2 * Copyright (c) 1982, 1986, 1988, 1990, 1993
3 * The Regents of the University of California.
4 * Copyright (c) 2004 The FreeBSD Foundation
5 * Copyright (c) 2004-2008 Robert N. M. Watson
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 * 4. Neither the name of the University nor the names of its contributors
17 * may be used to endorse or promote products derived from this software
18 * without specific prior written permission.
20 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * @(#)uipc_socket.c 8.3 (Berkeley) 4/15/94
36 * Comments on the socket life cycle:
38 * soalloc() sets of socket layer state for a socket, called only by
39 * socreate() and sonewconn(). Socket layer private.
41 * sodealloc() tears down socket layer state for a socket, called only by
42 * sofree() and sonewconn(). Socket layer private.
44 * pru_attach() associates protocol layer state with an allocated socket;
45 * called only once, may fail, aborting socket allocation. This is called
46 * from socreate() and sonewconn(). Socket layer private.
48 * pru_detach() disassociates protocol layer state from an attached socket,
49 * and will be called exactly once for sockets in which pru_attach() has
50 * been successfully called. If pru_attach() returned an error,
51 * pru_detach() will not be called. Socket layer private.
53 * pru_abort() and pru_close() notify the protocol layer that the last
54 * consumer of a socket is starting to tear down the socket, and that the
55 * protocol should terminate the connection. Historically, pru_abort() also
56 * detached protocol state from the socket state, but this is no longer the
59 * socreate() creates a socket and attaches protocol state. This is a public
60 * interface that may be used by socket layer consumers to create new
63 * sonewconn() creates a socket and attaches protocol state. This is a
64 * public interface that may be used by protocols to create new sockets when
65 * a new connection is received and will be available for accept() on a
68 * soclose() destroys a socket after possibly waiting for it to disconnect.
69 * This is a public interface that socket consumers should use to close and
70 * release a socket when done with it.
72 * soabort() destroys a socket without waiting for it to disconnect (used
73 * only for incoming connections that are already partially or fully
74 * connected). This is used internally by the socket layer when clearing
75 * listen socket queues (due to overflow or close on the listen socket), but
76 * is also a public interface protocols may use to abort connections in
77 * their incomplete listen queues should they no longer be required. Sockets
78 * placed in completed connection listen queues should not be aborted for
79 * reasons described in the comment above the soclose() implementation. This
80 * is not a general purpose close routine, and except in the specific
81 * circumstances described here, should not be used.
83 * sofree() will free a socket and its protocol state if all references on
84 * the socket have been released, and is the public interface to attempt to
85 * free a socket when a reference is removed. This is a socket layer private
88 * NOTE: In addition to socreate() and soclose(), which provide a single
89 * socket reference to the consumer to be managed as required, there are two
90 * calls to explicitly manage socket references, soref(), and sorele().
91 * Currently, these are generally required only when transitioning a socket
92 * from a listen queue to a file descriptor, in order to prevent garbage
93 * collection of the socket at an untimely moment. For a number of reasons,
94 * these interfaces are not preferred, and should be avoided.
96 * NOTE: With regard to VNETs the general rule is that callers do not set
97 * curvnet. Exceptions to this rule include soabort(), sodisconnect(),
98 * sofree() (and with that sorele(), sotryfree()), as well as sonewconn()
99 * and sorflush(), which are usually called from a pre-set VNET context.
100 * sopoll() currently does not need a VNET context to be set.
103 #include <sys/cdefs.h>
104 __FBSDID("$FreeBSD$");
106 #include "opt_inet.h"
107 #include "opt_inet6.h"
108 #include "opt_zero.h"
109 #include "opt_compat.h"
111 #include <sys/param.h>
112 #include <sys/systm.h>
113 #include <sys/fcntl.h>
114 #include <sys/limits.h>
115 #include <sys/lock.h>
117 #include <sys/malloc.h>
118 #include <sys/mbuf.h>
119 #include <sys/mutex.h>
120 #include <sys/domain.h>
121 #include <sys/file.h> /* for struct knote */
122 #include <sys/kernel.h>
123 #include <sys/event.h>
124 #include <sys/eventhandler.h>
125 #include <sys/poll.h>
126 #include <sys/proc.h>
127 #include <sys/protosw.h>
128 #include <sys/socket.h>
129 #include <sys/socketvar.h>
130 #include <sys/resourcevar.h>
131 #include <net/route.h>
132 #include <sys/signalvar.h>
133 #include <sys/stat.h>
135 #include <sys/sysctl.h>
137 #include <sys/jail.h>
139 #include <net/vnet.h>
141 #include <security/mac/mac_framework.h>
145 #ifdef COMPAT_FREEBSD32
146 #include <sys/mount.h>
147 #include <sys/sysent.h>
148 #include <compat/freebsd32/freebsd32.h>
151 static int soreceive_rcvoob(struct socket *so, struct uio *uio,
154 static void filt_sordetach(struct knote *kn);
155 static int filt_soread(struct knote *kn, long hint);
156 static void filt_sowdetach(struct knote *kn);
157 static int filt_sowrite(struct knote *kn, long hint);
158 static int filt_solisten(struct knote *kn, long hint);
160 static struct filterops solisten_filtops = {
162 .f_detach = filt_sordetach,
163 .f_event = filt_solisten,
165 static struct filterops soread_filtops = {
167 .f_detach = filt_sordetach,
168 .f_event = filt_soread,
170 static struct filterops sowrite_filtops = {
172 .f_detach = filt_sowdetach,
173 .f_event = filt_sowrite,
176 uma_zone_t socket_zone;
177 so_gen_t so_gencnt; /* generation count for sockets */
181 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
182 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
184 #define VNET_SO_ASSERT(so) \
185 VNET_ASSERT(curvnet != NULL, \
186 ("%s:%d curvnet is NULL, so=%p", __func__, __LINE__, (so)));
188 static int somaxconn = SOMAXCONN;
189 static int sysctl_somaxconn(SYSCTL_HANDLER_ARGS);
190 /* XXX: we dont have SYSCTL_USHORT */
191 SYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn, CTLTYPE_UINT | CTLFLAG_RW,
192 0, sizeof(int), sysctl_somaxconn, "I", "Maximum pending socket connection "
194 static int numopensockets;
195 SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD,
196 &numopensockets, 0, "Number of open sockets");
197 #ifdef ZERO_COPY_SOCKETS
198 /* These aren't static because they're used in other files. */
199 int so_zero_copy_send = 1;
200 int so_zero_copy_receive = 1;
201 SYSCTL_NODE(_kern_ipc, OID_AUTO, zero_copy, CTLFLAG_RD, 0,
202 "Zero copy controls");
203 SYSCTL_INT(_kern_ipc_zero_copy, OID_AUTO, receive, CTLFLAG_RW,
204 &so_zero_copy_receive, 0, "Enable zero copy receive");
205 SYSCTL_INT(_kern_ipc_zero_copy, OID_AUTO, send, CTLFLAG_RW,
206 &so_zero_copy_send, 0, "Enable zero copy send");
207 #endif /* ZERO_COPY_SOCKETS */
210 * accept_mtx locks down per-socket fields relating to accept queues. See
211 * socketvar.h for an annotation of the protected fields of struct socket.
213 struct mtx accept_mtx;
214 MTX_SYSINIT(accept_mtx, &accept_mtx, "accept", MTX_DEF);
217 * so_global_mtx protects so_gencnt, numopensockets, and the per-socket
220 static struct mtx so_global_mtx;
221 MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF);
224 * General IPC sysctl name space, used by sockets and a variety of other IPC
227 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC");
230 * Sysctl to get and set the maximum global sockets limit. Notify protocols
231 * of the change so that they can update their dependent limits as required.
234 sysctl_maxsockets(SYSCTL_HANDLER_ARGS)
236 int error, newmaxsockets;
238 newmaxsockets = maxsockets;
239 error = sysctl_handle_int(oidp, &newmaxsockets, 0, req);
240 if (error == 0 && req->newptr) {
241 if (newmaxsockets > maxsockets) {
242 maxsockets = newmaxsockets;
243 if (maxsockets > ((maxfiles / 4) * 3)) {
244 maxfiles = (maxsockets * 5) / 4;
245 maxfilesperproc = (maxfiles * 9) / 10;
247 EVENTHANDLER_INVOKE(maxsockets_change);
254 SYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets, CTLTYPE_INT|CTLFLAG_RW,
255 &maxsockets, 0, sysctl_maxsockets, "IU",
256 "Maximum number of sockets avaliable");
259 * Initialise maxsockets. This SYSINIT must be run after
263 init_maxsockets(void *ignored)
266 TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
267 maxsockets = imax(maxsockets, imax(maxfiles, nmbclusters));
269 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL);
272 * Socket operation routines. These routines are called by the routines in
273 * sys_socket.c or from a system process, and implement the semantics of
274 * socket operations by switching out to the protocol specific routines.
278 * Get a socket structure from our zone, and initialize it. Note that it
279 * would probably be better to allocate socket and PCB at the same time, but
280 * I'm not convinced that all the protocols can be easily modified to do
283 * soalloc() returns a socket with a ref count of 0.
285 static struct socket *
286 soalloc(struct vnet *vnet)
290 so = uma_zalloc(socket_zone, M_NOWAIT | M_ZERO);
294 if (mac_socket_init(so, M_NOWAIT) != 0) {
295 uma_zfree(socket_zone, so);
299 SOCKBUF_LOCK_INIT(&so->so_snd, "so_snd");
300 SOCKBUF_LOCK_INIT(&so->so_rcv, "so_rcv");
301 sx_init(&so->so_snd.sb_sx, "so_snd_sx");
302 sx_init(&so->so_rcv.sb_sx, "so_rcv_sx");
303 TAILQ_INIT(&so->so_aiojobq);
304 mtx_lock(&so_global_mtx);
305 so->so_gencnt = ++so_gencnt;
308 VNET_ASSERT(vnet != NULL, ("%s:%d vnet is NULL, so=%p",
309 __func__, __LINE__, so));
310 vnet->vnet_sockcnt++;
313 mtx_unlock(&so_global_mtx);
318 * Free the storage associated with a socket at the socket layer, tear down
319 * locks, labels, etc. All protocol state is assumed already to have been
320 * torn down (and possibly never set up) by the caller.
323 sodealloc(struct socket *so)
326 KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count));
327 KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL"));
329 mtx_lock(&so_global_mtx);
330 so->so_gencnt = ++so_gencnt;
331 --numopensockets; /* Could be below, but faster here. */
333 VNET_ASSERT(so->so_vnet != NULL, ("%s:%d so_vnet is NULL, so=%p",
334 __func__, __LINE__, so));
335 so->so_vnet->vnet_sockcnt--;
337 mtx_unlock(&so_global_mtx);
338 if (so->so_rcv.sb_hiwat)
339 (void)chgsbsize(so->so_cred->cr_uidinfo,
340 &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY);
341 if (so->so_snd.sb_hiwat)
342 (void)chgsbsize(so->so_cred->cr_uidinfo,
343 &so->so_snd.sb_hiwat, 0, RLIM_INFINITY);
345 /* remove acccept filter if one is present. */
346 if (so->so_accf != NULL)
347 do_setopt_accept_filter(so, NULL);
350 mac_socket_destroy(so);
353 sx_destroy(&so->so_snd.sb_sx);
354 sx_destroy(&so->so_rcv.sb_sx);
355 SOCKBUF_LOCK_DESTROY(&so->so_snd);
356 SOCKBUF_LOCK_DESTROY(&so->so_rcv);
357 uma_zfree(socket_zone, so);
361 * socreate returns a socket with a ref count of 1. The socket should be
362 * closed with soclose().
