xref: /freebsd/sys/kern/uipc_socket.c (revision 8ef24a0d4b28fe230e20637f56869cc4148cd2ca)
1 /*-
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
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
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.
19  *
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
30  * SUCH DAMAGE.
31  *
32  *	@(#)uipc_socket.c	8.3 (Berkeley) 4/15/94
33  */
34 
35 /*
36  * Comments on the socket life cycle:
37  *
38  * soalloc() sets of socket layer state for a socket, called only by
39  * socreate() and sonewconn().  Socket layer private.
40  *
41  * sodealloc() tears down socket layer state for a socket, called only by
42  * sofree() and sonewconn().  Socket layer private.
43  *
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.
47  *
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.
52  *
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
57  * case.
58  *
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
61  * sockets.
62  *
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
66  * listen socket.
67  *
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.
71  *
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.
82  *
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
86  * interface.
87  *
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.
95  *
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.
101  */
102 
103 #include <sys/cdefs.h>
104 __FBSDID("$FreeBSD$");
105 
106 #include "opt_inet.h"
107 #include "opt_inet6.h"
108 #include "opt_compat.h"
109 
110 #include <sys/param.h>
111 #include <sys/systm.h>
112 #include <sys/fcntl.h>
113 #include <sys/limits.h>
114 #include <sys/lock.h>
115 #include <sys/mac.h>
116 #include <sys/malloc.h>
117 #include <sys/mbuf.h>
118 #include <sys/mutex.h>
119 #include <sys/domain.h>
120 #include <sys/file.h>			/* for struct knote */
121 #include <sys/hhook.h>
122 #include <sys/kernel.h>
123 #include <sys/khelp.h>
124 #include <sys/event.h>
125 #include <sys/eventhandler.h>
126 #include <sys/poll.h>
127 #include <sys/proc.h>
128 #include <sys/protosw.h>
129 #include <sys/socket.h>
130 #include <sys/socketvar.h>
131 #include <sys/resourcevar.h>
132 #include <net/route.h>
133 #include <sys/signalvar.h>
134 #include <sys/stat.h>
135 #include <sys/sx.h>
136 #include <sys/sysctl.h>
137 #include <sys/taskqueue.h>
138 #include <sys/uio.h>
139 #include <sys/jail.h>
140 #include <sys/syslog.h>
141 #include <netinet/in.h>
142 
143 #include <net/vnet.h>
144 
145 #include <security/mac/mac_framework.h>
146 
147 #include <vm/uma.h>
148 
149 #ifdef COMPAT_FREEBSD32
150 #include <sys/mount.h>
151 #include <sys/sysent.h>
152 #include <compat/freebsd32/freebsd32.h>
153 #endif
154 
155 static int	soreceive_rcvoob(struct socket *so, struct uio *uio,
156 		    int flags);
157 
158 static void	filt_sordetach(struct knote *kn);
159 static int	filt_soread(struct knote *kn, long hint);
160 static void	filt_sowdetach(struct knote *kn);
161 static int	filt_sowrite(struct knote *kn, long hint);
162 static int	filt_solisten(struct knote *kn, long hint);
163 static int inline hhook_run_socket(struct socket *so, void *hctx, int32_t h_id);
164 fo_kqfilter_t	soo_kqfilter;
165 
166 static struct filterops solisten_filtops = {
167 	.f_isfd = 1,
168 	.f_detach = filt_sordetach,
169 	.f_event = filt_solisten,
170 };
171 static struct filterops soread_filtops = {
172 	.f_isfd = 1,
173 	.f_detach = filt_sordetach,
174 	.f_event = filt_soread,
175 };
176 static struct filterops sowrite_filtops = {
177 	.f_isfd = 1,
178 	.f_detach = filt_sowdetach,
179 	.f_event = filt_sowrite,
180 };
181 
182 so_gen_t	so_gencnt;	/* generation count for sockets */
183 
184 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
185 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
186 
187 #define	VNET_SO_ASSERT(so)						\
188 	VNET_ASSERT(curvnet != NULL,					\
189 	    ("%s:%d curvnet is NULL, so=%p", __func__, __LINE__, (so)));
190 
191 VNET_DEFINE(struct hhook_head *, socket_hhh[HHOOK_SOCKET_LAST + 1]);
192 #define	V_socket_hhh		VNET(socket_hhh)
193 
194 /*
195  * Limit on the number of connections in the listen queue waiting
196  * for accept(2).
197  * NB: The original sysctl somaxconn is still available but hidden
198  * to prevent confusion about the actual purpose of this number.
199  */
200 static u_int somaxconn = SOMAXCONN;
201 
202 static int
203 sysctl_somaxconn(SYSCTL_HANDLER_ARGS)
204 {
205 	int error;
206 	int val;
207 
208 	val = somaxconn;
209 	error = sysctl_handle_int(oidp, &val, 0, req);
210 	if (error || !req->newptr )
211 		return (error);
212 
213 	/*
214 	 * The purpose of the UINT_MAX / 3 limit, is so that the formula
215 	 *   3 * so_qlimit / 2
216 	 * below, will not overflow.
217          */
218 
219 	if (val < 1 || val > UINT_MAX / 3)
220 		return (EINVAL);
221 
222 	somaxconn = val;
223 	return (0);
224 }
225 SYSCTL_PROC(_kern_ipc, OID_AUTO, soacceptqueue, CTLTYPE_UINT | CTLFLAG_RW,
226     0, sizeof(int), sysctl_somaxconn, "I",
227     "Maximum listen socket pending connection accept queue size");
228 SYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn,
229     CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_SKIP,
230     0, sizeof(int), sysctl_somaxconn, "I",
231     "Maximum listen socket pending connection accept queue size (compat)");
232 
233 static int numopensockets;
234 SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD,
235     &numopensockets, 0, "Number of open sockets");
236 
237 /*
238  * accept_mtx locks down per-socket fields relating to accept queues.  See
239  * socketvar.h for an annotation of the protected fields of struct socket.
240  */
241 struct mtx accept_mtx;
242 MTX_SYSINIT(accept_mtx, &accept_mtx, "accept", MTX_DEF);
243 
244 /*
245  * so_global_mtx protects so_gencnt, numopensockets, and the per-socket
246  * so_gencnt field.
247  */
248 static struct mtx so_global_mtx;
249 MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF);
250 
251 /*
252  * General IPC sysctl name space, used by sockets and a variety of other IPC
253  * types.
254  */
255 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC");
256 
257 /*
258  * Initialize the socket subsystem and set up the socket
259  * memory allocator.
260  */
261 static uma_zone_t socket_zone;
262 int	maxsockets;
263 
264 static void
265 socket_zone_change(void *tag)
266 {
267 
268 	maxsockets = uma_zone_set_max(socket_zone, maxsockets);
269 }
270 
271 static void
272 socket_hhook_register(int subtype)
273 {
274 
275 	if (hhook_head_register(HHOOK_TYPE_SOCKET, subtype,
276 	    &V_socket_hhh[subtype],
277 	    HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
278 		printf("%s: WARNING: unable to register hook\n", __func__);
279 }
280 
281 static void
282 socket_hhook_deregister(int subtype)
283 {
284 
285 	if (hhook_head_deregister(V_socket_hhh[subtype]) != 0)
286 		printf("%s: WARNING: unable to deregister hook\n", __func__);
287 }
288 
289 static void
290 socket_init(void *tag)
291 {
292 
293 	socket_zone = uma_zcreate("socket", sizeof(struct socket), NULL, NULL,
294 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
295 	maxsockets = uma_zone_set_max(socket_zone, maxsockets);
296 	uma_zone_set_warning(socket_zone, "kern.ipc.maxsockets limit reached");
297 	EVENTHANDLER_REGISTER(maxsockets_change, socket_zone_change, NULL,
298 	    EVENTHANDLER_PRI_FIRST);
299 }
300 SYSINIT(socket, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, socket_init, NULL);
301 
302 static void
303 socket_vnet_init(const void *unused __unused)
304 {
305 	int i;
306 
307 	/* We expect a contiguous range */
308 	for (i = 0; i <= HHOOK_SOCKET_LAST; i++)
309 		socket_hhook_register(i);
310 }
311 VNET_SYSINIT(socket_vnet_init, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
312     socket_vnet_init, NULL);
313 
314 static void
315 socket_vnet_uninit(const void *unused __unused)
316 {
317 	int i;
318 
319 	for (i = 0; i <= HHOOK_SOCKET_LAST; i++)
320 		socket_hhook_deregister(i);
321 }
322 VNET_SYSUNINIT(socket_vnet_uninit, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
323     socket_vnet_uninit, NULL);
324 
325 /*
326  * Initialise maxsockets.  This SYSINIT must be run after
327  * tunable_mbinit().
328  */
329 static void
330 init_maxsockets(void *ignored)
331 {
332 
333 	TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
334 	maxsockets = imax(maxsockets, maxfiles);
335 }
336 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL);
337 
338 /*
339  * Sysctl to get and set the maximum global sockets limit.  Notify protocols
340  * of the change so that they can update their dependent limits as required.
341  */
342 static int
343 sysctl_maxsockets(SYSCTL_HANDLER_ARGS)
344 {
345 	int error, newmaxsockets;
346 
347 	newmaxsockets = maxsockets;
348 	error = sysctl_handle_int(oidp, &newmaxsockets, 0, req);
349 	if (error == 0 && req->newptr) {
350 		if (newmaxsockets > maxsockets &&
351 		    newmaxsockets <= maxfiles) {
352 			maxsockets = newmaxsockets;
353 			EVENTHANDLER_INVOKE(maxsockets_change);
354 		} else
355 			error = EINVAL;
356 	}
357 	return (error);
358 }
359 SYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets, CTLTYPE_INT|CTLFLAG_RW,
360     &maxsockets, 0, sysctl_maxsockets, "IU",
361     "Maximum number of sockets available");
362 
363 /*
364  * Socket operation routines.  These routines are called by the routines in
365  * sys_socket.c or from a system process, and implement the semantics of
366  * socket operations by switching out to the protocol specific routines.
367  */
368 
369 /*
370  * Get a socket structure from our zone, and initialize it.  Note that it
371  * would probably be better to allocate socket and PCB at the same time, but
372  * I'm not convinced that all the protocols can be easily modified to do
373  * this.
374  *
375  * soalloc() returns a socket with a ref count of 0.
376  */
377 static struct socket *
378 soalloc(struct vnet *vnet)
379 {
380 	struct socket *so;
381 
382 	so = uma_zalloc(socket_zone, M_NOWAIT | M_ZERO);
383 	if (so == NULL)
384 		return (NULL);
385 #ifdef MAC
386 	if (mac_socket_init(so, M_NOWAIT) != 0) {
387 		uma_zfree(socket_zone, so);
388 		return (NULL);
389 	}
390 #endif
391 	if (khelp_init_osd(HELPER_CLASS_SOCKET, &so->osd)) {
392 		uma_zfree(socket_zone, so);
393 		return (NULL);
394 	}
395 
396 	SOCKBUF_LOCK_INIT(&so->so_snd, "so_snd");
397 	SOCKBUF_LOCK_INIT(&so->so_rcv, "so_rcv");
398 	sx_init(&so->so_snd.sb_sx, "so_snd_sx");
399 	sx_init(&so->so_rcv.sb_sx, "so_rcv_sx");
400 	TAILQ_INIT(&so->so_snd.sb_aiojobq);
401 	TAILQ_INIT(&so->so_rcv.sb_aiojobq);
402 	TASK_INIT(&so->so_snd.sb_aiotask, 0, soaio_snd, so);
403 	TASK_INIT(&so->so_rcv.sb_aiotask, 0, soaio_rcv, so);
404 #ifdef VIMAGE
405 	VNET_ASSERT(vnet != NULL, ("%s:%d vnet is NULL, so=%p",
406 	    __func__, __LINE__, so));
407 	so->so_vnet = vnet;
408 #endif
409 	/* We shouldn't need the so_global_mtx */
410 	if (hhook_run_socket(so, NULL, HHOOK_SOCKET_CREATE)) {
411 		/* Do we need more comprehensive error returns? */
412 		uma_zfree(socket_zone, so);
413 		return (NULL);
414 	}
415 	mtx_lock(&so_global_mtx);
416 	so->so_gencnt = ++so_gencnt;
417 	++numopensockets;
418 #ifdef VIMAGE
419 	vnet->vnet_sockcnt++;
420 #endif
421 	mtx_unlock(&so_global_mtx);
422 
423 	return (so);
424 }
425 
426 /*
427  * Free the storage associated with a socket at the socket layer, tear down
428  * locks, labels, etc.  All protocol state is assumed already to have been
429  * torn down (and possibly never set up) by the caller.
430  */
431 static void
432 sodealloc(struct socket *so)
433 {
434 
435 	KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count));
436 	KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL"));
437 
438 	mtx_lock(&so_global_mtx);
439 	so->so_gencnt = ++so_gencnt;
440 	--numopensockets;	/* Could be below, but faster here. */
441 #ifdef VIMAGE
442 	VNET_ASSERT(so->so_vnet != NULL, ("%s:%d so_vnet is NULL, so=%p",
443 	    __func__, __LINE__, so));
444 	so->so_vnet->vnet_sockcnt--;
445 #endif
446 	mtx_unlock(&so_global_mtx);
447 	if (so->so_rcv.sb_hiwat)
448 		(void)chgsbsize(so->so_cred->cr_uidinfo,
449 		    &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY);
450 	if (so->so_snd.sb_hiwat)
451 		(void)chgsbsize(so->so_cred->cr_uidinfo,
452 		    &so->so_snd.sb_hiwat, 0, RLIM_INFINITY);
453 	/* remove accept filter if one is present. */
454 	if (so->so_accf != NULL)
455 		do_setopt_accept_filter(so, NULL);
456 #ifdef MAC
457 	mac_socket_destroy(so);
458 #endif
459 	hhook_run_socket(so, NULL, HHOOK_SOCKET_CLOSE);
460 
461 	crfree(so->so_cred);
462 	khelp_destroy_osd(&so->osd);
463 	sx_destroy(&so->so_snd.sb_sx);
464 	sx_destroy(&so->so_rcv.sb_sx);
465 	SOCKBUF_LOCK_DESTROY(&so->so_snd);
466 	SOCKBUF_LOCK_DESTROY(&so->so_rcv);
467 	uma_zfree(socket_zone, so);
468 }
469 
470 /*
471  * socreate returns a socket with a ref count of 1.  The socket should be
472  * closed with soclose().
