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.
97 #include <sys/cdefs.h>
98 __FBSDID("$FreeBSD$");
100 #include "opt_inet.h"
101 #include "opt_inet6.h"
102 #include "opt_zero.h"
103 #include "opt_compat.h"
105 #include <sys/param.h>
106 #include <sys/systm.h>
107 #include <sys/fcntl.h>
108 #include <sys/limits.h>
109 #include <sys/lock.h>
111 #include <sys/malloc.h>
112 #include <sys/mbuf.h>
113 #include <sys/mutex.h>
114 #include <sys/domain.h>
115 #include <sys/file.h> /* for struct knote */
116 #include <sys/kernel.h>
117 #include <sys/event.h>
118 #include <sys/eventhandler.h>
119 #include <sys/poll.h>
120 #include <sys/proc.h>
121 #include <sys/protosw.h>
122 #include <sys/socket.h>
123 #include <sys/socketvar.h>
124 #include <sys/resourcevar.h>
125 #include <net/route.h>
126 #include <sys/signalvar.h>
127 #include <sys/stat.h>
129 #include <sys/sysctl.h>
131 #include <sys/jail.h>
133 #include <net/vnet.h>
135 #include <security/mac/mac_framework.h>
140 #include <sys/mount.h>
141 #include <sys/sysent.h>
142 #include <compat/freebsd32/freebsd32.h>
145 static int soreceive_rcvoob(struct socket *so, struct uio *uio,
148 static void filt_sordetach(struct knote *kn);
149 static int filt_soread(struct knote *kn, long hint);
150 static void filt_sowdetach(struct knote *kn);
151 static int filt_sowrite(struct knote *kn, long hint);
152 static int filt_solisten(struct knote *kn, long hint);
154 static struct filterops solisten_filtops =
155 { 1, NULL, filt_sordetach, filt_solisten };
156 static struct filterops soread_filtops =
157 { 1, NULL, filt_sordetach, filt_soread };
158 static struct filterops sowrite_filtops =
159 { 1, NULL, filt_sowdetach, filt_sowrite };
161 uma_zone_t socket_zone;
162 so_gen_t so_gencnt; /* generation count for sockets */
166 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
167 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
169 static int somaxconn = SOMAXCONN;
170 static int sysctl_somaxconn(SYSCTL_HANDLER_ARGS);
171 /* XXX: we dont have SYSCTL_USHORT */
172 SYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn, CTLTYPE_UINT | CTLFLAG_RW,
173 0, sizeof(int), sysctl_somaxconn, "I", "Maximum pending socket connection "
175 static int numopensockets;
176 SYSCTL_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. */
180 int so_zero_copy_send = 1;
181 int so_zero_copy_receive = 1;
182 SYSCTL_NODE(_kern_ipc, OID_AUTO, zero_copy, CTLFLAG_RD, 0,
183 "Zero copy controls");
184 SYSCTL_INT(_kern_ipc_zero_copy, OID_AUTO, receive, CTLFLAG_RW,
185 &so_zero_copy_receive, 0, "Enable zero copy receive");
186 SYSCTL_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 */
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.
194 struct mtx accept_mtx;
195 MTX_SYSINIT(accept_mtx, &accept_mtx, "accept", MTX_DEF);
198 * so_global_mtx protects so_gencnt, numopensockets, and the per-socket
201 static struct mtx so_global_mtx;
202 MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF);
205 * General IPC sysctl name space, used by sockets and a variety of other IPC
208 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC");
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.
215 sysctl_maxsockets(SYSCTL_HANDLER_ARGS)
217 int error, newmaxsockets;
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;
228 EVENTHANDLER_INVOKE(maxsockets_change);
235 SYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets, CTLTYPE_INT|CTLFLAG_RW,
236 &maxsockets, 0, sysctl_maxsockets, "IU",
237 "Maximum number of sockets avaliable");
240 * Initialise maxsockets. This SYSINIT must be run after
244 init_maxsockets(void *ignored)
247 TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
248 maxsockets = imax(maxsockets, imax(maxfiles, nmbclusters));
250 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL);
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.
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
264 * soalloc() returns a socket with a ref count of 0.
266 static struct socket *
267 soalloc(struct vnet *vnet)
271 so = uma_zalloc(socket_zone, M_NOWAIT | M_ZERO);
275 if (mac_socket_init(so, M_NOWAIT) != 0) {
276 uma_zfree(socket_zone, so);
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;
289 vnet->vnet_sockcnt++;
292 mtx_unlock(&so_global_mtx);
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.
302 sodealloc(struct socket *so)
305 KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count));
306 KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL"));
308 mtx_lock(&so_global_mtx);
309 so->so_gencnt = ++so_gencnt;
310 --numopensockets; /* Could be below, but faster here. */
312 so->so_vnet->vnet_sockcnt--;
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);
322 /* remove acccept filter if one is present. */
323 if (so->so_accf != NULL)
324 do_setopt_accept_filter(so, NULL);
327 mac_socket_destroy(so);
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);
338 * socreate returns a socket with a ref count of 1. The socket should be
339 * closed with soclose().
342 socreate(int dom, struct socket **aso, int type, int proto,
343 struct ucred *cred, struct thread *td)
350 prp = pffindproto(dom, proto, type);
352 prp = pffindtype(dom, type);
354 if (prp == NULL || prp->pr_usrreqs->pru_attach == NULL ||
355 prp->pr_usrreqs->pru_attach == pru_attach_notsupp)
356 return (EPROTONOSUPPORT);
358 if (prison_check_af(cred, prp->pr_domain->dom_family) != 0)
359 return (EPROTONOSUPPORT);
361 if (prp->pr_type != type)
363 so = soalloc(CRED_TO_VNET(cred));
367 TAILQ_INIT(&so->so_incomp);
368 TAILQ_INIT(&so->so_comp);
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;
378 mac_socket_create(cred, so);
380 knlist_init_mtx(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv));
381 knlist_init_mtx(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd));
384 * Auto-sizing of socket buffers is managed by the protocols and
385 * the appropriate flags must be set in the pru_attach function.
387 CURVNET_SET(so->so_vnet);
388 error = (*prp->pr_usrreqs->pru_attach)(so, proto, td);
391 KASSERT(so->so_count == 1, ("socreate: so_count %d",
402 static int regression_sonewconn_earlytest = 1;
403 SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW,
404 ®ression_sonewconn_earlytest, 0, "Perform early sonewconn limit test");
408 * When an attempt at a new connection is noted on a socket which accepts
409 * connections, sonewconn is called. If the connection is possible (subject
410 * to space constraints, etc.) then we allocate a new structure, propoerly
411 * linked into the data structure of the original socket, and return this.
412 * Connstatus may be 0, or SO_ISCONFIRMING, or SO_ISCONNECTED.
414 * Note: the ref count on the socket is 0 on return.
417 sonewconn(struct socket *head, int connstatus)
423 over = (head->so_qlen > 3 * head->so_qlimit / 2);
426 if (regression_sonewconn_earlytest && over)
431 VNET_ASSERT(head->so_vnet);
432 so = soalloc(head->so_vnet);
435 if ((head->so_options & SO_ACCEPTFILTER) != 0)
438 so->so_type = head->so_type;
439 so->so_options = head->so_options &~ SO_ACCEPTCONN;
440 so->so_linger = head->so_linger;
441 so->so_state = head->so_state | SS_NOFDREF;
442 so->so_fibnum = head->so_fibnum;
443 so->so_proto = head->so_proto;
444 so->so_cred = crhold(head->so_cred);
446 mac_socket_newconn(head, so);
448 knlist_init_mtx(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv));
449 knlist_init_mtx(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd));
450 if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat) ||
451 (*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
455 so->so_rcv.sb_lowat = head->so_rcv.sb_lowat;
456 so->so_snd.sb_lowat = head->so_snd.sb_lowat;
457 so->so_rcv.sb_timeo = head->so_rcv.sb_timeo;
458 so->so_snd.sb_timeo = head->so_snd.sb_timeo;
459 so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE;
460 so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE;
461 so->so_state |= connstatus;
464 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
465 so->so_qstate |= SQ_COMP;
469 * Keep removing sockets from the head until there's room for
470 * us to insert on the tail. In pre-locking revisions, this
471 * was a simple if(), but as we could be racing with other
472 * threads and soabort() requires dropping locks, we must
473 * loop waiting for the condition to be true.
475 while (head->so_incqlen > head->so_qlimit) {
477 sp = TAILQ_FIRST(&head->so_incomp);
478 TAILQ_REMOVE(&head->so_incomp, sp, so_list);
480 sp->so_qstate &= ~SQ_INCOMP;
486 TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list);
487 so->so_qstate |= SQ_INCOMP;
493 wakeup_one(&head->so_timeo);
499 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
503 CURVNET_SET(so->so_vnet);
504 error = (*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td);
510 * solisten() transitions a socket from a non-listening state to a listening
511 * state, but can also be used to update the listen queue depth on an
512 * existing listen socket. The protocol will call back into the sockets
513 * layer using solisten_proto_check() and solisten_proto() to check and set
514 * socket-layer listen state. Call backs are used so that the protocol can
515 * acquire both protocol and socket layer locks in whatever order is required
518 * Protocol implementors are advised to hold the socket lock across the
519 * socket-layer test and set to avoid races at the socket layer.
522 solisten(struct socket *so, int backlog, struct thread *td)
525 return ((*so->so_proto->pr_usrreqs->pru_listen)(so, backlog, td));
529 solisten_proto_check(struct socket *so)
532 SOCK_LOCK_ASSERT(so);
534 if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
541 solisten_proto(struct socket *so, int backlog)
544 SOCK_LOCK_ASSERT(so);
546 if (backlog < 0 || backlog > somaxconn)
548 so->so_qlimit = backlog;
549 so->so_options |= SO_ACCEPTCONN;
553 * Attempt to free a socket. This should really be sotryfree().
555 * sofree() will succeed if:
557 * - There are no outstanding file descriptor references or related consumers
560 * - The socket has been closed by user space, if ever open (SS_NOFDREF).
