POLLWRNORM)) {
2956 selrecord(td, &so->so_snd.sb_sel);
2957 so->so_snd.sb_flags |= SB_SEL;
2961 SOCKBUF_UNLOCK(&so->so_rcv);
2962 SOCKBUF_UNLOCK(&so->so_snd);
2967 soo_kqfilter(struct file *fp, struct knote *kn)
2969 struct socket *so = kn->kn_fp->f_data;
2972 switch (kn->kn_filter) {
2974 if (so->so_options & SO_ACCEPTCONN)
2975 kn->kn_fop = &solisten_filtops;
2977 kn->kn_fop = &soread_filtops;
2981 kn->kn_fop = &sowrite_filtops;
2989 knlist_add(&sb->sb_sel.si_note, kn, 1);
2990 sb->sb_flags |= SB_KNOTE;
2996 * Some routines that return EOPNOTSUPP for entry points that are not
2997 * supported by a protocol. Fill in as needed.
3000 pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
3007 pru_attach_notsupp(struct socket *so, int proto, struct thread *td)
3014 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3021 pru_bindat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3029 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3036 pru_connectat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3044 pru_connect2_notsupp(struct socket *so1, struct socket *so2)
3051 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
3052 struct ifnet *ifp, struct thread *td)
3059 pru_disconnect_notsupp(struct socket *so)
3066 pru_listen_notsupp(struct socket *so, int backlog, struct thread *td)
3073 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam)
3080 pru_rcvd_notsupp(struct socket *so, int flags)
3087 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
3094 pru_send_notsupp(struct socket *so, int flags, struct mbuf *m,
3095 struct sockaddr *addr, struct mbuf *control, struct thread *td)
3102 * This isn't really a ``null'' operation, but it's the default one and
3103 * doesn't do anything destructive.
3106 pru_sense_null(struct socket *so, struct stat *sb)
3109 sb->st_blksize = so->so_snd.sb_hiwat;
3114 pru_shutdown_notsupp(struct socket *so)
3121 pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam)
3128 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio,
3129 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
3136 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr,
3137 struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
3144 pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred,
3152 filt_sordetach(struct knote *kn)
3154 struct socket *so = kn->kn_fp->f_data;
3156 SOCKBUF_LOCK(&so->so_rcv);
3157 knlist_remove(&so->so_rcv.sb_sel.si_note, kn, 1);
3158 if (knlist_empty(&so->so_rcv.sb_sel.si_note))
3159 so->so_rcv.sb_flags &= ~SB_KNOTE;
3160 SOCKBUF_UNLOCK(&so->so_rcv);
3165 filt_soread(struct knote *kn, long hint)
3169 so = kn->kn_fp->f_data;
3170 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
3172 kn->kn_data = so->so_rcv.sb_cc - so->so_rcv.sb_ctl;
3173 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3174 kn->kn_flags |= EV_EOF;
3175 kn->kn_fflags = so->so_error;
3177 } else if (so->so_error) /* temporary udp error */
3179 else if (kn->kn_sfflags & NOTE_LOWAT)
3180 return (kn->kn_data >= kn->kn_sdata);
3182 return (so->so_rcv.sb_cc >= so->so_rcv.sb_lowat);
3186 filt_sowdetach(struct knote *kn)
3188 struct socket *so = kn->kn_fp->f_data;
3190 SOCKBUF_LOCK(&so->so_snd);
3191 knlist_remove(&so->so_snd.sb_sel.si_note, kn, 1);
3192 if (knlist_empty(&so->so_snd.sb_sel.si_note))
3193 so->so_snd.sb_flags &= ~SB_KNOTE;
3194 SOCKBUF_UNLOCK(&so->so_snd);
3199 filt_sowrite(struct knote *kn, long hint)
3203 so = kn->kn_fp->f_data;
3204 SOCKBUF_LOCK_ASSERT(&so->so_snd);
3205 kn->kn_data = sbspace(&so->so_snd);
3206 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
3207 kn->kn_flags |= EV_EOF;
3208 kn->kn_fflags = so->so_error;
3210 } else if (so->so_error) /* temporary udp error */
3212 else if (((so->so_state & SS_ISCONNECTED) == 0) &&
3213 (so->so_proto->pr_flags & PR_CONNREQUIRED))
3215 else if (kn->kn_sfflags & NOTE_LOWAT)
3216 return (kn->kn_data >= kn->kn_sdata);
3218 return (kn->kn_data >= so->so_snd.sb_lowat);
3223 filt_solisten(struct knote *kn, long hint)
3225 struct socket *so = kn->kn_fp->f_data;
3227 kn->kn_data = so->so_qlen;
3228 return (!TAILQ_EMPTY(&so->so_comp));
3232 socheckuid(struct socket *so, uid_t uid)
3237 if (so->so_cred->cr_uid != uid)
3243 * These functions are used by protocols to notify the socket layer (and its
3244 * consumers) of state changes in the sockets driven by protocol-side events.