365 socreate(int dom, struct socket **aso, int type, int proto,
366 struct ucred *cred, struct thread *td)
373 prp = pffindproto(dom, proto, type);
375 prp = pffindtype(dom, type);
377 if (prp == NULL || prp->pr_usrreqs->pru_attach == NULL ||
378 prp->pr_usrreqs->pru_attach == pru_attach_notsupp)
379 return (EPROTONOSUPPORT);
381 if (prison_check_af(cred, prp->pr_domain->dom_family) != 0)
382 return (EPROTONOSUPPORT);
384 if (prp->pr_type != type)
386 so = soalloc(CRED_TO_VNET(cred));
390 TAILQ_INIT(&so->so_incomp);
391 TAILQ_INIT(&so->so_comp);
393 so->so_cred = crhold(cred);
394 if ((prp->pr_domain->dom_family == PF_INET) ||
395 (prp->pr_domain->dom_family == PF_INET6) ||
396 (prp->pr_domain->dom_family == PF_ROUTE))
397 so->so_fibnum = td->td_proc->p_fibnum;
402 mac_socket_create(cred, so);
404 knlist_init_mtx(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv));
405 knlist_init_mtx(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd));
408 * Auto-sizing of socket buffers is managed by the protocols and
409 * the appropriate flags must be set in the pru_attach function.
411 CURVNET_SET(so->so_vnet);
412 error = (*prp->pr_usrreqs->pru_attach)(so, proto, td);
415 KASSERT(so->so_count == 1, ("socreate: so_count %d",
426 static int regression_sonewconn_earlytest = 1;
427 SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW,
428 ®ression_sonewconn_earlytest, 0, "Perform early sonewconn limit test");
432 * When an attempt at a new connection is noted on a socket which accepts
433 * connections, sonewconn is called. If the connection is possible (subject
434 * to space constraints, etc.) then we allocate a new structure, propoerly
435 * linked into the data structure of the original socket, and return this.
436 * Connstatus may be 0, or SO_ISCONFIRMING, or SO_ISCONNECTED.
438 * Note: the ref count on the socket is 0 on return.
441 sonewconn(struct socket *head, int connstatus)
447 over = (head->so_qlen > 3 * head->so_qlimit / 2);
450 if (regression_sonewconn_earlytest && over)
455 VNET_ASSERT(head->so_vnet != NULL, ("%s:%d so_vnet is NULL, head=%p",
456 __func__, __LINE__, head));
457 so = soalloc(head->so_vnet);
460 if ((head->so_options & SO_ACCEPTFILTER) != 0)
463 so->so_type = head->so_type;
464 so->so_options = head->so_options &~ SO_ACCEPTCONN;
465 so->so_linger = head->so_linger;
466 so->so_state = head->so_state | SS_NOFDREF;
467 so->so_fibnum = head->so_fibnum;
468 so->so_proto = head->so_proto;
469 so->so_cred = crhold(head->so_cred);
471 mac_socket_newconn(head, so);
473 knlist_init_mtx(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv));
474 knlist_init_mtx(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd));
475 VNET_SO_ASSERT(head);
476 if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat) ||
477 (*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
481 so->so_rcv.sb_lowat = head->so_rcv.sb_lowat;
482 so->so_snd.sb_lowat = head->so_snd.sb_lowat;
483 so->so_rcv.sb_timeo = head->so_rcv.sb_timeo;
484 so->so_snd.sb_timeo = head->so_snd.sb_timeo;
485 so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE;
486 so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE;
487 so->so_state |= connstatus;
490 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
491 so->so_qstate |= SQ_COMP;
495 * Keep removing sockets from the head until there's room for
496 * us to insert on the tail. In pre-locking revisions, this
497 * was a simple if(), but as we could be racing with other
498 * threads and soabort() requires dropping locks, we must
499 * loop waiting for the condition to be true.
501 while (head->so_incqlen > head->so_qlimit) {
503 sp = TAILQ_FIRST(&head->so_incomp);
504 TAILQ_REMOVE(&head->so_incomp, sp, so_list);
506 sp->so_qstate &= ~SQ_INCOMP;
512 TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list);
513 so->so_qstate |= SQ_INCOMP;
519 wakeup_one(&head->so_timeo);
525 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
529 CURVNET_SET(so->so_vnet);
530 error = (*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td);
536 * solisten() transitions a socket from a non-listening state to a listening
537 * state, but can also be used to update the listen queue depth on an
538 * existing listen socket. The protocol will call back into the sockets
539 * layer using solisten_proto_check() and solisten_proto() to check and set
540 * socket-layer listen state. Call backs are used so that the protocol can
541 * acquire both protocol and socket layer locks in whatever order is required
544 * Protocol implementors are advised to hold the socket lock across the
545 * socket-layer test and set to avoid races at the socket layer.
548 solisten(struct socket *so, int backlog, struct thread *td)
552 CURVNET_SET(so->so_vnet);
553 error = (*so->so_proto->pr_usrreqs->pru_listen)(so, backlog, td);
559 solisten_proto_check(struct socket *so)
562 SOCK_LOCK_ASSERT(so);
564 if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
571 solisten_proto(struct socket *so, int backlog)
574 SOCK_LOCK_ASSERT(so);
576 if (backlog < 0 || backlog > somaxconn)
578 so->so_qlimit = backlog;
579 so->so_options |= SO_ACCEPTCONN;
583 * Evaluate the reference count and named references on a socket; if no
584 * references remain, free it. This should be called whenever a reference is
585 * released, such as in sorele(), but also when named reference flags are
586 * cleared in socket or protocol code.
588 * sofree() will free the socket if:
590 * - There are no outstanding file descriptor references or related consumers
593 * - The socket has been closed by user space, if ever open (SS_NOFDREF).
595 * - The protocol does not have an outstanding strong reference on the socket
598 * - The socket is not in a completed connection queue, so a process has been
599 * notified that it is present. If it is removed, the user process may
600 * block in accept() despite select() saying the socket was ready.
603 sofree(struct socket *so)
605 struct protosw *pr = so->so_proto;
608 ACCEPT_LOCK_ASSERT();
609 SOCK_LOCK_ASSERT(so);
611 if ((so->so_state & SS_NOFDREF) == 0 || so->so_count != 0 ||
612 (so->so_state & SS_PROTOREF) || (so->so_qstate & SQ_COMP)) {
620 KASSERT((so->so_qstate & SQ_COMP) != 0 ||
621 (so->so_qstate & SQ_INCOMP) != 0,
622 ("sofree: so_head != NULL, but neither SQ_COMP nor "
624 KASSERT((so->so_qstate & SQ_COMP) == 0 ||
625 (so->so_qstate & SQ_INCOMP) == 0,
626 ("sofree: so->so_qstate is SQ_COMP and also SQ_INCOMP"));
627 TAILQ_REMOVE(&head->so_incomp, so, so_list);
629 so->so_qstate &= ~SQ_INCOMP;
632 KASSERT((so->so_qstate & SQ_COMP) == 0 &&
633 (so->so_qstate & SQ_INCOMP) == 0,
634 ("sofree: so_head == NULL, but still SQ_COMP(%d) or SQ_INCOMP(%d)",
635 so->so_qstate & SQ_COMP, so->so_qstate & SQ_INCOMP));
636 if (so->so_options & SO_ACCEPTCONN) {
637 KASSERT((TAILQ_EMPTY(&so->so_comp)), ("sofree: so_comp populated"));
638 KASSERT((TAILQ_EMPTY(&so->so_incomp)), ("sofree: so_comp populated"));
644 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
645 (*pr->pr_domain->dom_dispose)(so->so_rcv.sb_mb);
646 if (pr->pr_usrreqs->pru_detach != NULL)
647 (*pr->pr_usrreqs->pru_detach)(so);
650 * From this point on, we assume that no other references to this
651 * socket exist anywhere else in the stack. Therefore, no locks need
652 * to be acquired or held.
654 * We used to do a lot of socket buffer and socket locking here, as
655 * well as invoke sorflush() and perform wakeups. The direct call to
656 * dom_dispose() and sbrelease_internal() are an inlining of what was
657 * necessary from sorflush().
659 * Notice that the socket buffer and kqueue state are torn down
660 * before calling pru_detach. This means that protocols shold not
661 * assume they can perform socket wakeups, etc, in their detach code.
663 sbdestroy(&so->so_snd, so);
664 sbdestroy(&so->so_rcv, so);
665 seldrain(&so->so_snd.sb_sel);
666 seldrain(&so->so_rcv.sb_sel);
667 knlist_destroy(&so->so_rcv.sb_sel.si_note);
668 knlist_destroy(&so->so_snd.sb_sel.si_note);
673 * Close a socket on last file table reference removal. Initiate disconnect
674 * if connected. Free socket when disconnect complete.
676 * This function will sorele() the socket. Note that soclose() may be called
677 * prior to the ref count reaching zero. The actual socket structure will
678 * not be freed until the ref count reaches zero.
681 soclose(struct socket *so)
685 KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter"));
687 CURVNET_SET(so->so_vnet);
688 funsetown(&so->so_sigio);
689 if (so->so_state & SS_ISCONNECTED) {
690 if ((so->so_state & SS_ISDISCONNECTING) == 0) {
691 error = sodisconnect(so);
693 if (error == ENOTCONN)
698 if (so->so_options & SO_LINGER) {
699 if ((so->so_state & SS_ISDISCONNECTING) &&
700 (so->so_state & SS_NBIO))
702 while (so->so_state & SS_ISCONNECTED) {
703 error = tsleep(&so->so_timeo,
704 PSOCK | PCATCH, "soclos", so->so_linger * hz);
712 if (so->so_proto->pr_usrreqs->pru_close != NULL)
713 (*so->so_proto->pr_usrreqs->pru_close)(so);
714 if (so->so_options & SO_ACCEPTCONN) {
717 while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
718 TAILQ_REMOVE(&so->so_incomp, sp, so_list);
720 sp->so_qstate &= ~SQ_INCOMP;
726 while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
727 TAILQ_REMOVE(&so->so_comp, sp, so_list);
729 sp->so_qstate &= ~SQ_COMP;
739 KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF"));
740 so->so_state |= SS_NOFDREF;
747 * soabort() is used to abruptly tear down a connection, such as when a
748 * resource limit is reached (listen queue depth exceeded), or if a listen
749 * socket is closed while there are sockets waiting to be accepted.
751 * This interface is tricky, because it is called on an unreferenced socket,
752 * and must be called only by a thread that has actually removed the socket
753 * from the listen queue it was on, or races with other threads are risked.