473  */
474 int
475 socreate(int dom, struct socket **aso, int type, int proto,
476     struct ucred *cred, struct thread *td)
477 {
478 	struct protosw *prp;
479 	struct socket *so;
480 	int error;
481 
482 	if (proto)
483 		prp = pffindproto(dom, proto, type);
484 	else
485 		prp = pffindtype(dom, type);
486 
487 	if (prp == NULL) {
488 		/* No support for domain. */
489 		if (pffinddomain(dom) == NULL)
490 			return (EAFNOSUPPORT);
491 		/* No support for socket type. */
492 		if (proto == 0 && type != 0)
493 			return (EPROTOTYPE);
494 		return (EPROTONOSUPPORT);
495 	}
496 	if (prp->pr_usrreqs->pru_attach == NULL ||
497 	    prp->pr_usrreqs->pru_attach == pru_attach_notsupp)
498 		return (EPROTONOSUPPORT);
499 
500 	if (prison_check_af(cred, prp->pr_domain->dom_family) != 0)
501 		return (EPROTONOSUPPORT);
502 
503 	if (prp->pr_type != type)
504 		return (EPROTOTYPE);
505 	so = soalloc(CRED_TO_VNET(cred));
506 	if (so == NULL)
507 		return (ENOBUFS);
508 
509 	TAILQ_INIT(&so->so_incomp);
510 	TAILQ_INIT(&so->so_comp);
511 	so->so_type = type;
512 	so->so_cred = crhold(cred);
513 	if ((prp->pr_domain->dom_family == PF_INET) ||
514 	    (prp->pr_domain->dom_family == PF_INET6) ||
515 	    (prp->pr_domain->dom_family == PF_ROUTE))
516 		so->so_fibnum = td->td_proc->p_fibnum;
517 	else
518 		so->so_fibnum = 0;
519 	so->so_proto = prp;
520 #ifdef MAC
521 	mac_socket_create(cred, so);
522 #endif
523 	knlist_init_mtx(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv));
524 	knlist_init_mtx(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd));
525 	so->so_count = 1;
526 	/*
527 	 * Auto-sizing of socket buffers is managed by the protocols and
528 	 * the appropriate flags must be set in the pru_attach function.
529 	 */
530 	CURVNET_SET(so->so_vnet);
531 	error = (*prp->pr_usrreqs->pru_attach)(so, proto, td);
532 	CURVNET_RESTORE();
533 	if (error) {
534 		KASSERT(so->so_count == 1, ("socreate: so_count %d",
535 		    so->so_count));
536 		so->so_count = 0;
537 		sodealloc(so);
538 		return (error);
539 	}
540 	*aso = so;
541 	return (0);
542 }
543 
544 #ifdef REGRESSION
545 static int regression_sonewconn_earlytest = 1;
546 SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW,
547     &regression_sonewconn_earlytest, 0, "Perform early sonewconn limit test");
548 #endif
549 
550 /*
551  * When an attempt at a new connection is noted on a socket which accepts
552  * connections, sonewconn is called.  If the connection is possible (subject
553  * to space constraints, etc.) then we allocate a new structure, propoerly
554  * linked into the data structure of the original socket, and return this.
555  * Connstatus may be 0, or SS_ISCONFIRMING, or SS_ISCONNECTED.
556  *
557  * Note: the ref count on the socket is 0 on return.
558  */
559 struct socket *
560 sonewconn(struct socket *head, int connstatus)
561 {
562 	static struct timeval lastover;
563 	static struct timeval overinterval = { 60, 0 };
564 	static int overcount;
565 
566 	struct socket *so;
567 	int over;
568 
569 	ACCEPT_LOCK();
570 	over = (head->so_qlen > 3 * head->so_qlimit / 2);
571 	ACCEPT_UNLOCK();
572 #ifdef REGRESSION
573 	if (regression_sonewconn_earlytest && over) {
574 #else
575 	if (over) {
576 #endif
577 		overcount++;
578 
579 		if (ratecheck(&lastover, &overinterval)) {
580 			log(LOG_DEBUG, "%s: pcb %p: Listen queue overflow: "
581 			    "%i already in queue awaiting acceptance "
582 			    "(%d occurrences)\n",
583 			    __func__, head->so_pcb, head->so_qlen, overcount);
584 
585 			overcount = 0;
586 		}
587 
588 		return (NULL);
589 	}
590 	VNET_ASSERT(head->so_vnet != NULL, ("%s:%d so_vnet is NULL, head=%p",
591 	    __func__, __LINE__, head));
592 	so = soalloc(head->so_vnet);
593 	if (so == NULL) {
594 		log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: "
595 		    "limit reached or out of memory\n",
596 		    __func__, head->so_pcb);
597 		return (NULL);
598 	}
599 	if ((head->so_options & SO_ACCEPTFILTER) != 0)
600 		connstatus = 0;
601 	so->so_head = head;
602 	so->so_type = head->so_type;
603 	so->so_options = head->so_options &~ SO_ACCEPTCONN;
604 	so->so_linger = head->so_linger;
605 	so->so_state = head->so_state | SS_NOFDREF;
606 	so->so_fibnum = head->so_fibnum;
607 	so->so_proto = head->so_proto;
608 	so->so_cred = crhold(head->so_cred);
609 #ifdef MAC
610 	mac_socket_newconn(head, so);
611 #endif
612 	knlist_init_mtx(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv));
613 	knlist_init_mtx(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd));
614 	VNET_SO_ASSERT(head);
615 	if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat)) {
616 		sodealloc(so);
617 		log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n",
618 		    __func__, head->so_pcb);
619 		return (NULL);
620 	}
621 	if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
622 		sodealloc(so);
623 		log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n",
624 		    __func__, head->so_pcb);
625 		return (NULL);
626 	}
627 	so->so_rcv.sb_lowat = head->so_rcv.sb_lowat;
628 	so->so_snd.sb_lowat = head->so_snd.sb_lowat;
629 	so->so_rcv.sb_timeo = head->so_rcv.sb_timeo;
630 	so->so_snd.sb_timeo = head->so_snd.sb_timeo;
631 	so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE;
632 	so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE;
633 	so->so_state |= connstatus;
634 	ACCEPT_LOCK();
635 	/*
636 	 * The accept socket may be tearing down but we just
637 	 * won a race on the ACCEPT_LOCK.
638 	 * However, if sctp_peeloff() is called on a 1-to-many
639 	 * style socket, the SO_ACCEPTCONN doesn't need to be set.
640 	 */
641 	if (!(head->so_options & SO_ACCEPTCONN) &&
642 	    ((head->so_proto->pr_protocol != IPPROTO_SCTP) ||
643 	     (head->so_type != SOCK_SEQPACKET))) {
644 		SOCK_LOCK(so);
645 		so->so_head = NULL;
646 		sofree(so);		/* NB: returns ACCEPT_UNLOCK'ed. */
647 		return (NULL);
648 	}
649 	if (connstatus) {
650 		TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
651 		so->so_qstate |= SQ_COMP;
652 		head->so_qlen++;
653 	} else {
654 		/*
655 		 * Keep removing sockets from the head until there's room for
656 		 * us to insert on the tail.  In pre-locking revisions, this
657 		 * was a simple if(), but as we could be racing with other
658 		 * threads and soabort() requires dropping locks, we must
659 		 * loop waiting for the condition to be true.
660 		 */
661 		while (head->so_incqlen > head->so_qlimit) {
662 			struct socket *sp;
663 			sp = TAILQ_FIRST(&head->so_incomp);
664 			TAILQ_REMOVE(&head->so_incomp, sp, so_list);
665 			head->so_incqlen--;
666 			sp->so_qstate &= ~SQ_INCOMP;
667 			sp->so_head = NULL;
668 			ACCEPT_UNLOCK();
669 			soabort(sp);
670 			ACCEPT_LOCK();
671 		}
672 		TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list);
673 		so->so_qstate |= SQ_INCOMP;
674 		head->so_incqlen++;
675 	}
676 	ACCEPT_UNLOCK();
677 	if (connstatus) {
678 		sorwakeup(head);
679 		wakeup_one(&head->so_timeo);
680 	}
681 	return (so);
682 }
683 
684 int
685 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
686 {
687 	int error;
688 
689 	CURVNET_SET(so->so_vnet);
690 	error = (*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td);
691 	CURVNET_RESTORE();
692 	return (error);
693 }
694 
695 int
696 sobindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
697 {
698 	int error;
699 
700 	CURVNET_SET(so->so_vnet);
701 	error = (*so->so_proto->pr_usrreqs->pru_bindat)(fd, so, nam, td);
702 	CURVNET_RESTORE();
703 	return (error);
704 }
705 
706 /*
707  * solisten() transitions a socket from a non-listening state to a listening
708  * state, but can also be used to update the listen queue depth on an
709  * existing listen socket.  The protocol will call back into the sockets
710  * layer using solisten_proto_check() and solisten_proto() to check and set
711  * socket-layer listen state.  Call backs are used so that the protocol can
712  * acquire both protocol and socket layer locks in whatever order is required
713  * by the protocol.
714  *
715  * Protocol implementors are advised to hold the socket lock across the
716  * socket-layer test and set to avoid races at the socket layer.
717  */
718 int
719 solisten(struct socket *so, int backlog, struct thread *td)
720 {
721 	int error;
722 
723 	CURVNET_SET(so->so_vnet);
724 	error = (*so->so_proto->pr_usrreqs->pru_listen)(so, backlog, td);
725 	CURVNET_RESTORE();
726 	return (error);
727 }
728 
729 int
730 solisten_proto_check(struct socket *so)
731 {
732 
733 	SOCK_LOCK_ASSERT(so);
734 
735 	if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
736 	    SS_ISDISCONNECTING))
737 		return (EINVAL);
738 	return (0);
739 }
740 
741 void
742 solisten_proto(struct socket *so, int backlog)
743 {
744 
745 	SOCK_LOCK_ASSERT(so);
746 
747 	if (backlog < 0 || backlog > somaxconn)
748 		backlog = somaxconn;
749 	so->so_qlimit = backlog;
750 	so->so_options |= SO_ACCEPTCONN;
751 }
752 
753 /*
754  * Evaluate the reference count and named references on a socket; if no
755  * references remain, free it.  This should be called whenever a reference is
756  * released, such as in sorele(), but also when named reference flags are
757  * cleared in socket or protocol code.
758  *
759  * sofree() will free the socket if:
760  *
761  * - There are no outstanding file descriptor references or related consumers
762  *   (so_count == 0).
763  *
764  * - The socket has been closed by user space, if ever open (SS_NOFDREF).
765  *
766  * - The protocol does not have an outstanding strong reference on the socket
767  *   (SS_PROTOREF).
768  *
769  * - The socket is not in a completed connection queue, so a process has been
770  *   notified that it is present.  If it is removed, the user process may
771  *   block in accept() despite select() saying the socket was ready.
772  */
773 void
774 sofree(struct socket *so)
775 {
776 	struct protosw *pr = so->so_proto;
777 	struct socket *head;
778 
779 	ACCEPT_LOCK_ASSERT();
780 	SOCK_LOCK_ASSERT(so);
781 
782 	if ((so->so_state & SS_NOFDREF) == 0 || so->so_count != 0 ||
783 	    (so->so_state & SS_PROTOREF) || (so->so_qstate & SQ_COMP)) {
784 		SOCK_UNLOCK(so);
785 		ACCEPT_UNLOCK();
786 		return;
787 	}
788 
789 	head = so->so_head;
790 	if (head != NULL) {
791 		KASSERT((so->so_qstate & SQ_COMP) != 0 ||
792 		    (so->so_qstate & SQ_INCOMP) != 0,
793 		    ("sofree: so_head != NULL, but neither SQ_COMP nor "
794 		    "SQ_INCOMP"));
795 		KASSERT((so->so_qstate & SQ_COMP) == 0 ||
796 		    (so->so_qstate & SQ_INCOMP) == 0,
797 		    ("sofree: so->so_qstate is SQ_COMP and also SQ_INCOMP"));
798 		TAILQ_REMOVE(&head->so_incomp, so, so_list);
799 		head->so_incqlen--;
800 		so->so_qstate &= ~SQ_INCOMP;
801 		so->so_head = NULL;
802 	}
803 	KASSERT((so->so_qstate & SQ_COMP) == 0 &&
804 	    (so->so_qstate & SQ_INCOMP) == 0,
805 	    ("sofree: so_head == NULL, but still SQ_COMP(%d) or SQ_INCOMP(%d)",
806 	    so->so_qstate & SQ_COMP, so->so_qstate & SQ_INCOMP));
807 	if (so->so_options & SO_ACCEPTCONN) {
808 		KASSERT((TAILQ_EMPTY(&so->so_comp)),
809 		    ("sofree: so_comp populated"));
810 		KASSERT((TAILQ_EMPTY(&so->so_incomp)),
811 		    ("sofree: so_incomp populated"));
812 	}
813 	SOCK_UNLOCK(so);
814 	ACCEPT_UNLOCK();
815 
816 	VNET_SO_ASSERT(so);
817 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
818 		(*pr->pr_domain->dom_dispose)(so);
819 	if (pr->pr_usrreqs->pru_detach != NULL)
820 		(*pr->pr_usrreqs->pru_detach)(so);
821 
822 	/*
823 	 * From this point on, we assume that no other references to this
824 	 * socket exist anywhere else in the stack.  Therefore, no locks need
825 	 * to be acquired or held.