3248 * Procedures to manipulate state flags of socket and do appropriate wakeups.
3250 * Normal sequence from the active (originating) side is that
3251 * soisconnecting() is called during processing of connect() call, resulting
3252 * in an eventual call to soisconnected() if/when the connection is
3253 * established. When the connection is torn down soisdisconnecting() is
3254 * called during processing of disconnect() call, and soisdisconnected() is
3255 * called when the connection to the peer is totally severed. The semantics
3256 * of these routines are such that connectionless protocols can call
3257 * soisconnected() and soisdisconnected() only, bypassing the in-progress
3258 * calls when setting up a ``connection'' takes no time.
3260 * From the passive side, a socket is created with two queues of sockets:
3261 * so_incomp for connections in progress and so_comp for connections already
3262 * made and awaiting user acceptance. As a protocol is preparing incoming
3263 * connections, it creates a socket structure queued on so_incomp by calling
3264 * sonewconn(). When the connection is established, soisconnected() is
3265 * called, and transfers the socket structure to so_comp, making it available
3268 * If a socket is closed with sockets on either so_incomp or so_comp, these
3269 * sockets are dropped.
3271 * If higher-level protocols are implemented in the kernel, the wakeups done
3272 * here will sometimes cause software-interrupt process scheduling.
3275 soisconnecting(struct socket *so)
3279 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
3280 so->so_state |= SS_ISCONNECTING;
3285 soisconnected(struct socket *so)
3287 struct socket *head;
3293 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
3294 so->so_state |= SS_ISCONNECTED;
3296 if (head != NULL && (so->so_qstate & SQ_INCOMP)) {
3297 if ((so->so_options & SO_ACCEPTFILTER) == 0) {
3299 TAILQ_REMOVE(&head->so_incomp, so, so_list);
3301 so->so_qstate &= ~SQ_INCOMP;
3302 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
3304 so->so_qstate |= SQ_COMP;
3307 wakeup_one(&head->so_timeo);
3310 soupcall_set(so, SO_RCV,
3311 head->so_accf->so_accept_filter->accf_callback,
3312 head->so_accf->so_accept_filter_arg);
3313 so->so_options &= ~SO_ACCEPTFILTER;
3314 ret = head->so_accf->so_accept_filter->accf_callback(so,
3315 head->so_accf->so_accept_filter_arg, M_NOWAIT);
3316 if (ret == SU_ISCONNECTED)
3317 soupcall_clear(so, SO_RCV);
3319 if (ret == SU_ISCONNECTED)
3326 wakeup(&so->so_timeo);
3332 soisdisconnecting(struct socket *so)
3336 * Note: This code assumes that SOCK_LOCK(so) and
3337 * SOCKBUF_LOCK(&so->so_rcv) are the same.
3339 SOCKBUF_LOCK(&so->so_rcv);
3340 so->so_state &= ~SS_ISCONNECTING;
3341 so->so_state |= SS_ISDISCONNECTING;
3342 so->so_rcv.sb_state |= SBS_CANTRCVMORE;
3343 sorwakeup_locked(so);
3344 SOCKBUF_LOCK(&so->so_snd);
3345 so->so_snd.sb_state |= SBS_CANTSENDMORE;
3346 sowwakeup_locked(so);
3347 wakeup(&so->so_timeo);
3351 soisdisconnected(struct socket *so)
3355 * Note: This code assumes that SOCK_LOCK(so) and
3356 * SOCKBUF_LOCK(&so->so_rcv) are the same.