755 * This interface will call into the protocol code, so must not be called
756 * with any socket locks held. Protocols do call it while holding their own
757 * recursible protocol mutexes, but this is something that should be subject
758 * to review in the future.
761 soabort(struct socket *so)
765 * In as much as is possible, assert that no references to this
766 * socket are held. This is not quite the same as asserting that the
767 * current thread is responsible for arranging for no references, but
768 * is as close as we can get for now.
770 KASSERT(so->so_count == 0, ("soabort: so_count"));
771 KASSERT((so->so_state & SS_PROTOREF) == 0, ("soabort: SS_PROTOREF"));
772 KASSERT(so->so_state & SS_NOFDREF, ("soabort: !SS_NOFDREF"));
773 KASSERT((so->so_state & SQ_COMP) == 0, ("soabort: SQ_COMP"));
774 KASSERT((so->so_state & SQ_INCOMP) == 0, ("soabort: SQ_INCOMP"));
777 if (so->so_proto->pr_usrreqs->pru_abort != NULL)
778 (*so->so_proto->pr_usrreqs->pru_abort)(so);
785 soaccept(struct socket *so, struct sockaddr **nam)
790 KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF"));
791 so->so_state &= ~SS_NOFDREF;
794 CURVNET_SET(so->so_vnet);
795 error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam);
801 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
805 if (so->so_options & SO_ACCEPTCONN)
808 CURVNET_SET(so->so_vnet);
810 * If protocol is connection-based, can only connect once.
811 * Otherwise, if connected, try to disconnect first. This allows
812 * user to disconnect by connecting to, e.g., a null address.
814 if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
815 ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
816 (error = sodisconnect(so)))) {
820 * Prevent accumulated error from previous connection from
824 error = (*so->so_proto->pr_usrreqs->pru_connect)(so, nam, td);
832 soconnect2(struct socket *so1, struct socket *so2)
836 CURVNET_SET(so1->so_vnet);
837 error = (*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2);
843 sodisconnect(struct socket *so)
847 if ((so->so_state & SS_ISCONNECTED) == 0)
849 if (so->so_state & SS_ISDISCONNECTING)
852 error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so);
856 #ifdef ZERO_COPY_SOCKETS
857 struct so_zerocopy_stats{
862 struct so_zerocopy_stats so_zerocp_stats = {0,0,0};
863 #include <netinet/in.h>
864 #include <net/route.h>
865 #include <netinet/in_pcb.h>
867 #include <vm/vm_page.h>
868 #include <vm/vm_object.h>
871 * sosend_copyin() is only used if zero copy sockets are enabled. Otherwise
872 * sosend_dgram() and sosend_generic() use m_uiotombuf().
874 * sosend_copyin() accepts a uio and prepares an mbuf chain holding part or
875 * all of the data referenced by the uio. If desired, it uses zero-copy.
876 * *space will be updated to reflect data copied in.
878 * NB: If atomic I/O is requested, the caller must already have checked that
879 * space can hold resid bytes.
881 * NB: In the event of an error, the caller may need to free the partial
882 * chain pointed to by *mpp. The contents of both *uio and *space may be
883 * modified even in the case of an error.
886 sosend_copyin(struct uio *uio, struct mbuf **retmp, int atomic, long *space,
889 struct mbuf *m, **mp, *top;
892 #ifdef ZERO_COPY_SOCKETS
899 resid = uio->uio_resid;
902 #ifdef ZERO_COPY_SOCKETS
904 #endif /* ZERO_COPY_SOCKETS */
905 if (resid >= MINCLSIZE) {
906 #ifdef ZERO_COPY_SOCKETS
908 m = m_gethdr(M_WAITOK, MT_DATA);
910 m->m_pkthdr.rcvif = NULL;
912 m = m_get(M_WAITOK, MT_DATA);
913 if (so_zero_copy_send &&
916 uio->uio_iov->iov_len>=PAGE_SIZE) {
917 so_zerocp_stats.size_ok++;
918 so_zerocp_stats.align_ok++;
919 cow_send = socow_setup(m, uio);
923 m_clget(m, M_WAITOK);
924 len = min(min(MCLBYTES, resid), *space);
926 #else /* ZERO_COPY_SOCKETS */
928 m = m_getcl(M_WAIT, MT_DATA, M_PKTHDR);
930 m->m_pkthdr.rcvif = NULL;
932 m = m_getcl(M_WAIT, MT_DATA, 0);
933 len = min(min(MCLBYTES, resid), *space);
934 #endif /* ZERO_COPY_SOCKETS */
937 m = m_gethdr(M_WAIT, MT_DATA);
939 m->m_pkthdr.rcvif = NULL;
941 len = min(min(MHLEN, resid), *space);
943 * For datagram protocols, leave room
944 * for protocol headers in first mbuf.
946 if (atomic && m && len < MHLEN)
949 m = m_get(M_WAIT, MT_DATA);
950 len = min(min(MLEN, resid), *space);
959 #ifdef ZERO_COPY_SOCKETS
963 #endif /* ZERO_COPY_SOCKETS */
964 error = uiomove(mtod(m, void *), (int)len, uio);
965 resid = uio->uio_resid;
968 top->m_pkthdr.len += len;
974 top->m_flags |= M_EOR;
977 } while (*space > 0 && atomic);
982 #endif /*ZERO_COPY_SOCKETS*/
984 #define SBLOCKWAIT(f) (((f) & MSG_DONTWAIT) ? 0 : SBL_WAIT)
987 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio,
988 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
991 int clen = 0, error, dontroute;
992 #ifdef ZERO_COPY_SOCKETS
993 int atomic = sosendallatonce(so) || top;
996 KASSERT(so->so_type == SOCK_DGRAM, ("sodgram_send: !SOCK_DGRAM"));
997 KASSERT(so->so_proto->pr_flags & PR_ATOMIC,
998 ("sodgram_send: !PR_ATOMIC"));
1001 resid = uio->uio_resid;
1003 resid = top->m_pkthdr.len;
1005 * In theory resid should be unsigned. However, space must be
1006 * signed, as it might be less than 0 if we over-committed, and we
1007 * must use a signed comparison of space and resid. On the other
1008 * hand, a negative resid causes us to loop sending 0-length
1009 * segments to the protocol.
1017 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0;
1019 td->td_ru.ru_msgsnd++;
1020 if (control != NULL)
1021 clen = control->m_len;
1023 SOCKBUF_LOCK(&so->so_snd);
1024 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1025 SOCKBUF_UNLOCK(&so->so_snd);
1030 error = so->so_error;
1032 SOCKBUF_UNLOCK(&so->so_snd);
1035 if ((so->so_state & SS_ISCONNECTED) == 0) {
1037 * `sendto' and `sendmsg' is allowed on a connection-based
1038 * socket if it supports implied connect. Return ENOTCONN if
1039 * not connected and no address is supplied.
1041 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1042 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1043 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1044 !(resid == 0 && clen != 0)) {
1045 SOCKBUF_UNLOCK(&so->so_snd);
1049 } else if (addr == NULL) {
1050 if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1053 error = EDESTADDRREQ;
1054 SOCKBUF_UNLOCK(&so->so_snd);
1060 * Do we need MSG_OOB support in SOCK_DGRAM? Signs here may be a
1061 * problem and need fixing.
1063 space = sbspace(&so->so_snd);
1064 if (flags & MSG_OOB)
1067 SOCKBUF_UNLOCK(&so->so_snd);
1068 if (resid > space) {
1074 if (flags & MSG_EOR)
1075 top->m_flags |= M_EOR;
1077 #ifdef ZERO_COPY_SOCKETS
1078 error = sosend_copyin(uio, &top, atomic, &space, flags);
1083 * Copy the data from userland into a mbuf chain.
1084 * If no data is to be copied in, a single empty mbuf
1087 top = m_uiotombuf(uio, M_WAITOK, space, max_hdr,
1088 (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0)));
1090 error = EFAULT; /* only possible error */
1093 space -= resid - uio->uio_resid;
1095 resid = uio->uio_resid;
1097 KASSERT(resid == 0, ("sosend_dgram: resid != 0"));
1099 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock
1104 so->so_options |= SO_DONTROUTE;
1108 * XXX all the SBS_CANTSENDMORE checks previously done could be out
1109 * of date. We could have recieved a reset packet in an interrupt or
1110 * maybe we slept while doing page faults in uiomove() etc. We could
1111 * probably recheck again inside the locking protection here, but
1112 * there are probably other places that this also happens. We must
1116 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1117 (flags & MSG_OOB) ? PRUS_OOB :
1119 * If the user set MSG_EOF, the protocol understands this flag and
1120 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND.
1122 ((flags & MSG_EOF) &&
1123 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1126 /* If there is more to send set PRUS_MORETOCOME */
1127 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1128 top, addr, control, td);
1131 so->so_options &= ~SO_DONTROUTE;
1140 if (control != NULL)
1146 * Send on a socket. If send must go all at once and message is larger than
1147 * send buffering, then hard error. Lock against other senders. If must go
1148 * all at once and not enough room now, then inform user that this would
1149 * block and do nothing. Otherwise, if nonblocking, send as much as
1150 * possible. The data to be sent is described by "uio" if nonzero, otherwise
1151 * by the mbuf chain "top" (which must be null if uio is not). Data provided
1152 * in mbuf chain must be small enough to send all at once.
1154 * Returns nonzero on error, timeout or signal; callers must check for short
1155 * counts if EINTR/ERESTART are returned. Data and control buffers are freed
1159 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio,
1160 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1163 int clen = 0, error, dontroute;
1164 int atomic = sosendallatonce(so) || top;
1167 resid = uio->uio_resid;
1169 resid = top->m_pkthdr.len;
1171 * In theory resid should be unsigned. However, space must be
1172 * signed, as it might be less than 0 if we over-committed, and we
1173 * must use a signed comparison of space and resid. On the other
1174 * hand, a negative resid causes us to loop sending 0-length
1175 * segments to the protocol.
1177 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
1178 * type sockets since that's an error.
1180 if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
1186 (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
1187 (so->so_proto->pr_flags & PR_ATOMIC);
1189 td->td_ru.ru_msgsnd++;
1190 if (control != NULL)
1191 clen = control->m_len;
1193 error = sblock(&so->so_snd, SBLOCKWAIT(flags));
1199 SOCKBUF_LOCK(&so->so_snd);
1200 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1201 SOCKBUF_UNLOCK(&so->so_snd);
1206 error = so->so_error;
1208 SOCKBUF_UNLOCK(&so->so_snd);
1211 if ((so->so_state & SS_ISCONNECTED) == 0) {
1213 * `sendto' and `sendmsg' is allowed on a connection-
1214 * based socket if it supports implied connect.