826 	 *
827 	 * We used to do a lot of socket buffer and socket locking here, as
828 	 * well as invoke sorflush() and perform wakeups.  The direct call to
829 	 * dom_dispose() and sbrelease_internal() are an inlining of what was
830 	 * necessary from sorflush().
831 	 *
832 	 * Notice that the socket buffer and kqueue state are torn down
833 	 * before calling pru_detach.  This means that protocols shold not
834 	 * assume they can perform socket wakeups, etc, in their detach code.
835 	 */
836 	sbdestroy(&so->so_snd, so);
837 	sbdestroy(&so->so_rcv, so);
838 	seldrain(&so->so_snd.sb_sel);
839 	seldrain(&so->so_rcv.sb_sel);
840 	knlist_destroy(&so->so_rcv.sb_sel.si_note);
841 	knlist_destroy(&so->so_snd.sb_sel.si_note);
842 	sodealloc(so);
843 }
844 
845 /*
846  * Close a socket on last file table reference removal.  Initiate disconnect
847  * if connected.  Free socket when disconnect complete.
848  *
849  * This function will sorele() the socket.  Note that soclose() may be called
850  * prior to the ref count reaching zero.  The actual socket structure will
851  * not be freed until the ref count reaches zero.
852  */
853 int
854 soclose(struct socket *so)
855 {
856 	int error = 0;
857 
858 	KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter"));
859 
860 	CURVNET_SET(so->so_vnet);
861 	funsetown(&so->so_sigio);
862 	if (so->so_state & SS_ISCONNECTED) {
863 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
864 			error = sodisconnect(so);
865 			if (error) {
866 				if (error == ENOTCONN)
867 					error = 0;
868 				goto drop;
869 			}
870 		}
871 		if (so->so_options & SO_LINGER) {
872 			if ((so->so_state & SS_ISDISCONNECTING) &&
873 			    (so->so_state & SS_NBIO))
874 				goto drop;
875 			while (so->so_state & SS_ISCONNECTED) {
876 				error = tsleep(&so->so_timeo,
877 				    PSOCK | PCATCH, "soclos",
878 				    so->so_linger * hz);
879 				if (error)
880 					break;
881 			}
882 		}
883 	}
884 
885 drop:
886 	if (so->so_proto->pr_usrreqs->pru_close != NULL)
887 		(*so->so_proto->pr_usrreqs->pru_close)(so);
888 	ACCEPT_LOCK();
889 	if (so->so_options & SO_ACCEPTCONN) {
890 		struct socket *sp;
891 		/*
892 		 * Prevent new additions to the accept queues due
893 		 * to ACCEPT_LOCK races while we are draining them.
894 		 */
895 		so->so_options &= ~SO_ACCEPTCONN;
896 		while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
897 			TAILQ_REMOVE(&so->so_incomp, sp, so_list);
898 			so->so_incqlen--;
899 			sp->so_qstate &= ~SQ_INCOMP;
900 			sp->so_head = NULL;
901 			ACCEPT_UNLOCK();
902 			soabort(sp);
903 			ACCEPT_LOCK();
904 		}
905 		while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
906 			TAILQ_REMOVE(&so->so_comp, sp, so_list);
907 			so->so_qlen--;
908 			sp->so_qstate &= ~SQ_COMP;
909 			sp->so_head = NULL;
910 			ACCEPT_UNLOCK();
911 			soabort(sp);
912 			ACCEPT_LOCK();
913 		}
914 		KASSERT((TAILQ_EMPTY(&so->so_comp)),
915 		    ("%s: so_comp populated", __func__));
916 		KASSERT((TAILQ_EMPTY(&so->so_incomp)),
917 		    ("%s: so_incomp populated", __func__));
918 	}
919 	SOCK_LOCK(so);
920 	KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF"));
921 	so->so_state |= SS_NOFDREF;
922 	sorele(so);			/* NB: Returns with ACCEPT_UNLOCK(). */
923 	CURVNET_RESTORE();
924 	return (error);
925 }
926 
927 /*
928  * soabort() is used to abruptly tear down a connection, such as when a
929  * resource limit is reached (listen queue depth exceeded), or if a listen
930  * socket is closed while there are sockets waiting to be accepted.
931  *
932  * This interface is tricky, because it is called on an unreferenced socket,
933  * and must be called only by a thread that has actually removed the socket
934  * from the listen queue it was on, or races with other threads are risked.
935  *
936  * This interface will call into the protocol code, so must not be called
937  * with any socket locks held.  Protocols do call it while holding their own
938  * recursible protocol mutexes, but this is something that should be subject
939  * to review in the future.
940  */
941 void
942 soabort(struct socket *so)
943 {
944 
945 	/*
946 	 * In as much as is possible, assert that no references to this
947 	 * socket are held.  This is not quite the same as asserting that the
948 	 * current thread is responsible for arranging for no references, but
949 	 * is as close as we can get for now.
950 	 */
951 	KASSERT(so->so_count == 0, ("soabort: so_count"));
952 	KASSERT((so->so_state & SS_PROTOREF) == 0, ("soabort: SS_PROTOREF"));
953 	KASSERT(so->so_state & SS_NOFDREF, ("soabort: !SS_NOFDREF"));
954 	KASSERT((so->so_state & SQ_COMP) == 0, ("soabort: SQ_COMP"));
955 	KASSERT((so->so_state & SQ_INCOMP) == 0, ("soabort: SQ_INCOMP"));
956 	VNET_SO_ASSERT(so);
957 
958 	if (so->so_proto->pr_usrreqs->pru_abort != NULL)
959 		(*so->so_proto->pr_usrreqs->pru_abort)(so);
960 	ACCEPT_LOCK();
961 	SOCK_LOCK(so);
962 	sofree(so);
963 }
964 
965 int
966 soaccept(struct socket *so, struct sockaddr **nam)
967 {
968 	int error;
969 
970 	SOCK_LOCK(so);
971 	KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF"));
972 	so->so_state &= ~SS_NOFDREF;
973 	SOCK_UNLOCK(so);
974 
975 	CURVNET_SET(so->so_vnet);
976 	error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam);
977 	CURVNET_RESTORE();
978 	return (error);
979 }
980 
981 int
982 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
983 {
984 
985 	return (soconnectat(AT_FDCWD, so, nam, td));
986 }
987 
988 int
989 soconnectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
990 {
991 	int error;
992 
993 	if (so->so_options & SO_ACCEPTCONN)
994 		return (EOPNOTSUPP);
995 
996 	CURVNET_SET(so->so_vnet);
997 	/*
998 	 * If protocol is connection-based, can only connect once.
999 	 * Otherwise, if connected, try to disconnect first.  This allows
1000 	 * user to disconnect by connecting to, e.g., a null address.
1001 	 */
1002 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
1003 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
1004 	    (error = sodisconnect(so)))) {
1005 		error = EISCONN;
1006 	} else {
1007 		/*
1008 		 * Prevent accumulated error from previous connection from
1009 		 * biting us.
1010 		 */
1011 		so->so_error = 0;
1012 		if (fd == AT_FDCWD) {
1013 			error = (*so->so_proto->pr_usrreqs->pru_connect)(so,
1014 			    nam, td);
1015 		} else {
1016 			error = (*so->so_proto->pr_usrreqs->pru_connectat)(fd,
1017 			    so, nam, td);
1018 		}
1019 	}
1020 	CURVNET_RESTORE();
1021 
1022 	return (error);
1023 }
1024 
1025 int
1026 soconnect2(struct socket *so1, struct socket *so2)
1027 {
1028 	int error;
1029 
1030 	CURVNET_SET(so1->so_vnet);
1031 	error = (*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2);
1032 	CURVNET_RESTORE();
1033 	return (error);
1034 }
1035 
1036 int
1037 sodisconnect(struct socket *so)
1038 {
1039 	int error;
1040 
1041 	if ((so->so_state & SS_ISCONNECTED) == 0)
1042 		return (ENOTCONN);
1043 	if (so->so_state & SS_ISDISCONNECTING)
1044 		return (EALREADY);
1045 	VNET_SO_ASSERT(so);
1046 	error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so);
1047 	return (error);
1048 }
1049 
1050 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? 0 : SBL_WAIT)
1051 
1052 int
1053 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio,
1054     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1055 {
1056 	long space;
1057 	ssize_t resid;
1058 	int clen = 0, error, dontroute;
1059 
1060 	KASSERT(so->so_type == SOCK_DGRAM, ("sosend_dgram: !SOCK_DGRAM"));
1061 	KASSERT(so->so_proto->pr_flags & PR_ATOMIC,
1062 	    ("sosend_dgram: !PR_ATOMIC"));
1063 
1064 	if (uio != NULL)
1065 		resid = uio->uio_resid;
1066 	else
1067 		resid = top->m_pkthdr.len;
1068 	/*
1069 	 * In theory resid should be unsigned.  However, space must be
1070 	 * signed, as it might be less than 0 if we over-committed, and we
1071 	 * must use a signed comparison of space and resid.  On the other
1072 	 * hand, a negative resid causes us to loop sending 0-length
1073 	 * segments to the protocol.
1074 	 */
1075 	if (resid < 0) {
1076 		error = EINVAL;
1077 		goto out;
1078 	}
1079 
1080 	dontroute =
1081 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0;
1082 	if (td != NULL)
1083 		td->td_ru.ru_msgsnd++;
1084 	if (control != NULL)
1085 		clen = control->m_len;
1086 
1087 	SOCKBUF_LOCK(&so->so_snd);
1088 	if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1089 		SOCKBUF_UNLOCK(&so->so_snd);
1090 		error = EPIPE;
1091 		goto out;
1092 	}
1093 	if (so->so_error) {
1094 		error = so->so_error;
1095 		so->so_error = 0;
1096 		SOCKBUF_UNLOCK(&so->so_snd);
1097 		goto out;
1098 	}
1099 	if ((so->so_state & SS_ISCONNECTED) == 0) {
1100 		/*
1101 		 * `sendto' and `sendmsg' is allowed on a connection-based
1102 		 * socket if it supports implied connect.  Return ENOTCONN if
1103 		 * not connected and no address is supplied.
1104 		 */
1105 		if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1106 		    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1107 			if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1108 			    !(resid == 0 && clen != 0)) {
1109 				SOCKBUF_UNLOCK(&so->so_snd);
1110 				error = ENOTCONN;
1111 				goto out;
1112 			}
1113 		} else if (addr == NULL) {
1114 			if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1115 				error = ENOTCONN;
1116 			else
1117 				error = EDESTADDRREQ;
1118 			SOCKBUF_UNLOCK(&so->so_snd);
1119 			goto out;
1120 		}
1121 	}
1122 
1123 	/*
1124 	 * Do we need MSG_OOB support in SOCK_DGRAM?  Signs here may be a
1125 	 * problem and need fixing.
1126 	 */
1127 	space = sbspace(&so->so_snd);
1128 	if (flags & MSG_OOB)
1129 		space += 1024;
1130 	space -= clen;
1131 	SOCKBUF_UNLOCK(&so->so_snd);
1132 	if (resid > space) {
1133 		error = EMSGSIZE;
1134 		goto out;
1135 	}
1136 	if (uio == NULL) {
1137 		resid = 0;
1138 		if (flags & MSG_EOR)
1139 			top->m_flags |= M_EOR;
1140 	} else {
1141 		/*
1142 		 * Copy the data from userland into a mbuf chain.
1143 		 * If no data is to be copied in, a single empty mbuf
1144 		 * is returned.
1145 		 */
1146 		top = m_uiotombuf(uio, M_WAITOK, space, max_hdr,
1147 		    (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0)));
1148 		if (top == NULL) {
1149 			error = EFAULT;	/* only possible error */
1150 			goto out;
1151 		}
1152 		space -= resid - uio->uio_resid;
1153 		resid = uio->uio_resid;
1154 	}
1155 	KASSERT(resid == 0, ("sosend_dgram: resid != 0"));
1156 	/*
1157 	 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock
1158 	 * than with.
1159 	 */
1160 	if (dontroute) {
1161 		SOCK_LOCK(so);
1162 		so->so_options |= SO_DONTROUTE;
1163 		SOCK_UNLOCK(so);
1164 	}
1165 	/*
1166 	 * XXX all the SBS_CANTSENDMORE checks previously done could be out
1167 	 * of date.  We could have received a reset packet in an interrupt or
1168 	 * maybe we slept while doing page faults in uiomove() etc.  We could
1169 	 * probably recheck again inside the locking protection here, but
1170 	 * there are probably other places that this also happens.  We must
1171 	 * rethink this.
1172 	 */
1173 	VNET_SO_ASSERT(so);
1174 	error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1175 	    (flags & MSG_OOB) ? PRUS_OOB :
1176 	/*
1177 	 * If the user set MSG_EOF, the protocol understands this flag and
1178 	 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND.
1179 	 */
1180 	    ((flags & MSG_EOF) &&
1181 	     (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1182 	     (resid <= 0)) ?
