2 * Copyright (c) 1982, 1986, 1988, 1990, 1993
3 * The Regents of the University of California. All rights reserved.
4 * Copyright (c) 2004 The FreeBSD Foundation
5 * Copyright (c) 2004-2006 Robert N. M. Watson
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 4. Neither the name of the University nor the names of its contributors
16 * may be used to endorse or promote products derived from this software
17 * without specific prior written permission.
19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * @(#)uipc_socket.c 8.3 (Berkeley) 4/15/94
35 * Comments on the socket life cycle:
37 * soalloc() sets of socket layer state for a socket, called only by
38 * socreate() and sonewconn(). Socket layer private.
40 * sodealloc() tears down socket layer state for a socket, called only by
41 * sofree() and sonewconn(). Socket layer private.
43 * pru_attach() associates protocol layer state with an allocated socket;
44 * called only once, may fail, aborting socket allocation. This is called
45 * from socreate() and sonewconn(). Socket layer private.
47 * pru_detach() disassociates protocol layer state from an attached socket,
48 * and will be called exactly once for sockets in which pru_attach() has
49 * been successfully called. If pru_attach() returned an error,
50 * pru_detach() will not be called. Socket layer private.
52 * pru_abort() and pru_close() notify the protocol layer that the last
53 * consumer of a socket is starting to tear down the socket, and that the
54 * protocol should terminate the connection. Historically, pru_abort() also
55 * detached protocol state from the socket state, but this is no longer the
58 * socreate() creates a socket and attaches protocol state. This is a public
59 * interface that may be used by socket layer consumers to create new
62 * sonewconn() creates a socket and attaches protocol state. This is a
63 * public interface that may be used by protocols to create new sockets when
64 * a new connection is received and will be available for accept() on a
67 * soclose() destroys a socket after possibly waiting for it to disconnect.
68 * This is a public interface that socket consumers should use to close and
69 * release a socket when done with it.
71 * soabort() destroys a socket without waiting for it to disconnect (used
72 * only for incoming connections that are already partially or fully
73 * connected). This is used internally by the socket layer when clearing
74 * listen socket queues (due to overflow or close on the listen socket), but
75 * is also a public interface protocols may use to abort connections in
76 * their incomplete listen queues should they no longer be required. Sockets
77 * placed in completed connection listen queues should not be aborted for
78 * reasons described in the comment above the soclose() implementation. This
79 * is not a general purpose close routine, and except in the specific
80 * circumstances described here, should not be used.
82 * sofree() will free a socket and its protocol state if all references on
83 * the socket have been released, and is the public interface to attempt to
84 * free a socket when a reference is removed. This is a socket layer private
87 * NOTE: In addition to socreate() and soclose(), which provide a single
88 * socket reference to the consumer to be managed as required, there are two
89 * calls to explicitly manage socket references, soref(), and sorele().
90 * Currently, these are generally required only when transitioning a socket
91 * from a listen queue to a file descriptor, in order to prevent garbage
92 * collection of the socket at an untimely moment. For a number of reasons,
93 * these interfaces are not preferred, and should be avoided.
96 #include <sys/cdefs.h>
97 __FBSDID("$FreeBSD$");
101 #include "opt_zero.h"
102 #include "opt_compat.h"
104 #include <sys/param.h>
105 #include <sys/systm.h>
106 #include <sys/fcntl.h>
107 #include <sys/limits.h>
108 #include <sys/lock.h>
110 #include <sys/malloc.h>
111 #include <sys/mbuf.h>
112 #include <sys/mutex.h>
113 #include <sys/domain.h>
114 #include <sys/file.h> /* for struct knote */
115 #include <sys/kernel.h>
116 #include <sys/event.h>
117 #include <sys/eventhandler.h>
118 #include <sys/poll.h>
119 #include <sys/proc.h>
120 #include <sys/protosw.h>
121 #include <sys/socket.h>
122 #include <sys/socketvar.h>
123 #include <sys/resourcevar.h>
124 #include <sys/signalvar.h>
125 #include <sys/sysctl.h>
127 #include <sys/jail.h>
132 #include <sys/mount.h>
133 #include <compat/freebsd32/freebsd32.h>
135 extern struct sysentvec ia32_freebsd_sysvec;
138 static int soreceive_rcvoob(struct socket *so, struct uio *uio,
141 static void filt_sordetach(struct knote *kn);
142 static int filt_soread(struct knote *kn, long hint);
143 static void filt_sowdetach(struct knote *kn);
144 static int filt_sowrite(struct knote *kn, long hint);
145 static int filt_solisten(struct knote *kn, long hint);
147 static struct filterops solisten_filtops =
148 { 1, NULL, filt_sordetach, filt_solisten };
149 static struct filterops soread_filtops =
150 { 1, NULL, filt_sordetach, filt_soread };
151 static struct filterops sowrite_filtops =
152 { 1, NULL, filt_sowdetach, filt_sowrite };
154 uma_zone_t socket_zone;
155 so_gen_t so_gencnt; /* generation count for sockets */
159 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
160 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
162 static int somaxconn = SOMAXCONN;
163 static int somaxconn_sysctl(SYSCTL_HANDLER_ARGS);
164 /* XXX: we dont have SYSCTL_USHORT */
165 SYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn, CTLTYPE_UINT | CTLFLAG_RW,
166 0, sizeof(int), somaxconn_sysctl, "I", "Maximum pending socket connection "
168 static int numopensockets;
169 SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD,
170 &numopensockets, 0, "Number of open sockets");
171 #ifdef ZERO_COPY_SOCKETS
172 /* These aren't static because they're used in other files. */
173 int so_zero_copy_send = 1;
174 int so_zero_copy_receive = 1;
175 SYSCTL_NODE(_kern_ipc, OID_AUTO, zero_copy, CTLFLAG_RD, 0,
176 "Zero copy controls");
177 SYSCTL_INT(_kern_ipc_zero_copy, OID_AUTO, receive, CTLFLAG_RW,
178 &so_zero_copy_receive, 0, "Enable zero copy receive");
179 SYSCTL_INT(_kern_ipc_zero_copy, OID_AUTO, send, CTLFLAG_RW,
180 &so_zero_copy_send, 0, "Enable zero copy send");
181 #endif /* ZERO_COPY_SOCKETS */
184 * accept_mtx locks down per-socket fields relating to accept queues. See
185 * socketvar.h for an annotation of the protected fields of struct socket.
187 struct mtx accept_mtx;
188 MTX_SYSINIT(accept_mtx, &accept_mtx, "accept", MTX_DEF);
191 * so_global_mtx protects so_gencnt, numopensockets, and the per-socket
194 static struct mtx so_global_mtx;
195 MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF);
198 * General IPC sysctl name space, used by sockets and a variety of other IPC
201 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC");
204 * Sysctl to get and set the maximum global sockets limit. Notify protocols
205 * of the change so that they can update their dependent limits as required.
208 sysctl_maxsockets(SYSCTL_HANDLER_ARGS)
210 int error, newmaxsockets;
212 newmaxsockets = maxsockets;
213 error = sysctl_handle_int(oidp, &newmaxsockets, sizeof(int), req);
214 if (error == 0 && req->newptr) {
215 if (newmaxsockets > maxsockets) {
216 maxsockets = newmaxsockets;
217 if (maxsockets > ((maxfiles / 4) * 3)) {
218 maxfiles = (maxsockets * 5) / 4;
219 maxfilesperproc = (maxfiles * 9) / 10;
221 EVENTHANDLER_INVOKE(maxsockets_change);
228 SYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets, CTLTYPE_INT|CTLFLAG_RW,
229 &maxsockets, 0, sysctl_maxsockets, "IU",
230 "Maximum number of sockets avaliable");
233 * Initialise maxsockets.
235 static void init_maxsockets(void *ignored)
237 TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
238 maxsockets = imax(maxsockets, imax(maxfiles, nmbclusters));
240 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL);
243 * Socket operation routines. These routines are called by the routines in
244 * sys_socket.c or from a system process, and implement the semantics of
245 * socket operations by switching out to the protocol specific routines.
249 * Get a socket structure from our zone, and initialize it. Note that it
250 * would probably be better to allocate socket and PCB at the same time, but
251 * I'm not convinced that all the protocols can be easily modified to do
254 * soalloc() returns a socket with a ref count of 0.