562 * - The protocol does not have an outstanding strong reference on the socket
565 * - The socket is not in a completed connection queue, so a process has been
566 * notified that it is present. If it is removed, the user process may
567 * block in accept() despite select() saying the socket was ready.
569 * Otherwise, it will quietly abort so that a future call to sofree(), when
570 * conditions are right, can succeed.
573 sofree(struct socket *so)
575 struct protosw *pr = so->so_proto;
578 ACCEPT_LOCK_ASSERT();
579 SOCK_LOCK_ASSERT(so);
581 if ((so->so_state & SS_NOFDREF) == 0 || so->so_count != 0 ||
582 (so->so_state & SS_PROTOREF) || (so->so_qstate & SQ_COMP)) {
590 KASSERT((so->so_qstate & SQ_COMP) != 0 ||
591 (so->so_qstate & SQ_INCOMP) != 0,
592 ("sofree: so_head != NULL, but neither SQ_COMP nor "
594 KASSERT((so->so_qstate & SQ_COMP) == 0 ||
595 (so->so_qstate & SQ_INCOMP) == 0,
596 ("sofree: so->so_qstate is SQ_COMP and also SQ_INCOMP"));
597 TAILQ_REMOVE(&head->so_incomp, so, so_list);
599 so->so_qstate &= ~SQ_INCOMP;
602 KASSERT((so->so_qstate & SQ_COMP) == 0 &&
603 (so->so_qstate & SQ_INCOMP) == 0,
604 ("sofree: so_head == NULL, but still SQ_COMP(%d) or SQ_INCOMP(%d)",
605 so->so_qstate & SQ_COMP, so->so_qstate & SQ_INCOMP));
606 if (so->so_options & SO_ACCEPTCONN) {
607 KASSERT((TAILQ_EMPTY(&so->so_comp)), ("sofree: so_comp populated"));
608 KASSERT((TAILQ_EMPTY(&so->so_incomp)), ("sofree: so_comp populated"));
613 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
614 (*pr->pr_domain->dom_dispose)(so->so_rcv.sb_mb);
615 if (pr->pr_usrreqs->pru_detach != NULL)
616 (*pr->pr_usrreqs->pru_detach)(so);
619 * From this point on, we assume that no other references to this
620 * socket exist anywhere else in the stack. Therefore, no locks need
621 * to be acquired or held.
623 * We used to do a lot of socket buffer and socket locking here, as
624 * well as invoke sorflush() and perform wakeups. The direct call to
625 * dom_dispose() and sbrelease_internal() are an inlining of what was
626 * necessary from sorflush().
628 * Notice that the socket buffer and kqueue state are torn down
629 * before calling pru_detach. This means that protocols shold not
630 * assume they can perform socket wakeups, etc, in their detach code.
632 sbdestroy(&so->so_snd, so);
633 sbdestroy(&so->so_rcv, so);
634 knlist_destroy(&so->so_rcv.sb_sel.si_note);
635 knlist_destroy(&so->so_snd.sb_sel.si_note);
640 * Close a socket on last file table reference removal. Initiate disconnect
641 * if connected. Free socket when disconnect complete.
643 * This function will sorele() the socket. Note that soclose() may be called
644 * prior to the ref count reaching zero. The actual socket structure will
645 * not be freed until the ref count reaches zero.
648 soclose(struct socket *so)
652 KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter"));
654 CURVNET_SET(so->so_vnet);
655 funsetown(&so->so_sigio);
656 if (so->so_state & SS_ISCONNECTED) {
657 if ((so->so_state & SS_ISDISCONNECTING) == 0) {
658 error = sodisconnect(so);
662 if (so->so_options & SO_LINGER) {
663 if ((so->so_state & SS_ISDISCONNECTING) &&
664 (so->so_state & SS_NBIO))
666 while (so->so_state & SS_ISCONNECTED) {
667 error = tsleep(&so->so_timeo,
668 PSOCK | PCATCH, "soclos", so->so_linger * hz);
676 if (so->so_proto->pr_usrreqs->pru_close != NULL)
677 (*so->so_proto->pr_usrreqs->pru_close)(so);
678 if (so->so_options & SO_ACCEPTCONN) {
681 while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
682 TAILQ_REMOVE(&so->so_incomp, sp, so_list);
684 sp->so_qstate &= ~SQ_INCOMP;
690 while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
691 TAILQ_REMOVE(&so->so_comp, sp, so_list);
693 sp->so_qstate &= ~SQ_COMP;
703 KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF"));
704 so->so_state |= SS_NOFDREF;
711 * soabort() is used to abruptly tear down a connection, such as when a
712 * resource limit is reached (listen queue depth exceeded), or if a listen
713 * socket is closed while there are sockets waiting to be accepted.
715 * This interface is tricky, because it is called on an unreferenced socket,
716 * and must be called only by a thread that has actually removed the socket
717 * from the listen queue it was on, or races with other threads are risked.
719 * This interface will call into the protocol code, so must not be called
720 * with any socket locks held. Protocols do call it while holding their own
721 * recursible protocol mutexes, but this is something that should be subject
722 * to review in the future.
725 soabort(struct socket *so)
729 * In as much as is possible, assert that no references to this
730 * socket are held. This is not quite the same as asserting that the
731 * current thread is responsible for arranging for no references, but
732 * is as close as we can get for now.
734 KASSERT(so->so_count == 0, ("soabort: so_count"));
735 KASSERT((so->so_state & SS_PROTOREF) == 0, ("soabort: SS_PROTOREF"));
736 KASSERT(so->so_state & SS_NOFDREF, ("soabort: !SS_NOFDREF"));
737 KASSERT((so->so_state & SQ_COMP) == 0, ("soabort: SQ_COMP"));
738 KASSERT((so->so_state & SQ_INCOMP) == 0, ("soabort: SQ_INCOMP"));
740 if (so->so_proto->pr_usrreqs->pru_abort != NULL)
741 (*so->so_proto->pr_usrreqs->pru_abort)(so);
748 soaccept(struct socket *so, struct sockaddr **nam)
753 KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF"));
754 so->so_state &= ~SS_NOFDREF;
756 error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam);
761 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
765 if (so->so_options & SO_ACCEPTCONN)
768 * If protocol is connection-based, can only connect once.
769 * Otherwise, if connected, try to disconnect first. This allows
770 * user to disconnect by connecting to, e.g., a null address.
772 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
773 ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
774 (error = sodisconnect(so)))) {
778 * Prevent accumulated error from previous connection from
782 CURVNET_SET(so->so_vnet);
783 error = (*so->so_proto->pr_usrreqs->pru_connect)(so, nam, td);
791 soconnect2(struct socket *so1, struct socket *so2)
794 return ((*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2));
798 sodisconnect(struct socket *so)
802 if ((so->so_state & SS_ISCONNECTED) == 0)
804 if (so->so_state & SS_ISDISCONNECTING)
806 error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so);
810 #ifdef ZERO_COPY_SOCKETS
811 struct so_zerocopy_stats{
816 struct so_zerocopy_stats so_zerocp_stats = {0,0,0};
817 #include <netinet/in.h>
818 #include <net/route.h>
819 #include <netinet/in_pcb.h>
821 #include <vm/vm_page.h>
822 #include <vm/vm_object.h>
825 * sosend_copyin() is only used if zero copy sockets are enabled. Otherwise
826 * sosend_dgram() and sosend_generic() use m_uiotombuf().
828 * sosend_copyin() accepts a uio and prepares an mbuf chain holding part or
829 * all of the data referenced by the uio. If desired, it uses zero-copy.
830 * *space will be updated to reflect data copied in.
832 * NB: If atomic I/O is requested, the caller must already have checked that
833 * space can hold resid bytes.
835 * NB: In the event of an error, the caller may need to free the partial
836 * chain pointed to by *mpp. The contents of both *uio and *space may be
837 * modified even in the case of an error.
840 sosend_copyin(struct uio *uio, struct mbuf **retmp, int atomic, long *space,
843 struct mbuf *m, **mp, *top;
846 #ifdef ZERO_COPY_SOCKETS
853 resid = uio->uio_resid;
856 #ifdef ZERO_COPY_SOCKETS
858 #endif /* ZERO_COPY_SOCKETS */
859 if (resid >= MINCLSIZE) {
860 #ifdef ZERO_COPY_SOCKETS
862 m = m_gethdr(M_WAITOK, MT_DATA);
864 m->m_pkthdr.rcvif = NULL;
866 m = m_get(M_WAITOK, MT_DATA);
867 if (so_zero_copy_send &&
870 uio->uio_iov->iov_len>=PAGE_SIZE) {
871 so_zerocp_stats.size_ok++;
872 so_zerocp_stats.align_ok++;
873 cow_send = socow_setup(m, uio);
877 m_clget(m, M_WAITOK);
878 len = min(min(MCLBYTES, resid), *space);
880 #else /* ZERO_COPY_SOCKETS */
882 m = m_getcl(M_WAIT, MT_DATA, M_PKTHDR);
884 m->m_pkthdr.rcvif = NULL;
886 m = m_getcl(M_WAIT, MT_DATA, 0);
887 len = min(min(MCLBYTES, resid), *space);
888 #endif /* ZERO_COPY_SOCKETS */
891 m = m_gethdr(M_WAIT, MT_DATA);
893 m->m_pkthdr.rcvif = NULL;
895 len = min(min(MHLEN, resid), *space);
897 * For datagram protocols, leave room
898 * for protocol headers in first mbuf.
900 if (atomic && m && len < MHLEN)
903 m = m_get(M_WAIT, MT_DATA);
904 len = min(min(MLEN, resid), *space);
913 #ifdef ZERO_COPY_SOCKETS
917 #endif /* ZERO_COPY_SOCKETS */
918 error = uiomove(mtod(m, void *), (int)len, uio);
919 resid = uio->uio_resid;
922 top->m_pkthdr.len += len;
928 top->m_flags |= M_EOR;
931 } while (*space > 0 && atomic);
936 #endif /*ZERO_COPY_SOCKETS*/
938 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? 0 : SBL_WAIT)
941 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio,
942 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
945 int clen = 0, error, dontroute;
946 #ifdef ZERO_COPY_SOCKETS
947 int atomic = sosendallatonce(so) || top;
950 KASSERT(so->so_type == SOCK_DGRAM, ("sodgram_send: !SOCK_DGRAM"));
951 KASSERT(so->so_proto->pr_flags & PR_ATOMIC,
952 ("sodgram_send: !PR_ATOMIC"));
955 resid = uio->uio_resid;
957 resid = top->m_pkthdr.len;
959 * In theory resid should be unsigned. However, space must be
960 * signed, as it might be less than 0 if we over-committed, and we
961 * must use a signed comparison of space and resid. On the other
962 * hand, a negative resid causes us to loop sending 0-length
963 * segments to the protocol.