1183 		PRUS_EOF :
1184 		/* If there is more to send set PRUS_MORETOCOME */
1185 		(resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1186 		top, addr, control, td);
1187 	if (dontroute) {
1188 		SOCK_LOCK(so);
1189 		so->so_options &= ~SO_DONTROUTE;
1190 		SOCK_UNLOCK(so);
1191 	}
1192 	clen = 0;
1193 	control = NULL;
1194 	top = NULL;
1195 out:
1196 	if (top != NULL)
1197 		m_freem(top);
1198 	if (control != NULL)
1199 		m_freem(control);
1200 	return (error);
1201 }
1202 
1203 /*
1204  * Send on a socket.  If send must go all at once and message is larger than
1205  * send buffering, then hard error.  Lock against other senders.  If must go
1206  * all at once and not enough room now, then inform user that this would
1207  * block and do nothing.  Otherwise, if nonblocking, send as much as
1208  * possible.  The data to be sent is described by "uio" if nonzero, otherwise
1209  * by the mbuf chain "top" (which must be null if uio is not).  Data provided
1210  * in mbuf chain must be small enough to send all at once.
1211  *
1212  * Returns nonzero on error, timeout or signal; callers must check for short
1213  * counts if EINTR/ERESTART are returned.  Data and control buffers are freed
1214  * on return.
1215  */
1216 int
1217 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio,
1218     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1219 {
1220 	long space;
1221 	ssize_t resid;
1222 	int clen = 0, error, dontroute;
1223 	int atomic = sosendallatonce(so) || top;
1224 
1225 	if (uio != NULL)
1226 		resid = uio->uio_resid;
1227 	else
1228 		resid = top->m_pkthdr.len;
1229 	/*
1230 	 * In theory resid should be unsigned.  However, space must be
1231 	 * signed, as it might be less than 0 if we over-committed, and we
1232 	 * must use a signed comparison of space and resid.  On the other
1233 	 * hand, a negative resid causes us to loop sending 0-length
1234 	 * segments to the protocol.
1235 	 *
1236 	 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
1237 	 * type sockets since that's an error.
1238 	 */
1239 	if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
1240 		error = EINVAL;
1241 		goto out;
1242 	}
1243 
1244 	dontroute =
1245 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
1246 	    (so->so_proto->pr_flags & PR_ATOMIC);
1247 	if (td != NULL)
1248 		td->td_ru.ru_msgsnd++;
1249 	if (control != NULL)
1250 		clen = control->m_len;
1251 
1252 	error = sblock(&so->so_snd, SBLOCKWAIT(flags));
1253 	if (error)
1254 		goto out;
1255 
1256 restart:
1257 	do {
1258 		SOCKBUF_LOCK(&so->so_snd);
1259 		if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1260 			SOCKBUF_UNLOCK(&so->so_snd);
1261 			error = EPIPE;
1262 			goto release;
1263 		}
1264 		if (so->so_error) {
1265 			error = so->so_error;
1266 			so->so_error = 0;
1267 			SOCKBUF_UNLOCK(&so->so_snd);
1268 			goto release;
1269 		}
1270 		if ((so->so_state & SS_ISCONNECTED) == 0) {
1271 			/*
1272 			 * `sendto' and `sendmsg' is allowed on a connection-
1273 			 * based socket if it supports implied connect.
1274 			 * Return ENOTCONN if not connected and no address is
1275 			 * supplied.
1276 			 */
1277 			if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1278 			    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1279 				if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1280 				    !(resid == 0 && clen != 0)) {
1281 					SOCKBUF_UNLOCK(&so->so_snd);
1282 					error = ENOTCONN;
1283 					goto release;
1284 				}
1285 			} else if (addr == NULL) {
1286 				SOCKBUF_UNLOCK(&so->so_snd);
1287 				if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1288 					error = ENOTCONN;
1289 				else
1290 					error = EDESTADDRREQ;
1291 				goto release;
1292 			}
1293 		}
1294 		space = sbspace(&so->so_snd);
1295 		if (flags & MSG_OOB)
1296 			space += 1024;
1297 		if ((atomic && resid > so->so_snd.sb_hiwat) ||
1298 		    clen > so->so_snd.sb_hiwat) {
1299 			SOCKBUF_UNLOCK(&so->so_snd);
1300 			error = EMSGSIZE;
1301 			goto release;
1302 		}
1303 		if (space < resid + clen &&
1304 		    (atomic || space < so->so_snd.sb_lowat || space < clen)) {
1305 			if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO)) {
1306 				SOCKBUF_UNLOCK(&so->so_snd);
1307 				error = EWOULDBLOCK;
1308 				goto release;
1309 			}
1310 			error = sbwait(&so->so_snd);
1311 			SOCKBUF_UNLOCK(&so->so_snd);
1312 			if (error)
1313 				goto release;
1314 			goto restart;
1315 		}
1316 		SOCKBUF_UNLOCK(&so->so_snd);
1317 		space -= clen;
1318 		do {
1319 			if (uio == NULL) {
1320 				resid = 0;
1321 				if (flags & MSG_EOR)
1322 					top->m_flags |= M_EOR;
1323 			} else {
1324 				/*
1325 				 * Copy the data from userland into a mbuf
1326 				 * chain.  If resid is 0, which can happen
1327 				 * only if we have control to send, then
1328 				 * a single empty mbuf is returned.  This
1329 				 * is a workaround to prevent protocol send
1330 				 * methods to panic.
1331 				 */
1332 				top = m_uiotombuf(uio, M_WAITOK, space,
1333 				    (atomic ? max_hdr : 0),
1334 				    (atomic ? M_PKTHDR : 0) |
1335 				    ((flags & MSG_EOR) ? M_EOR : 0));
1336 				if (top == NULL) {
1337 					error = EFAULT; /* only possible error */
1338 					goto release;
1339 				}
1340 				space -= resid - uio->uio_resid;
1341 				resid = uio->uio_resid;
1342 			}
1343 			if (dontroute) {
1344 				SOCK_LOCK(so);
1345 				so->so_options |= SO_DONTROUTE;
1346 				SOCK_UNLOCK(so);
1347 			}
1348 			/*
1349 			 * XXX all the SBS_CANTSENDMORE checks previously
1350 			 * done could be out of date.  We could have received
1351 			 * a reset packet in an interrupt or maybe we slept
1352 			 * while doing page faults in uiomove() etc.  We
1353 			 * could probably recheck again inside the locking
1354 			 * protection here, but there are probably other
1355 			 * places that this also happens.  We must rethink
1356 			 * this.
1357 			 */
1358 			VNET_SO_ASSERT(so);
1359 			error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1360 			    (flags & MSG_OOB) ? PRUS_OOB :
1361 			/*
1362 			 * If the user set MSG_EOF, the protocol understands
1363 			 * this flag and nothing left to send then use
1364 			 * PRU_SEND_EOF instead of PRU_SEND.
1365 			 */
1366 			    ((flags & MSG_EOF) &&
1367 			     (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1368 			     (resid <= 0)) ?
1369 				PRUS_EOF :
1370 			/* If there is more to send set PRUS_MORETOCOME. */
1371 			    (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1372 			    top, addr, control, td);
1373 			if (dontroute) {
1374 				SOCK_LOCK(so);
1375 				so->so_options &= ~SO_DONTROUTE;
1376 				SOCK_UNLOCK(so);
1377 			}
1378 			clen = 0;
1379 			control = NULL;
1380 			top = NULL;
1381 			if (error)
1382 				goto release;
1383 		} while (resid && space > 0);
1384 	} while (resid);
1385 
1386 release:
1387 	sbunlock(&so->so_snd);
1388 out:
1389 	if (top != NULL)
1390 		m_freem(top);
1391 	if (control != NULL)
1392 		m_freem(control);
1393 	return (error);
1394 }
1395 
1396 int
1397 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
1398     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1399 {
1400 	int error;
1401 
1402 	CURVNET_SET(so->so_vnet);
1403 	error = so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio, top,
1404 	    control, flags, td);
1405 	CURVNET_RESTORE();
1406 	return (error);
1407 }
1408 
1409 /*
1410  * The part of soreceive() that implements reading non-inline out-of-band
1411  * data from a socket.  For more complete comments, see soreceive(), from
1412  * which this code originated.
1413  *
1414  * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
1415  * unable to return an mbuf chain to the caller.
1416  */
1417 static int
1418 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags)
1419 {
1420 	struct protosw *pr = so->so_proto;
1421 	struct mbuf *m;
1422 	int error;
1423 
1424 	KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
1425 	VNET_SO_ASSERT(so);
1426 
1427 	m = m_get(M_WAITOK, MT_DATA);
1428 	error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK);
1429 	if (error)
1430 		goto bad;
1431 	do {
1432 		error = uiomove(mtod(m, void *),
1433 		    (int) min(uio->uio_resid, m->m_len), uio);
1434 		m = m_free(m);
1435 	} while (uio->uio_resid && error == 0 && m);
1436 bad:
1437 	if (m != NULL)
1438 		m_freem(m);
1439 	return (error);
1440 }
1441 
1442 /*
1443  * Following replacement or removal of the first mbuf on the first mbuf chain
1444  * of a socket buffer, push necessary state changes back into the socket
1445  * buffer so that other consumers see the values consistently.  'nextrecord'
1446  * is the callers locally stored value of the original value of
1447  * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes.
1448  * NOTE: 'nextrecord' may be NULL.
1449  */
1450 static __inline void
1451 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord)
1452 {
1453 
1454 	SOCKBUF_LOCK_ASSERT(sb);
1455 	/*
1456 	 * First, update for the new value of nextrecord.  If necessary, make
1457 	 * it the first record.
1458 	 */
1459 	if (sb->sb_mb != NULL)
1460 		sb->sb_mb->m_nextpkt = nextrecord;
1461 	else
1462 		sb->sb_mb = nextrecord;
1463 
1464 	/*
1465 	 * Now update any dependent socket buffer fields to reflect the new
1466 	 * state.  This is an expanded inline of SB_EMPTY_FIXUP(), with the
1467 	 * addition of a second clause that takes care of the case where
1468 	 * sb_mb has been updated, but remains the last record.
1469 	 */
1470 	if (sb->sb_mb == NULL) {
1471 		sb->sb_mbtail = NULL;
1472 		sb->sb_lastrecord = NULL;
1473 	} else if (sb->sb_mb->m_nextpkt == NULL)
1474 		sb->sb_lastrecord = sb->sb_mb;
1475 }
1476 
1477 /*
1478  * Implement receive operations on a socket.  We depend on the way that
1479  * records are added to the sockbuf by sbappend.  In particular, each record
1480  * (mbufs linked through m_next) must begin with an address if the protocol
1481  * so specifies, followed by an optional mbuf or mbufs containing ancillary
1482  * data, and then zero or more mbufs of data.  In order to allow parallelism
1483  * between network receive and copying to user space, as well as avoid
1484  * sleeping with a mutex held, we release the socket buffer mutex during the
1485  * user space copy.  Although the sockbuf is locked, new data may still be
1486  * appended, and thus we must maintain consistency of the sockbuf during that
1487  * time.
1488  *
1489  * The caller may receive the data as a single mbuf chain by supplying an
1490  * mbuf **mp0 for use in returning the chain.  The uio is then used only for
1491  * the count in uio_resid.
1492  */
1493 int
1494 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio,
1495     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1496 {
1497 	struct mbuf *m, **mp;
1498 	int flags, error, offset;
1499 	ssize_t len;
1500 	struct protosw *pr = so->so_proto;
1501 	struct mbuf *nextrecord;
1502 	int moff, type = 0;
1503 	ssize_t orig_resid = uio->uio_resid;
1504 
1505 	mp = mp0;
1506 	if (psa != NULL)
1507 		*psa = NULL;
1508 	if (controlp != NULL)
1509 		*controlp = NULL;
1510 	if (flagsp != NULL)
1511 		flags = *flagsp &~ MSG_EOR;
1512 	else
1513 		flags = 0;
1514 	if (flags & MSG_OOB)
1515 		return (soreceive_rcvoob(so, uio, flags));
1516 	if (mp != NULL)
1517 		*mp = NULL;
1518 	if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING)
1519 	    && uio->uio_resid) {
1520 		VNET_SO_ASSERT(so);
1521 		(*pr->pr_usrreqs->pru_rcvd)(so, 0);
1522 	}
1523 
1524 	error = sblock(&so->so_rcv, SBLOCKWAIT(flags));
1525 	if (error)
1526 		return (error);
1527 
1528 restart:
1529 	SOCKBUF_LOCK(&so->so_rcv);
1530 	m = so->so_rcv.sb_mb;
1531 	/*
1532 	 * If we have less data than requested, block awaiting more (subject
1533 	 * to any timeout) if:
1534 	 *   1. the current count is less than the low water mark, or
1535 	 *   2. MSG_DONTWAIT is not set
1536 	 */
1537 	if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1538 	    sbavail(&so->so_rcv) < uio->uio_resid) &&
1539 	    sbavail(&so->so_rcv) < so->so_rcv.sb_lowat &&
1540 	    m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
1541 		KASSERT(m != NULL || !sbavail(&so->so_rcv),
1542 		    ("receive: m == %p sbavail == %u",
1543 		    m, sbavail(&so->so_rcv)));
1544 		if (so->so_error) {
1545 			if (m != NULL)
1546 				goto dontblock;
1547 			error = so->so_error;
1548 			if ((flags & MSG_PEEK) == 0)
1549 				so->so_error = 0;
1550 			SOCKBUF_UNLOCK(&so->so_rcv);
1551 			goto release;
1552 		}
1553 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1554 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1555 			if (m == NULL) {
1556 				SOCKBUF_UNLOCK(&so->so_rcv);
1557 				goto release;
1558 			} else
1559 				goto dontblock;
1560 		}
1561 		for (; m != NULL; m = m->m_next)
1562 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
1563 				m = so->so_rcv.sb_mb;
1564 				goto dontblock;
1565 			}
1566 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1567 		    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1568 			SOCKBUF_UNLOCK(&so->so_rcv);
1569 			error = ENOTCONN;
1570 			goto release;
1571 		}
1572 		if (uio->uio_resid == 0) {
1573 			SOCKBUF_UNLOCK(&so->so_rcv);
1574 			goto release;
1575 		}
1576 		if ((so->so_state & SS_NBIO) ||
1577 		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
1578 			SOCKBUF_UNLOCK(&so->so_rcv);
1579 			error = EWOULDBLOCK;
1580 			goto release;
1581 		}
1582 		SBLASTRECORDCHK(&so->so_rcv);
1583 		SBLASTMBUFCHK(&so->so_rcv);
1584 		error = sbwait(&so->so_rcv);
1585 		SOCKBUF_UNLOCK(&so->so_rcv);
1586 		if (error)
1587 			goto release;
1588 		goto restart;
1589 	}
1590 dontblock:
1591 	/*
1592 	 * From this point onward, we maintain 'nextrecord' as a cache of the
1593 	 * pointer to the next record in the socket buffer.  We must keep the
1594 	 * various socket buffer pointers and local stack versions of the
1595 	 * pointers in sync, pushing out modifications before dropping the
1596 	 * socket buffer mutex, and re-reading them when picking it up.