256 static struct socket *
261 so = uma_zalloc(socket_zone, mflags | M_ZERO);
265 if (mac_init_socket(so, mflags) != 0) {
266 uma_zfree(socket_zone, so);
270 SOCKBUF_LOCK_INIT(&so->so_snd, "so_snd");
271 SOCKBUF_LOCK_INIT(&so->so_rcv, "so_rcv");
272 TAILQ_INIT(&so->so_aiojobq);
273 mtx_lock(&so_global_mtx);
274 so->so_gencnt = ++so_gencnt;
276 mtx_unlock(&so_global_mtx);
281 * Free the storage associated with a socket at the socket layer, tear down
282 * locks, labels, etc. All protocol state is assumed already to have been
283 * torn down (and possibly never set up) by the caller.
286 sodealloc(struct socket *so)
289 KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count));
290 KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL"));
292 mtx_lock(&so_global_mtx);
293 so->so_gencnt = ++so_gencnt;
294 --numopensockets; /* Could be below, but faster here. */
295 mtx_unlock(&so_global_mtx);
296 if (so->so_rcv.sb_hiwat)
297 (void)chgsbsize(so->so_cred->cr_uidinfo,
298 &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY);
299 if (so->so_snd.sb_hiwat)
300 (void)chgsbsize(so->so_cred->cr_uidinfo,
301 &so->so_snd.sb_hiwat, 0, RLIM_INFINITY);
303 /* remove acccept filter if one is present. */
304 if (so->so_accf != NULL)
305 do_setopt_accept_filter(so, NULL);
308 mac_destroy_socket(so);
311 SOCKBUF_LOCK_DESTROY(&so->so_snd);
312 SOCKBUF_LOCK_DESTROY(&so->so_rcv);
313 uma_zfree(socket_zone, so);
317 * socreate returns a socket with a ref count of 1. The socket should be
318 * closed with soclose().
321 socreate(dom, aso, type, proto, cred, td)
334 prp = pffindproto(dom, proto, type);
336 prp = pffindtype(dom, type);
338 if (prp == NULL || prp->pr_usrreqs->pru_attach == NULL ||
339 prp->pr_usrreqs->pru_attach == pru_attach_notsupp)
340 return (EPROTONOSUPPORT);
342 if (jailed(cred) && jail_socket_unixiproute_only &&
343 prp->pr_domain->dom_family != PF_LOCAL &&
344 prp->pr_domain->dom_family != PF_INET &&
345 prp->pr_domain->dom_family != PF_ROUTE) {
346 return (EPROTONOSUPPORT);
349 if (prp->pr_type != type)
351 so = soalloc(M_WAITOK);
355 TAILQ_INIT(&so->so_incomp);
356 TAILQ_INIT(&so->so_comp);
358 so->so_cred = crhold(cred);
361 mac_create_socket(cred, so);
363 knlist_init(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv),
365 knlist_init(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd),
368 error = (*prp->pr_usrreqs->pru_attach)(so, proto, td);
370 KASSERT(so->so_count == 1, ("socreate: so_count %d",
381 static int regression_sonewconn_earlytest = 1;
382 SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW,
383 ®ression_sonewconn_earlytest, 0, "Perform early sonewconn limit test");
387 * When an attempt at a new connection is noted on a socket which accepts
388 * connections, sonewconn is called. If the connection is possible (subject
389 * to space constraints, etc.) then we allocate a new structure, propoerly
390 * linked into the data structure of the original socket, and return this.
391 * Connstatus may be 0, or SO_ISCONFIRMING, or SO_ISCONNECTED.
393 * Note: the ref count on the socket is 0 on return.
396 sonewconn(head, connstatus)
397 register struct socket *head;
400 register struct socket *so;
404 over = (head->so_qlen > 3 * head->so_qlimit / 2);
407 if (regression_sonewconn_earlytest && over)
412 so = soalloc(M_NOWAIT);
415 if ((head->so_options & SO_ACCEPTFILTER) != 0)
418 so->so_type = head->so_type;
419 so->so_options = head->so_options &~ SO_ACCEPTCONN;
420 so->so_linger = head->so_linger;
421 so->so_state = head->so_state | SS_NOFDREF;
422 so->so_proto = head->so_proto;
423 so->so_timeo = head->so_timeo;
424 so->so_cred = crhold(head->so_cred);
427 mac_create_socket_from_socket(head, so);
430 knlist_init(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv),
432 knlist_init(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd),
434 if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat) ||
435 (*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
439 so->so_state |= connstatus;
442 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
443 so->so_qstate |= SQ_COMP;
447 * Keep removing sockets from the head until there's room for
448 * us to insert on the tail. In pre-locking revisions, this
449 * was a simple if(), but as we could be racing with other
450 * threads and soabort() requires dropping locks, we must
451 * loop waiting for the condition to be true.
453 while (head->so_incqlen > head->so_qlimit) {
455 sp = TAILQ_FIRST(&head->so_incomp);
456 TAILQ_REMOVE(&head->so_incomp, sp, so_list);
458 sp->so_qstate &= ~SQ_INCOMP;
464 TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list);
465 so->so_qstate |= SQ_INCOMP;
471 wakeup_one(&head->so_timeo);
479 struct sockaddr *nam;
483 return ((*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td));
487 * solisten() transitions a socket from a non-listening state to a listening
488 * state, but can also be used to update the listen queue depth on an
489 * existing listen socket. The protocol will call back into the sockets
490 * layer using solisten_proto_check() and solisten_proto() to check and set
491 * socket-layer listen state. Call backs are used so that the protocol can
492 * acquire both protocol and socket layer locks in whatever order is required
495 * Protocol implementors are advised to hold the socket lock across the
496 * socket-layer test and set to avoid races at the socket layer.
499 solisten(so, backlog, td)
505 return ((*so->so_proto->pr_usrreqs->pru_listen)(so, backlog, td));
509 solisten_proto_check(so)
513 SOCK_LOCK_ASSERT(so);
515 if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
522 solisten_proto(so, backlog)
527 SOCK_LOCK_ASSERT(so);
529 if (backlog < 0 || backlog > somaxconn)
531 so->so_qlimit = backlog;
532 so->so_options |= SO_ACCEPTCONN;
536 * Attempt to free a socket. This should really be sotryfree().
538 * sofree() will succeed if:
540 * - There are no outstanding file descriptor references or related consumers
543 * - The socket has been closed by user space, if ever open (SS_NOFDREF).
545 * - The protocol does not have an outstanding strong reference on the socket
548 * - The socket is not in a completed connection queue, so a process has been
549 * notified that it is present. If it is removed, the user process may
550 * block in accept() despite select() saying the socket was ready.
552 * Otherwise, it will quietly abort so that a future call to sofree(), when
553 * conditions are right, can succeed.
559 struct protosw *pr = so->so_proto;
562 ACCEPT_LOCK_ASSERT();
563 SOCK_LOCK_ASSERT(so);
565 if ((so->so_state & SS_NOFDREF) == 0 || so->so_count != 0 ||
566 (so->so_state & SS_PROTOREF) || (so->so_qstate & SQ_COMP)) {
574 KASSERT((so->so_qstate & SQ_COMP) != 0 ||
575 (so->so_qstate & SQ_INCOMP) != 0,
576 ("sofree: so_head != NULL, but neither SQ_COMP nor "
578 KASSERT((so->so_qstate & SQ_COMP) == 0 ||
579 (so->so_qstate & SQ_INCOMP) == 0,
580 ("sofree: so->so_qstate is SQ_COMP and also SQ_INCOMP"));
581 TAILQ_REMOVE(&head->so_incomp, so, so_list);
583 so->so_qstate &= ~SQ_INCOMP;
586 KASSERT((so->so_qstate & SQ_COMP) == 0 &&
587 (so->so_qstate & SQ_INCOMP) == 0,
588 ("sofree: so_head == NULL, but still SQ_COMP(%d) or SQ_INCOMP(%d)",
589 so->so_qstate & SQ_COMP, so->so_qstate & SQ_INCOMP));
594 * From this point on, we assume that no other references to this
595 * socket exist anywhere else in the stack. Therefore, no locks need
596 * to be acquired or held.