965 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
966 * type sockets since that's an error.
974 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0;
976 td->td_ru.ru_msgsnd++;
978 clen = control->m_len;
980 SOCKBUF_LOCK(&so->so_snd);
981 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
982 SOCKBUF_UNLOCK(&so->so_snd);
987 error = so->so_error;
989 SOCKBUF_UNLOCK(&so->so_snd);
992 if ((so->so_state & SS_ISCONNECTED) == 0) {
994 * `sendto' and `sendmsg' is allowed on a connection-based
995 * socket if it supports implied connect. Return ENOTCONN if
996 * not connected and no address is supplied.
998 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
999 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1000 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1001 !(resid == 0 && clen != 0)) {
1002 SOCKBUF_UNLOCK(&so->so_snd);
1006 } else if (addr == NULL) {
1007 if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1010 error = EDESTADDRREQ;
1011 SOCKBUF_UNLOCK(&so->so_snd);
1017 * Do we need MSG_OOB support in SOCK_DGRAM? Signs here may be a
1018 * problem and need fixing.
1020 space = sbspace(&so->so_snd);
1021 if (flags & MSG_OOB)
1024 SOCKBUF_UNLOCK(&so->so_snd);
1025 if (resid > space) {
1031 if (flags & MSG_EOR)
1032 top->m_flags |= M_EOR;
1034 #ifdef ZERO_COPY_SOCKETS
1035 error = sosend_copyin(uio, &top, atomic, &space, flags);
1040 * Copy the data from userland into a mbuf chain.
1041 * If no data is to be copied in, a single empty mbuf
1044 top = m_uiotombuf(uio, M_WAITOK, space, max_hdr,
1045 (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0)));
1047 error = EFAULT; /* only possible error */
1050 space -= resid - uio->uio_resid;
1052 resid = uio->uio_resid;
1054 KASSERT(resid == 0, ("sosend_dgram: resid != 0"));
1056 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock
1061 so->so_options |= SO_DONTROUTE;
1065 * XXX all the SBS_CANTSENDMORE checks previously done could be out
1066 * of date. We could have recieved a reset packet in an interrupt or
1067 * maybe we slept while doing page faults in uiomove() etc. We could
1068 * probably recheck again inside the locking protection here, but
1069 * there are probably other places that this also happens. We must
1072 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1073 (flags & MSG_OOB) ? PRUS_OOB :
1075 * If the user set MSG_EOF, the protocol understands this flag and
1076 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND.
1078 ((flags & MSG_EOF) &&
1079 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1082 /* If there is more to send set PRUS_MORETOCOME */
1083 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1084 top, addr, control, td);
1087 so->so_options &= ~SO_DONTROUTE;
1096 if (control != NULL)
1102 * Send on a socket. If send must go all at once and message is larger than
1103 * send buffering, then hard error. Lock against other senders. If must go
1104 * all at once and not enough room now, then inform user that this would
1105 * block and do nothing. Otherwise, if nonblocking, send as much as
1106 * possible. The data to be sent is described by "uio" if nonzero, otherwise
1107 * by the mbuf chain "top" (which must be null if uio is not). Data provided
1108 * in mbuf chain must be small enough to send all at once.
1110 * Returns nonzero on error, timeout or signal; callers must check for short
1111 * counts if EINTR/ERESTART are returned. Data and control buffers are freed
1115 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio,
1116 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1119 int clen = 0, error, dontroute;
1120 int atomic = sosendallatonce(so) || top;
1123 resid = uio->uio_resid;
1125 resid = top->m_pkthdr.len;
1127 * In theory resid should be unsigned. However, space must be
1128 * signed, as it might be less than 0 if we over-committed, and we
1129 * must use a signed comparison of space and resid. On the other
1130 * hand, a negative resid causes us to loop sending 0-length
1131 * segments to the protocol.
1133 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
1134 * type sockets since that's an error.
1136 if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
1142 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
1143 (so->so_proto->pr_flags & PR_ATOMIC);
1145 td->td_ru.ru_msgsnd++;
1146 if (control != NULL)
1147 clen = control->m_len;
1149 error = sblock(&so->so_snd, SBLOCKWAIT(flags));
1155 SOCKBUF_LOCK(&so->so_snd);
1156 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1157 SOCKBUF_UNLOCK(&so->so_snd);
1162 error = so->so_error;
1164 SOCKBUF_UNLOCK(&so->so_snd);
1167 if ((so->so_state & SS_ISCONNECTED) == 0) {
1169 * `sendto' and `sendmsg' is allowed on a connection-
1170 * based socket if it supports implied connect.
1171 * Return ENOTCONN if not connected and no address is
1174 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1175 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1176 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1177 !(resid == 0 && clen != 0)) {
1178 SOCKBUF_UNLOCK(&so->so_snd);
1182 } else if (addr == NULL) {
1183 SOCKBUF_UNLOCK(&so->so_snd);
1184 if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1187 error = EDESTADDRREQ;
1191 space = sbspace(&so->so_snd);
1192 if (flags & MSG_OOB)
1194 if ((atomic && resid > so->so_snd.sb_hiwat) ||
1195 clen > so->so_snd.sb_hiwat) {
1196 SOCKBUF_UNLOCK(&so->so_snd);
1200 if (space < resid + clen &&
1201 (atomic || space < so->so_snd.sb_lowat || space < clen)) {
1202 if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO)) {
1203 SOCKBUF_UNLOCK(&so->so_snd);
1204 error = EWOULDBLOCK;
1207 error = sbwait(&so->so_snd);
1208 SOCKBUF_UNLOCK(&so->so_snd);
1213 SOCKBUF_UNLOCK(&so->so_snd);
1218 if (flags & MSG_EOR)
1219 top->m_flags |= M_EOR;
1221 #ifdef ZERO_COPY_SOCKETS
1222 error = sosend_copyin(uio, &top, atomic,
1228 * Copy the data from userland into a mbuf
1229 * chain. If no data is to be copied in,
1230 * a single empty mbuf is returned.
1232 top = m_uiotombuf(uio, M_WAITOK, space,
1233 (atomic ? max_hdr : 0),
1234 (atomic ? M_PKTHDR : 0) |
1235 ((flags & MSG_EOR) ? M_EOR : 0));
1237 error = EFAULT; /* only possible error */
1240 space -= resid - uio->uio_resid;
1242 resid = uio->uio_resid;
1246 so->so_options |= SO_DONTROUTE;
1250 * XXX all the SBS_CANTSENDMORE checks previously
1251 * done could be out of date. We could have recieved
1252 * a reset packet in an interrupt or maybe we slept
1253 * while doing page faults in uiomove() etc. We
1254 * could probably recheck again inside the locking
1255 * protection here, but there are probably other
1256 * places that this also happens. We must rethink
1259 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1260 (flags & MSG_OOB) ? PRUS_OOB :
1262 * If the user set MSG_EOF, the protocol understands
1263 * this flag and nothing left to send then use
1264 * PRU_SEND_EOF instead of PRU_SEND.
1266 ((flags & MSG_EOF) &&
1267 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1270 /* If there is more to send set PRUS_MORETOCOME. */
1271 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1272 top, addr, control, td);
1275 so->so_options &= ~SO_DONTROUTE;
1283 } while (resid && space > 0);
1287 sbunlock(&so->so_snd);
1291 if (control != NULL)
1297 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
1298 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1302 CURVNET_SET(so->so_vnet);
1303 error = so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio, top,
1304 control, flags, td);
1310 * The part of soreceive() that implements reading non-inline out-of-band
1311 * data from a socket. For more complete comments, see soreceive(), from
1312 * which this code originated.
1314 * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
1315 * unable to return an mbuf chain to the caller.
1318 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags)
1320 struct protosw *pr = so->so_proto;
1324 KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
1326 m = m_get(M_WAIT, MT_DATA);
1327 error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK);
1331 #ifdef ZERO_COPY_SOCKETS
1332 if (so_zero_copy_receive) {
1335 if ((m->m_flags & M_EXT)
1336 && (m->m_ext.ext_type == EXT_DISPOSABLE))
1341 error = uiomoveco(mtod(m, void *),
1342 min(uio->uio_resid, m->m_len),
1345 #endif /* ZERO_COPY_SOCKETS */
1346 error = uiomove(mtod(m, void *),
1347 (int) min(uio->uio_resid, m->m_len), uio);
1349 } while (uio->uio_resid && error == 0 && m);
1357 * Following replacement or removal of the first mbuf on the first mbuf chain
1358 * of a socket buffer, push necessary state changes back into the socket
1359 * buffer so that other consumers see the values consistently. 'nextrecord'
1360 * is the callers locally stored value of the original value of
1361 * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes.
1362 * NOTE: 'nextrecord' may be NULL.
1364 static __inline void
1365 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord)
1368 SOCKBUF_LOCK_ASSERT(sb);
1370 * First, update for the new value of nextrecord. If necessary, make
1371 * it the first record.
1373 if (sb->sb_mb != NULL)
1374 sb->sb_mb->m_nextpkt = nextrecord;
1376 sb->sb_mb = nextrecord;
1379 * Now update any dependent socket buffer fields to reflect the new
1380 * state. This is an expanded inline of SB_EMPTY_FIXUP(), with the
1381 * addition of a second clause that takes care of the case where
1382 * sb_mb has been updated, but remains the last record.