1597 	 *
1598 	 * Otherwise, we will race with the network stack appending new data
1599 	 * or records onto the socket buffer by using inconsistent/stale
1600 	 * versions of the field, possibly resulting in socket buffer
1601 	 * corruption.
1602 	 *
1603 	 * By holding the high-level sblock(), we prevent simultaneous
1604 	 * readers from pulling off the front of the socket buffer.
1605 	 */
1606 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1607 	if (uio->uio_td)
1608 		uio->uio_td->td_ru.ru_msgrcv++;
1609 	KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb"));
1610 	SBLASTRECORDCHK(&so->so_rcv);
1611 	SBLASTMBUFCHK(&so->so_rcv);
1612 	nextrecord = m->m_nextpkt;
1613 	if (pr->pr_flags & PR_ADDR) {
1614 		KASSERT(m->m_type == MT_SONAME,
1615 		    ("m->m_type == %d", m->m_type));
1616 		orig_resid = 0;
1617 		if (psa != NULL)
1618 			*psa = sodupsockaddr(mtod(m, struct sockaddr *),
1619 			    M_NOWAIT);
1620 		if (flags & MSG_PEEK) {
1621 			m = m->m_next;
1622 		} else {
1623 			sbfree(&so->so_rcv, m);
1624 			so->so_rcv.sb_mb = m_free(m);
1625 			m = so->so_rcv.sb_mb;
1626 			sockbuf_pushsync(&so->so_rcv, nextrecord);
1627 		}
1628 	}
1629 
1630 	/*
1631 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
1632 	 * in the first mbuf chain on the socket buffer.  If MSG_PEEK, we
1633 	 * just copy the data; if !MSG_PEEK, we call into the protocol to
1634 	 * perform externalization (or freeing if controlp == NULL).
1635 	 */
1636 	if (m != NULL && m->m_type == MT_CONTROL) {
1637 		struct mbuf *cm = NULL, *cmn;
1638 		struct mbuf **cme = &cm;
1639 
1640 		do {
1641 			if (flags & MSG_PEEK) {
1642 				if (controlp != NULL) {
1643 					*controlp = m_copy(m, 0, m->m_len);
1644 					controlp = &(*controlp)->m_next;
1645 				}
1646 				m = m->m_next;
1647 			} else {
1648 				sbfree(&so->so_rcv, m);
1649 				so->so_rcv.sb_mb = m->m_next;
1650 				m->m_next = NULL;
1651 				*cme = m;
1652 				cme = &(*cme)->m_next;
1653 				m = so->so_rcv.sb_mb;
1654 			}
1655 		} while (m != NULL && m->m_type == MT_CONTROL);
1656 		if ((flags & MSG_PEEK) == 0)
1657 			sockbuf_pushsync(&so->so_rcv, nextrecord);
1658 		while (cm != NULL) {
1659 			cmn = cm->m_next;
1660 			cm->m_next = NULL;
1661 			if (pr->pr_domain->dom_externalize != NULL) {
1662 				SOCKBUF_UNLOCK(&so->so_rcv);
1663 				VNET_SO_ASSERT(so);
1664 				error = (*pr->pr_domain->dom_externalize)
1665 				    (cm, controlp, flags);
1666 				SOCKBUF_LOCK(&so->so_rcv);
1667 			} else if (controlp != NULL)
1668 				*controlp = cm;
1669 			else
1670 				m_freem(cm);
1671 			if (controlp != NULL) {
1672 				orig_resid = 0;
1673 				while (*controlp != NULL)
1674 					controlp = &(*controlp)->m_next;
1675 			}
1676 			cm = cmn;
1677 		}
1678 		if (m != NULL)
1679 			nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1680 		else
1681 			nextrecord = so->so_rcv.sb_mb;
1682 		orig_resid = 0;
1683 	}
1684 	if (m != NULL) {
1685 		if ((flags & MSG_PEEK) == 0) {
1686 			KASSERT(m->m_nextpkt == nextrecord,
1687 			    ("soreceive: post-control, nextrecord !sync"));
1688 			if (nextrecord == NULL) {
1689 				KASSERT(so->so_rcv.sb_mb == m,
1690 				    ("soreceive: post-control, sb_mb!=m"));
1691 				KASSERT(so->so_rcv.sb_lastrecord == m,
1692 				    ("soreceive: post-control, lastrecord!=m"));
1693 			}
1694 		}
1695 		type = m->m_type;
1696 		if (type == MT_OOBDATA)
1697 			flags |= MSG_OOB;
1698 	} else {
1699 		if ((flags & MSG_PEEK) == 0) {
1700 			KASSERT(so->so_rcv.sb_mb == nextrecord,
1701 			    ("soreceive: sb_mb != nextrecord"));
1702 			if (so->so_rcv.sb_mb == NULL) {
1703 				KASSERT(so->so_rcv.sb_lastrecord == NULL,
1704 				    ("soreceive: sb_lastercord != NULL"));
1705 			}
1706 		}
1707 	}
1708 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1709 	SBLASTRECORDCHK(&so->so_rcv);
1710 	SBLASTMBUFCHK(&so->so_rcv);
1711 
1712 	/*
1713 	 * Now continue to read any data mbufs off of the head of the socket
1714 	 * buffer until the read request is satisfied.  Note that 'type' is
1715 	 * used to store the type of any mbuf reads that have happened so far
1716 	 * such that soreceive() can stop reading if the type changes, which
1717 	 * causes soreceive() to return only one of regular data and inline
1718 	 * out-of-band data in a single socket receive operation.
1719 	 */
1720 	moff = 0;
1721 	offset = 0;
1722 	while (m != NULL && !(m->m_flags & M_NOTAVAIL) && uio->uio_resid > 0
1723 	    && error == 0) {
1724 		/*
1725 		 * If the type of mbuf has changed since the last mbuf
1726 		 * examined ('type'), end the receive operation.
1727 		 */
1728 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1729 		if (m->m_type == MT_OOBDATA || m->m_type == MT_CONTROL) {
1730 			if (type != m->m_type)
1731 				break;
1732 		} else if (type == MT_OOBDATA)
1733 			break;
1734 		else
1735 		    KASSERT(m->m_type == MT_DATA,
1736 			("m->m_type == %d", m->m_type));
1737 		so->so_rcv.sb_state &= ~SBS_RCVATMARK;
1738 		len = uio->uio_resid;
1739 		if (so->so_oobmark && len > so->so_oobmark - offset)
1740 			len = so->so_oobmark - offset;
1741 		if (len > m->m_len - moff)
1742 			len = m->m_len - moff;
1743 		/*
1744 		 * If mp is set, just pass back the mbufs.  Otherwise copy
1745 		 * them out via the uio, then free.  Sockbuf must be
1746 		 * consistent here (points to current mbuf, it points to next
1747 		 * record) when we drop priority; we must note any additions
1748 		 * to the sockbuf when we block interrupts again.
1749 		 */
1750 		if (mp == NULL) {
1751 			SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1752 			SBLASTRECORDCHK(&so->so_rcv);
1753 			SBLASTMBUFCHK(&so->so_rcv);
1754 			SOCKBUF_UNLOCK(&so->so_rcv);
1755 			error = uiomove(mtod(m, char *) + moff, (int)len, uio);
1756 			SOCKBUF_LOCK(&so->so_rcv);
1757 			if (error) {
1758 				/*
1759 				 * The MT_SONAME mbuf has already been removed
1760 				 * from the record, so it is necessary to
1761 				 * remove the data mbufs, if any, to preserve
1762 				 * the invariant in the case of PR_ADDR that
1763 				 * requires MT_SONAME mbufs at the head of
1764 				 * each record.
1765 				 */
1766 				if (m && pr->pr_flags & PR_ATOMIC &&
1767 				    ((flags & MSG_PEEK) == 0))
1768 					(void)sbdroprecord_locked(&so->so_rcv);
1769 				SOCKBUF_UNLOCK(&so->so_rcv);
1770 				goto release;
1771 			}
1772 		} else
1773 			uio->uio_resid -= len;
1774 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1775 		if (len == m->m_len - moff) {
1776 			if (m->m_flags & M_EOR)
1777 				flags |= MSG_EOR;
1778 			if (flags & MSG_PEEK) {
1779 				m = m->m_next;
1780 				moff = 0;
1781 			} else {
1782 				nextrecord = m->m_nextpkt;
1783 				sbfree(&so->so_rcv, m);
1784 				if (mp != NULL) {
1785 					m->m_nextpkt = NULL;
1786 					*mp = m;
1787 					mp = &m->m_next;
1788 					so->so_rcv.sb_mb = m = m->m_next;
1789 					*mp = NULL;
1790 				} else {
1791 					so->so_rcv.sb_mb = m_free(m);
1792 					m = so->so_rcv.sb_mb;
1793 				}
1794 				sockbuf_pushsync(&so->so_rcv, nextrecord);
1795 				SBLASTRECORDCHK(&so->so_rcv);
1796 				SBLASTMBUFCHK(&so->so_rcv);
1797 			}
1798 		} else {
1799 			if (flags & MSG_PEEK)
1800 				moff += len;
1801 			else {
1802 				if (mp != NULL) {
1803 					if (flags & MSG_DONTWAIT) {
1804 						*mp = m_copym(m, 0, len,
1805 						    M_NOWAIT);
1806 						if (*mp == NULL) {
1807 							/*
1808 							 * m_copym() couldn't
1809 							 * allocate an mbuf.
1810 							 * Adjust uio_resid back
1811 							 * (it was adjusted
1812 							 * down by len bytes,
1813 							 * which we didn't end
1814 							 * up "copying" over).
1815 							 */
1816 							uio->uio_resid += len;
1817 							break;
1818 						}
1819 					} else {
1820 						SOCKBUF_UNLOCK(&so->so_rcv);
1821 						*mp = m_copym(m, 0, len,
1822 						    M_WAITOK);
1823 						SOCKBUF_LOCK(&so->so_rcv);
1824 					}
1825 				}
1826 				sbcut_locked(&so->so_rcv, len);
1827 			}
1828 		}
1829 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1830 		if (so->so_oobmark) {
1831 			if ((flags & MSG_PEEK) == 0) {
1832 				so->so_oobmark -= len;
1833 				if (so->so_oobmark == 0) {
1834 					so->so_rcv.sb_state |= SBS_RCVATMARK;
1835 					break;
1836 				}
1837 			} else {
1838 				offset += len;
1839 				if (offset == so->so_oobmark)
1840 					break;
1841 			}
1842 		}
1843 		if (flags & MSG_EOR)
1844 			break;
1845 		/*
1846 		 * If the MSG_WAITALL flag is set (for non-atomic socket), we
1847 		 * must not quit until "uio->uio_resid == 0" or an error
1848 		 * termination.  If a signal/timeout occurs, return with a
1849 		 * short count but without error.  Keep sockbuf locked
1850 		 * against other readers.
1851 		 */
1852 		while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1853 		    !sosendallatonce(so) && nextrecord == NULL) {
1854 			SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1855 			if (so->so_error ||
1856 			    so->so_rcv.sb_state & SBS_CANTRCVMORE)
1857 				break;
1858 			/*
1859 			 * Notify the protocol that some data has been
1860 			 * drained before blocking.
1861 			 */
1862 			if (pr->pr_flags & PR_WANTRCVD) {
1863 				SOCKBUF_UNLOCK(&so->so_rcv);
1864 				VNET_SO_ASSERT(so);
1865 				(*pr->pr_usrreqs->pru_rcvd)(so, flags);
1866 				SOCKBUF_LOCK(&so->so_rcv);
1867 			}
1868 			SBLASTRECORDCHK(&so->so_rcv);
1869 			SBLASTMBUFCHK(&so->so_rcv);
1870 			/*
1871 			 * We could receive some data while was notifying
1872 			 * the protocol. Skip blocking in this case.