598 * We used to do a lot of socket buffer and socket locking here, as
599 * well as invoke sorflush() and perform wakeups. The direct call to
600 * dom_dispose() and sbrelease_internal() are an inlining of what was
601 * necessary from sorflush().
603 * Notice that the socket buffer and kqueue state are torn down
604 * before calling pru_detach. This means that protocols shold not
605 * assume they can perform socket wakeups, etc, in their detach
608 KASSERT((so->so_snd.sb_flags & SB_LOCK) == 0, ("sofree: snd sblock"));
609 KASSERT((so->so_rcv.sb_flags & SB_LOCK) == 0, ("sofree: rcv sblock"));
610 sbdestroy(&so->so_snd, so);
611 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
612 (*pr->pr_domain->dom_dispose)(so->so_rcv.sb_mb);
613 sbdestroy(&so->so_rcv, so);
614 if (pr->pr_usrreqs->pru_detach != NULL)
615 (*pr->pr_usrreqs->pru_detach)(so);
616 knlist_destroy(&so->so_rcv.sb_sel.si_note);
617 knlist_destroy(&so->so_snd.sb_sel.si_note);
622 * Close a socket on last file table reference removal. Initiate disconnect
623 * if connected. Free socket when disconnect complete.
625 * This function will sorele() the socket. Note that soclose() may be called
626 * prior to the ref count reaching zero. The actual socket structure will
627 * not be freed until the ref count reaches zero.
635 KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter"));
637 funsetown(&so->so_sigio);
638 if (so->so_options & SO_ACCEPTCONN) {
641 while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
642 TAILQ_REMOVE(&so->so_incomp, sp, so_list);
644 sp->so_qstate &= ~SQ_INCOMP;
650 while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
651 TAILQ_REMOVE(&so->so_comp, sp, so_list);
653 sp->so_qstate &= ~SQ_COMP;
661 if (so->so_state & SS_ISCONNECTED) {
662 if ((so->so_state & SS_ISDISCONNECTING) == 0) {
663 error = sodisconnect(so);
667 if (so->so_options & SO_LINGER) {
668 if ((so->so_state & SS_ISDISCONNECTING) &&
669 (so->so_state & SS_NBIO))
671 while (so->so_state & SS_ISCONNECTED) {
672 error = tsleep(&so->so_timeo,
673 PSOCK | PCATCH, "soclos", so->so_linger * hz);
681 if (so->so_proto->pr_usrreqs->pru_close != NULL)
682 (*so->so_proto->pr_usrreqs->pru_close)(so);
685 KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF"));
686 so->so_state |= SS_NOFDREF;
692 * soabort() is used to abruptly tear down a connection, such as when a
693 * resource limit is reached (listen queue depth exceeded), or if a listen
694 * socket is closed while there are sockets waiting to be accepted.
696 * This interface is tricky, because it is called on an unreferenced socket,
697 * and must be called only by a thread that has actually removed the socket
698 * from the listen queue it was on, or races with other threads are risked.
700 * This interface will call into the protocol code, so must not be called
701 * with any socket locks held. Protocols do call it while holding their own
702 * recursible protocol mutexes, but this is something that should be subject
703 * to review in the future.
711 * In as much as is possible, assert that no references to this
712 * socket are held. This is not quite the same as asserting that the
713 * current thread is responsible for arranging for no references, but
714 * is as close as we can get for now.
716 KASSERT(so->so_count == 0, ("soabort: so_count"));
717 KASSERT((so->so_state & SS_PROTOREF) == 0, ("soabort: SS_PROTOREF"));
718 KASSERT(so->so_state & SS_NOFDREF, ("soabort: !SS_NOFDREF"));
719 KASSERT((so->so_state & SQ_COMP) == 0, ("soabort: SQ_COMP"));
720 KASSERT((so->so_state & SQ_INCOMP) == 0, ("soabort: SQ_INCOMP"));
722 if (so->so_proto->pr_usrreqs->pru_abort != NULL)
723 (*so->so_proto->pr_usrreqs->pru_abort)(so);
732 struct sockaddr **nam;
737 KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF"));
738 so->so_state &= ~SS_NOFDREF;
740 error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam);
745 soconnect(so, nam, td)
747 struct sockaddr *nam;
752 if (so->so_options & SO_ACCEPTCONN)
755 * If protocol is connection-based, can only connect once.
756 * Otherwise, if connected, try to disconnect first. This allows
757 * user to disconnect by connecting to, e.g., a null address.
759 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
760 ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
761 (error = sodisconnect(so)))) {
765 * Prevent accumulated error from previous connection from
769 error = (*so->so_proto->pr_usrreqs->pru_connect)(so, nam, td);
781 return ((*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2));
790 if ((so->so_state & SS_ISCONNECTED) == 0)
792 if (so->so_state & SS_ISDISCONNECTING)
794 error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so);
798 #ifdef ZERO_COPY_SOCKETS
799 struct so_zerocopy_stats{
804 struct so_zerocopy_stats so_zerocp_stats = {0,0,0};
805 #include <netinet/in.h>
806 #include <net/route.h>
807 #include <netinet/in_pcb.h>
809 #include <vm/vm_page.h>
810 #include <vm/vm_object.h>
811 #endif /*ZERO_COPY_SOCKETS*/
814 * sosend_copyin() accepts a uio and prepares an mbuf chain holding part or
815 * all of the data referenced by the uio. If desired, it uses zero-copy.
816 * *space will be updated to reflect data copied in.
818 * NB: If atomic I/O is requested, the caller must already have checked that
819 * space can hold resid bytes.
821 * NB: In the event of an error, the caller may need to free the partial
822 * chain pointed to by *mpp. The contents of both *uio and *space may be
823 * modified even in the case of an error.
826 sosend_copyin(struct uio *uio, struct mbuf **retmp, int atomic, long *space,
829 struct mbuf *m, **mp, *top;
832 #ifdef ZERO_COPY_SOCKETS
839 resid = uio->uio_resid;
842 #ifdef ZERO_COPY_SOCKETS
844 #endif /* ZERO_COPY_SOCKETS */
845 if (resid >= MINCLSIZE) {
846 #ifdef ZERO_COPY_SOCKETS
848 MGETHDR(m, M_TRYWAIT, MT_DATA);
854 m->m_pkthdr.rcvif = NULL;
856 MGET(m, M_TRYWAIT, MT_DATA);
862 if (so_zero_copy_send &&
865 uio->uio_iov->iov_len>=PAGE_SIZE) {
866 so_zerocp_stats.size_ok++;
867 so_zerocp_stats.align_ok++;
868 cow_send = socow_setup(m, uio);
872 MCLGET(m, M_TRYWAIT);
873 if ((m->m_flags & M_EXT) == 0) {
877 len = min(min(MCLBYTES, resid),
881 #else /* ZERO_COPY_SOCKETS */
883 m = m_getcl(M_TRYWAIT, MT_DATA, M_PKTHDR);
885 m->m_pkthdr.rcvif = NULL;
887 m = m_getcl(M_TRYWAIT, MT_DATA, 0);
888 len = min(min(MCLBYTES, resid), *space);
889 #endif /* ZERO_COPY_SOCKETS */
892 m = m_gethdr(M_TRYWAIT, MT_DATA);
894 m->m_pkthdr.rcvif = NULL;
896 len = min(min(MHLEN, resid), *space);
898 * For datagram protocols, leave room
899 * for protocol headers in first mbuf.
901 if (atomic && m && len < MHLEN)
904 m = m_get(M_TRYWAIT, MT_DATA);
905 len = min(min(MLEN, resid), *space);
914 #ifdef ZERO_COPY_SOCKETS
918 #endif /* ZERO_COPY_SOCKETS */
919 error = uiomove(mtod(m, void *), (int)len, uio);
920 resid = uio->uio_resid;
923 top->m_pkthdr.len += len;
929 top->m_flags |= M_EOR;
932 } while (*space > 0 && atomic);
938 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
941 sosend_dgram(so, addr, uio, top, control, flags, td)
943 struct sockaddr *addr;
946 struct mbuf *control;
951 int clen = 0, error, dontroute;
952 int atomic = sosendallatonce(so) || top;
954 KASSERT(so->so_type == SOCK_DGRAM, ("sodgram_send: !SOCK_DGRAM"));
955 KASSERT(so->so_proto->pr_flags & PR_ATOMIC,
956 ("sodgram_send: !PR_ATOMIC"));
959 resid = uio->uio_resid;
961 resid = top->m_pkthdr.len;
963 * In theory resid should be unsigned. However, space must be
964 * signed, as it might be less than 0 if we over-committed, and we
965 * must use a signed comparison of space and resid. On the other
966 * hand, a negative resid causes us to loop sending 0-length
967 * segments to the protocol.