1215 * Return ENOTCONN if not connected and no address is
1218 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1219 (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1220 if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1221 !(resid == 0 && clen != 0)) {
1222 SOCKBUF_UNLOCK(&so->so_snd);
1226 } else if (addr == NULL) {
1227 SOCKBUF_UNLOCK(&so->so_snd);
1228 if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1231 error = EDESTADDRREQ;
1235 space = sbspace(&so->so_snd);
1236 if (flags & MSG_OOB)
1238 if ((atomic && resid > so->so_snd.sb_hiwat) ||
1239 clen > so->so_snd.sb_hiwat) {
1240 SOCKBUF_UNLOCK(&so->so_snd);
1244 if (space < resid + clen &&
1245 (atomic || space < so->so_snd.sb_lowat || space < clen)) {
1246 if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO)) {
1247 SOCKBUF_UNLOCK(&so->so_snd);
1248 error = EWOULDBLOCK;
1251 error = sbwait(&so->so_snd);
1252 SOCKBUF_UNLOCK(&so->so_snd);
1257 SOCKBUF_UNLOCK(&so->so_snd);
1262 if (flags & MSG_EOR)
1263 top->m_flags |= M_EOR;
1265 #ifdef ZERO_COPY_SOCKETS
1266 error = sosend_copyin(uio, &top, atomic,
1272 * Copy the data from userland into a mbuf
1273 * chain. If no data is to be copied in,
1274 * a single empty mbuf is returned.
1276 top = m_uiotombuf(uio, M_WAITOK, space,
1277 (atomic ? max_hdr : 0),
1278 (atomic ? M_PKTHDR : 0) |
1279 ((flags & MSG_EOR) ? M_EOR : 0));
1281 error = EFAULT; /* only possible error */
1284 space -= resid - uio->uio_resid;
1286 resid = uio->uio_resid;
1290 so->so_options |= SO_DONTROUTE;
1294 * XXX all the SBS_CANTSENDMORE checks previously
1295 * done could be out of date. We could have recieved
1296 * a reset packet in an interrupt or maybe we slept
1297 * while doing page faults in uiomove() etc. We
1298 * could probably recheck again inside the locking
1299 * protection here, but there are probably other
1300 * places that this also happens. We must rethink
1304 error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1305 (flags & MSG_OOB) ? PRUS_OOB :
1307 * If the user set MSG_EOF, the protocol understands
1308 * this flag and nothing left to send then use
1309 * PRU_SEND_EOF instead of PRU_SEND.
1311 ((flags & MSG_EOF) &&
1312 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1315 /* If there is more to send set PRUS_MORETOCOME. */
1316 (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1317 top, addr, control, td);
1320 so->so_options &= ~SO_DONTROUTE;
1328 } while (resid && space > 0);
1332 sbunlock(&so->so_snd);
1336 if (control != NULL)
1342 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
1343 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1347 CURVNET_SET(so->so_vnet);
1348 error = so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio, top,
1349 control, flags, td);
1355 * The part of soreceive() that implements reading non-inline out-of-band
1356 * data from a socket. For more complete comments, see soreceive(), from
1357 * which this code originated.
1359 * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
1360 * unable to return an mbuf chain to the caller.
1363 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags)
1365 struct protosw *pr = so->so_proto;
1369 KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
1372 m = m_get(M_WAIT, MT_DATA);
1373 error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK);
1377 #ifdef ZERO_COPY_SOCKETS
1378 if (so_zero_copy_receive) {
1381 if ((m->m_flags & M_EXT)
1382 && (m->m_ext.ext_type == EXT_DISPOSABLE))
1387 error = uiomoveco(mtod(m, void *),
1388 min(uio->uio_resid, m->m_len),
1391 #endif /* ZERO_COPY_SOCKETS */
1392 error = uiomove(mtod(m, void *),
1393 (int) min(uio->uio_resid, m->m_len), uio);
1395 } while (uio->uio_resid && error == 0 && m);
1403 * Following replacement or removal of the first mbuf on the first mbuf chain
1404 * of a socket buffer, push necessary state changes back into the socket
1405 * buffer so that other consumers see the values consistently. 'nextrecord'
1406 * is the callers locally stored value of the original value of
1407 * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes.
1408 * NOTE: 'nextrecord' may be NULL.
1410 static __inline void
1411 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord)
1414 SOCKBUF_LOCK_ASSERT(sb);
1416 * First, update for the new value of nextrecord. If necessary, make
1417 * it the first record.
1419 if (sb->sb_mb != NULL)
1420 sb->sb_mb->m_nextpkt = nextrecord;
1422 sb->sb_mb = nextrecord;
1425 * Now update any dependent socket buffer fields to reflect the new
1426 * state. This is an expanded inline of SB_EMPTY_FIXUP(), with the
1427 * addition of a second clause that takes care of the case where
1428 * sb_mb has been updated, but remains the last record.
1430 if (sb->sb_mb == NULL) {
1431 sb->sb_mbtail = NULL;
1432 sb->sb_lastrecord = NULL;
1433 } else if (sb->sb_mb->m_nextpkt == NULL)
1434 sb->sb_lastrecord = sb->sb_mb;
1439 * Implement receive operations on a socket. We depend on the way that
1440 * records are added to the sockbuf by sbappend. In particular, each record
1441 * (mbufs linked through m_next) must begin with an address if the protocol
1442 * so specifies, followed by an optional mbuf or mbufs containing ancillary
1443 * data, and then zero or more mbufs of data. In order to allow parallelism
1444 * between network receive and copying to user space, as well as avoid
1445 * sleeping with a mutex held, we release the socket buffer mutex during the
1446 * user space copy. Although the sockbuf is locked, new data may still be
1447 * appended, and thus we must maintain consistency of the sockbuf during that
1450 * The caller may receive the data as a single mbuf chain by supplying an
1451 * mbuf **mp0 for use in returning the chain. The uio is then used only for
1452 * the count in uio_resid.
1455 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio,
1456 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1458 struct mbuf *m, **mp;
1459 int flags, len, error, offset;
1460 struct protosw *pr = so->so_proto;
1461 struct mbuf *nextrecord;
1463 int orig_resid = uio->uio_resid;
1468 if (controlp != NULL)
1471 flags = *flagsp &~ MSG_EOR;
1474 if (flags & MSG_OOB)
1475 return (soreceive_rcvoob(so, uio, flags));
1478 if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING)
1479 && uio->uio_resid) {
1481 (*pr->pr_usrreqs->pru_rcvd)(so, 0);
1484 error = sblock(&so->so_rcv, SBLOCKWAIT(flags));
1489 SOCKBUF_LOCK(&so->so_rcv);
1490 m = so->so_rcv.sb_mb;
1492 * If we have less data than requested, block awaiting more (subject
1493 * to any timeout) if:
1494 * 1. the current count is less than the low water mark, or
1495 * 2. MSG_WAITALL is set, and it is possible to do the entire
1496 * receive operation at once if we block (resid <= hiwat).
1497 * 3. MSG_DONTWAIT is not set
1498 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1499 * we have to do the receive in sections, and thus risk returning a
1500 * short count if a timeout or signal occurs after we start.
1502 if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1503 so->so_rcv.sb_cc < uio->uio_resid) &&
1504 (so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
1505 ((flags & MSG_WAITALL) && uio->uio_resid <= so->so_rcv.sb_hiwat)) &&
1506 m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
1507 KASSERT(m != NULL || !so->so_rcv.sb_cc,
1508 ("receive: m == %p so->so_rcv.sb_cc == %u",
1509 m, so->so_rcv.sb_cc));
1513 error = so->so_error;
1514 if ((flags & MSG_PEEK) == 0)
1516 SOCKBUF_UNLOCK(&so->so_rcv);
1519 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1520 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1522 SOCKBUF_UNLOCK(&so->so_rcv);
1527 for (; m != NULL; m = m->m_next)
1528 if (m->m_type == MT_OOBDATA || (m->m_flags & M_EOR)) {
1529 m = so->so_rcv.sb_mb;
1532 if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1533 (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1534 SOCKBUF_UNLOCK(&so->so_rcv);
1538 if (uio->uio_resid == 0) {
1539 SOCKBUF_UNLOCK(&so->so_rcv);
1542 if ((so->so_state & SS_NBIO) ||
1543 (flags & (MSG_DONTWAIT|MSG_NBIO))) {
1544 SOCKBUF_UNLOCK(&so->so_rcv);
1545 error = EWOULDBLOCK;
1548 SBLASTRECORDCHK(&so->so_rcv);
1549 SBLASTMBUFCHK(&so->so_rcv);
1550 error = sbwait(&so->so_rcv);
1551 SOCKBUF_UNLOCK(&so->so_rcv);
1558 * From this point onward, we maintain 'nextrecord' as a cache of the
1559 * pointer to the next record in the socket buffer. We must keep the
1560 * various socket buffer pointers and local stack versions of the
1561 * pointers in sync, pushing out modifications before dropping the
1562 * socket buffer mutex, and re-reading them when picking it up.
1564 * Otherwise, we will race with the network stack appending new data
1565 * or records onto the socket buffer by using inconsistent/stale
1566 * versions of the field, possibly resulting in socket buffer
1569 * By holding the high-level sblock(), we prevent simultaneous
1570 * readers from pulling off the front of the socket buffer.
1572 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1574 uio->uio_td->td_ru.ru_msgrcv++;
1575 KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb"));
1576 SBLASTRECORDCHK(&so->so_rcv);
1577 SBLASTMBUFCHK(&so->so_rcv);
1578 nextrecord = m->m_nextpkt;
1579 if (pr->pr_flags & PR_ADDR) {
1580 KASSERT(m->m_type == MT_SONAME,
1581 ("m->m_type == %d", m->m_type));
1584 *psa = sodupsockaddr(mtod(m, struct sockaddr *),
1586 if (flags & MSG_PEEK) {
1589 sbfree(&so->so_rcv, m);
1590 so->so_rcv.sb_mb = m_free(m);
1591 m = so->so_rcv.sb_mb;
1592 sockbuf_pushsync(&so->so_rcv, nextrecord);
1597 * Process one or more MT_CONTROL mbufs present before any data mbufs
1598 * in the first mbuf chain on the socket buffer. If MSG_PEEK, we
1599 * just copy the data; if !MSG_PEEK, we call into the protocol to
1600 * perform externalization (or freeing if controlp == NULL).