1873 			 */
1874 			if (so->so_rcv.sb_mb == NULL) {
1875 				error = sbwait(&so->so_rcv);
1876 				if (error) {
1877 					SOCKBUF_UNLOCK(&so->so_rcv);
1878 					goto release;
1879 				}
1880 			}
1881 			m = so->so_rcv.sb_mb;
1882 			if (m != NULL)
1883 				nextrecord = m->m_nextpkt;
1884 		}
1885 	}
1886 
1887 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1888 	if (m != NULL && pr->pr_flags & PR_ATOMIC) {
1889 		flags |= MSG_TRUNC;
1890 		if ((flags & MSG_PEEK) == 0)
1891 			(void) sbdroprecord_locked(&so->so_rcv);
1892 	}
1893 	if ((flags & MSG_PEEK) == 0) {
1894 		if (m == NULL) {
1895 			/*
1896 			 * First part is an inline SB_EMPTY_FIXUP().  Second
1897 			 * part makes sure sb_lastrecord is up-to-date if
1898 			 * there is still data in the socket buffer.
1899 			 */
1900 			so->so_rcv.sb_mb = nextrecord;
1901 			if (so->so_rcv.sb_mb == NULL) {
1902 				so->so_rcv.sb_mbtail = NULL;
1903 				so->so_rcv.sb_lastrecord = NULL;
1904 			} else if (nextrecord->m_nextpkt == NULL)
1905 				so->so_rcv.sb_lastrecord = nextrecord;
1906 		}
1907 		SBLASTRECORDCHK(&so->so_rcv);
1908 		SBLASTMBUFCHK(&so->so_rcv);
1909 		/*
1910 		 * If soreceive() is being done from the socket callback,
1911 		 * then don't need to generate ACK to peer to update window,
1912 		 * since ACK will be generated on return to TCP.
1913 		 */
1914 		if (!(flags & MSG_SOCALLBCK) &&
1915 		    (pr->pr_flags & PR_WANTRCVD)) {
1916 			SOCKBUF_UNLOCK(&so->so_rcv);
1917 			VNET_SO_ASSERT(so);
1918 			(*pr->pr_usrreqs->pru_rcvd)(so, flags);
1919 			SOCKBUF_LOCK(&so->so_rcv);
1920 		}
1921 	}
1922 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1923 	if (orig_resid == uio->uio_resid && orig_resid &&
1924 	    (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
1925 		SOCKBUF_UNLOCK(&so->so_rcv);
1926 		goto restart;
1927 	}
1928 	SOCKBUF_UNLOCK(&so->so_rcv);
1929 
1930 	if (flagsp != NULL)
1931 		*flagsp |= flags;
1932 release:
1933 	sbunlock(&so->so_rcv);
1934 	return (error);
1935 }
1936 
1937 /*
1938  * Optimized version of soreceive() for stream (TCP) sockets.
1939  * XXXAO: (MSG_WAITALL | MSG_PEEK) isn't properly handled.
1940  */
1941 int
1942 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio,
1943     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1944 {
1945 	int len = 0, error = 0, flags, oresid;
1946 	struct sockbuf *sb;
1947 	struct mbuf *m, *n = NULL;
1948 
1949 	/* We only do stream sockets. */
1950 	if (so->so_type != SOCK_STREAM)
1951 		return (EINVAL);
1952 	if (psa != NULL)
1953 		*psa = NULL;
1954 	if (controlp != NULL)
1955 		return (EINVAL);
1956 	if (flagsp != NULL)
1957 		flags = *flagsp &~ MSG_EOR;
1958 	else
1959 		flags = 0;
1960 	if (flags & MSG_OOB)
1961 		return (soreceive_rcvoob(so, uio, flags));
1962 	if (mp0 != NULL)
1963 		*mp0 = NULL;
1964 
1965 	sb = &so->so_rcv;
1966 
1967 	/* Prevent other readers from entering the socket. */
1968 	error = sblock(sb, SBLOCKWAIT(flags));
1969 	if (error)
1970 		goto out;
1971 	SOCKBUF_LOCK(sb);
1972 
1973 	/* Easy one, no space to copyout anything. */
1974 	if (uio->uio_resid == 0) {
1975 		error = EINVAL;
1976 		goto out;
1977 	}
1978 	oresid = uio->uio_resid;
1979 
1980 	/* We will never ever get anything unless we are or were connected. */
1981 	if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) {
1982 		error = ENOTCONN;
1983 		goto out;
1984 	}
1985 
1986 restart:
1987 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1988 
1989 	/* Abort if socket has reported problems. */
1990 	if (so->so_error) {
1991 		if (sbavail(sb) > 0)
1992 			goto deliver;
1993 		if (oresid > uio->uio_resid)
1994 			goto out;
1995 		error = so->so_error;
1996 		if (!(flags & MSG_PEEK))
1997 			so->so_error = 0;
1998 		goto out;
1999 	}
2000 
2001 	/* Door is closed.  Deliver what is left, if any. */
2002 	if (sb->sb_state & SBS_CANTRCVMORE) {
2003 		if (sbavail(sb) > 0)
2004 			goto deliver;
2005 		else
2006 			goto out;
2007 	}
2008 
2009 	/* Socket buffer is empty and we shall not block. */
2010 	if (sbavail(sb) == 0 &&
2011 	    ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) {
2012 		error = EAGAIN;
2013 		goto out;
2014 	}
2015 
2016 	/* Socket buffer got some data that we shall deliver now. */
2017 	if (sbavail(sb) > 0 && !(flags & MSG_WAITALL) &&
2018 	    ((so->so_state & SS_NBIO) ||
2019 	     (flags & (MSG_DONTWAIT|MSG_NBIO)) ||
2020 	     sbavail(sb) >= sb->sb_lowat ||
2021 	     sbavail(sb) >= uio->uio_resid ||
2022 	     sbavail(sb) >= sb->sb_hiwat) ) {
2023 		goto deliver;
2024 	}
2025 
2026 	/* On MSG_WAITALL we must wait until all data or error arrives. */
2027 	if ((flags & MSG_WAITALL) &&
2028 	    (sbavail(sb) >= uio->uio_resid || sbavail(sb) >= sb->sb_hiwat))
2029 		goto deliver;
2030 
2031 	/*
2032 	 * Wait and block until (more) data comes in.
2033 	 * NB: Drops the sockbuf lock during wait.
2034 	 */
2035 	error = sbwait(sb);
2036 	if (error)
2037 		goto out;
2038 	goto restart;
2039 
2040 deliver:
2041 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2042 	KASSERT(sbavail(sb) > 0, ("%s: sockbuf empty", __func__));
2043 	KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__));
2044 
2045 	/* Statistics. */
2046 	if (uio->uio_td)
2047 		uio->uio_td->td_ru.ru_msgrcv++;
2048 
2049 	/* Fill uio until full or current end of socket buffer is reached. */
2050 	len = min(uio->uio_resid, sbavail(sb));
2051 	if (mp0 != NULL) {
2052 		/* Dequeue as many mbufs as possible. */
2053 		if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) {
2054 			if (*mp0 == NULL)
2055 				*mp0 = sb->sb_mb;
2056 			else
2057 				m_cat(*mp0, sb->sb_mb);
2058 			for (m = sb->sb_mb;
2059 			     m != NULL && m->m_len <= len;
2060 			     m = m->m_next) {
2061 				KASSERT(!(m->m_flags & M_NOTAVAIL),
2062 				    ("%s: m %p not available", __func__, m));
2063 				len -= m->m_len;
2064 				uio->uio_resid -= m->m_len;
2065 				sbfree(sb, m);
2066 				n = m;
2067 			}
2068 			n->m_next = NULL;
2069 			sb->sb_mb = m;
2070 			sb->sb_lastrecord = sb->sb_mb;
2071 			if (sb->sb_mb == NULL)
2072 				SB_EMPTY_FIXUP(sb);
2073 		}
2074 		/* Copy the remainder. */
2075 		if (len > 0) {
2076 			KASSERT(sb->sb_mb != NULL,
2077 			    ("%s: len > 0 && sb->sb_mb empty", __func__));
2078 
2079 			m = m_copym(sb->sb_mb, 0, len, M_NOWAIT);
2080 			if (m == NULL)
2081 				len = 0;	/* Don't flush data from sockbuf. */
2082 			else
2083 				uio->uio_resid -= len;
2084 			if (*mp0 != NULL)
2085 				m_cat(*mp0, m);
2086 			else
2087 				*mp0 = m;
2088 			if (*mp0 == NULL) {
2089 				error = ENOBUFS;
2090 				goto out;
2091 			}
2092 		}
2093 	} else {
2094 		/* NB: Must unlock socket buffer as uiomove may sleep. */
2095 		SOCKBUF_UNLOCK(sb);
2096 		error = m_mbuftouio(uio, sb->sb_mb, len);
2097 		SOCKBUF_LOCK(sb);
2098 		if (error)
2099 			goto out;
2100 	}
2101 	SBLASTRECORDCHK(sb);
2102 	SBLASTMBUFCHK(sb);
2103 
2104 	/*
2105 	 * Remove the delivered data from the socket buffer unless we
2106 	 * were only peeking.
2107 	 */
2108 	if (!(flags & MSG_PEEK)) {
2109 		if (len > 0)
2110 			sbdrop_locked(sb, len);
2111 
2112 		/* Notify protocol that we drained some data. */
2113 		if ((so->so_proto->pr_flags & PR_WANTRCVD) &&
2114 		    (((flags & MSG_WAITALL) && uio->uio_resid > 0) ||
2115 		     !(flags & MSG_SOCALLBCK))) {
2116 			SOCKBUF_UNLOCK(sb);
2117 			VNET_SO_ASSERT(so);
2118 			(*so->so_proto->pr_usrreqs->pru_rcvd)(so, flags);
2119 			SOCKBUF_LOCK(sb);
2120 		}
2121 	}
2122 
2123 	/*
2124 	 * For MSG_WAITALL we may have to loop again and wait for
2125 	 * more data to come in.
2126 	 */
2127 	if ((flags & MSG_WAITALL) && uio->uio_resid > 0)
2128 		goto restart;
2129 out:
2130 	SOCKBUF_LOCK_ASSERT(sb);
2131 	SBLASTRECORDCHK(sb);
2132 	SBLASTMBUFCHK(sb);
2133 	SOCKBUF_UNLOCK(sb);
2134 	sbunlock(sb);
2135 	return (error);
2136 }
2137 
2138 /*
2139  * Optimized version of soreceive() for simple datagram cases from userspace.
2140  * Unlike in the stream case, we're able to drop a datagram if copyout()
2141  * fails, and because we handle datagrams atomically, we don't need to use a
2142  * sleep lock to prevent I/O interlacing.
2143  */
2144 int
2145 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio,
2146     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2147 {
2148 	struct mbuf *m, *m2;
2149 	int flags, error;
2150 	ssize_t len;
2151 	struct protosw *pr = so->so_proto;
2152 	struct mbuf *nextrecord;
2153 
2154 	if (psa != NULL)
2155 		*psa = NULL;
2156 	if (controlp != NULL)
2157 		*controlp = NULL;
2158 	if (flagsp != NULL)
2159 		flags = *flagsp &~ MSG_EOR;
2160 	else
2161 		flags = 0;
2162 
2163 	/*
2164 	 * For any complicated cases, fall back to the full
2165 	 * soreceive_generic().
2166 	 */
2167 	if (mp0 != NULL || (flags & MSG_PEEK) || (flags & MSG_OOB))
2168 		return (soreceive_generic(so, psa, uio, mp0, controlp,
2169 		    flagsp));
2170 
2171 	/*
2172 	 * Enforce restrictions on use.
2173 	 */
2174 	KASSERT((pr->pr_flags & PR_WANTRCVD) == 0,
2175 	    ("soreceive_dgram: wantrcvd"));
2176 	KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic"));
2177 	KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0,
2178 	    ("soreceive_dgram: SBS_RCVATMARK"));
2179 	KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0,
2180 	    ("soreceive_dgram: P_CONNREQUIRED"));
2181 
2182 	/*
2183 	 * Loop blocking while waiting for a datagram.
2184 	 */
2185 	SOCKBUF_LOCK(&so->so_rcv);
2186 	while ((m = so->so_rcv.sb_mb) == NULL) {
2187 		KASSERT(sbavail(&so->so_rcv) == 0,
2188 		    ("soreceive_dgram: sb_mb NULL but sbavail %u",
2189 		    sbavail(&so->so_rcv)));
2190 		if (so->so_error) {
2191 			error = so->so_error;
2192 			so->so_error = 0;
2193 			SOCKBUF_UNLOCK(&so->so_rcv);
2194 			return (error);
2195 		}
2196 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE ||
2197 		    uio->uio_resid == 0) {
2198 			SOCKBUF_UNLOCK(&so->so_rcv);
2199 			return (0);
2200 		}
2201 		if ((so->so_state & SS_NBIO) ||
2202 		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2203 			SOCKBUF_UNLOCK(&so->so_rcv);
2204 			return (EWOULDBLOCK);
2205 		}
2206 		SBLASTRECORDCHK(&so->so_rcv);
2207 		SBLASTMBUFCHK(&so->so_rcv);
2208 		error = sbwait(&so->so_rcv);
2209 		if (error) {
2210 			SOCKBUF_UNLOCK(&so->so_rcv);
2211 			return (error);
2212 		}
2213 	}
2214 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2215 
2216 	if (uio->uio_td)
2217 		uio->uio_td->td_ru.ru_msgrcv++;
2218 	SBLASTRECORDCHK(&so->so_rcv);
2219 	SBLASTMBUFCHK(&so->so_rcv);
2220 	nextrecord = m->m_nextpkt;
2221 	if (nextrecord == NULL) {
2222 		KASSERT(so->so_rcv.sb_lastrecord == m,
2223 		    ("soreceive_dgram: lastrecord != m"));
2224 	}
2225 
2226 	KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord,
2227 	    ("soreceive_dgram: m_nextpkt != nextrecord"));
2228 
2229 	/*
2230 	 * Pull 'm' and its chain off the front of the packet queue.