969 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
970 * type sockets since that's an error.
978 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0;
980 td->td_proc->p_stats->p_ru.ru_msgsnd++;
982 clen = control->m_len;
984 SOCKBUF_LOCK(&so->so_snd);
985 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
986 SOCKBUF_UNLOCK(&so->so_snd);
991 error = so->so_error;
993 SOCKBUF_UNLOCK(&so->so_snd);
996 if ((so->so_state & SS_ISCONNECTED) == 0) {
998 * `sendto' and `sendmsg' is allowed on a connection-based
999 * socket if it supports implied connect. Return ENOTCONN if
1000 * not connected and no address is supplied.
1002 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1003 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1004 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1005 !(resid == 0 && clen != 0)) {
1006 SOCKBUF_UNLOCK(&so->so_snd);
1010 } else if (addr == NULL) {
1011 if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1014 error = EDESTADDRREQ;
1015 SOCKBUF_UNLOCK(&so->so_snd);
1021 * Do we need MSG_OOB support in SOCK_DGRAM? Signs here may be a
1022 * problem and need fixing.
1024 space = sbspace(&so->so_snd);
1025 if (flags & MSG_OOB)
1028 if (resid > space) {
1032 SOCKBUF_UNLOCK(&so->so_snd);
1035 if (flags & MSG_EOR)
1036 top->m_flags |= M_EOR;
1038 error = sosend_copyin(uio, &top, atomic, &space, flags);
1041 resid = uio->uio_resid;
1043 KASSERT(resid == 0, ("sosend_dgram: resid != 0"));
1045 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock
1050 so->so_options |= SO_DONTROUTE;
1054 * XXX all the SBS_CANTSENDMORE checks previously done could be out
1055 * of date. We could have recieved a reset packet in an interrupt or
1056 * maybe we slept while doing page faults in uiomove() etc. We could
1057 * probably recheck again inside the locking protection here, but
1058 * there are probably other places that this also happens. We must
1061 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1062 (flags & MSG_OOB) ? PRUS_OOB :
1064 * If the user set MSG_EOF, the protocol understands this flag and
1065 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND.
1067 ((flags & MSG_EOF) &&
1068 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1071 /* If there is more to send set PRUS_MORETOCOME */
1072 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1073 top, addr, control, td);
1076 so->so_options &= ~SO_DONTROUTE;
1085 if (control != NULL)
1091 * Send on a socket. If send must go all at once and message is larger than
1092 * send buffering, then hard error. Lock against other senders. If must go
1093 * all at once and not enough room now, then inform user that this would
1094 * block and do nothing. Otherwise, if nonblocking, send as much as
1095 * possible. The data to be sent is described by "uio" if nonzero, otherwise
1096 * by the mbuf chain "top" (which must be null if uio is not). Data provided
1097 * in mbuf chain must be small enough to send all at once.
1099 * Returns nonzero on error, timeout or signal; callers must check for short
1100 * counts if EINTR/ERESTART are returned. Data and control buffers are freed
1103 #define snderr(errno) { error = (errno); goto release; }
1105 sosend_generic(so, addr, uio, top, control, flags, td)
1107 struct sockaddr *addr;
1110 struct mbuf *control;
1115 int clen = 0, error, dontroute;
1116 int atomic = sosendallatonce(so) || top;
1119 resid = uio->uio_resid;
1121 resid = top->m_pkthdr.len;
1123 * In theory resid should be unsigned. However, space must be
1124 * signed, as it might be less than 0 if we over-committed, and we
1125 * must use a signed comparison of space and resid. On the other
1126 * hand, a negative resid causes us to loop sending 0-length
1127 * segments to the protocol.
1129 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
1130 * type sockets since that's an error.
1132 if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
1138 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
1139 (so->so_proto->pr_flags & PR_ATOMIC);
1141 td->td_proc->p_stats->p_ru.ru_msgsnd++;
1142 if (control != NULL)
1143 clen = control->m_len;
1145 SOCKBUF_LOCK(&so->so_snd);
1147 SOCKBUF_LOCK_ASSERT(&so->so_snd);
1148 error = sblock(&so->so_snd, SBLOCKWAIT(flags));
1152 SOCKBUF_LOCK_ASSERT(&so->so_snd);
1153 if (so->so_snd.sb_state & SBS_CANTSENDMORE)
1156 error = so->so_error;
1160 if ((so->so_state & SS_ISCONNECTED) == 0) {
1162 * `sendto' and `sendmsg' is allowed on a connection-
1163 * based socket if it supports implied connect.
1164 * Return ENOTCONN if not connected and no address is
1167 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1168 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1169 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1170 !(resid == 0 && clen != 0))
1172 } else if (addr == NULL)
1173 snderr(so->so_proto->pr_flags & PR_CONNREQUIRED ?
1174 ENOTCONN : EDESTADDRREQ);
1176 space = sbspace(&so->so_snd);
1177 if (flags & MSG_OOB)
1179 if ((atomic && resid > so->so_snd.sb_hiwat) ||
1180 clen > so->so_snd.sb_hiwat)
1182 if (space < resid + clen &&
1183 (atomic || space < so->so_snd.sb_lowat || space < clen)) {
1184 if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO))
1185 snderr(EWOULDBLOCK);
1186 sbunlock(&so->so_snd);
1187 error = sbwait(&so->so_snd);
1192 SOCKBUF_UNLOCK(&so->so_snd);
1197 if (flags & MSG_EOR)
1198 top->m_flags |= M_EOR;
1200 error = sosend_copyin(uio, &top, atomic,
1203 SOCKBUF_LOCK(&so->so_snd);
1206 resid = uio->uio_resid;
1210 so->so_options |= SO_DONTROUTE;
1214 * XXX all the SBS_CANTSENDMORE checks previously
1215 * done could be out of date. We could have recieved
1216 * a reset packet in an interrupt or maybe we slept
1217 * while doing page faults in uiomove() etc. We
1218 * could probably recheck again inside the locking
1219 * protection here, but there are probably other
1220 * places that this also happens. We must rethink
1223 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1224 (flags & MSG_OOB) ? PRUS_OOB :
1226 * If the user set MSG_EOF, the protocol understands
1227 * this flag and nothing left to send then use
1228 * PRU_SEND_EOF instead of PRU_SEND.
1230 ((flags & MSG_EOF) &&
1231 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1234 /* If there is more to send set PRUS_MORETOCOME. */
1235 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1236 top, addr, control, td);
1239 so->so_options &= ~SO_DONTROUTE;
1246 SOCKBUF_LOCK(&so->so_snd);
1249 } while (resid && space > 0);
1250 SOCKBUF_LOCK(&so->so_snd);
1254 SOCKBUF_LOCK_ASSERT(&so->so_snd);
1255 sbunlock(&so->so_snd);
1257 SOCKBUF_LOCK_ASSERT(&so->so_snd);
1258 SOCKBUF_UNLOCK(&so->so_snd);
1262 if (control != NULL)
1269 sosend(so, addr, uio, top, control, flags, td)
1271 struct sockaddr *addr;
1274 struct mbuf *control;
1279 /* XXXRW: Temporary debugging. */
1280 KASSERT(so->so_proto->pr_usrreqs->pru_sosend != sosend,
1281 ("sosend: protocol calls sosend"));
1283 return (so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio, top,
1284 control, flags, td));
1288 * The part of soreceive() that implements reading non-inline out-of-band
1289 * data from a socket. For more complete comments, see soreceive(), from
1290 * which this code originated.
1292 * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
1293 * unable to return an mbuf chain to the caller.