1602 if (m != NULL && m->m_type == MT_CONTROL) {
1603 struct mbuf *cm = NULL, *cmn;
1604 struct mbuf **cme = &cm;
1607 if (flags & MSG_PEEK) {
1608 if (controlp != NULL) {
1609 *controlp = m_copy(m, 0, m->m_len);
1610 controlp = &(*controlp)->m_next;
1614 sbfree(&so->so_rcv, m);
1615 so->so_rcv.sb_mb = m->m_next;
1618 cme = &(*cme)->m_next;
1619 m = so->so_rcv.sb_mb;
1621 } while (m != NULL && m->m_type == MT_CONTROL);
1622 if ((flags & MSG_PEEK) == 0)
1623 sockbuf_pushsync(&so->so_rcv, nextrecord);
1624 while (cm != NULL) {
1627 if (pr->pr_domain->dom_externalize != NULL) {
1628 SOCKBUF_UNLOCK(&so->so_rcv);
1630 error = (*pr->pr_domain->dom_externalize)
1632 SOCKBUF_LOCK(&so->so_rcv);
1633 } else if (controlp != NULL)
1637 if (controlp != NULL) {
1639 while (*controlp != NULL)
1640 controlp = &(*controlp)->m_next;
1645 nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1647 nextrecord = so->so_rcv.sb_mb;
1651 if ((flags & MSG_PEEK) == 0) {
1652 KASSERT(m->m_nextpkt == nextrecord,
1653 ("soreceive: post-control, nextrecord !sync"));
1654 if (nextrecord == NULL) {
1655 KASSERT(so->so_rcv.sb_mb == m,
1656 ("soreceive: post-control, sb_mb!=m"));
1657 KASSERT(so->so_rcv.sb_lastrecord == m,
1658 ("soreceive: post-control, lastrecord!=m"));
1662 if (type == MT_OOBDATA)
1665 if ((flags & MSG_PEEK) == 0) {
1666 KASSERT(so->so_rcv.sb_mb == nextrecord,
1667 ("soreceive: sb_mb != nextrecord"));
1668 if (so->so_rcv.sb_mb == NULL) {
1669 KASSERT(so->so_rcv.sb_lastrecord == NULL,
1670 ("soreceive: sb_lastercord != NULL"));
1674 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1675 SBLASTRECORDCHK(&so->so_rcv);
1676 SBLASTMBUFCHK(&so->so_rcv);
1679 * Now continue to read any data mbufs off of the head of the socket
1680 * buffer until the read request is satisfied. Note that 'type' is
1681 * used to store the type of any mbuf reads that have happened so far
1682 * such that soreceive() can stop reading if the type changes, which
1683 * causes soreceive() to return only one of regular data and inline
1684 * out-of-band data in a single socket receive operation.
1688 while (m != NULL && uio->uio_resid > 0 && error == 0) {
1690 * If the type of mbuf has changed since the last mbuf
1691 * examined ('type'), end the receive operation.
1693 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1694 if (m->m_type == MT_OOBDATA) {
1695 if (type != MT_OOBDATA)
1697 } else if (type == MT_OOBDATA)
1700 KASSERT(m->m_type == MT_DATA,
1701 ("m->m_type == %d", m->m_type));
1702 so->so_rcv.sb_state &= ~SBS_RCVATMARK;
1703 len = uio->uio_resid;
1704 if (so->so_oobmark && len > so->so_oobmark - offset)
1705 len = so->so_oobmark - offset;
1706 if (len > m->m_len - moff)
1707 len = m->m_len - moff;
1709 * If mp is set, just pass back the mbufs. Otherwise copy
1710 * them out via the uio, then free. Sockbuf must be
1711 * consistent here (points to current mbuf, it points to next
1712 * record) when we drop priority; we must note any additions
1713 * to the sockbuf when we block interrupts again.
1716 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1717 SBLASTRECORDCHK(&so->so_rcv);
1718 SBLASTMBUFCHK(&so->so_rcv);
1719 SOCKBUF_UNLOCK(&so->so_rcv);
1720 #ifdef ZERO_COPY_SOCKETS
1721 if (so_zero_copy_receive) {
1724 if ((m->m_flags & M_EXT)
1725 && (m->m_ext.ext_type == EXT_DISPOSABLE))
1730 error = uiomoveco(mtod(m, char *) + moff,
1734 #endif /* ZERO_COPY_SOCKETS */
1735 error = uiomove(mtod(m, char *) + moff, (int)len, uio);
1736 SOCKBUF_LOCK(&so->so_rcv);
1739 * The MT_SONAME mbuf has already been removed
1740 * from the record, so it is necessary to
1741 * remove the data mbufs, if any, to preserve
1742 * the invariant in the case of PR_ADDR that
1743 * requires MT_SONAME mbufs at the head of
1746 if (m && pr->pr_flags & PR_ATOMIC &&
1747 ((flags & MSG_PEEK) == 0))
1748 (void)sbdroprecord_locked(&so->so_rcv);
1749 SOCKBUF_UNLOCK(&so->so_rcv);
1753 uio->uio_resid -= len;
1754 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1755 if (len == m->m_len - moff) {
1756 if (m->m_flags & M_EOR)
1758 if (flags & MSG_PEEK) {
1762 nextrecord = m->m_nextpkt;
1763 sbfree(&so->so_rcv, m);
1767 so->so_rcv.sb_mb = m = m->m_next;
1770 so->so_rcv.sb_mb = m_free(m);
1771 m = so->so_rcv.sb_mb;
1773 sockbuf_pushsync(&so->so_rcv, nextrecord);
1774 SBLASTRECORDCHK(&so->so_rcv);
1775 SBLASTMBUFCHK(&so->so_rcv);
1778 if (flags & MSG_PEEK)
1784 if (flags & MSG_DONTWAIT)
1785 copy_flag = M_DONTWAIT;
1788 if (copy_flag == M_WAIT)
1789 SOCKBUF_UNLOCK(&so->so_rcv);
1790 *mp = m_copym(m, 0, len, copy_flag);
1791 if (copy_flag == M_WAIT)
1792 SOCKBUF_LOCK(&so->so_rcv);
1795 * m_copym() couldn't
1796 * allocate an mbuf. Adjust
1797 * uio_resid back (it was
1798 * adjusted down by len
1799 * bytes, which we didn't end
1800 * up "copying" over).
1802 uio->uio_resid += len;
1808 so->so_rcv.sb_cc -= len;
1811 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1812 if (so->so_oobmark) {
1813 if ((flags & MSG_PEEK) == 0) {
1814 so->so_oobmark -= len;
1815 if (so->so_oobmark == 0) {
1816 so->so_rcv.sb_state |= SBS_RCVATMARK;
1821 if (offset == so->so_oobmark)
1825 if (flags & MSG_EOR)
1828 * If the MSG_WAITALL flag is set (for non-atomic socket), we
1829 * must not quit until "uio->uio_resid == 0" or an error
1830 * termination. If a signal/timeout occurs, return with a
1831 * short count but without error. Keep sockbuf locked
1832 * against other readers.
1834 while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1835 !sosendallatonce(so) && nextrecord == NULL) {
1836 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1837 if (so->so_error || so->so_rcv.sb_state & SBS_CANTRCVMORE)
1840 * Notify the protocol that some data has been
1841 * drained before blocking.
1843 if (pr->pr_flags & PR_WANTRCVD) {
1844 SOCKBUF_UNLOCK(&so->so_rcv);
1846 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
1847 SOCKBUF_LOCK(&so->so_rcv);
1849 SBLASTRECORDCHK(&so->so_rcv);
1850 SBLASTMBUFCHK(&so->so_rcv);
1852 * We could receive some data while was notifying
1853 * the protocol. Skip blocking in this case.
1855 if (so->so_rcv.sb_mb == NULL) {
1856 error = sbwait(&so->so_rcv);
1858 SOCKBUF_UNLOCK(&so->so_rcv);
1862 m = so->so_rcv.sb_mb;
1864 nextrecord = m->m_nextpkt;
1868 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1869 if (m != NULL && pr->pr_flags & PR_ATOMIC) {
1871 if ((flags & MSG_PEEK) == 0)
1872 (void) sbdroprecord_locked(&so->so_rcv);
1874 if ((flags & MSG_PEEK) == 0) {
1877 * First part is an inline SB_EMPTY_FIXUP(). Second
1878 * part makes sure sb_lastrecord is up-to-date if
1879 * there is still data in the socket buffer.
1881 so->so_rcv.sb_mb = nextrecord;
1882 if (so->so_rcv.sb_mb == NULL) {
1883 so->so_rcv.sb_mbtail = NULL;
1884 so->so_rcv.sb_lastrecord = NULL;
1885 } else if (nextrecord->m_nextpkt == NULL)
1886 so->so_rcv.sb_lastrecord = nextrecord;
1888 SBLASTRECORDCHK(&so->so_rcv);
1889 SBLASTMBUFCHK(&so->so_rcv);
1891 * If soreceive() is being done from the socket callback,
1892 * then don't need to generate ACK to peer to update window,
1893 * since ACK will be generated on return to TCP.
1895 if (!(flags & MSG_SOCALLBCK) &&
1896 (pr->pr_flags & PR_WANTRCVD)) {
1897 SOCKBUF_UNLOCK(&so->so_rcv);
1899 (*pr->pr_usrreqs->pru_rcvd)(so, flags);
1900 SOCKBUF_LOCK(&so->so_rcv);
1903 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1904 if (orig_resid == uio->uio_resid && orig_resid &&
1905 (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
1906 SOCKBUF_UNLOCK(&so->so_rcv);
1909 SOCKBUF_UNLOCK(&so->so_rcv);
1914 sbunlock(&so->so_rcv);
1919 * Optimized version of soreceive() for stream (TCP) sockets.
1922 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio,
1923 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1925 int len = 0, error = 0, flags, oresid;
1927 struct mbuf *m, *n = NULL;
1929 /* We only do stream sockets. */
1930 if (so->so_type != SOCK_STREAM)
1934 if (controlp != NULL)
1937 flags = *flagsp &~ MSG_EOR;
1940 if (flags & MSG_OOB)
1941 return (soreceive_rcvoob(so, uio, flags));
1947 /* Prevent other readers from entering the socket. */
1948 error = sblock(sb, SBLOCKWAIT(flags));
1953 /* Easy one, no space to copyout anything. */
1954 if (uio->uio_resid == 0) {
1958 oresid = uio->uio_resid;
1960 /* We will never ever get anything unless we are or were connected. */
1961 if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) {
1967 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1969 /* Abort if socket has reported problems. */
1973 if (oresid > uio->uio_resid)
1975 error = so->so_error;
1976 if (!(flags & MSG_PEEK))
1981 /* Door is closed. Deliver what is left, if any. */
1982 if (sb->sb_state & SBS_CANTRCVMORE) {
1989 /* Socket buffer is empty and we shall not block. */
1990 if (sb->sb_cc == 0 &&
1991 ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) {
1996 /* Socket buffer got some data that we shall deliver now. */
1997 if (sb->sb_cc > 0 && !(flags & MSG_WAITALL) &&
1998 ((sb->sb_flags & SS_NBIO) ||
1999 (flags & (MSG_DONTWAIT|MSG_NBIO)) ||
2000 sb->sb_cc >= sb->sb_lowat ||
2001 sb->sb_cc >= uio->uio_resid ||
2002 sb->sb_cc >= sb->sb_hiwat) ) {
2006 /* On MSG_WAITALL we must wait until all data or error arrives. */
2007 if ((flags & MSG_WAITALL) &&
2008 (sb->sb_cc >= uio->uio_resid || sb->sb_cc >= sb->sb_lowat))
2012 * Wait and block until (more) data comes in.
2013 * NB: Drops the sockbuf lock during wait.
2021 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2022 KASSERT(sb->sb_cc > 0, ("%s: sockbuf empty", __func__));
2023 KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__));
2027 uio->uio_td->td_ru.ru_msgrcv++;
2029 /* Fill uio until full or current end of socket buffer is reached. */
2030 len = min(uio->uio_resid, sb->sb_cc);
2032 /* Dequeue as many mbufs as possible. */
2033 if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) {
2034 for (*mp0 = m = sb->sb_mb;
2035 m != NULL && m->m_len <= len;
2038 uio->uio_resid -= m->m_len;
2043 if (sb->sb_mb == NULL)
2047 /* Copy the remainder. */
2049 KASSERT(sb->sb_mb != NULL,
2050 ("%s: len > 0 && sb->sb_mb empty", __func__));
2052 m = m_copym(sb->sb_mb, 0, len, M_DONTWAIT);
2054 len = 0; /* Don't flush data from sockbuf. */
2056 uio->uio_resid -= m->m_len;
2067 /* NB: Must unlock socket buffer as uiomove may sleep. */
2069 error = m_mbuftouio(uio, sb->sb_mb, len);
2074 SBLASTRECORDCHK(sb);
2078 * Remove the delivered data from the socket buffer unless we
2079 * were only peeking.