2231 	 */
2232 	so->so_rcv.sb_mb = NULL;
2233 	sockbuf_pushsync(&so->so_rcv, nextrecord);
2234 
2235 	/*
2236 	 * Walk 'm's chain and free that many bytes from the socket buffer.
2237 	 */
2238 	for (m2 = m; m2 != NULL; m2 = m2->m_next)
2239 		sbfree(&so->so_rcv, m2);
2240 
2241 	/*
2242 	 * Do a few last checks before we let go of the lock.
2243 	 */
2244 	SBLASTRECORDCHK(&so->so_rcv);
2245 	SBLASTMBUFCHK(&so->so_rcv);
2246 	SOCKBUF_UNLOCK(&so->so_rcv);
2247 
2248 	if (pr->pr_flags & PR_ADDR) {
2249 		KASSERT(m->m_type == MT_SONAME,
2250 		    ("m->m_type == %d", m->m_type));
2251 		if (psa != NULL)
2252 			*psa = sodupsockaddr(mtod(m, struct sockaddr *),
2253 			    M_NOWAIT);
2254 		m = m_free(m);
2255 	}
2256 	if (m == NULL) {
2257 		/* XXXRW: Can this happen? */
2258 		return (0);
2259 	}
2260 
2261 	/*
2262 	 * Packet to copyout() is now in 'm' and it is disconnected from the
2263 	 * queue.
2264 	 *
2265 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
2266 	 * in the first mbuf chain on the socket buffer.  We call into the
2267 	 * protocol to perform externalization (or freeing if controlp ==
2268 	 * NULL). In some cases there can be only MT_CONTROL mbufs without
2269 	 * MT_DATA mbufs.
2270 	 */
2271 	if (m->m_type == MT_CONTROL) {
2272 		struct mbuf *cm = NULL, *cmn;
2273 		struct mbuf **cme = &cm;
2274 
2275 		do {
2276 			m2 = m->m_next;
2277 			m->m_next = NULL;
2278 			*cme = m;
2279 			cme = &(*cme)->m_next;
2280 			m = m2;
2281 		} while (m != NULL && m->m_type == MT_CONTROL);
2282 		while (cm != NULL) {
2283 			cmn = cm->m_next;
2284 			cm->m_next = NULL;
2285 			if (pr->pr_domain->dom_externalize != NULL) {
2286 				error = (*pr->pr_domain->dom_externalize)
2287 				    (cm, controlp, flags);
2288 			} else if (controlp != NULL)
2289 				*controlp = cm;
2290 			else
2291 				m_freem(cm);
2292 			if (controlp != NULL) {
2293 				while (*controlp != NULL)
2294 					controlp = &(*controlp)->m_next;
2295 			}
2296 			cm = cmn;
2297 		}
2298 	}
2299 	KASSERT(m == NULL || m->m_type == MT_DATA,
2300 	    ("soreceive_dgram: !data"));
2301 	while (m != NULL && uio->uio_resid > 0) {
2302 		len = uio->uio_resid;
2303 		if (len > m->m_len)
2304 			len = m->m_len;
2305 		error = uiomove(mtod(m, char *), (int)len, uio);
2306 		if (error) {
2307 			m_freem(m);
2308 			return (error);
2309 		}
2310 		if (len == m->m_len)
2311 			m = m_free(m);
2312 		else {
2313 			m->m_data += len;
2314 			m->m_len -= len;
2315 		}
2316 	}
2317 	if (m != NULL) {
2318 		flags |= MSG_TRUNC;
2319 		m_freem(m);
2320 	}
2321 	if (flagsp != NULL)
2322 		*flagsp |= flags;
2323 	return (0);
2324 }
2325 
2326 int
2327 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
2328     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2329 {
2330 	int error;
2331 
2332 	CURVNET_SET(so->so_vnet);
2333 	error = (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio, mp0,
2334 	    controlp, flagsp));
2335 	CURVNET_RESTORE();
2336 	return (error);
2337 }
2338 
2339 int
2340 soshutdown(struct socket *so, int how)
2341 {
2342 	struct protosw *pr = so->so_proto;
2343 	int error;
2344 
2345 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
2346 		return (EINVAL);
2347 	if ((so->so_state &
2348 	    (SS_ISCONNECTED | SS_ISCONNECTING | SS_ISDISCONNECTING)) == 0)
2349 		return (ENOTCONN);
2350 
2351 	CURVNET_SET(so->so_vnet);
2352 	if (pr->pr_usrreqs->pru_flush != NULL)
2353 		(*pr->pr_usrreqs->pru_flush)(so, how);
2354 	if (how != SHUT_WR)
2355 		sorflush(so);
2356 	if (how != SHUT_RD) {
2357 		error = (*pr->pr_usrreqs->pru_shutdown)(so);
2358 		wakeup(&so->so_timeo);
2359 		CURVNET_RESTORE();
2360 		return (error);
2361 	}
2362 	wakeup(&so->so_timeo);
2363 	CURVNET_RESTORE();
2364 	return (0);
2365 }
2366 
2367 void
2368 sorflush(struct socket *so)
2369 {
2370 	struct sockbuf *sb = &so->so_rcv;
2371 	struct protosw *pr = so->so_proto;
2372 	struct socket aso;
2373 
2374 	VNET_SO_ASSERT(so);
2375 
2376 	/*
2377 	 * In order to avoid calling dom_dispose with the socket buffer mutex
2378 	 * held, and in order to generally avoid holding the lock for a long
2379 	 * time, we make a copy of the socket buffer and clear the original
2380 	 * (except locks, state).  The new socket buffer copy won't have
2381 	 * initialized locks so we can only call routines that won't use or
2382 	 * assert those locks.
2383 	 *
2384 	 * Dislodge threads currently blocked in receive and wait to acquire
2385 	 * a lock against other simultaneous readers before clearing the
2386 	 * socket buffer.  Don't let our acquire be interrupted by a signal
2387 	 * despite any existing socket disposition on interruptable waiting.
2388 	 */
2389 	socantrcvmore(so);
2390 	(void) sblock(sb, SBL_WAIT | SBL_NOINTR);
2391 
2392 	/*
2393 	 * Invalidate/clear most of the sockbuf structure, but leave selinfo
2394 	 * and mutex data unchanged.
2395 	 */
2396 	SOCKBUF_LOCK(sb);
2397 	bzero(&aso, sizeof(aso));
2398 	aso.so_pcb = so->so_pcb;
2399 	bcopy(&sb->sb_startzero, &aso.so_rcv.sb_startzero,
2400 	    sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2401 	bzero(&sb->sb_startzero,
2402 	    sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2403 	SOCKBUF_UNLOCK(sb);
2404 	sbunlock(sb);
2405 
2406 	/*
2407 	 * Dispose of special rights and flush the copied socket.  Don't call
2408 	 * any unsafe routines (that rely on locks being initialized) on aso.
2409 	 */
2410 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
2411 		(*pr->pr_domain->dom_dispose)(&aso);
2412 	sbrelease_internal(&aso.so_rcv, so);
2413 }
2414 
2415 /*
2416  * Wrapper for Socket established helper hook.
2417  * Parameters: socket, context of the hook point, hook id.
2418  */
2419 static int inline
2420 hhook_run_socket(struct socket *so, void *hctx, int32_t h_id)
2421 {
2422 	struct socket_hhook_data hhook_data = {
2423 		.so = so,
2424 		.hctx = hctx,
2425 		.m = NULL,
2426 		.status = 0
2427 	};
2428 
2429 	CURVNET_SET(so->so_vnet);
2430 	HHOOKS_RUN_IF(V_socket_hhh[h_id], &hhook_data, &so->osd);
2431 	CURVNET_RESTORE();
2432 
2433 	/* Ugly but needed, since hhooks return void for now */
2434 	return (hhook_data.status);
2435 }
2436 
2437 /*
2438  * Perhaps this routine, and sooptcopyout(), below, ought to come in an
2439  * additional variant to handle the case where the option value needs to be
2440  * some kind of integer, but not a specific size.  In addition to their use
2441  * here, these functions are also called by the protocol-level pr_ctloutput()
2442  * routines.
2443  */
2444 int
2445 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2446 {
2447 	size_t	valsize;
2448 
2449 	/*
2450 	 * If the user gives us more than we wanted, we ignore it, but if we
2451 	 * don't get the minimum length the caller wants, we return EINVAL.
2452 	 * On success, sopt->sopt_valsize is set to however much we actually
2453 	 * retrieved.
2454 	 */
2455 	if ((valsize = sopt->sopt_valsize) < minlen)
2456 		return EINVAL;
2457 	if (valsize > len)
2458 		sopt->sopt_valsize = valsize = len;
2459 
2460 	if (sopt->sopt_td != NULL)
2461 		return (copyin(sopt->sopt_val, buf, valsize));
2462 
2463 	bcopy(sopt->sopt_val, buf, valsize);
2464 	return (0);
2465 }
2466 
2467 /*
2468  * Kernel version of setsockopt(2).
2469  *
2470  * XXX: optlen is size_t, not socklen_t
2471  */
2472 int
2473 so_setsockopt(struct socket *so, int level, int optname, void *optval,
2474     size_t optlen)
2475 {
2476 	struct sockopt sopt;
2477 
2478 	sopt.sopt_level = level;
2479 	sopt.sopt_name = optname;
2480 	sopt.sopt_dir = SOPT_SET;
2481 	sopt.sopt_val = optval;
2482 	sopt.sopt_valsize = optlen;
2483 	sopt.sopt_td = NULL;
2484 	return (sosetopt(so, &sopt));
2485 }
2486 
2487 int
2488 sosetopt(struct socket *so, struct sockopt *sopt)
2489 {
2490 	int	error, optval;
2491 	struct	linger l;
2492 	struct	timeval tv;
2493 	sbintime_t val;
2494 	uint32_t val32;
2495 #ifdef MAC
2496 	struct mac extmac;
2497 #endif
2498 
2499 	CURVNET_SET(so->so_vnet);
2500 	error = 0;
2501 	if (sopt->sopt_level != SOL_SOCKET) {
2502 		if (so->so_proto->pr_ctloutput != NULL) {
2503 			error = (*so->so_proto->pr_ctloutput)(so, sopt);
2504 			CURVNET_RESTORE();
2505 			return (error);
2506 		}
2507 		error = ENOPROTOOPT;
2508 	} else {
2509 		switch (sopt->sopt_name) {
2510 		case SO_ACCEPTFILTER:
2511 			error = do_setopt_accept_filter(so, sopt);
2512 			if (error)
2513 				goto bad;
2514 			break;
2515 
2516 		case SO_LINGER:
2517 			error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
2518 			if (error)
2519 				goto bad;
2520 
2521 			SOCK_LOCK(so);
2522 			so->so_linger = l.l_linger;
2523 			if (l.l_onoff)
2524 				so->so_options |= SO_LINGER;
2525 			else
2526 				so->so_options &= ~SO_LINGER;
2527 			SOCK_UNLOCK(so);
2528 			break;
2529 
2530 		case SO_DEBUG:
2531 		case SO_KEEPALIVE:
2532 		case SO_DONTROUTE:
2533 		case SO_USELOOPBACK:
2534 		case SO_BROADCAST:
2535 		case SO_REUSEADDR:
2536 		case SO_REUSEPORT:
2537 		case SO_OOBINLINE:
2538 		case SO_TIMESTAMP:
2539 		case SO_BINTIME:
2540 		case SO_NOSIGPIPE:
2541 		case SO_NO_DDP:
2542 		case SO_NO_OFFLOAD:
2543 			error = sooptcopyin(sopt, &optval, sizeof optval,
2544 			    sizeof optval);
2545 			if (error)
2546 				goto bad;
2547 			SOCK_LOCK(so);
2548 			if (optval)
2549 				so->so_options |= sopt->sopt_name;
2550 			else
2551 				so->so_options &= ~sopt->sopt_name;
2552 			SOCK_UNLOCK(so);
2553 			break;
2554 
2555 		case SO_SETFIB:
2556 			error = sooptcopyin(sopt, &optval, sizeof optval,
2557 			    sizeof optval);
2558 			if (error)
2559 				goto bad;
2560 
2561 			if (optval < 0 || optval >= rt_numfibs) {
2562 				error = EINVAL;
2563 				goto bad;
2564 			}
2565 			if (((so->so_proto->pr_domain->dom_family == PF_INET) ||
2566 			   (so->so_proto->pr_domain->dom_family == PF_INET6) ||
2567 			   (so->so_proto->pr_domain->dom_family == PF_ROUTE)))
2568 				so->so_fibnum = optval;
2569 			else
2570 				so->so_fibnum = 0;
2571 			break;
2572 
2573 		case SO_USER_COOKIE:
2574 			error = sooptcopyin(sopt, &val32, sizeof val32,
2575 			    sizeof val32);
2576 			if (error)
2577 				goto bad;
2578 			so->so_user_cookie = val32;
2579 			break;
2580 
2581 		case SO_SNDBUF:
2582 		case SO_RCVBUF:
2583 		case SO_SNDLOWAT:
2584 		case SO_RCVLOWAT:
2585 			error = sooptcopyin(sopt, &optval, sizeof optval,
2586 			    sizeof optval);
2587 			if (error)
2588 				goto bad;
2589 
2590 			/*
2591 			 * Values < 1 make no sense for any of these options,
2592 			 * so disallow them.
2593 			 */
2594 			if (optval < 1) {
2595 				error = EINVAL;
2596 				goto bad;
2597 			}
2598 
2599 			switch (sopt->sopt_name) {
2600 			case SO_SNDBUF:
2601 			case SO_RCVBUF:
2602 				if (sbreserve(sopt->sopt_name == SO_SNDBUF ?