1296 soreceive_rcvoob(so, uio, flags)
1301 struct protosw *pr = so->so_proto;
1305 KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
1307 m = m_get(M_TRYWAIT, MT_DATA);
1310 error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK);
1314 #ifdef ZERO_COPY_SOCKETS
1315 if (so_zero_copy_receive) {
1318 if ((m->m_flags & M_EXT)
1319 && (m->m_ext.ext_type == EXT_DISPOSABLE))
1324 error = uiomoveco(mtod(m, void *),
1325 min(uio->uio_resid, m->m_len),
1328 #endif /* ZERO_COPY_SOCKETS */
1329 error = uiomove(mtod(m, void *),
1330 (int) min(uio->uio_resid, m->m_len), uio);
1332 } while (uio->uio_resid && error == 0 && m);
1340 * Following replacement or removal of the first mbuf on the first mbuf chain
1341 * of a socket buffer, push necessary state changes back into the socket
1342 * buffer so that other consumers see the values consistently. 'nextrecord'
1343 * is the callers locally stored value of the original value of
1344 * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes.
1345 * NOTE: 'nextrecord' may be NULL.
1347 static __inline void
1348 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord)
1351 SOCKBUF_LOCK_ASSERT(sb);
1353 * First, update for the new value of nextrecord. If necessary, make
1354 * it the first record.
1356 if (sb->sb_mb != NULL)
1357 sb->sb_mb->m_nextpkt = nextrecord;
1359 sb->sb_mb = nextrecord;
1362 * Now update any dependent socket buffer fields to reflect the new
1363 * state. This is an expanded inline of SB_EMPTY_FIXUP(), with the
1364 * addition of a second clause that takes care of the case where
1365 * sb_mb has been updated, but remains the last record.
1367 if (sb->sb_mb == NULL) {
1368 sb->sb_mbtail = NULL;
1369 sb->sb_lastrecord = NULL;
1370 } else if (sb->sb_mb->m_nextpkt == NULL)
1371 sb->sb_lastrecord = sb->sb_mb;
1376 * Implement receive operations on a socket. We depend on the way that
1377 * records are added to the sockbuf by sbappend. In particular, each record
1378 * (mbufs linked through m_next) must begin with an address if the protocol
1379 * so specifies, followed by an optional mbuf or mbufs containing ancillary
1380 * data, and then zero or more mbufs of data. In order to allow parallelism
1381 * between network receive and copying to user space, as well as avoid
1382 * sleeping with a mutex held, we release the socket buffer mutex during the
1383 * user space copy. Although the sockbuf is locked, new data may still be
1384 * appended, and thus we must maintain consistency of the sockbuf during that
1387 * The caller may receive the data as a single mbuf chain by supplying an
1388 * mbuf **mp0 for use in returning the chain. The uio is then used only for
1389 * the count in uio_resid.
1392 soreceive_generic(so, psa, uio, mp0, controlp, flagsp)
1394 struct sockaddr **psa;
1397 struct mbuf **controlp;
1400 struct mbuf *m, **mp;
1401 int flags, len, error, offset;
1402 struct protosw *pr = so->so_proto;
1403 struct mbuf *nextrecord;
1405 int orig_resid = uio->uio_resid;
1410 if (controlp != NULL)
1413 flags = *flagsp &~ MSG_EOR;
1416 if (flags & MSG_OOB)
1417 return (soreceive_rcvoob(so, uio, flags));
1420 if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING)
1422 (*pr->pr_usrreqs->pru_rcvd)(so, 0);
1424 SOCKBUF_LOCK(&so->so_rcv);
1426 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1427 error = sblock(&so->so_rcv, SBLOCKWAIT(flags));
1431 m = so->so_rcv.sb_mb;
1433 * If we have less data than requested, block awaiting more (subject
1434 * to any timeout) if:
1435 * 1. the current count is less than the low water mark, or
1436 * 2. MSG_WAITALL is set, and it is possible to do the entire
1437 * receive operation at once if we block (resid <= hiwat).
1438 * 3. MSG_DONTWAIT is not set
1439 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1440 * we have to do the receive in sections, and thus risk returning a
1441 * short count if a timeout or signal occurs after we start.
1443 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1444 so->so_rcv.sb_cc < uio->uio_resid) &&
1445 (so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
1446 ((flags & MSG_WAITALL) && uio->uio_resid <= so->so_rcv.sb_hiwat)) &&
1447 m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
1448 KASSERT(m != NULL || !so->so_rcv.sb_cc,
1449 ("receive: m == %p so->so_rcv.sb_cc == %u",
1450 m, so->so_rcv.sb_cc));
1454 error = so->so_error;
1455 if ((flags & MSG_PEEK) == 0)
1459 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1460 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1466 for (; m != NULL; m = m->m_next)
1467 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) {
1468 m = so->so_rcv.sb_mb;
1471 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1472 (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1476 if (uio->uio_resid == 0)
1478 if ((so->so_state & SS_NBIO) ||
1479 (flags & (MSG_DONTWAIT|MSG_NBIO))) {
1480 error = EWOULDBLOCK;
1483 SBLASTRECORDCHK(&so->so_rcv);
1484 SBLASTMBUFCHK(&so->so_rcv);
1485 sbunlock(&so->so_rcv);
1486 error = sbwait(&so->so_rcv);
1493 * From this point onward, we maintain 'nextrecord' as a cache of the
1494 * pointer to the next record in the socket buffer. We must keep the
1495 * various socket buffer pointers and local stack versions of the
1496 * pointers in sync, pushing out modifications before dropping the
1497 * socket buffer mutex, and re-reading them when picking it up.
1499 * Otherwise, we will race with the network stack appending new data
1500 * or records onto the socket buffer by using inconsistent/stale
1501 * versions of the field, possibly resulting in socket buffer
1504 * By holding the high-level sblock(), we prevent simultaneous
1505 * readers from pulling off the front of the socket buffer.
1507 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1509 uio->uio_td->td_proc->p_stats->p_ru.ru_msgrcv++;
1510 KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb"));
1511 SBLASTRECORDCHK(&so->so_rcv);
1512 SBLASTMBUFCHK(&so->so_rcv);
1513 nextrecord = m->m_nextpkt;
1514 if (pr->pr_flags & PR_ADDR) {
1515 KASSERT(m->m_type == MT_SONAME,
1516 ("m->m_type == %d", m->m_type));
1519 *psa = sodupsockaddr(mtod(m, struct sockaddr *),
1521 if (flags & MSG_PEEK) {
1524 sbfree(&so->so_rcv, m);
1525 so->so_rcv.sb_mb = m_free(m);
1526 m = so->so_rcv.sb_mb;
1527 sockbuf_pushsync(&so->so_rcv, nextrecord);
1532 * Process one or more MT_CONTROL mbufs present before any data mbufs
1533 * in the first mbuf chain on the socket buffer. If MSG_PEEK, we
1534 * just copy the data; if !MSG_PEEK, we call into the protocol to
1535 * perform externalization (or freeing if controlp == NULL).
1537 if (m != NULL && m->m_type == MT_CONTROL) {
1538 struct mbuf *cm = NULL, *cmn;
1539 struct mbuf **cme = &cm;
1542 if (flags & MSG_PEEK) {
1543 if (controlp != NULL) {
1544 *controlp = m_copy(m, 0, m->m_len);
1545 controlp = &(*controlp)->m_next;
1549 sbfree(&so->so_rcv, m);
1550 so->so_rcv.sb_mb = m->m_next;
1553 cme = &(*cme)->m_next;
1554 m = so->so_rcv.sb_mb;
1556 } while (m != NULL && m->m_type == MT_CONTROL);
1557 if ((flags & MSG_PEEK) == 0)
1558 sockbuf_pushsync(&so->so_rcv, nextrecord);
1559 while (cm != NULL) {
1562 if (pr->pr_domain->dom_externalize != NULL) {
1563 SOCKBUF_UNLOCK(&so->so_rcv);
1564 error = (*pr->pr_domain->dom_externalize)
1566 SOCKBUF_LOCK(&so->so_rcv);
1567 } else if (controlp != NULL)
1571 if (controlp != NULL) {
1573 while (*controlp != NULL)
1574 controlp = &(*controlp)->m_next;
1579 nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1581 nextrecord = so->so_rcv.sb_mb;
1585 if ((flags & MSG_PEEK) == 0) {
1586 KASSERT(m->m_nextpkt == nextrecord,
1587 ("soreceive: post-control, nextrecord !sync"));
1588 if (nextrecord == NULL) {
1589 KASSERT(so->so_rcv.sb_mb == m,
1590 ("soreceive: post-control, sb_mb!=m"));
1591 KASSERT(so->so_rcv.sb_lastrecord == m,
1592 ("soreceive: post-control, lastrecord!=m"));
1596 if (type == MT_OOBDATA)
1599 if ((flags & MSG_PEEK) == 0) {
1600 KASSERT(so->so_rcv.sb_mb == nextrecord,
1601 ("soreceive: sb_mb != nextrecord"));
1602 if (so->so_rcv.sb_mb == NULL) {
1603 KASSERT(so->so_rcv.sb_lastrecord == NULL,
1604 ("soreceive: sb_lastercord != NULL"));
1608 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1609 SBLASTRECORDCHK(&so->so_rcv);
1610 SBLASTMBUFCHK(&so->so_rcv);
1613 * Now continue to read any data mbufs off of the head of the socket
1614 * buffer until the read request is satisfied. Note that 'type' is
1615 * used to store the type of any mbuf reads that have happened so far
1616 * such that soreceive() can stop reading if the type changes, which
1617 * causes soreceive() to return only one of regular data and inline
1618 * out-of-band data in a single socket receive operation.