2081 if (!(flags & MSG_PEEK)) {
2083 sbdrop_locked(sb, len);
2085 /* Notify protocol that we drained some data. */
2086 if ((so->so_proto->pr_flags & PR_WANTRCVD) &&
2087 (((flags & MSG_WAITALL) && uio->uio_resid > 0) ||
2088 !(flags & MSG_SOCALLBCK))) {
2091 (*so->so_proto->pr_usrreqs->pru_rcvd)(so, flags);
2097 * For MSG_WAITALL we may have to loop again and wait for
2098 * more data to come in.
2100 if ((flags & MSG_WAITALL) && uio->uio_resid > 0)
2103 SOCKBUF_LOCK_ASSERT(sb);
2104 SBLASTRECORDCHK(sb);
2112 * Optimized version of soreceive() for simple datagram cases from userspace.
2113 * Unlike in the stream case, we're able to drop a datagram if copyout()
2114 * fails, and because we handle datagrams atomically, we don't need to use a
2115 * sleep lock to prevent I/O interlacing.
2118 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio,
2119 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2121 struct mbuf *m, *m2;
2122 int flags, len, error;
2123 struct protosw *pr = so->so_proto;
2124 struct mbuf *nextrecord;
2128 if (controlp != NULL)
2131 flags = *flagsp &~ MSG_EOR;
2136 * For any complicated cases, fall back to the full
2137 * soreceive_generic().
2139 if (mp0 != NULL || (flags & MSG_PEEK) || (flags & MSG_OOB))
2140 return (soreceive_generic(so, psa, uio, mp0, controlp,
2144 * Enforce restrictions on use.
2146 KASSERT((pr->pr_flags & PR_WANTRCVD) == 0,
2147 ("soreceive_dgram: wantrcvd"));
2148 KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic"));
2149 KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0,
2150 ("soreceive_dgram: SBS_RCVATMARK"));
2151 KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0,
2152 ("soreceive_dgram: P_CONNREQUIRED"));
2155 * Loop blocking while waiting for a datagram.
2157 SOCKBUF_LOCK(&so->so_rcv);
2158 while ((m = so->so_rcv.sb_mb) == NULL) {
2159 KASSERT(so->so_rcv.sb_cc == 0,
2160 ("soreceive_dgram: sb_mb NULL but sb_cc %u",
2163 error = so->so_error;
2165 SOCKBUF_UNLOCK(&so->so_rcv);
2168 if (so->so_rcv.sb_state & SBS_CANTRCVMORE ||
2169 uio->uio_resid == 0) {
2170 SOCKBUF_UNLOCK(&so->so_rcv);
2173 if ((so->so_state & SS_NBIO) ||
2174 (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2175 SOCKBUF_UNLOCK(&so->so_rcv);
2176 return (EWOULDBLOCK);
2178 SBLASTRECORDCHK(&so->so_rcv);
2179 SBLASTMBUFCHK(&so->so_rcv);
2180 error = sbwait(&so->so_rcv);
2182 SOCKBUF_UNLOCK(&so->so_rcv);
2186 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2189 uio->uio_td->td_ru.ru_msgrcv++;
2190 SBLASTRECORDCHK(&so->so_rcv);
2191 SBLASTMBUFCHK(&so->so_rcv);
2192 nextrecord = m->m_nextpkt;
2193 if (nextrecord == NULL) {
2194 KASSERT(so->so_rcv.sb_lastrecord == m,
2195 ("soreceive_dgram: lastrecord != m"));
2198 KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord,
2199 ("soreceive_dgram: m_nextpkt != nextrecord"));
2202 * Pull 'm' and its chain off the front of the packet queue.
2204 so->so_rcv.sb_mb = NULL;
2205 sockbuf_pushsync(&so->so_rcv, nextrecord);
2208 * Walk 'm's chain and free that many bytes from the socket buffer.
2210 for (m2 = m; m2 != NULL; m2 = m2->m_next)
2211 sbfree(&so->so_rcv, m2);
2214 * Do a few last checks before we let go of the lock.
2216 SBLASTRECORDCHK(&so->so_rcv);
2217 SBLASTMBUFCHK(&so->so_rcv);
2218 SOCKBUF_UNLOCK(&so->so_rcv);
2220 if (pr->pr_flags & PR_ADDR) {
2221 KASSERT(m->m_type == MT_SONAME,
2222 ("m->m_type == %d", m->m_type));
2224 *psa = sodupsockaddr(mtod(m, struct sockaddr *),
2229 /* XXXRW: Can this happen? */
2234 * Packet to copyout() is now in 'm' and it is disconnected from the
2237 * Process one or more MT_CONTROL mbufs present before any data mbufs
2238 * in the first mbuf chain on the socket buffer. We call into the
2239 * protocol to perform externalization (or freeing if controlp ==
2242 if (m->m_type == MT_CONTROL) {
2243 struct mbuf *cm = NULL, *cmn;
2244 struct mbuf **cme = &cm;
2250 cme = &(*cme)->m_next;
2252 } while (m != NULL && m->m_type == MT_CONTROL);
2253 while (cm != NULL) {
2256 if (pr->pr_domain->dom_externalize != NULL) {
2257 error = (*pr->pr_domain->dom_externalize)
2259 } else if (controlp != NULL)
2263 if (controlp != NULL) {
2264 while (*controlp != NULL)
2265 controlp = &(*controlp)->m_next;
2270 KASSERT(m->m_type == MT_DATA, ("soreceive_dgram: !data"));
2272 while (m != NULL && uio->uio_resid > 0) {
2273 len = uio->uio_resid;
2276 error = uiomove(mtod(m, char *), (int)len, uio);
2281 if (len == m->m_len)
2297 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
2298 struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2302 CURVNET_SET(so->so_vnet);
2303 error = (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio, mp0,
2310 soshutdown(struct socket *so, int how)
2312 struct protosw *pr = so->so_proto;
2315 if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
2318 CURVNET_SET(so->so_vnet);
2319 if (pr->pr_usrreqs->pru_flush != NULL) {
2320 (*pr->pr_usrreqs->pru_flush)(so, how);
2324 if (how != SHUT_RD) {
2325 error = (*pr->pr_usrreqs->pru_shutdown)(so);
2334 sorflush(struct socket *so)
2336 struct sockbuf *sb = &so->so_rcv;
2337 struct protosw *pr = so->so_proto;
2343 * In order to avoid calling dom_dispose with the socket buffer mutex
2344 * held, and in order to generally avoid holding the lock for a long
2345 * time, we make a copy of the socket buffer and clear the original
2346 * (except locks, state). The new socket buffer copy won't have
2347 * initialized locks so we can only call routines that won't use or
2348 * assert those locks.
2350 * Dislodge threads currently blocked in receive and wait to acquire
2351 * a lock against other simultaneous readers before clearing the
2352 * socket buffer. Don't let our acquire be interrupted by a signal
2353 * despite any existing socket disposition on interruptable waiting.
2356 (void) sblock(sb, SBL_WAIT | SBL_NOINTR);
2359 * Invalidate/clear most of the sockbuf structure, but leave selinfo
2360 * and mutex data unchanged.
2363 bzero(&asb, offsetof(struct sockbuf, sb_startzero));
2364 bcopy(&sb->sb_startzero, &asb.sb_startzero,
2365 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2366 bzero(&sb->sb_startzero,
2367 sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2372 * Dispose of special rights and flush the socket buffer. Don't call
2373 * any unsafe routines (that rely on locks being initialized) on asb.
2375 if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
2376 (*pr->pr_domain->dom_dispose)(asb.sb_mb);
2377 sbrelease_internal(&asb, so);
2381 * Perhaps this routine, and sooptcopyout(), below, ought to come in an
2382 * additional variant to handle the case where the option value needs to be
2383 * some kind of integer, but not a specific size. In addition to their use
2384 * here, these functions are also called by the protocol-level pr_ctloutput()
2388 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2393 * If the user gives us more than we wanted, we ignore it, but if we
2394 * don't get the minimum length the caller wants, we return EINVAL.
2395 * On success, sopt->sopt_valsize is set to however much we actually
2398 if ((valsize = sopt->sopt_valsize) < minlen)
2401 sopt->sopt_valsize = valsize = len;
2403 if (sopt->sopt_td != NULL)
2404 return (copyin(sopt->sopt_val, buf, valsize));
2406 bcopy(sopt->sopt_val, buf, valsize);
2411 * Kernel version of setsockopt(2).
2413 * XXX: optlen is size_t, not socklen_t
2416 so_setsockopt(struct socket *so, int level, int optname, void *optval,
2419 struct sockopt sopt;
2421 sopt.sopt_level = level;
2422 sopt.sopt_name = optname;
2423 sopt.sopt_dir = SOPT_SET;
2424 sopt.sopt_val = optval;
2425 sopt.sopt_valsize = optlen;
2426 sopt.sopt_td = NULL;
2427 return (sosetopt(so, &sopt));
2431 sosetopt(struct socket *so, struct sockopt *sopt)
2442 CURVNET_SET(so->so_vnet);
2444 if (sopt->sopt_level != SOL_SOCKET) {
2445 if (so->so_proto->pr_ctloutput != NULL) {
2446 error = (*so->so_proto->pr_ctloutput)(so, sopt);
2450 error = ENOPROTOOPT;
2452 switch (sopt->sopt_name) {
2454 case SO_ACCEPTFILTER:
2455 error = do_setopt_accept_filter(so, sopt);
2461 error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
2466 so->so_linger = l.l_linger;
2468 so->so_options |= SO_LINGER;
2470 so->so_options &= ~SO_LINGER;
2477 case SO_USELOOPBACK:
2487 error = sooptcopyin(sopt, &optval, sizeof optval,
2493 so->so_options |= sopt->sopt_name;
2495 so->so_options &= ~sopt->sopt_name;
2500 error = sooptcopyin(sopt, &optval, sizeof optval,
2502 if (optval < 0 || optval >= rt_numfibs) {
2506 if (((so->so_proto->pr_domain->dom_family == PF_INET) ||
2507 (so->so_proto->pr_domain->dom_family == PF_INET6) ||
2508 (so->so_proto->pr_domain->dom_family == PF_ROUTE))) {
2509 so->so_fibnum = optval;
2510 /* Note: ignore error */
2511 if (so->so_proto->pr_ctloutput)
2512 (*so->so_proto->pr_ctloutput)(so, sopt);
2518 case SO_USER_COOKIE:
2519 error = sooptcopyin(sopt, &val32, sizeof val32,
2523 so->so_user_cookie = val32;
2530 error = sooptcopyin(sopt, &optval, sizeof optval,
2536 * Values < 1 make no sense for any of these options,
2544 switch (sopt->sopt_name) {
2547 if (sbreserve(sopt->sopt_name == SO_SNDBUF ?