1384 if (sb->sb_mb == NULL) {
1385 sb->sb_mbtail = NULL;
1386 sb->sb_lastrecord = NULL;
1387 } else if (sb->sb_mb->m_nextpkt == NULL)
1388 sb->sb_lastrecord = sb->sb_mb;
1393 * Implement receive operations on a socket. We depend on the way that
1394 * records are added to the sockbuf by sbappend. In particular, each record
1395 * (mbufs linked through m_next) must begin with an address if the protocol
1396 * so specifies, followed by an optional mbuf or mbufs containing ancillary
1397 * data, and then zero or more mbufs of data. In order to allow parallelism
1398 * between network receive and copying to user space, as well as avoid
1399 * sleeping with a mutex held, we release the socket buffer mutex during the
1400 * user space copy. Although the sockbuf is locked, new data may still be
1401 * appended, and thus we must maintain consistency of the sockbuf during that
1404 * The caller may receive the data as a single mbuf chain by supplying an
1405 * mbuf **mp0 for use in returning the chain. The uio is then used only for
1406 * the count in uio_resid.
1409 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio,
1410 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1412 struct mbuf *m, **mp;
1413 int flags, len, error, offset;
1414 struct protosw *pr = so->so_proto;
1415 struct mbuf *nextrecord;
1417 int orig_resid = uio->uio_resid;
1422 if (controlp != NULL)
1425 flags = *flagsp &~ MSG_EOR;
1428 if (flags & MSG_OOB)
1429 return (soreceive_rcvoob(so, uio, flags));
1432 if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING)
1434 (*pr->pr_usrreqs->pru_rcvd)(so, 0);
1436 error = sblock(&so->so_rcv, SBLOCKWAIT(flags));
1441 SOCKBUF_LOCK(&so->so_rcv);
1442 m = so->so_rcv.sb_mb;
1444 * If we have less data than requested, block awaiting more (subject
1445 * to any timeout) if:
1446 * 1. the current count is less than the low water mark, or
1447 * 2. MSG_WAITALL is set, and it is possible to do the entire
1448 * receive operation at once if we block (resid <= hiwat).
1449 * 3. MSG_DONTWAIT is not set
1450 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1451 * we have to do the receive in sections, and thus risk returning a
1452 * short count if a timeout or signal occurs after we start.
1454 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1455 so->so_rcv.sb_cc < uio->uio_resid) &&
1456 (so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
1457 ((flags & MSG_WAITALL) && uio->uio_resid <= so->so_rcv.sb_hiwat)) &&
1458 m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
1459 KASSERT(m != NULL || !so->so_rcv.sb_cc,
1460 ("receive: m == %p so->so_rcv.sb_cc == %u",
1461 m, so->so_rcv.sb_cc));
1465 error = so->so_error;
1466 if ((flags & MSG_PEEK) == 0)
1468 SOCKBUF_UNLOCK(&so->so_rcv);
1471 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1472 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1474 SOCKBUF_UNLOCK(&so->so_rcv);
1479 for (; m != NULL; m = m->m_next)
1480 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) {
1481 m = so->so_rcv.sb_mb;
1484 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1485 (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1486 SOCKBUF_UNLOCK(&so->so_rcv);
1490 if (uio->uio_resid == 0) {
1491 SOCKBUF_UNLOCK(&so->so_rcv);
1494 if ((so->so_state & SS_NBIO) ||
1495 (flags & (MSG_DONTWAIT|MSG_NBIO))) {
1496 SOCKBUF_UNLOCK(&so->so_rcv);
1497 error = EWOULDBLOCK;
1500 SBLASTRECORDCHK(&so->so_rcv);
1501 SBLASTMBUFCHK(&so->so_rcv);
1502 error = sbwait(&so->so_rcv);
1503 SOCKBUF_UNLOCK(&so->so_rcv);
1510 * From this point onward, we maintain 'nextrecord' as a cache of the
1511 * pointer to the next record in the socket buffer. We must keep the
1512 * various socket buffer pointers and local stack versions of the
1513 * pointers in sync, pushing out modifications before dropping the
1514 * socket buffer mutex, and re-reading them when picking it up.
1516 * Otherwise, we will race with the network stack appending new data
1517 * or records onto the socket buffer by using inconsistent/stale
1518 * versions of the field, possibly resulting in socket buffer
1521 * By holding the high-level sblock(), we prevent simultaneous
1522 * readers from pulling off the front of the socket buffer.
1524 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1526 uio->uio_td->td_ru.ru_msgrcv++;
1527 KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb"));
1528 SBLASTRECORDCHK(&so->so_rcv);
1529 SBLASTMBUFCHK(&so->so_rcv);
1530 nextrecord = m->m_nextpkt;
1531 if (pr->pr_flags & PR_ADDR) {
1532 KASSERT(m->m_type == MT_SONAME,
1533 ("m->m_type == %d", m->m_type));
1536 *psa = sodupsockaddr(mtod(m, struct sockaddr *),
1538 if (flags & MSG_PEEK) {
1541 sbfree(&so->so_rcv, m);
1542 so->so_rcv.sb_mb = m_free(m);
1543 m = so->so_rcv.sb_mb;
1544 sockbuf_pushsync(&so->so_rcv, nextrecord);
1549 * Process one or more MT_CONTROL mbufs present before any data mbufs
1550 * in the first mbuf chain on the socket buffer. If MSG_PEEK, we
1551 * just copy the data; if !MSG_PEEK, we call into the protocol to
1552 * perform externalization (or freeing if controlp == NULL).
1554 if (m != NULL && m->m_type == MT_CONTROL) {
1555 struct mbuf *cm = NULL, *cmn;
1556 struct mbuf **cme = &cm;
1559 if (flags & MSG_PEEK) {
1560 if (controlp != NULL) {
1561 *controlp = m_copy(m, 0, m->m_len);
1562 controlp = &(*controlp)->m_next;
1566 sbfree(&so->so_rcv, m);
1567 so->so_rcv.sb_mb = m->m_next;
1570 cme = &(*cme)->m_next;
1571 m = so->so_rcv.sb_mb;
1573 } while (m != NULL && m->m_type == MT_CONTROL);
1574 if ((flags & MSG_PEEK) == 0)
1575 sockbuf_pushsync(&so->so_rcv, nextrecord);
1576 while (cm != NULL) {
1579 if (pr->pr_domain->dom_externalize != NULL) {
1580 SOCKBUF_UNLOCK(&so->so_rcv);
1581 error = (*pr->pr_domain->dom_externalize)
1583 SOCKBUF_LOCK(&so->so_rcv);
1584 } else if (controlp != NULL)
1588 if (controlp != NULL) {
1590 while (*controlp != NULL)
1591 controlp = &(*controlp)->m_next;
1596 nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1598 nextrecord = so->so_rcv.sb_mb;
1602 if ((flags & MSG_PEEK) == 0) {
1603 KASSERT(m->m_nextpkt == nextrecord,
1604 ("soreceive: post-control, nextrecord !sync"));
1605 if (nextrecord == NULL) {
1606 KASSERT(so->so_rcv.sb_mb == m,
1607 ("soreceive: post-control, sb_mb!=m"));
1608 KASSERT(so->so_rcv.sb_lastrecord == m,
1609 ("soreceive: post-control, lastrecord!=m"));
1613 if (type == MT_OOBDATA)
1616 if ((flags & MSG_PEEK) == 0) {
1617 KASSERT(so->so_rcv.sb_mb == nextrecord,
1618 ("soreceive: sb_mb != nextrecord"));
1619 if (so->so_rcv.sb_mb == NULL) {
1620 KASSERT(so->so_rcv.sb_lastrecord == NULL,
1621 ("soreceive: sb_lastercord != NULL"));
1625 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1626 SBLASTRECORDCHK(&so->so_rcv);
1627 SBLASTMBUFCHK(&so->so_rcv);
1630 * Now continue to read any data mbufs off of the head of the socket
1631 * buffer until the read request is satisfied. Note that 'type' is
1632 * used to store the type of any mbuf reads that have happened so far
1633 * such that soreceive() can stop reading if the type changes, which
1634 * causes soreceive() to return only one of regular data and inline
1635 * out-of-band data in a single socket receive operation.
1639 while (m != NULL && uio->uio_resid > 0 && error == 0) {
1641 * If the type of mbuf has changed since the last mbuf
1642 * examined ('type'), end the receive operation.
1644 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1645 if (m->m_type == MT_OOBDATA) {
1646 if (type != MT_OOBDATA)
1648 } else if (type == MT_OOBDATA)
1651 KASSERT(m->m_type == MT_DATA,
1652 ("m->m_type == %d", m->m_type));
1653 so->so_rcv.sb_state &= ~SBS_RCVATMARK;
1654 len = uio->uio_resid;
1655 if (so->so_oobmark && len > so->so_oobmark - offset)
1656 len = so->so_oobmark - offset;
1657 if (len > m->m_len - moff)
1658 len = m->m_len - moff;
1660 * If mp is set, just pass back the mbufs. Otherwise copy
1661 * them out via the uio, then free. Sockbuf must be
1662 * consistent here (points to current mbuf, it points to next
1663 * record) when we drop priority; we must note any additions
1664 * to the sockbuf when we block interrupts again.