1622 while (m != NULL && uio->uio_resid > 0 && error == 0) {
1624 * If the type of mbuf has changed since the last mbuf
1625 * examined ('type'), end the receive operation.
1627 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1628 if (m->m_type == MT_OOBDATA) {
1629 if (type != MT_OOBDATA)
1631 } else if (type == MT_OOBDATA)
1634 KASSERT(m->m_type == MT_DATA,
1635 ("m->m_type == %d", m->m_type));
1636 so->so_rcv.sb_state &= ~SBS_RCVATMARK;
1637 len = uio->uio_resid;
1638 if (so->so_oobmark && len > so->so_oobmark - offset)
1639 len = so->so_oobmark - offset;
1640 if (len > m->m_len - moff)
1641 len = m->m_len - moff;
1643 * If mp is set, just pass back the mbufs. Otherwise copy
1644 * them out via the uio, then free. Sockbuf must be
1645 * consistent here (points to current mbuf, it points to next
1646 * record) when we drop priority; we must note any additions
1647 * to the sockbuf when we block interrupts again.
1650 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1651 SBLASTRECORDCHK(&so->so_rcv);
1652 SBLASTMBUFCHK(&so->so_rcv);
1653 SOCKBUF_UNLOCK(&so->so_rcv);
1654 #ifdef ZERO_COPY_SOCKETS
1655 if (so_zero_copy_receive) {
1658 if ((m->m_flags & M_EXT)
1659 && (m->m_ext.ext_type == EXT_DISPOSABLE))
1664 error = uiomoveco(mtod(m, char *) + moff,
1668 #endif /* ZERO_COPY_SOCKETS */
1669 error = uiomove(mtod(m, char *) + moff, (int)len, uio);
1670 SOCKBUF_LOCK(&so->so_rcv);
1674 uio->uio_resid -= len;
1675 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1676 if (len == m->m_len - moff) {
1677 if (m->m_flags & M_EOR)
1679 if (flags & MSG_PEEK) {
1683 nextrecord = m->m_nextpkt;
1684 sbfree(&so->so_rcv, m);
1688 so->so_rcv.sb_mb = m = m->m_next;
1691 so->so_rcv.sb_mb = m_free(m);
1692 m = so->so_rcv.sb_mb;
1694 sockbuf_pushsync(&so->so_rcv, nextrecord);
1695 SBLASTRECORDCHK(&so->so_rcv);
1696 SBLASTMBUFCHK(&so->so_rcv);
1699 if (flags & MSG_PEEK)
1705 if (flags & MSG_DONTWAIT)
1706 copy_flag = M_DONTWAIT;
1708 copy_flag = M_TRYWAIT;
1709 if (copy_flag == M_TRYWAIT)
1710 SOCKBUF_UNLOCK(&so->so_rcv);
1711 *mp = m_copym(m, 0, len, copy_flag);
1712 if (copy_flag == M_TRYWAIT)
1713 SOCKBUF_LOCK(&so->so_rcv);
1716 * m_copym() couldn't
1717 * allocate an mbuf. Adjust
1718 * uio_resid back (it was
1719 * adjusted down by len
1720 * bytes, which we didn't end
1721 * up "copying" over).
1723 uio->uio_resid += len;
1729 so->so_rcv.sb_cc -= len;
1732 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1733 if (so->so_oobmark) {
1734 if ((flags & MSG_PEEK) == 0) {
1735 so->so_oobmark -= len;
1736 if (so->so_oobmark == 0) {
1737 so->so_rcv.sb_state |= SBS_RCVATMARK;
1742 if (offset == so->so_oobmark)
1746 if (flags & MSG_EOR)
1749 * If the MSG_WAITALL flag is set (for non-atomic socket), we
1750 * must not quit until "uio->uio_resid == 0" or an error
1751 * termination. If a signal/timeout occurs, return with a
1752 * short count but without error. Keep sockbuf locked
1753 * against other readers.
1755 while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1756 !sosendallatonce(so) && nextrecord == NULL) {
1757 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1758 if (so->so_error || so->so_rcv.sb_state & SBS_CANTRCVMORE)
1761 * Notify the protocol that some data has been
1762 * drained before blocking.
1764 if (pr->pr_flags & PR_WANTRCVD) {
1765 SOCKBUF_UNLOCK(&so->so_rcv);
1766 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
1767 SOCKBUF_LOCK(&so->so_rcv);
1769 SBLASTRECORDCHK(&so->so_rcv);
1770 SBLASTMBUFCHK(&so->so_rcv);
1771 error = sbwait(&so->so_rcv);
1774 m = so->so_rcv.sb_mb;
1776 nextrecord = m->m_nextpkt;
1780 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1781 if (m != NULL && pr->pr_flags & PR_ATOMIC) {
1783 if ((flags & MSG_PEEK) == 0)
1784 (void) sbdroprecord_locked(&so->so_rcv);
1786 if ((flags & MSG_PEEK) == 0) {
1789 * First part is an inline SB_EMPTY_FIXUP(). Second
1790 * part makes sure sb_lastrecord is up-to-date if
1791 * there is still data in the socket buffer.
1793 so->so_rcv.sb_mb = nextrecord;
1794 if (so->so_rcv.sb_mb == NULL) {
1795 so->so_rcv.sb_mbtail = NULL;
1796 so->so_rcv.sb_lastrecord = NULL;
1797 } else if (nextrecord->m_nextpkt == NULL)
1798 so->so_rcv.sb_lastrecord = nextrecord;
1800 SBLASTRECORDCHK(&so->so_rcv);
1801 SBLASTMBUFCHK(&so->so_rcv);
1803 * If soreceive() is being done from the socket callback,
1804 * then don't need to generate ACK to peer to update window,
1805 * since ACK will be generated on return to TCP.
1807 if (!(flags & MSG_SOCALLBCK) &&
1808 (pr->pr_flags & PR_WANTRCVD)) {
1809 SOCKBUF_UNLOCK(&so->so_rcv);
1810 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
1811 SOCKBUF_LOCK(&so->so_rcv);
1814 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1815 if (orig_resid == uio->uio_resid && orig_resid &&
1816 (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
1817 sbunlock(&so->so_rcv);
1824 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1825 sbunlock(&so->so_rcv);
1827 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1828 SOCKBUF_UNLOCK(&so->so_rcv);
1833 soreceive(so, psa, uio, mp0, controlp, flagsp)
1835 struct sockaddr **psa;
1838 struct mbuf **controlp;
1842 /* XXXRW: Temporary debugging. */
1843 KASSERT(so->so_proto->pr_usrreqs->pru_soreceive != soreceive,
1844 ("soreceive: protocol calls soreceive"));
1846 return (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio, mp0,
1855 struct protosw *pr = so->so_proto;
1857 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1863 return ((*pr->pr_usrreqs->pru_shutdown)(so));
1871 struct sockbuf *sb = &so->so_rcv;
1872 struct protosw *pr = so->so_proto;
1876 * XXXRW: This is quite ugly. Previously, this code made a copy of
1877 * the socket buffer, then zero'd the original to clear the buffer
1878 * fields. However, with mutexes in the socket buffer, this causes
1879 * problems. We only clear the zeroable bits of the original;
1880 * however, we have to initialize and destroy the mutex in the copy
1881 * so that dom_dispose() and sbrelease() can lock t as needed.