2548 &so->so_snd : &so->so_rcv, (u_long)optval,
2549 so, curthread) == 0) {
2553 (sopt->sopt_name == SO_SNDBUF ? &so->so_snd :
2554 &so->so_rcv)->sb_flags &= ~SB_AUTOSIZE;
2558 * Make sure the low-water is never greater than the
2562 SOCKBUF_LOCK(&so->so_snd);
2563 so->so_snd.sb_lowat =
2564 (optval > so->so_snd.sb_hiwat) ?
2565 so->so_snd.sb_hiwat : optval;
2566 SOCKBUF_UNLOCK(&so->so_snd);
2569 SOCKBUF_LOCK(&so->so_rcv);
2570 so->so_rcv.sb_lowat =
2571 (optval > so->so_rcv.sb_hiwat) ?
2572 so->so_rcv.sb_hiwat : optval;
2573 SOCKBUF_UNLOCK(&so->so_rcv);
2580 #ifdef COMPAT_FREEBSD32
2581 if (SV_CURPROC_FLAG(SV_ILP32)) {
2582 struct timeval32 tv32;
2584 error = sooptcopyin(sopt, &tv32, sizeof tv32,
2586 CP(tv32, tv, tv_sec);
2587 CP(tv32, tv, tv_usec);
2590 error = sooptcopyin(sopt, &tv, sizeof tv,
2595 /* assert(hz > 0); */
2596 if (tv.tv_sec < 0 || tv.tv_sec > INT_MAX / hz ||
2597 tv.tv_usec < 0 || tv.tv_usec >= 1000000) {
2601 /* assert(tick > 0); */
2602 /* assert(ULONG_MAX - INT_MAX >= 1000000); */
2603 val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / tick;
2604 if (val > INT_MAX) {
2608 if (val == 0 && tv.tv_usec != 0)
2611 switch (sopt->sopt_name) {
2613 so->so_snd.sb_timeo = val;
2616 so->so_rcv.sb_timeo = val;
2623 error = sooptcopyin(sopt, &extmac, sizeof extmac,
2627 error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
2635 error = ENOPROTOOPT;
2638 if (error == 0 && so->so_proto->pr_ctloutput != NULL)
2639 (void)(*so->so_proto->pr_ctloutput)(so, sopt);
2647 * Helper routine for getsockopt.
2650 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
2658 * Documented get behavior is that we always return a value, possibly
2659 * truncated to fit in the user's buffer. Traditional behavior is
2660 * that we always tell the user precisely how much we copied, rather
2661 * than something useful like the total amount we had available for
2662 * her. Note that this interface is not idempotent; the entire
2663 * answer must generated ahead of time.
2665 valsize = min(len, sopt->sopt_valsize);
2666 sopt->sopt_valsize = valsize;
2667 if (sopt->sopt_val != NULL) {
2668 if (sopt->sopt_td != NULL)
2669 error = copyout(buf, sopt->sopt_val, valsize);
2671 bcopy(buf, sopt->sopt_val, valsize);
2677 sogetopt(struct socket *so, struct sockopt *sopt)
2686 CURVNET_SET(so->so_vnet);
2688 if (sopt->sopt_level != SOL_SOCKET) {
2689 if (so->so_proto->pr_ctloutput != NULL)
2690 error = (*so->so_proto->pr_ctloutput)(so, sopt);
2692 error = ENOPROTOOPT;
2696 switch (sopt->sopt_name) {
2698 case SO_ACCEPTFILTER:
2699 error = do_getopt_accept_filter(so, sopt);
2704 l.l_onoff = so->so_options & SO_LINGER;
2705 l.l_linger = so->so_linger;
2707 error = sooptcopyout(sopt, &l, sizeof l);
2710 case SO_USELOOPBACK:
2722 optval = so->so_options & sopt->sopt_name;
2724 error = sooptcopyout(sopt, &optval, sizeof optval);
2728 optval = so->so_type;
2732 optval = so->so_proto->pr_protocol;
2737 optval = so->so_error;
2743 optval = so->so_snd.sb_hiwat;
2747 optval = so->so_rcv.sb_hiwat;
2751 optval = so->so_snd.sb_lowat;
2755 optval = so->so_rcv.sb_lowat;
2760 optval = (sopt->sopt_name == SO_SNDTIMEO ?
2761 so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
2763 tv.tv_sec = optval / hz;
2764 tv.tv_usec = (optval % hz) * tick;
2765 #ifdef COMPAT_FREEBSD32
2766 if (SV_CURPROC_FLAG(SV_ILP32)) {
2767 struct timeval32 tv32;
2769 CP(tv, tv32, tv_sec);
2770 CP(tv, tv32, tv_usec);
2771 error = sooptcopyout(sopt, &tv32, sizeof tv32);
2774 error = sooptcopyout(sopt, &tv, sizeof tv);
2779 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
2783 error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
2787 error = sooptcopyout(sopt, &extmac, sizeof extmac);
2795 error = sooptcopyin(sopt, &extmac, sizeof(extmac),
2799 error = mac_getsockopt_peerlabel(
2800 sopt->sopt_td->td_ucred, so, &extmac);
2803 error = sooptcopyout(sopt, &extmac, sizeof extmac);
2809 case SO_LISTENQLIMIT:
2810 optval = so->so_qlimit;
2814 optval = so->so_qlen;
2817 case SO_LISTENINCQLEN:
2818 optval = so->so_incqlen;
2822 error = ENOPROTOOPT;
2833 /* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */
2835 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
2837 struct mbuf *m, *m_prev;
2838 int sopt_size = sopt->sopt_valsize;
2840 MGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT, MT_DATA);
2843 if (sopt_size > MLEN) {
2844 MCLGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT);
2845 if ((m->m_flags & M_EXT) == 0) {
2849 m->m_len = min(MCLBYTES, sopt_size);
2851 m->m_len = min(MLEN, sopt_size);
2853 sopt_size -= m->m_len;
2858 MGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT, MT_DATA);
2863 if (sopt_size > MLEN) {
2864 MCLGET(m, sopt->sopt_td != NULL ? M_WAIT :
2866 if ((m->m_flags & M_EXT) == 0) {
2871 m->m_len = min(MCLBYTES, sopt_size);
2873 m->m_len = min(MLEN, sopt_size);
2875 sopt_size -= m->m_len;
2882 /* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */
2884 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
2886 struct mbuf *m0 = m;
2888 if (sopt->sopt_val == NULL)
2890 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
2891 if (sopt->sopt_td != NULL) {
2894 error = copyin(sopt->sopt_val, mtod(m, char *),
2901 bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
2902 sopt->sopt_valsize -= m->m_len;
2903 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
2906 if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
2907 panic("ip6_sooptmcopyin");
2911 /* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */
2913 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
2915 struct mbuf *m0 = m;
2918 if (sopt->sopt_val == NULL)
2920 while (m != NULL && sopt->sopt_valsize >= m->m_len) {
2921 if (sopt->sopt_td != NULL) {
2924 error = copyout(mtod(m, char *), sopt->sopt_val,
2931 bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
2932 sopt->sopt_valsize -= m->m_len;
2933 sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
2934 valsize += m->m_len;
2938 /* enough soopt buffer should be given from user-land */
2942 sopt->sopt_valsize = valsize;
2947 * sohasoutofband(): protocol notifies socket layer of the arrival of new
2948 * out-of-band data, which will then notify socket consumers.
2951 sohasoutofband(struct socket *so)
2954 if (so->so_sigio != NULL)
2955 pgsigio(&so->so_sigio, SIGURG, 0);
2956 selwakeuppri(&so->so_rcv.sb_sel, PSOCK);
2960 sopoll(struct socket *so, int events, struct ucred *active_cred,
2965 * We do not need to set or assert curvnet as long as everyone uses
2968 return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred,
2973 sopoll_generic(struct socket *so, int events, struct ucred *active_cred,
2978 SOCKBUF_LOCK(&so->so_snd);
2979 SOCKBUF_LOCK(&so->so_rcv);
2980 if (events & (POLLIN | POLLRDNORM))
2981 if (soreadabledata(so))
2982 revents |= events & (POLLIN | POLLRDNORM);
2984 if (events & (POLLOUT | POLLWRNORM))
2985 if (sowriteable(so))
2986 revents |= events & (POLLOUT | POLLWRNORM);
2988 if (events & (POLLPRI | POLLRDBAND))
2989 if (so->so_oobmark || (so->so_rcv.sb_state & SBS_RCVATMARK))
2990 revents |= events & (POLLPRI | POLLRDBAND);
2992 if ((events & POLLINIGNEOF) == 0) {
2993 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2994 revents |= events & (POLLIN | POLLRDNORM);
2995 if (so->so_snd.sb_state & SBS_CANTSENDMORE)
3001 if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
3002 selrecord(td, &so->so_rcv.sb_sel);
3003 so->so_rcv.sb_flags |= SB_SEL;
3006 if (events & (POLLOUT | POLLWRNORM)) {
3007 selrecord(td, &so->so_snd.sb_sel);
3008 so->so_snd.sb_flags |= SB_SEL;
3012 SOCKBUF_UNLOCK(&so->so_rcv);
3013 SOCKBUF_UNLOCK(&so->so_snd);
3018 soo_kqfilter(struct file *fp, struct knote *kn)
3020 struct socket *so = kn->kn_fp->f_data;
3023 switch (kn->kn_filter) {
3025 if (so->so_options & SO_ACCEPTCONN)
3026 kn->kn_fop = &solisten_filtops;
3028 kn->kn_fop = &soread_filtops;
3032 kn->kn_fop = &sowrite_filtops;
3040 knlist_add(&sb->sb_sel.si_note, kn, 1);
3041 sb->sb_flags |= SB_KNOTE;
3047 * Some routines that return EOPNOTSUPP for entry points that are not
3048 * supported by a protocol. Fill in as needed.
3051 pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
3058 pru_attach_notsupp(struct socket *so, int proto, struct thread *td)
3065 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3072 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3079 pru_connect2_notsupp(struct socket *so1, struct socket *so2)
3086 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
3087 struct ifnet *ifp, struct thread *td)
3094 pru_disconnect_notsupp(struct socket *so)
3101 pru_listen_notsupp(struct socket *so, int backlog, struct thread *td)
3108 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam)
3115 pru_rcvd_notsupp(struct socket *so, int flags)
3122 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
3129 pru_send_notsupp(struct socket *so, int flags, struct mbuf *m,
3130 struct sockaddr *addr, struct mbuf *control, struct thread *td)
3137 * This isn't really a ``null'' operation, but it's the default one and
3138 * doesn't do anything destructive.