1667 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1668 SBLASTRECORDCHK(&so->so_rcv);
1669 SBLASTMBUFCHK(&so->so_rcv);
1670 SOCKBUF_UNLOCK(&so->so_rcv);
1671 #ifdef ZERO_COPY_SOCKETS
1672 if (so_zero_copy_receive) {
1675 if ((m->m_flags & M_EXT)
1676 && (m->m_ext.ext_type == EXT_DISPOSABLE))
1681 error = uiomoveco(mtod(m, char *) + moff,
1685 #endif /* ZERO_COPY_SOCKETS */
1686 error = uiomove(mtod(m, char *) + moff, (int)len, uio);
1687 SOCKBUF_LOCK(&so->so_rcv);
1690 * The MT_SONAME mbuf has already been removed
1691 * from the record, so it is necessary to
1692 * remove the data mbufs, if any, to preserve
1693 * the invariant in the case of PR_ADDR that
1694 * requires MT_SONAME mbufs at the head of
1697 if (m && pr->pr_flags & PR_ATOMIC &&
1698 ((flags & MSG_PEEK) == 0))
1699 (void)sbdroprecord_locked(&so->so_rcv);
1700 SOCKBUF_UNLOCK(&so->so_rcv);
1704 uio->uio_resid -= len;
1705 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1706 if (len == m->m_len - moff) {
1707 if (m->m_flags & M_EOR)
1709 if (flags & MSG_PEEK) {
1713 nextrecord = m->m_nextpkt;
1714 sbfree(&so->so_rcv, m);
1718 so->so_rcv.sb_mb = m = m->m_next;
1721 so->so_rcv.sb_mb = m_free(m);
1722 m = so->so_rcv.sb_mb;
1724 sockbuf_pushsync(&so->so_rcv, nextrecord);
1725 SBLASTRECORDCHK(&so->so_rcv);
1726 SBLASTMBUFCHK(&so->so_rcv);
1729 if (flags & MSG_PEEK)
1735 if (flags & MSG_DONTWAIT)
1736 copy_flag = M_DONTWAIT;
1739 if (copy_flag == M_WAIT)
1740 SOCKBUF_UNLOCK(&so->so_rcv);
1741 *mp = m_copym(m, 0, len, copy_flag);
1742 if (copy_flag == M_WAIT)
1743 SOCKBUF_LOCK(&so->so_rcv);
1746 * m_copym() couldn't
1747 * allocate an mbuf. Adjust
1748 * uio_resid back (it was
1749 * adjusted down by len
1750 * bytes, which we didn't end
1751 * up "copying" over).
1753 uio->uio_resid += len;
1759 so->so_rcv.sb_cc -= len;
1762 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1763 if (so->so_oobmark) {
1764 if ((flags & MSG_PEEK) == 0) {
1765 so->so_oobmark -= len;
1766 if (so->so_oobmark == 0) {
1767 so->so_rcv.sb_state |= SBS_RCVATMARK;
1772 if (offset == so->so_oobmark)
1776 if (flags & MSG_EOR)
1779 * If the MSG_WAITALL flag is set (for non-atomic socket), we
1780 * must not quit until "uio->uio_resid == 0" or an error
1781 * termination. If a signal/timeout occurs, return with a
1782 * short count but without error. Keep sockbuf locked
1783 * against other readers.
1785 while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1786 !sosendallatonce(so) && nextrecord == NULL) {
1787 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1788 if (so->so_error || so->so_rcv.sb_state & SBS_CANTRCVMORE)
1791 * Notify the protocol that some data has been
1792 * drained before blocking.
1794 if (pr->pr_flags & PR_WANTRCVD) {
1795 SOCKBUF_UNLOCK(&so->so_rcv);
1796 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
1797 SOCKBUF_LOCK(&so->so_rcv);
1799 SBLASTRECORDCHK(&so->so_rcv);
1800 SBLASTMBUFCHK(&so->so_rcv);
1801 error = sbwait(&so->so_rcv);
1803 SOCKBUF_UNLOCK(&so->so_rcv);
1806 m = so->so_rcv.sb_mb;
1808 nextrecord = m->m_nextpkt;
1812 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1813 if (m != NULL && pr->pr_flags & PR_ATOMIC) {
1815 if ((flags & MSG_PEEK) == 0)
1816 (void) sbdroprecord_locked(&so->so_rcv);
1818 if ((flags & MSG_PEEK) == 0) {
1821 * First part is an inline SB_EMPTY_FIXUP(). Second
1822 * part makes sure sb_lastrecord is up-to-date if
1823 * there is still data in the socket buffer.
1825 so->so_rcv.sb_mb = nextrecord;
1826 if (so->so_rcv.sb_mb == NULL) {
1827 so->so_rcv.sb_mbtail = NULL;
1828 so->so_rcv.sb_lastrecord = NULL;
1829 } else if (nextrecord->m_nextpkt == NULL)
1830 so->so_rcv.sb_lastrecord = nextrecord;
1832 SBLASTRECORDCHK(&so->so_rcv);
1833 SBLASTMBUFCHK(&so->so_rcv);
1835 * If soreceive() is being done from the socket callback,
1836 * then don't need to generate ACK to peer to update window,
1837 * since ACK will be generated on return to TCP.
1839 if (!(flags & MSG_SOCALLBCK) &&
1840 (pr->pr_flags & PR_WANTRCVD)) {
1841 SOCKBUF_UNLOCK(&so->so_rcv);
1842 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
1843 SOCKBUF_LOCK(&so->so_rcv);
1846 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1847 if (orig_resid == uio->uio_resid && orig_resid &&
1848 (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
1849 SOCKBUF_UNLOCK(&so->so_rcv);
1852 SOCKBUF_UNLOCK(&so->so_rcv);
1857 sbunlock(&so->so_rcv);
1862 * Optimized version of soreceive() for stream (TCP) sockets.
1865 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio,
1866 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1868 int len = 0, error = 0, flags, oresid;
1870 struct mbuf *m, *n = NULL;
1872 /* We only do stream sockets. */
1873 if (so->so_type != SOCK_STREAM)
1877 if (controlp != NULL)
1880 flags = *flagsp &~ MSG_EOR;
1883 if (flags & MSG_OOB)
1884 return (soreceive_rcvoob(so, uio, flags));
1890 /* Prevent other readers from entering the socket. */
1891 error = sblock(sb, SBLOCKWAIT(flags));
1896 /* Easy one, no space to copyout anything. */
1897 if (uio->uio_resid == 0) {
1901 oresid = uio->uio_resid;
1903 /* We will never ever get anything unless we are connected. */
1904 if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) {
1905 /* When disconnecting there may be still some data left. */
1908 if (!(so->so_state & SS_ISDISCONNECTED))
1913 /* Socket buffer is empty and we shall not block. */
1914 if (sb->sb_cc == 0 &&
1915 ((sb->sb_flags & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) {
1921 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1923 /* Abort if socket has reported problems. */
1927 if (oresid > uio->uio_resid)
1929 error = so->so_error;
1930 if (!(flags & MSG_PEEK))
1935 /* Door is closed. Deliver what is left, if any. */
1936 if (sb->sb_state & SBS_CANTRCVMORE) {
1943 /* Socket buffer got some data that we shall deliver now. */
1944 if (sb->sb_cc > 0 && !(flags & MSG_WAITALL) &&
1945 ((sb->sb_flags & SS_NBIO) ||
1946 (flags & (MSG_DONTWAIT|MSG_NBIO)) ||
1947 sb->sb_cc >= sb->sb_lowat ||
1948 sb->sb_cc >= uio->uio_resid ||
1949 sb->sb_cc >= sb->sb_hiwat) ) {
1953 /* On MSG_WAITALL we must wait until all data or error arrives. */
1954 if ((flags & MSG_WAITALL) &&
1955 (sb->sb_cc >= uio->uio_resid || sb->sb_cc >= sb->sb_lowat))
1959 * Wait and block until (more) data comes in.
1960 * NB: Drops the sockbuf lock during wait.
1968 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1969 KASSERT(sb->sb_cc > 0, ("%s: sockbuf empty", __func__));
1970 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__));
1974 uio->uio_td->td_ru.ru_msgrcv++;
1976 /* Fill uio until full or current end of socket buffer is reached. */
1977 len = min(uio->uio_resid, sb->sb_cc);
1979 /* Dequeue as many mbufs as possible. */
1980 if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) {
1981 for (*mp0 = m = sb->sb_mb;
1982 m != NULL && m->m_len <= len;
1985 uio->uio_resid -= m->m_len;
1990 if (sb->sb_mb == NULL)
1994 /* Copy the remainder. */
1996 KASSERT(sb->sb_mb != NULL,
1997 ("%s: len > 0 && sb->sb_mb empty", __func__));
1999 m = m_copym(sb->sb_mb, 0, len, M_DONTWAIT);
2001 len = 0; /* Don't flush data from sockbuf. */
2003 uio->uio_resid -= m->m_len;
2014 /* NB: Must unlock socket buffer as uiomove may sleep. */
2016 error = m_mbuftouio(uio, sb->sb_mb, len);
2021 SBLASTRECORDCHK(sb);
2025 * Remove the delivered data from the socket buffer unless we
2026 * were only peeking.
2028 if (!(flags & MSG_PEEK)) {
2030 sbdrop_locked(sb, len);
2032 /* Notify protocol that we drained some data. */
2033 if ((so->so_proto->pr_flags & PR_WANTRCVD) &&
2034 (((flags & MSG_WAITALL) && uio->uio_resid > 0) ||
2035 !(flags & MSG_SOCALLBCK))) {
2037 (*so->so_proto->pr_usrreqs->pru_rcvd)(so, flags);
2043 * For MSG_WAITALL we may have to loop again and wait for
2044 * more data to come in.
2046 if ((flags & MSG_WAITALL) && uio->uio_resid > 0)
2049 SOCKBUF_LOCK_ASSERT(sb);
2050 SBLASTRECORDCHK(sb);
2058 * Optimized version of soreceive() for simple datagram cases from userspace.
2059 * Unlike in the stream case, we're able to drop a datagram if copyout()
2060 * fails, and because we handle datagrams atomically, we don't need to use a
2061 * sleep lock to prevent I/O interlacing.
2064 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio,
2065 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2067 struct mbuf *m, *m2;
2068 int flags, len, error;
2069 struct protosw *pr = so->so_proto;
2070 struct mbuf *nextrecord;
2074 if (controlp != NULL)
2077 flags = *flagsp &~ MSG_EOR;
2082 * For any complicated cases, fall back to the full
2083 * soreceive_generic().
2085 if (mp0 != NULL || (flags & MSG_PEEK) || (flags & MSG_OOB))
2086 return (soreceive_generic(so, psa, uio, mp0, controlp,
2090 * Enforce restrictions on use.
2092 KASSERT((pr->pr_flags & PR_WANTRCVD) == 0,
2093 ("soreceive_dgram: wantrcvd"));
2094 KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic"));
2095 KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0,
2096 ("soreceive_dgram: SBS_RCVATMARK"));
2097 KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0,
2098 ("soreceive_dgram: P_CONNREQUIRED"));
2101 * Loop blocking while waiting for a datagram.