1884 sb->sb_flags |= SB_NOINTR;
1885 (void) sblock(sb, M_WAITOK);
1887 * socantrcvmore_locked() drops the socket buffer mutex so that it
1888 * can safely perform wakeups. Re-acquire the mutex before
1891 socantrcvmore_locked(so);
1895 * Invalidate/clear most of the sockbuf structure, but leave selinfo
1896 * and mutex data unchanged.
1898 bzero(&asb, offsetof(struct sockbuf, sb_startzero));
1899 bcopy(&sb->sb_startzero, &asb.sb_startzero,
1900 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
1901 bzero(&sb->sb_startzero,
1902 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
1905 SOCKBUF_LOCK_INIT(&asb, "so_rcv");
1906 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
1907 (*pr->pr_domain->dom_dispose)(asb.sb_mb);
1908 sbrelease(&asb, so);
1909 SOCKBUF_LOCK_DESTROY(&asb);
1913 * Perhaps this routine, and sooptcopyout(), below, ought to come in an
1914 * additional variant to handle the case where the option value needs to be
1915 * some kind of integer, but not a specific size. In addition to their use
1916 * here, these functions are also called by the protocol-level pr_ctloutput()
1920 sooptcopyin(sopt, buf, len, minlen)
1921 struct sockopt *sopt;
1929 * If the user gives us more than we wanted, we ignore it, but if we
1930 * don't get the minimum length the caller wants, we return EINVAL.
1931 * On success, sopt->sopt_valsize is set to however much we actually
1934 if ((valsize = sopt->sopt_valsize) < minlen)
1937 sopt->sopt_valsize = valsize = len;
1939 if (sopt->sopt_td != NULL)
1940 return (copyin(sopt->sopt_val, buf, valsize));
1942 bcopy(sopt->sopt_val, buf, valsize);
1947 * Kernel version of setsockopt(2).
1949 * XXX: optlen is size_t, not socklen_t
1952 so_setsockopt(struct socket *so, int level, int optname, void *optval,
1955 struct sockopt sopt;
1957 sopt.sopt_level = level;
1958 sopt.sopt_name = optname;
1959 sopt.sopt_dir = SOPT_SET;
1960 sopt.sopt_val = optval;
1961 sopt.sopt_valsize = optlen;
1962 sopt.sopt_td = NULL;
1963 return (sosetopt(so, &sopt));
1969 struct sockopt *sopt;
1980 if (sopt->sopt_level != SOL_SOCKET) {
1981 if (so->so_proto && so->so_proto->pr_ctloutput)
1982 return ((*so->so_proto->pr_ctloutput)
1984 error = ENOPROTOOPT;
1986 switch (sopt->sopt_name) {
1988 case SO_ACCEPTFILTER:
1989 error = do_setopt_accept_filter(so, sopt);
1995 error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
2000 so->so_linger = l.l_linger;
2002 so->so_options |= SO_LINGER;
2004 so->so_options &= ~SO_LINGER;
2011 case SO_USELOOPBACK:
2019 error = sooptcopyin(sopt, &optval, sizeof optval,
2025 so->so_options |= sopt->sopt_name;
2027 so->so_options &= ~sopt->sopt_name;
2035 error = sooptcopyin(sopt, &optval, sizeof optval,
2041 * Values < 1 make no sense for any of these options,
2049 switch (sopt->sopt_name) {
2052 if (sbreserve(sopt->sopt_name == SO_SNDBUF ?
2053 &so->so_snd : &so->so_rcv, (u_long)optval,
2054 so, curthread) == 0) {
2061 * Make sure the low-water is never greater than the
2065 SOCKBUF_LOCK(&so->so_snd);
2066 so->so_snd.sb_lowat =
2067 (optval > so->so_snd.sb_hiwat) ?
2068 so->so_snd.sb_hiwat : optval;
2069 SOCKBUF_UNLOCK(&so->so_snd);
2072 SOCKBUF_LOCK(&so->so_rcv);
2073 so->so_rcv.sb_lowat =
2074 (optval > so->so_rcv.sb_hiwat) ?
2075 so->so_rcv.sb_hiwat : optval;
2076 SOCKBUF_UNLOCK(&so->so_rcv);
2084 if (curthread->td_proc->p_sysent == &ia32_freebsd_sysvec) {
2085 struct timeval32 tv32;
2087 error = sooptcopyin(sopt, &tv32, sizeof tv32,
2089 CP(tv32, tv, tv_sec);
2090 CP(tv32, tv, tv_usec);
2093 error = sooptcopyin(sopt, &tv, sizeof tv,
2098 /* assert(hz > 0); */
2099 if (tv.tv_sec < 0 || tv.tv_sec > INT_MAX / hz ||
2100 tv.tv_usec < 0 || tv.tv_usec >= 1000000) {
2104 /* assert(tick > 0); */
2105 /* assert(ULONG_MAX - INT_MAX >= 1000000); */
2106 val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / tick;
2107 if (val > INT_MAX) {
2111 if (val == 0 && tv.tv_usec != 0)
2114 switch (sopt->sopt_name) {
2116 so->so_snd.sb_timeo = val;
2119 so->so_rcv.sb_timeo = val;
2126 error = sooptcopyin(sopt, &extmac, sizeof extmac,
2130 error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
2138 error = ENOPROTOOPT;
2141 if (error == 0 && so->so_proto != NULL &&
2142 so->so_proto->pr_ctloutput != NULL) {
2143 (void) ((*so->so_proto->pr_ctloutput)
2152 * Helper routine for getsockopt.
2155 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
2163 * Documented get behavior is that we always return a value, possibly
2164 * truncated to fit in the user's buffer. Traditional behavior is
2165 * that we always tell the user precisely how much we copied, rather
2166 * than something useful like the total amount we had available for
2167 * her. Note that this interface is not idempotent; the entire
2168 * answer must generated ahead of time.
2170 valsize = min(len, sopt->sopt_valsize);
2171 sopt->sopt_valsize = valsize;
2172 if (sopt->sopt_val != NULL) {
2173 if (sopt->sopt_td != NULL)
2174 error = copyout(buf, sopt->sopt_val, valsize);
2176 bcopy(buf, sopt->sopt_val, valsize);
2184 struct sockopt *sopt;
2194 if (sopt->sopt_level != SOL_SOCKET) {
2195 if (so->so_proto && so->so_proto->pr_ctloutput) {
2196 return ((*so->so_proto->pr_ctloutput)
2199 return (ENOPROTOOPT);
2201 switch (sopt->sopt_name) {
2203 case SO_ACCEPTFILTER:
2204 error = do_getopt_accept_filter(so, sopt);
2209 l.l_onoff = so->so_options & SO_LINGER;
2210 l.l_linger = so->so_linger;
2212 error = sooptcopyout(sopt, &l, sizeof l);
2215 case SO_USELOOPBACK:
2227 optval = so->so_options & sopt->sopt_name;
2229 error = sooptcopyout(sopt, &optval, sizeof optval);
2233 optval = so->so_type;
2238 optval = so->so_error;
2244 optval = so->so_snd.sb_hiwat;
2248 optval = so->so_rcv.sb_hiwat;
2252 optval = so->so_snd.sb_lowat;
2256 optval = so->so_rcv.sb_lowat;
2261 optval = (sopt->sopt_name == SO_SNDTIMEO ?