3141 pru_sense_null(struct socket *so, struct stat *sb)
3144 sb->st_blksize = so->so_snd.sb_hiwat;
3149 pru_shutdown_notsupp(struct socket *so)
3156 pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam)
3163 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio,
3164 struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
3171 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr,
3172 struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
3179 pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred,
3187 filt_sordetach(struct knote *kn)
3189 struct socket *so = kn->kn_fp->f_data;
3191 SOCKBUF_LOCK(&so->so_rcv);
3192 knlist_remove(&so->so_rcv.sb_sel.si_note, kn, 1);
3193 if (knlist_empty(&so->so_rcv.sb_sel.si_note))
3194 so->so_rcv.sb_flags &= ~SB_KNOTE;
3195 SOCKBUF_UNLOCK(&so->so_rcv);
3200 filt_soread(struct knote *kn, long hint)
3204 so = kn->kn_fp->f_data;
3205 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
3207 kn->kn_data = so->so_rcv.sb_cc - so->so_rcv.sb_ctl;
3208 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3209 kn->kn_flags |= EV_EOF;
3210 kn->kn_fflags = so->so_error;
3212 } else if (so->so_error) /* temporary udp error */
3214 else if (kn->kn_sfflags & NOTE_LOWAT)
3215 return (kn->kn_data >= kn->kn_sdata);
3217 return (so->so_rcv.sb_cc >= so->so_rcv.sb_lowat);
3221 filt_sowdetach(struct knote *kn)
3223 struct socket *so = kn->kn_fp->f_data;
3225 SOCKBUF_LOCK(&so->so_snd);
3226 knlist_remove(&so->so_snd.sb_sel.si_note, kn, 1);
3227 if (knlist_empty(&so->so_snd.sb_sel.si_note))
3228 so->so_snd.sb_flags &= ~SB_KNOTE;
3229 SOCKBUF_UNLOCK(&so->so_snd);
3234 filt_sowrite(struct knote *kn, long hint)
3238 so = kn->kn_fp->f_data;
3239 SOCKBUF_LOCK_ASSERT(&so->so_snd);
3240 kn->kn_data = sbspace(&so->so_snd);
3241 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
3242 kn->kn_flags |= EV_EOF;
3243 kn->kn_fflags = so->so_error;
3245 } else if (so->so_error) /* temporary udp error */
3247 else if (((so->so_state & SS_ISCONNECTED) == 0) &&
3248 (so->so_proto->pr_flags & PR_CONNREQUIRED))
3250 else if (kn->kn_sfflags & NOTE_LOWAT)
3251 return (kn->kn_data >= kn->kn_sdata);
3253 return (kn->kn_data >= so->so_snd.sb_lowat);
3258 filt_solisten(struct knote *kn, long hint)
3260 struct socket *so = kn->kn_fp->f_data;
3262 kn->kn_data = so->so_qlen;
3263 return (! TAILQ_EMPTY(&so->so_comp));
3267 socheckuid(struct socket *so, uid_t uid)
3272 if (so->so_cred->cr_uid != uid)
3278 sysctl_somaxconn(SYSCTL_HANDLER_ARGS)
3284 error = sysctl_handle_int(oidp, &val, 0, req);
3285 if (error || !req->newptr )
3288 if (val < 1 || val > USHRT_MAX)
3296 * These functions are used by protocols to notify the socket layer (and its
3297 * consumers) of state changes in the sockets driven by protocol-side events.
3301 * Procedures to manipulate state flags of socket and do appropriate wakeups.
3303 * Normal sequence from the active (originating) side is that
3304 * soisconnecting() is called during processing of connect() call, resulting
3305 * in an eventual call to soisconnected() if/when the connection is
3306 * established. When the connection is torn down soisdisconnecting() is
3307 * called during processing of disconnect() call, and soisdisconnected() is
3308 * called when the connection to the peer is totally severed. The semantics
3309 * of these routines are such that connectionless protocols can call
3310 * soisconnected() and soisdisconnected() only, bypassing the in-progress
3311 * calls when setting up a ``connection'' takes no time.
3313 * From the passive side, a socket is created with two queues of sockets:
3314 * so_incomp for connections in progress and so_comp for connections already
3315 * made and awaiting user acceptance. As a protocol is preparing incoming
3316 * connections, it creates a socket structure queued on so_incomp by calling
3317 * sonewconn(). When the connection is established, soisconnected() is
3318 * called, and transfers the socket structure to so_comp, making it available
3321 * If a socket is closed with sockets on either so_incomp or so_comp, these
3322 * sockets are dropped.
3324 * If higher-level protocols are implemented in the kernel, the wakeups done
3325 * here will sometimes cause software-interrupt process scheduling.
3328 soisconnecting(struct socket *so)
3332 so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
3333 so->so_state |= SS_ISCONNECTING;
3338 soisconnected(struct socket *so)
3340 struct socket *head;
3346 so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
3347 so->so_state |= SS_ISCONNECTED;
3349 if (head != NULL && (so->so_qstate & SQ_INCOMP)) {
3350 if ((so->so_options & SO_ACCEPTFILTER) == 0) {
3352 TAILQ_REMOVE(&head->so_incomp, so, so_list);
3354 so->so_qstate &= ~SQ_INCOMP;
3355 TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
3357 so->so_qstate |= SQ_COMP;
3360 wakeup_one(&head->so_timeo);
3363 soupcall_set(so, SO_RCV,
3364 head->so_accf->so_accept_filter->accf_callback,
3365 head->so_accf->so_accept_filter_arg);
3366 so->so_options &= ~SO_ACCEPTFILTER;
3367 ret = head->so_accf->so_accept_filter->accf_callback(so,
3368 head->so_accf->so_accept_filter_arg, M_DONTWAIT);
3369 if (ret == SU_ISCONNECTED)
3370 soupcall_clear(so, SO_RCV);
3372 if (ret == SU_ISCONNECTED)
3379 wakeup(&so->so_timeo);
3385 soisdisconnecting(struct socket *so)
3389 * Note: This code assumes that SOCK_LOCK(so) and
3390 * SOCKBUF_LOCK(&so->so_rcv) are the same.
3392 SOCKBUF_LOCK(&so->so_rcv);
3393 so->so_state &= ~SS_ISCONNECTING;
3394 so->so_state |= SS_ISDISCONNECTING;
3395 so->so_rcv.sb_state |= SBS_CANTRCVMORE;
3396 sorwakeup_locked(so);
3397 SOCKBUF_LOCK(&so->so_snd);
3398 so->so_snd.sb_state |= SBS_CANTSENDMORE;
3399 sowwakeup_locked(so);
3400 wakeup(&so->so_timeo);
3404 soisdisconnected(struct socket *so)
3408 * Note: This code assumes that SOCK_LOCK(so) and
3409 * SOCKBUF_LOCK(&so->so_rcv) are the same.
3411 SOCKBUF_LOCK(&so->so_rcv);
3412 so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
3413 so->so_state |= SS_ISDISCONNECTED;
3414 so->so_rcv.sb_state |= SBS_CANTRCVMORE;
3415 sorwakeup_locked(so);
3416 SOCKBUF_LOCK(&so->so_snd);
3417 so->so_snd.sb_state |= SBS_CANTSENDMORE;
3418 sbdrop_locked(&so->so_snd, so->so_snd.sb_cc);
3419 sowwakeup_locked(so);
3420 wakeup(&so->so_timeo);
3424 * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
3427 sodupsockaddr(const struct sockaddr *sa, int mflags)
3429 struct sockaddr *sa2;
3431 sa2 = malloc(sa->sa_len, M_SONAME, mflags);
3433 bcopy(sa, sa2, sa->sa_len);
3438 * Register per-socket buffer upcalls.
3441 soupcall_set(struct socket *so, int which,
3442 int (*func)(struct socket *, void *, int), void *arg)
3454 panic("soupcall_set: bad which");
3456 SOCKBUF_LOCK_ASSERT(sb);
3458 /* XXX: accf_http actually wants to do this on purpose. */
3459 KASSERT(sb->sb_upcall == NULL, ("soupcall_set: overwriting upcall"));
3461 sb->sb_upcall = func;
3462 sb->sb_upcallarg = arg;
3463 sb->sb_flags |= SB_UPCALL;
3467 soupcall_clear(struct socket *so, int which)
3479 panic("soupcall_clear: bad which");
3481 SOCKBUF_LOCK_ASSERT(sb);
3482 KASSERT(sb->sb_upcall != NULL, ("soupcall_clear: no upcall to clear"));
3483 sb->sb_upcall = NULL;
3484 sb->sb_upcallarg = NULL;
3485 sb->sb_flags &= ~SB_UPCALL;
3489 * Create an external-format (``xsocket'') structure using the information in
3490 * the kernel-format socket structure pointed to by so. This is done to
3491 * reduce the spew of irrelevant information over this interface, to isolate
3492 * user code from changes in the kernel structure, and potentially to provide
3493 * information-hiding if we decide that some of this information should be
3494 * hidden from users.
3497 sotoxsocket(struct socket *so, struct xsocket *xso)
3500 xso->xso_len = sizeof *xso;
3502 xso->so_type = so->so_type;
3503 xso->so_options = so->so_options;
3504 xso->so_linger = so->so_linger;
3505 xso->so_state = so->so_state;
3506 xso->so_pcb = so->so_pcb;
3507 xso->xso_protocol = so->so_proto->pr_protocol;
3508 xso->xso_family = so->so_proto->pr_domain->dom_family;
3509 xso->so_qlen = so->so_qlen;
3510 xso->so_incqlen = so->so_incqlen;
3511 xso->so_qlimit = so->so_qlimit;
3512 xso->so_timeo = so->so_timeo;
3513 xso->so_error = so->so_error;
3514 xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
3515 xso->so_oobmark = so->so_oobmark;
3516 sbtoxsockbuf(&so->so_snd, &xso->so_snd);
3517 sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
3518 xso->so_uid = so->so_cred->cr_uid;
3523 * Socket accessor functions to provide external consumers with
3524 * a safe interface to socket state
3529 so_listeners_apply_all(struct socket *so, void (*func)(struct socket *, void *), void *arg)
3532 TAILQ_FOREACH(so, &so->so_comp, so_list)
3537 so_sockbuf_rcv(struct socket *so)
3540 return (&so->so_rcv);
3544 so_sockbuf_snd(struct socket *so)
3547 return (&so->so_snd);
3551 so_state_get(const struct socket *so)
3554 return (so->so_state);
3558 so_state_set(struct socket *so, int val)
3565 so_options_get(const struct socket *so)
3568 return (so->so_options);
3572 so_options_set(struct socket *so, int val)
3575 so->so_options = val;
3579 so_error_get(const struct socket *so)
3582 return (so->so_error);
3586 so_error_set(struct socket *so, int val)
3593 so_linger_get(const struct socket *so)
3596 return (so->so_linger);
3600 so_linger_set(struct socket *so, int val)
3603 so->so_linger = val;
3607 so_protosw_get(const struct socket *so)
3610 return (so->so_proto);
3614 so_protosw_set(struct socket *so, struct protosw *val)
3621 so_sorwakeup(struct socket *so)
3628 so_sowwakeup(struct socket *so)
3635 so_sorwakeup_locked(struct socket *so)
3638 sorwakeup_locked(so);
3642 so_sowwakeup_locked(struct socket *so)
3645 sowwakeup_locked(so);
3649 so_lock(struct socket *so)
3655 so_unlock(struct socket *so)