2103 SOCKBUF_LOCK(&so->so_rcv);
2104 while ((m = so->so_rcv.sb_mb) == NULL) {
2105 KASSERT(so->so_rcv.sb_cc == 0,
2106 ("soreceive_dgram: sb_mb NULL but sb_cc %u",
2109 error = so->so_error;
2111 SOCKBUF_UNLOCK(&so->so_rcv);
2114 if (so->so_rcv.sb_state & SBS_CANTRCVMORE ||
2115 uio->uio_resid == 0) {
2116 SOCKBUF_UNLOCK(&so->so_rcv);
2119 if ((so->so_state & SS_NBIO) ||
2120 (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2121 SOCKBUF_UNLOCK(&so->so_rcv);
2122 return (EWOULDBLOCK);
2124 SBLASTRECORDCHK(&so->so_rcv);
2125 SBLASTMBUFCHK(&so->so_rcv);
2126 error = sbwait(&so->so_rcv);
2128 SOCKBUF_UNLOCK(&so->so_rcv);
2132 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2135 uio->uio_td->td_ru.ru_msgrcv++;
2136 SBLASTRECORDCHK(&so->so_rcv);
2137 SBLASTMBUFCHK(&so->so_rcv);
2138 nextrecord = m->m_nextpkt;
2139 if (nextrecord == NULL) {
2140 KASSERT(so->so_rcv.sb_lastrecord == m,
2141 ("soreceive_dgram: lastrecord != m"));
2144 KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord,
2145 ("soreceive_dgram: m_nextpkt != nextrecord"));
2148 * Pull 'm' and its chain off the front of the packet queue.
2150 so->so_rcv.sb_mb = NULL;
2151 sockbuf_pushsync(&so->so_rcv, nextrecord);
2154 * Walk 'm's chain and free that many bytes from the socket buffer.
2156 for (m2 = m; m2 != NULL; m2 = m2->m_next)
2157 sbfree(&so->so_rcv, m2);
2160 * Do a few last checks before we let go of the lock.
2162 SBLASTRECORDCHK(&so->so_rcv);
2163 SBLASTMBUFCHK(&so->so_rcv);
2164 SOCKBUF_UNLOCK(&so->so_rcv);
2166 if (pr->pr_flags & PR_ADDR) {
2167 KASSERT(m->m_type == MT_SONAME,
2168 ("m->m_type == %d", m->m_type));
2170 *psa = sodupsockaddr(mtod(m, struct sockaddr *),
2175 /* XXXRW: Can this happen? */
2180 * Packet to copyout() is now in 'm' and it is disconnected from the
2183 * Process one or more MT_CONTROL mbufs present before any data mbufs
2184 * in the first mbuf chain on the socket buffer. We call into the
2185 * protocol to perform externalization (or freeing if controlp ==
2188 if (m->m_type == MT_CONTROL) {
2189 struct mbuf *cm = NULL, *cmn;
2190 struct mbuf **cme = &cm;
2196 cme = &(*cme)->m_next;
2198 } while (m != NULL && m->m_type == MT_CONTROL);
2199 while (cm != NULL) {
2202 if (pr->pr_domain->dom_externalize != NULL) {
2203 error = (*pr->pr_domain->dom_externalize)
2205 } else if (controlp != NULL)
2209 if (controlp != NULL) {
2210 while (*controlp != NULL)
2211 controlp = &(*controlp)->m_next;
2216 KASSERT(m->m_type == MT_DATA, ("soreceive_dgram: !data"));
2218 while (m != NULL && uio->uio_resid > 0) {
2219 len = uio->uio_resid;
2222 error = uiomove(mtod(m, char *), (int)len, uio);
2238 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
2239 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2242 return (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio, mp0,
2247 soshutdown(struct socket *so, int how)
2249 struct protosw *pr = so->so_proto;
2252 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
2254 if (pr->pr_usrreqs->pru_flush != NULL) {
2255 (*pr->pr_usrreqs->pru_flush)(so, how);
2259 if (how != SHUT_RD) {
2260 CURVNET_SET(so->so_vnet);
2261 error = (*pr->pr_usrreqs->pru_shutdown)(so);
2269 sorflush(struct socket *so)
2271 struct sockbuf *sb = &so->so_rcv;
2272 struct protosw *pr = so->so_proto;
2276 * In order to avoid calling dom_dispose with the socket buffer mutex
2277 * held, and in order to generally avoid holding the lock for a long
2278 * time, we make a copy of the socket buffer and clear the original
2279 * (except locks, state). The new socket buffer copy won't have
2280 * initialized locks so we can only call routines that won't use or
2281 * assert those locks.
2283 * Dislodge threads currently blocked in receive and wait to acquire
2284 * a lock against other simultaneous readers before clearing the
2285 * socket buffer. Don't let our acquire be interrupted by a signal
2286 * despite any existing socket disposition on interruptable waiting.
2288 CURVNET_SET(so->so_vnet);
2290 (void) sblock(sb, SBL_WAIT | SBL_NOINTR);
2293 * Invalidate/clear most of the sockbuf structure, but leave selinfo
2294 * and mutex data unchanged.
2297 bzero(&asb, offsetof(struct sockbuf, sb_startzero));
2298 bcopy(&sb->sb_startzero, &asb.sb_startzero,
2299 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2300 bzero(&sb->sb_startzero,
2301 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2306 * Dispose of special rights and flush the socket buffer. Don't call
2307 * any unsafe routines (that rely on locks being initialized) on asb.
2309 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
2310 (*pr->pr_domain->dom_dispose)(asb.sb_mb);
2311 sbrelease_internal(&asb, so);
2316 * Perhaps this routine, and sooptcopyout(), below, ought to come in an
2317 * additional variant to handle the case where the option value needs to be
2318 * some kind of integer, but not a specific size. In addition to their use
2319 * here, these functions are also called by the protocol-level pr_ctloutput()
2323 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2328 * If the user gives us more than we wanted, we ignore it, but if we
2329 * don't get the minimum length the caller wants, we return EINVAL.
2330 * On success, sopt->sopt_valsize is set to however much we actually
2333 if ((valsize = sopt->sopt_valsize) < minlen)
2336 sopt->sopt_valsize = valsize = len;
2338 if (sopt->sopt_td != NULL)
2339 return (copyin(sopt->sopt_val, buf, valsize));
2341 bcopy(sopt->sopt_val, buf, valsize);
2346 * Kernel version of setsockopt(2).
2348 * XXX: optlen is size_t, not socklen_t
2351 so_setsockopt(struct socket *so, int level, int optname, void *optval,
2354 struct sockopt sopt;
2356 sopt.sopt_level = level;
2357 sopt.sopt_name = optname;
2358 sopt.sopt_dir = SOPT_SET;
2359 sopt.sopt_val = optval;
2360 sopt.sopt_valsize = optlen;
2361 sopt.sopt_td = NULL;
2362 return (sosetopt(so, &sopt));
2366 sosetopt(struct socket *so, struct sockopt *sopt)
2377 if (sopt->sopt_level != SOL_SOCKET) {
2378 if (so->so_proto && so->so_proto->pr_ctloutput)
2379 return ((*so->so_proto->pr_ctloutput)
2381 error = ENOPROTOOPT;
2383 switch (sopt->sopt_name) {
2385 case SO_ACCEPTFILTER:
2386 error = do_setopt_accept_filter(so, sopt);
2392 error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
2397 so->so_linger = l.l_linger;
2399 so->so_options |= SO_LINGER;
2401 so->so_options &= ~SO_LINGER;
2408 case SO_USELOOPBACK:
2418 error = sooptcopyin(sopt, &optval, sizeof optval,
2424 so->so_options |= sopt->sopt_name;
2426 so->so_options &= ~sopt->sopt_name;
2431 error = sooptcopyin(sopt, &optval, sizeof optval,
2433 if (optval < 1 || optval > rt_numfibs) {
2437 if ((so->so_proto->pr_domain->dom_family == PF_INET) ||
2438 (so->so_proto->pr_domain->dom_family == PF_ROUTE)) {
2439 so->so_fibnum = optval;
2440 /* Note: ignore error */
2441 if (so->so_proto && so->so_proto->pr_ctloutput)
2442 (*so->so_proto->pr_ctloutput)(so, sopt);
2451 error = sooptcopyin(sopt, &optval, sizeof optval,
2457 * Values < 1 make no sense for any of these options,
2465 switch (sopt->sopt_name) {
2468 if (sbreserve(sopt->sopt_name == SO_SNDBUF ?
2469 &so->so_snd : &so->so_rcv, (u_long)optval,
2470 so, curthread) == 0) {
2474 (sopt->sopt_name == SO_SNDBUF ? &so->so_snd :
2475 &so->so_rcv)->sb_flags &= ~SB_AUTOSIZE;
2479 * Make sure the low-water is never greater than the
2483 SOCKBUF_LOCK(&so->so_snd);
2484 so->so_snd.sb_lowat =
2485 (optval > so->so_snd.sb_hiwat) ?
2486 so->so_snd.sb_hiwat : optval;
2487 SOCKBUF_UNLOCK(&so->so_snd);
2490 SOCKBUF_LOCK(&so->so_rcv);
2491 so->so_rcv.sb_lowat =
2492 (optval > so->so_rcv.sb_hiwat) ?