2262 so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
2264 tv.tv_sec = optval / hz;
2265 tv.tv_usec = (optval % hz) * tick;
2267 if (curthread->td_proc->p_sysent == &ia32_freebsd_sysvec) {
2268 struct timeval32 tv32;
2270 CP(tv, tv32, tv_sec);
2271 CP(tv, tv32, tv_usec);
2272 error = sooptcopyout(sopt, &tv32, sizeof tv32);
2275 error = sooptcopyout(sopt, &tv, sizeof tv);
2280 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
2284 error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
2288 error = sooptcopyout(sopt, &extmac, sizeof extmac);
2296 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
2300 error = mac_getsockopt_peerlabel(
2301 sopt->sopt_td->td_ucred, so, &extmac);
2304 error = sooptcopyout(sopt, &extmac, sizeof extmac);
2310 case SO_LISTENQLIMIT:
2311 optval = so->so_qlimit;
2315 optval = so->so_qlen;
2318 case SO_LISTENINCQLEN:
2319 optval = so->so_incqlen;
2323 error = ENOPROTOOPT;
2330 /* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */
2332 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
2334 struct mbuf *m, *m_prev;
2335 int sopt_size = sopt->sopt_valsize;
2337 MGET(m, sopt->sopt_td ? M_TRYWAIT : M_DONTWAIT, MT_DATA);
2340 if (sopt_size > MLEN) {
2341 MCLGET(m, sopt->sopt_td ? M_TRYWAIT : M_DONTWAIT);
2342 if ((m->m_flags & M_EXT) == 0) {
2346 m->m_len = min(MCLBYTES, sopt_size);
2348 m->m_len = min(MLEN, sopt_size);
2350 sopt_size -= m->m_len;
2355 MGET(m, sopt->sopt_td ? M_TRYWAIT : M_DONTWAIT, MT_DATA);
2360 if (sopt_size > MLEN) {
2361 MCLGET(m, sopt->sopt_td != NULL ? M_TRYWAIT :
2363 if ((m->m_flags & M_EXT) == 0) {
2368 m->m_len = min(MCLBYTES, sopt_size);
2370 m->m_len = min(MLEN, sopt_size);
2372 sopt_size -= m->m_len;
2379 /* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */
2381 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
2383 struct mbuf *m0 = m;
2385 if (sopt->sopt_val == NULL)
2387 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
2388 if (sopt->sopt_td != NULL) {
2391 error = copyin(sopt->sopt_val, mtod(m, char *),
2398 bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
2399 sopt->sopt_valsize -= m->m_len;
2400 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
2403 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
2404 panic("ip6_sooptmcopyin");
2408 /* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */
2410 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
2412 struct mbuf *m0 = m;
2415 if (sopt->sopt_val == NULL)
2417 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
2418 if (sopt->sopt_td != NULL) {
2421 error = copyout(mtod(m, char *), sopt->sopt_val,
2428 bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
2429 sopt->sopt_valsize -= m->m_len;
2430 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
2431 valsize += m->m_len;
2435 /* enough soopt buffer should be given from user-land */
2439 sopt->sopt_valsize = valsize;
2444 * sohasoutofband(): protocol notifies socket layer of the arrival of new
2445 * out-of-band data, which will then notify socket consumers.
2451 if (so->so_sigio != NULL)
2452 pgsigio(&so->so_sigio, SIGURG, 0);
2453 selwakeuppri(&so->so_rcv.sb_sel, PSOCK);
2457 sopoll(struct socket *so, int events, struct ucred *active_cred,
2461 /* XXXRW: Temporary debugging. */
2462 KASSERT(so->so_proto->pr_usrreqs->pru_sopoll != sopoll,
2463 ("sopoll: protocol calls sopoll"));
2465 return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred,
2470 sopoll_generic(struct socket *so, int events, struct ucred *active_cred,
2475 SOCKBUF_LOCK(&so->so_snd);
2476 SOCKBUF_LOCK(&so->so_rcv);
2477 if (events & (POLLIN | POLLRDNORM))
2479 revents |= events & (POLLIN | POLLRDNORM);
2481 if (events & POLLINIGNEOF)
2482 if (so->so_rcv.sb_cc >= so->so_rcv.sb_lowat ||
2483 !TAILQ_EMPTY(&so->so_comp) || so->so_error)
2484 revents |= POLLINIGNEOF;
2486 if (events & (POLLOUT | POLLWRNORM))
2487 if (sowriteable(so))
2488 revents |= events & (POLLOUT | POLLWRNORM);
2490 if (events & (POLLPRI | POLLRDBAND))
2491 if (so->so_oobmark || (so->so_rcv.sb_state & SBS_RCVATMARK))
2492 revents |= events & (POLLPRI | POLLRDBAND);
2496 (POLLIN | POLLINIGNEOF | POLLPRI | POLLRDNORM |
2498 selrecord(td, &so->so_rcv.sb_sel);
2499 so->so_rcv.sb_flags |= SB_SEL;
2502 if (events & (POLLOUT | POLLWRNORM)) {
2503 selrecord(td, &so->so_snd.sb_sel);
2504 so->so_snd.sb_flags |= SB_SEL;
2508 SOCKBUF_UNLOCK(&so->so_rcv);
2509 SOCKBUF_UNLOCK(&so->so_snd);
2514 soo_kqfilter(struct file *fp, struct knote *kn)
2516 struct socket *so = kn->kn_fp->f_data;
2519 switch (kn->kn_filter) {
2521 if (so->so_options & SO_ACCEPTCONN)
2522 kn->kn_fop = &solisten_filtops;
2524 kn->kn_fop = &soread_filtops;
2528 kn->kn_fop = &sowrite_filtops;
2536 knlist_add(&sb->sb_sel.si_note, kn, 1);
2537 sb->sb_flags |= SB_KNOTE;
2543 filt_sordetach(struct knote *kn)
2545 struct socket *so = kn->kn_fp->f_data;
2547 SOCKBUF_LOCK(&so->so_rcv);
2548 knlist_remove(&so->so_rcv.sb_sel.si_note, kn, 1);
2549 if (knlist_empty(&so->so_rcv.sb_sel.si_note))
2550 so->so_rcv.sb_flags &= ~SB_KNOTE;
2551 SOCKBUF_UNLOCK(&so->so_rcv);
2556 filt_soread(struct knote *kn, long hint)
2560 so = kn->kn_fp->f_data;
2561 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2563 kn->kn_data = so->so_rcv.sb_cc - so->so_rcv.sb_ctl;
2564 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2565 kn->kn_flags |= EV_EOF;
2566 kn->kn_fflags = so->so_error;
2568 } else if (so->so_error) /* temporary udp error */
2570 else if (kn->kn_sfflags & NOTE_LOWAT)
2571 return (kn->kn_data >= kn->kn_sdata);
2573 return (so->so_rcv.sb_cc >= so->so_rcv.sb_lowat);
2577 filt_sowdetach(struct knote *kn)
2579 struct socket *so = kn->kn_fp->f_data;
2581 SOCKBUF_LOCK(&so->so_snd);
2582 knlist_remove(&so->so_snd.sb_sel.si_note, kn, 1);
2583 if (knlist_empty(&so->so_snd.sb_sel.si_note))
2584 so->so_snd.sb_flags &= ~SB_KNOTE;
2585 SOCKBUF_UNLOCK(&so->so_snd);
2590 filt_sowrite(struct knote *kn, long hint)
2594 so = kn->kn_fp->f_data;
2595 SOCKBUF_LOCK_ASSERT(&so->so_snd);
2596 kn->kn_data = sbspace(&so->so_snd);
2597 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
2598 kn->kn_flags |= EV_EOF;
2599 kn->kn_fflags = so->so_error;
2601 } else if (so->so_error) /* temporary udp error */
2603 else if (((so->so_state & SS_ISCONNECTED) == 0) &&
2604 (so->so_proto->pr_flags & PR_CONNREQUIRED))
2606 else if (kn->kn_sfflags & NOTE_LOWAT)
2607 return (kn->kn_data >= kn->kn_sdata);
2609 return (kn->kn_data >= so->so_snd.sb_lowat);
2614 filt_solisten(struct knote *kn, long hint)
2616 struct socket *so = kn->kn_fp->f_data;
2618 kn->kn_data = so->so_qlen;
2619 return (! TAILQ_EMPTY(&so->so_comp));
2623 socheckuid(struct socket *so, uid_t uid)
2628 if (so->so_cred->cr_uid != uid)
2634 somaxconn_sysctl(SYSCTL_HANDLER_ARGS)
2640 error = sysctl_handle_int(oidp, &val, sizeof(int), req);
2641 if (error || !req->newptr )
2644 if (val < 1 || val > USHRT_MAX)