3358 SOCKBUF_LOCK(&so->so_rcv);
3359 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
3360 so->so_state |= SS_ISDISCONNECTED;
3361 so->so_rcv.sb_state |= SBS_CANTRCVMORE;
3362 sorwakeup_locked(so);
3363 SOCKBUF_LOCK(&so->so_snd);
3364 so->so_snd.sb_state |= SBS_CANTSENDMORE;
3365 sbdrop_locked(&so->so_snd, so->so_snd.sb_cc);
3366 sowwakeup_locked(so);
3367 wakeup(&so->so_timeo);
3371 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
3374 sodupsockaddr(const struct sockaddr *sa, int mflags)
3376 struct sockaddr *sa2;
3378 sa2 = malloc(sa->sa_len, M_SONAME, mflags);
3380 bcopy(sa, sa2, sa->sa_len);
3385 * Register per-socket buffer upcalls.
3388 soupcall_set(struct socket *so, int which,
3389 int (*func)(struct socket *, void *, int), void *arg)
3401 panic("soupcall_set: bad which");
3403 SOCKBUF_LOCK_ASSERT(sb);
3405 /* XXX: accf_http actually wants to do this on purpose. */
3406 KASSERT(sb->sb_upcall == NULL, ("soupcall_set: overwriting upcall"));
3408 sb->sb_upcall = func;
3409 sb->sb_upcallarg = arg;
3410 sb->sb_flags |= SB_UPCALL;
3414 soupcall_clear(struct socket *so, int which)
3426 panic("soupcall_clear: bad which");
3428 SOCKBUF_LOCK_ASSERT(sb);
3429 KASSERT(sb->sb_upcall != NULL, ("soupcall_clear: no upcall to clear"));
3430 sb->sb_upcall = NULL;
3431 sb->sb_upcallarg = NULL;
3432 sb->sb_flags &= ~SB_UPCALL;
3436 * Create an external-format (``xsocket'') structure using the information in
3437 * the kernel-format socket structure pointed to by so. This is done to
3438 * reduce the spew of irrelevant information over this interface, to isolate
3439 * user code from changes in the kernel structure, and potentially to provide
3440 * information-hiding if we decide that some of this information should be
3441 * hidden from users.
3444 sotoxsocket(struct socket *so, struct xsocket *xso)
3447 xso->xso_len = sizeof *xso;
3449 xso->so_type = so->so_type;
3450 xso->so_options = so->so_options;
3451 xso->so_linger = so->so_linger;
3452 xso->so_state = so->so_state;
3453 xso->so_pcb = so->so_pcb;
3454 xso->xso_protocol = so->so_proto->pr_protocol;
3455 xso->xso_family = so->so_proto->pr_domain->dom_family;
3456 xso->so_qlen = so->so_qlen;
3457 xso->so_incqlen = so->so_incqlen;
3458 xso->so_qlimit = so->so_qlimit;
3459 xso->so_timeo = so->so_timeo;
3460 xso->so_error = so->so_error;
3461 xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
3462 xso->so_oobmark = so->so_oobmark;
3463 sbtoxsockbuf(&so->so_snd, &xso->so_snd);
3464 sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
3465 xso->so_uid = so->so_cred->cr_uid;
3470 * Socket accessor functions to provide external consumers with
3471 * a safe interface to socket state
3476 so_listeners_apply_all(struct socket *so, void (*func)(struct socket *, void *),
3480 TAILQ_FOREACH(so, &so->so_comp, so_list)
3485 so_sockbuf_rcv(struct socket *so)
3488 return (&so->so_rcv);
3492 so_sockbuf_snd(struct socket *so)
3495 return (&so->so_snd);
3499 so_state_get(const struct socket *so)
3502 return (so->so_state);
3506 so_state_set(struct socket *so, int val)
3513 so_options_get(const struct socket *so)
3516 return (so->so_options);
3520 so_options_set(struct socket *so, int val)
3523 so->so_options = val;
3527 so_error_get(const struct socket *so)
3530 return (so->so_error);
3534 so_error_set(struct socket *so, int val)
3541 so_linger_get(const struct socket *so)
3544 return (so->so_linger);
3548 so_linger_set(struct socket *so, int val)
3551 so->so_linger = val;
3555 so_protosw_get(const struct socket *so)
3558 return (so->so_proto);
3562 so_protosw_set(struct socket *so, struct protosw *val)
3569 so_sorwakeup(struct socket *so)
3576 so_sowwakeup(struct socket *so)
3583 so_sorwakeup_locked(struct socket *so)
3586 sorwakeup_locked(so);
3590 so_sowwakeup_locked(struct socket *so)
3593 sowwakeup_locked(so);
3597 so_lock(struct socket *so)
3604 so_unlock(struct socket *so)