2493 so->so_rcv.sb_hiwat : optval;
2494 SOCKBUF_UNLOCK(&so->so_rcv);
2502 if (SV_CURPROC_FLAG(SV_ILP32)) {
2503 struct timeval32 tv32;
2505 error = sooptcopyin(sopt, &tv32, sizeof tv32,
2507 CP(tv32, tv, tv_sec);
2508 CP(tv32, tv, tv_usec);
2511 error = sooptcopyin(sopt, &tv, sizeof tv,
2516 /* assert(hz > 0); */
2517 if (tv.tv_sec < 0 || tv.tv_sec > INT_MAX / hz ||
2518 tv.tv_usec < 0 || tv.tv_usec >= 1000000) {
2522 /* assert(tick > 0); */
2523 /* assert(ULONG_MAX - INT_MAX >= 1000000); */
2524 val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / tick;
2525 if (val > INT_MAX) {
2529 if (val == 0 && tv.tv_usec != 0)
2532 switch (sopt->sopt_name) {
2534 so->so_snd.sb_timeo = val;
2537 so->so_rcv.sb_timeo = val;
2544 error = sooptcopyin(sopt, &extmac, sizeof extmac,
2548 error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
2556 error = ENOPROTOOPT;
2559 if (error == 0 && so->so_proto != NULL &&
2560 so->so_proto->pr_ctloutput != NULL) {
2561 (void) ((*so->so_proto->pr_ctloutput)
2570 * Helper routine for getsockopt.
2573 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
2581 * Documented get behavior is that we always return a value, possibly
2582 * truncated to fit in the user's buffer. Traditional behavior is
2583 * that we always tell the user precisely how much we copied, rather
2584 * than something useful like the total amount we had available for
2585 * her. Note that this interface is not idempotent; the entire
2586 * answer must generated ahead of time.
2588 valsize = min(len, sopt->sopt_valsize);
2589 sopt->sopt_valsize = valsize;
2590 if (sopt->sopt_val != NULL) {
2591 if (sopt->sopt_td != NULL)
2592 error = copyout(buf, sopt->sopt_val, valsize);
2594 bcopy(buf, sopt->sopt_val, valsize);
2600 sogetopt(struct socket *so, struct sockopt *sopt)
2610 if (sopt->sopt_level != SOL_SOCKET) {
2611 if (so->so_proto && so->so_proto->pr_ctloutput) {
2612 return ((*so->so_proto->pr_ctloutput)
2615 return (ENOPROTOOPT);
2617 switch (sopt->sopt_name) {
2619 case SO_ACCEPTFILTER:
2620 error = do_getopt_accept_filter(so, sopt);
2625 l.l_onoff = so->so_options & SO_LINGER;
2626 l.l_linger = so->so_linger;
2628 error = sooptcopyout(sopt, &l, sizeof l);
2631 case SO_USELOOPBACK:
2643 optval = so->so_options & sopt->sopt_name;
2645 error = sooptcopyout(sopt, &optval, sizeof optval);
2649 optval = so->so_type;
2654 optval = so->so_error;
2660 optval = so->so_snd.sb_hiwat;
2664 optval = so->so_rcv.sb_hiwat;
2668 optval = so->so_snd.sb_lowat;
2672 optval = so->so_rcv.sb_lowat;
2677 optval = (sopt->sopt_name == SO_SNDTIMEO ?
2678 so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
2680 tv.tv_sec = optval / hz;
2681 tv.tv_usec = (optval % hz) * tick;
2683 if (SV_CURPROC_FLAG(SV_ILP32)) {
2684 struct timeval32 tv32;
2686 CP(tv, tv32, tv_sec);
2687 CP(tv, tv32, tv_usec);
2688 error = sooptcopyout(sopt, &tv32, sizeof tv32);
2691 error = sooptcopyout(sopt, &tv, sizeof tv);
2696 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
2700 error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
2704 error = sooptcopyout(sopt, &extmac, sizeof extmac);
2712 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
2716 error = mac_getsockopt_peerlabel(
2717 sopt->sopt_td->td_ucred, so, &extmac);
2720 error = sooptcopyout(sopt, &extmac, sizeof extmac);
2726 case SO_LISTENQLIMIT:
2727 optval = so->so_qlimit;
2731 optval = so->so_qlen;
2734 case SO_LISTENINCQLEN:
2735 optval = so->so_incqlen;
2739 error = ENOPROTOOPT;
2746 /* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */
2748 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
2750 struct mbuf *m, *m_prev;
2751 int sopt_size = sopt->sopt_valsize;
2753 MGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT, MT_DATA);
2756 if (sopt_size > MLEN) {
2757 MCLGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT);
2758 if ((m->m_flags & M_EXT) == 0) {
2762 m->m_len = min(MCLBYTES, sopt_size);
2764 m->m_len = min(MLEN, sopt_size);
2766 sopt_size -= m->m_len;
2771 MGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT, MT_DATA);
2776 if (sopt_size > MLEN) {
2777 MCLGET(m, sopt->sopt_td != NULL ? M_WAIT :
2779 if ((m->m_flags & M_EXT) == 0) {
2784 m->m_len = min(MCLBYTES, sopt_size);
2786 m->m_len = min(MLEN, sopt_size);
2788 sopt_size -= m->m_len;
2795 /* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */
2797 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
2799 struct mbuf *m0 = m;
2801 if (sopt->sopt_val == NULL)
2803 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
2804 if (sopt->sopt_td != NULL) {
2807 error = copyin(sopt->sopt_val, mtod(m, char *),
2814 bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
2815 sopt->sopt_valsize -= m->m_len;
2816 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
2819 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
2820 panic("ip6_sooptmcopyin");
2824 /* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */
2826 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
2828 struct mbuf *m0 = m;
2831 if (sopt->sopt_val == NULL)
2833 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
2834 if (sopt->sopt_td != NULL) {
2837 error = copyout(mtod(m, char *), sopt->sopt_val,
2844 bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
2845 sopt->sopt_valsize -= m->m_len;
2846 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
2847 valsize += m->m_len;
2851 /* enough soopt buffer should be given from user-land */
2855 sopt->sopt_valsize = valsize;
2860 * sohasoutofband(): protocol notifies socket layer of the arrival of new
2861 * out-of-band data, which will then notify socket consumers.
2864 sohasoutofband(struct socket *so)
2867 if (so->so_sigio != NULL)
2868 pgsigio(&so->so_sigio, SIGURG, 0);
2869 selwakeuppri(&so->so_rcv.sb_sel, PSOCK);
2873 sopoll(struct socket *so, int events, struct ucred *active_cred,
2877 return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred,
2882 sopoll_generic(struct socket *so, int events, struct ucred *active_cred,
2887 SOCKBUF_LOCK(&so->so_snd);
2888 SOCKBUF_LOCK(&so->so_rcv);
2889 if (events & (POLLIN | POLLRDNORM))
2890 if (soreadabledata(so))
2891 revents |= events & (POLLIN | POLLRDNORM);
2893 if (events & (POLLOUT | POLLWRNORM))
2894 if (sowriteable(so))
2895 revents |= events & (POLLOUT | POLLWRNORM);
2897 if (events & (POLLPRI | POLLRDBAND))
2898 if (so->so_oobmark || (so->so_rcv.sb_state & SBS_RCVATMARK))
2899 revents |= events & (POLLPRI | POLLRDBAND);
2901 if ((events & POLLINIGNEOF) == 0) {
2902 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2903 revents |= events & (POLLIN | POLLRDNORM);
2904 if (so->so_snd.sb_state & SBS_CANTSENDMORE)
2910 if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
2911 selrecord(td, &so->so_rcv.sb_sel);
2912 so->so_rcv.sb_flags |= SB_SEL;
2915 if (events & (POLLOUT | POLLWRNORM)) {
2916 selrecord(td, &so->so_snd.sb_sel);
2917 so->so_snd.sb_flags |= SB_SEL;
2921 SOCKBUF_UNLOCK(&so->so_rcv);
2922 SOCKBUF_UNLOCK(&so->so_snd);
2927 soo_kqfilter(struct file *fp, struct knote *kn)
2929 struct socket *so = kn->kn_fp->f_data;
2932 switch (kn->kn_filter) {
2934 if (so->so_options & SO_ACCEPTCONN)
2935 kn->kn_fop = &solisten_filtops;
2937 kn->kn_fop = &soread_filtops;
2941 kn->kn_fop = &sowrite_filtops;
2949 knlist_add(&sb->sb_sel.si_note, kn, 1);
2950 sb->sb_flags |= SB_KNOTE;
2956 * Some routines that return EOPNOTSUPP for entry points that are not
2957 * supported by a protocol. Fill in as needed.
2960 pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
2967 pru_attach_notsupp(struct socket *so, int proto, struct thread *td)
2974 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
2981 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
2988 pru_connect2_notsupp(struct socket *so1, struct socket *so2)
2995 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
2996 struct ifnet *ifp, struct thread *td)
3003 pru_disconnect_notsupp(struct socket *so)
3010 pru_listen_notsupp(struct socket *so, int backlog, struct thread *td)
3017 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam)
3024 pru_rcvd_notsupp(struct socket *so, int flags)
3031 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
3038 pru_send_notsupp(struct socket *so, int flags, struct mbuf *m,
3039 struct sockaddr *addr, struct mbuf *control, struct thread *td)
3046 * This isn't really a ``null'' operation, but it's the default one and
3047 * doesn't do anything destructive.
3050 pru_sense_null(struct socket *so, struct stat *sb)
3053 sb->st_blksize = so->so_snd.sb_hiwat;
3058 pru_shutdown_notsupp(struct socket *so)
3065 pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam)
3072 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio,
3073 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
3080 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr,
3081 struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
3088 pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred,
3096 filt_sordetach(struct knote *kn)
3098 struct socket *so = kn->kn_fp->f_data;
3100 SOCKBUF_LOCK(&so->so_rcv);
3101 knlist_remove(&so->so_rcv.sb_sel.si_note, kn, 1);
3102 if (knlist_empty(&so->so_rcv.sb_sel.si_note))
3103 so->so_rcv.sb_flags &= ~SB_KNOTE;
3104 SOCKBUF_UNLOCK(&so->so_rcv);
3109 filt_soread(struct knote *kn, long hint)
3113 so = kn->kn_fp->f_data;
3114 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
3116 kn->kn_data = so->so_rcv.sb_cc - so->so_rcv.sb_ctl;
3117 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3118 kn->kn_flags |= EV_EOF;
3119 kn->kn_fflags = so->so_error;
3121 } else if (so->so_error) /* temporary udp error */
3123 else if (kn->kn_sfflags & NOTE_LOWAT)
3124 return (kn->kn_data >= kn->kn_sdata);
3126 return (so->so_rcv.sb_cc >= so->so_rcv.sb_lowat);
3130 filt_sowdetach(struct knote *kn)
3132 struct socket *so = kn->kn_fp->f_data;
3134 SOCKBUF_LOCK(&so->so_snd);
3135 knlist_remove(&so->so_snd.sb_sel.si_note, kn, 1);
3136 if (knlist_empty(&so->so_snd.sb_sel.si_note))
3137 so->so_snd.sb_flags &= ~SB_KNOTE;
3138 SOCKBUF_UNLOCK(&so->so_snd);
3143 filt_sowrite(struct knote *kn, long hint)
3147 so = kn->kn_fp->f_data;
3148 SOCKBUF_LOCK_ASSERT(&so->so_snd);
3149 kn->kn_data = sbspace(&so->so_snd);
3150 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
3151 kn->kn_flags |= EV_EOF;
3152 kn->kn_fflags = so->so_error;
3154 } else if (so->so_error) /* temporary udp error */
3156 else if (((so->so_state & SS_ISCONNECTED) == 0) &&
3157 (so->so_proto->pr_flags & PR_CONNREQUIRED))
3159 else if (kn->kn_sfflags & NOTE_LOWAT)
3160 return (kn->kn_data >= kn->kn_sdata);
3162 return (kn->kn_data >= so->so_snd.sb_lowat);
3167 filt_solisten(struct knote *kn, long hint)
3169 struct socket *so = kn->kn_fp->f_data;
3171 kn->kn_data = so->so_qlen;
3172 return (! TAILQ_EMPTY(&so->so_comp));
3176 socheckuid(struct socket *so, uid_t uid)
3181 if (so->so_cred->cr_uid != uid)
3187 sysctl_somaxconn(SYSCTL_HANDLER_ARGS)
3193 error = sysctl_handle_int(oidp, &val, 0, req);
3194 if (error || !req->newptr )
3197 if (val < 1 || val > USHRT_MAX)
3205 * These functions are used by protocols to notify the socket layer (and its
3206 * consumers) of state changes in the sockets driven by protocol-side events.
3210 * Procedures to manipulate state flags of socket and do appropriate wakeups.
3212 * Normal sequence from the active (originating) side is that
3213 * soisconnecting() is called during processing of connect() call, resulting
3214 * in an eventual call to soisconnected() if/when the connection is
3215 * established. When the connection is torn down soisdisconnecting() is
3216 * called during processing of disconnect() call, and soisdisconnected() is
3217 * called when the connection to the peer is totally severed. The semantics
3218 * of these routines are such that connectionless protocols can call
3219 * soisconnected() and soisdisconnected() only, bypassing the in-progress
3220 * calls when setting up a ``connection'' takes no time.
3222 * From the passive side, a socket is created with two queues of sockets:
3223 * so_incomp for connections in progress and so_comp for connections already
3224 * made and awaiting user acceptance. As a protocol is preparing incoming
3225 * connections, it creates a socket structure queued on so_incomp by calling
3226 * sonewconn(). When the connection is established, soisconnected() is
3227 * called, and transfers the socket structure to so_comp, making it available
3230 * If a socket is closed with sockets on either so_incomp or so_comp, these
3231 * sockets are dropped.
3233 * If higher-level protocols are implemented in the kernel, the wakeups done
3234 * here will sometimes cause software-interrupt process scheduling.
3237 soisconnecting(struct socket *so)
3241 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
3242 so->so_state |= SS_ISCONNECTING;
3247 soisconnected(struct socket *so)
3249 struct socket *head;
3255 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
3256 so->so_state |= SS_ISCONNECTED;
3258 if (head != NULL && (so->so_qstate & SQ_INCOMP)) {
3259 if ((so->so_options & SO_ACCEPTFILTER) == 0) {
3261 TAILQ_REMOVE(&head->so_incomp, so, so_list);
3263 so->so_qstate &= ~SQ_INCOMP;
3264 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
3266 so->so_qstate |= SQ_COMP;
3269 wakeup_one(&head->so_timeo);
3272 soupcall_set(so, SO_RCV,
3273 head->so_accf->so_accept_filter->accf_callback,
3274 head->so_accf->so_accept_filter_arg);
3275 so->so_options &= ~SO_ACCEPTFILTER;
3276 ret = head->so_accf->so_accept_filter->accf_callback(so,
3277 head->so_accf->so_accept_filter_arg, M_DONTWAIT);
3278 if (ret == SU_ISCONNECTED)
3279 soupcall_clear(so, SO_RCV);
3281 if (ret == SU_ISCONNECTED)
3288 wakeup(&so->so_timeo);
3294 soisdisconnecting(struct socket *so)
3298 * Note: This code assumes that SOCK_LOCK(so) and
3299 * SOCKBUF_LOCK(&so->so_rcv) are the same.
3301 SOCKBUF_LOCK(&so->so_rcv);
3302 so->so_state &= ~SS_ISCONNECTING;
3303 so->so_state |= SS_ISDISCONNECTING;
3304 so->so_rcv.sb_state |= SBS_CANTRCVMORE;
3305 sorwakeup_locked(so);
3306 SOCKBUF_LOCK(&so->so_snd);
3307 so->so_snd.sb_state |= SBS_CANTSENDMORE;
3308 sowwakeup_locked(so);
3309 wakeup(&so->so_timeo);
3313 soisdisconnected(struct socket *so)
3317 * Note: This code assumes that SOCK_LOCK(so) and
3318 * SOCKBUF_LOCK(&so->so_rcv) are the same.
3320 SOCKBUF_LOCK(&so->so_rcv);
3321 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
3322 so->so_state |= SS_ISDISCONNECTED;
3323 so->so_rcv.sb_state |= SBS_CANTRCVMORE;
3324 sorwakeup_locked(so);
3325 SOCKBUF_LOCK(&so->so_snd);
3326 so->so_snd.sb_state |= SBS_CANTSENDMORE;
3327 sbdrop_locked(&so->so_snd, so->so_snd.sb_cc);
3328 sowwakeup_locked(so);
3329 wakeup(&so->so_timeo);
3333 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
3336 sodupsockaddr(const struct sockaddr *sa, int mflags)
3338 struct sockaddr *sa2;
3340 sa2 = malloc(sa->sa_len, M_SONAME, mflags);
3342 bcopy(sa, sa2, sa->sa_len);
3347 * Register per-socket buffer upcalls.
3350 soupcall_set(struct socket *so, int which,
3351 int (*func)(struct socket *, void *, int), void *arg)
3363 panic("soupcall_set: bad which");
3365 SOCKBUF_LOCK_ASSERT(sb);
3367 /* XXX: accf_http actually wants to do this on purpose. */
3368 KASSERT(sb->sb_upcall == NULL, ("soupcall_set: overwriting upcall"));
3370 sb->sb_upcall = func;
3371 sb->sb_upcallarg = arg;
3372 sb->sb_flags |= SB_UPCALL;
3376 soupcall_clear(struct socket *so, int which)
3388 panic("soupcall_clear: bad which");
3390 SOCKBUF_LOCK_ASSERT(sb);
3391 KASSERT(sb->sb_upcall != NULL, ("soupcall_clear: no upcall to clear"));
3392 sb->sb_upcall = NULL;
3393 sb->sb_upcallarg = NULL;
3394 sb->sb_flags &= ~SB_UPCALL;
3398 * Create an external-format (``xsocket'') structure using the information in
3399 * the kernel-format socket structure pointed to by so. This is done to
3400 * reduce the spew of irrelevant information over this interface, to isolate
3401 * user code from changes in the kernel structure, and potentially to provide
3402 * information-hiding if we decide that some of this information should be
3403 * hidden from users.
3406 sotoxsocket(struct socket *so, struct xsocket *xso)
3409 xso->xso_len = sizeof *xso;
3411 xso->so_type = so->so_type;
3412 xso->so_options = so->so_options;
3413 xso->so_linger = so->so_linger;
3414 xso->so_state = so->so_state;
3415 xso->so_pcb = so->so_pcb;
3416 xso->xso_protocol = so->so_proto->pr_protocol;
3417 xso->xso_family = so->so_proto->pr_domain->dom_family;
3418 xso->so_qlen = so->so_qlen;
3419 xso->so_incqlen = so->so_incqlen;
3420 xso->so_qlimit = so->so_qlimit;
3421 xso->so_timeo = so->so_timeo;
3422 xso->so_error = so->so_error;
3423 xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
3424 xso->so_oobmark = so->so_oobmark;
3425 sbtoxsockbuf(&so->so_snd, &xso->so_snd);
3426 sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
3427 xso->so_uid = so->so_cred->cr_uid;
3432 * Socket accessor functions to provide external consumers with
3433 * a safe interface to socket state
3438 so_listeners_apply_all(struct socket *so, void (*func)(struct socket *, void *), void *arg)
3441 TAILQ_FOREACH(so, &so->so_comp, so_list)
3446 so_sockbuf_rcv(struct socket *so)
3449 return (&so->so_rcv);
3453 so_sockbuf_snd(struct socket *so)
3456 return (&so->so_snd);
3460 so_state_get(const struct socket *so)
3463 return (so->so_state);
3467 so_state_set(struct socket *so, int val)
3474 so_options_get(const struct socket *so)
3477 return (so->so_options);
3481 so_options_set(struct socket *so, int val)
3484 so->so_options = val;
3488 so_error_get(const struct socket *so)
3491 return (so->so_error);
3495 so_error_set(struct socket *so, int val)
3502 so_linger_get(const struct socket *so)
3505 return (so->so_linger);
3509 so_linger_set(struct socket *so, int val)
3512 so->so_linger = val;
3516 so_protosw_get(const struct socket *so)
3519 return (so->so_proto);
3523 so_protosw_set(struct socket *so, struct protosw *val)
3530 so_sorwakeup(struct socket *so)
3537 so_sowwakeup(struct socket *so)
3544 so_sorwakeup_locked(struct socket *so)
3547 sorwakeup_locked(so);
3551 so_sowwakeup_locked(struct socket *so)
3554 sowwakeup_locked(so);
3558 so_lock(struct socket *so)
3564 so_unlock(struct socket *so)