2603 				    &so->so_snd : &so->so_rcv, (u_long)optval,
2604 				    so, curthread) == 0) {
2605 					error = ENOBUFS;
2606 					goto bad;
2607 				}
2608 				(sopt->sopt_name == SO_SNDBUF ? &so->so_snd :
2609 				    &so->so_rcv)->sb_flags &= ~SB_AUTOSIZE;
2610 				break;
2611 
2612 			/*
2613 			 * Make sure the low-water is never greater than the
2614 			 * high-water.
2615 			 */
2616 			case SO_SNDLOWAT:
2617 				SOCKBUF_LOCK(&so->so_snd);
2618 				so->so_snd.sb_lowat =
2619 				    (optval > so->so_snd.sb_hiwat) ?
2620 				    so->so_snd.sb_hiwat : optval;
2621 				SOCKBUF_UNLOCK(&so->so_snd);
2622 				break;
2623 			case SO_RCVLOWAT:
2624 				SOCKBUF_LOCK(&so->so_rcv);
2625 				so->so_rcv.sb_lowat =
2626 				    (optval > so->so_rcv.sb_hiwat) ?
2627 				    so->so_rcv.sb_hiwat : optval;
2628 				SOCKBUF_UNLOCK(&so->so_rcv);
2629 				break;
2630 			}
2631 			break;
2632 
2633 		case SO_SNDTIMEO:
2634 		case SO_RCVTIMEO:
2635 #ifdef COMPAT_FREEBSD32
2636 			if (SV_CURPROC_FLAG(SV_ILP32)) {
2637 				struct timeval32 tv32;
2638 
2639 				error = sooptcopyin(sopt, &tv32, sizeof tv32,
2640 				    sizeof tv32);
2641 				CP(tv32, tv, tv_sec);
2642 				CP(tv32, tv, tv_usec);
2643 			} else
2644 #endif
2645 				error = sooptcopyin(sopt, &tv, sizeof tv,
2646 				    sizeof tv);
2647 			if (error)
2648 				goto bad;
2649 			if (tv.tv_sec < 0 || tv.tv_usec < 0 ||
2650 			    tv.tv_usec >= 1000000) {
2651 				error = EDOM;
2652 				goto bad;
2653 			}
2654 			if (tv.tv_sec > INT32_MAX)
2655 				val = SBT_MAX;
2656 			else
2657 				val = tvtosbt(tv);
2658 			switch (sopt->sopt_name) {
2659 			case SO_SNDTIMEO:
2660 				so->so_snd.sb_timeo = val;
2661 				break;
2662 			case SO_RCVTIMEO:
2663 				so->so_rcv.sb_timeo = val;
2664 				break;
2665 			}
2666 			break;
2667 
2668 		case SO_LABEL:
2669 #ifdef MAC
2670 			error = sooptcopyin(sopt, &extmac, sizeof extmac,
2671 			    sizeof extmac);
2672 			if (error)
2673 				goto bad;
2674 			error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
2675 			    so, &extmac);
2676 #else
2677 			error = EOPNOTSUPP;
2678 #endif
2679 			break;
2680 
2681 		default:
2682 			if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
2683 				error = hhook_run_socket(so, sopt,
2684 				    HHOOK_SOCKET_OPT);
2685 			else
2686 				error = ENOPROTOOPT;
2687 			break;
2688 		}
2689 		if (error == 0 && so->so_proto->pr_ctloutput != NULL)
2690 			(void)(*so->so_proto->pr_ctloutput)(so, sopt);
2691 	}
2692 bad:
2693 	CURVNET_RESTORE();
2694 	return (error);
2695 }
2696 
2697 /*
2698  * Helper routine for getsockopt.
2699  */
2700 int
2701 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
2702 {
2703 	int	error;
2704 	size_t	valsize;
2705 
2706 	error = 0;
2707 
2708 	/*
2709 	 * Documented get behavior is that we always return a value, possibly
2710 	 * truncated to fit in the user's buffer.  Traditional behavior is
2711 	 * that we always tell the user precisely how much we copied, rather
2712 	 * than something useful like the total amount we had available for
2713 	 * her.  Note that this interface is not idempotent; the entire
2714 	 * answer must generated ahead of time.
2715 	 */
2716 	valsize = min(len, sopt->sopt_valsize);
2717 	sopt->sopt_valsize = valsize;
2718 	if (sopt->sopt_val != NULL) {
2719 		if (sopt->sopt_td != NULL)
2720 			error = copyout(buf, sopt->sopt_val, valsize);
2721 		else
2722 			bcopy(buf, sopt->sopt_val, valsize);
2723 	}
2724 	return (error);
2725 }
2726 
2727 int
2728 sogetopt(struct socket *so, struct sockopt *sopt)
2729 {
2730 	int	error, optval;
2731 	struct	linger l;
2732 	struct	timeval tv;
2733 #ifdef MAC
2734 	struct mac extmac;
2735 #endif
2736 
2737 	CURVNET_SET(so->so_vnet);
2738 	error = 0;
2739 	if (sopt->sopt_level != SOL_SOCKET) {
2740 		if (so->so_proto->pr_ctloutput != NULL)
2741 			error = (*so->so_proto->pr_ctloutput)(so, sopt);
2742 		else
2743 			error = ENOPROTOOPT;
2744 		CURVNET_RESTORE();
2745 		return (error);
2746 	} else {
2747 		switch (sopt->sopt_name) {
2748 		case SO_ACCEPTFILTER:
2749 			error = do_getopt_accept_filter(so, sopt);
2750 			break;
2751 
2752 		case SO_LINGER:
2753 			SOCK_LOCK(so);
2754 			l.l_onoff = so->so_options & SO_LINGER;
2755 			l.l_linger = so->so_linger;
2756 			SOCK_UNLOCK(so);
2757 			error = sooptcopyout(sopt, &l, sizeof l);
2758 			break;
2759 
2760 		case SO_USELOOPBACK:
2761 		case SO_DONTROUTE:
2762 		case SO_DEBUG:
2763 		case SO_KEEPALIVE:
2764 		case SO_REUSEADDR:
2765 		case SO_REUSEPORT:
2766 		case SO_BROADCAST:
2767 		case SO_OOBINLINE:
2768 		case SO_ACCEPTCONN:
2769 		case SO_TIMESTAMP:
2770 		case SO_BINTIME:
2771 		case SO_NOSIGPIPE:
2772 			optval = so->so_options & sopt->sopt_name;
2773 integer:
2774 			error = sooptcopyout(sopt, &optval, sizeof optval);
2775 			break;
2776 
2777 		case SO_TYPE:
2778 			optval = so->so_type;
2779 			goto integer;
2780 
2781 		case SO_PROTOCOL:
2782 			optval = so->so_proto->pr_protocol;
2783 			goto integer;
2784 
2785 		case SO_ERROR:
2786 			SOCK_LOCK(so);
2787 			optval = so->so_error;
2788 			so->so_error = 0;
2789 			SOCK_UNLOCK(so);
2790 			goto integer;
2791 
2792 		case SO_SNDBUF:
2793 			optval = so->so_snd.sb_hiwat;
2794 			goto integer;
2795 
2796 		case SO_RCVBUF:
2797 			optval = so->so_rcv.sb_hiwat;
2798 			goto integer;
2799 
2800 		case SO_SNDLOWAT:
2801 			optval = so->so_snd.sb_lowat;
2802 			goto integer;
2803 
2804 		case SO_RCVLOWAT:
2805 			optval = so->so_rcv.sb_lowat;
2806 			goto integer;
2807 
2808 		case SO_SNDTIMEO:
2809 		case SO_RCVTIMEO:
2810 			tv = sbttotv(sopt->sopt_name == SO_SNDTIMEO ?
2811 			    so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
2812 #ifdef COMPAT_FREEBSD32
2813 			if (SV_CURPROC_FLAG(SV_ILP32)) {
2814 				struct timeval32 tv32;
2815 
2816 				CP(tv, tv32, tv_sec);
2817 				CP(tv, tv32, tv_usec);
2818 				error = sooptcopyout(sopt, &tv32, sizeof tv32);
2819 			} else
2820 #endif
2821 				error = sooptcopyout(sopt, &tv, sizeof tv);
2822 			break;
2823 
2824 		case SO_LABEL:
2825 #ifdef MAC
2826 			error = sooptcopyin(sopt, &extmac, sizeof(extmac),
2827 			    sizeof(extmac));
2828 			if (error)
2829 				goto bad;
2830 			error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
2831 			    so, &extmac);
2832 			if (error)
2833 				goto bad;
2834 			error = sooptcopyout(sopt, &extmac, sizeof extmac);
2835 #else
2836 			error = EOPNOTSUPP;
2837 #endif
2838 			break;
2839 
2840 		case SO_PEERLABEL:
2841 #ifdef MAC
2842 			error = sooptcopyin(sopt, &extmac, sizeof(extmac),
2843 			    sizeof(extmac));
2844 			if (error)
2845 				goto bad;
2846 			error = mac_getsockopt_peerlabel(
2847 			    sopt->sopt_td->td_ucred, so, &extmac);
2848 			if (error)
2849 				goto bad;
2850 			error = sooptcopyout(sopt, &extmac, sizeof extmac);
2851 #else
2852 			error = EOPNOTSUPP;
2853 #endif
2854 			break;
2855 
2856 		case SO_LISTENQLIMIT:
2857 			optval = so->so_qlimit;
2858 			goto integer;
2859 
2860 		case SO_LISTENQLEN:
2861 			optval = so->so_qlen;
2862 			goto integer;
2863 
2864 		case SO_LISTENINCQLEN:
2865 			optval = so->so_incqlen;
2866 			goto integer;
2867 
2868 		default:
2869 			if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
2870 				error = hhook_run_socket(so, sopt,
2871 				    HHOOK_SOCKET_OPT);
2872 			else
2873 				error = ENOPROTOOPT;
2874 			break;
2875 		}
2876 	}
2877 #ifdef MAC
2878 bad:
2879 #endif
2880 	CURVNET_RESTORE();
2881 	return (error);
2882 }
2883 
2884 int
2885 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
2886 {
2887 	struct mbuf *m, *m_prev;
2888 	int sopt_size = sopt->sopt_valsize;
2889 
2890 	MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
2891 	if (m == NULL)
2892 		return ENOBUFS;
2893 	if (sopt_size > MLEN) {
2894 		MCLGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT);
2895 		if ((m->m_flags & M_EXT) == 0) {
2896 			m_free(m);
2897 			return ENOBUFS;
2898 		}
2899 		m->m_len = min(MCLBYTES, sopt_size);
2900 	} else {
2901 		m->m_len = min(MLEN, sopt_size);
2902 	}
2903 	sopt_size -= m->m_len;
2904 	*mp = m;
2905 	m_prev = m;
2906 
2907 	while (sopt_size) {
2908 		MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
2909 		if (m == NULL) {
2910 			m_freem(*mp);
2911 			return ENOBUFS;
2912 		}
2913 		if (sopt_size > MLEN) {
2914 			MCLGET(m, sopt->sopt_td != NULL ? M_WAITOK :
2915 			    M_NOWAIT);
2916 			if ((m->m_flags & M_EXT) == 0) {
2917 				m_freem(m);
2918 				m_freem(*mp);
2919 				return ENOBUFS;
2920 			}
2921 			m->m_len = min(MCLBYTES, sopt_size);
2922 		} else {
2923 			m->m_len = min(MLEN, sopt_size);
2924 		}
2925 		sopt_size -= m->m_len;
2926 		m_prev->m_next = m;
2927 		m_prev = m;
2928 	}
2929 	return (0);
2930 }
2931 
2932 int
2933 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
2934 {
2935 	struct mbuf *m0 = m;
2936 
2937 	if (sopt->sopt_val == NULL)
2938 		return (0);
2939 	while (m != NULL && sopt->sopt_valsize >= m->m_len) {
2940 		if (sopt->sopt_td != NULL) {
2941 			int error;
2942 
2943 			error = copyin(sopt->sopt_val, mtod(m, char *),
2944 			    m->m_len);
2945 			if (error != 0) {
2946 				m_freem(m0);
2947 				return(error);
2948 			}
2949 		} else
2950 			bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
2951 		sopt->sopt_valsize -= m->m_len;
2952 		sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
2953 		m = m->m_next;
2954 	}
2955 	if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
2956 		panic("ip6_sooptmcopyin");
2957 	return (0);
2958 }
2959 
2960 int
2961 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
2962 {
2963 	struct mbuf *m0 = m;
2964 	size_t valsize = 0;
2965 
2966 	if (sopt->sopt_val == NULL)
2967 		return (0);
2968 	while (m != NULL && sopt->sopt_valsize >= m->m_len) {
2969 		if (sopt->sopt_td != NULL) {
2970 			int error;
2971 
2972 			error = copyout(mtod(m, char *), sopt->sopt_val,
2973 			    m->m_len);
2974 			if (error != 0) {
2975 				m_freem(m0);
2976 				return(error);
2977 			}
2978 		} else
2979 			bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
2980 		sopt->sopt_valsize -= m->m_len;
2981 		sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
2982 		valsize += m->m_len;
2983 		m = m->m_next;
2984 	}
2985 	if (m != NULL) {
2986 		/* enough soopt buffer should be given from user-land */
2987 		m_freem(m0);
2988 		return(EINVAL);
2989 	}
2990 	sopt->sopt_valsize = valsize;
2991 	return (0);
2992 }
2993 
2994 /*
2995  * sohasoutofband(): protocol notifies socket layer of the arrival of new
2996  * out-of-band data, which will then notify socket consumers.
2997  */
2998 void
2999 sohasoutofband(struct socket *so)
3000 {
3001 
3002 	if (so->so_sigio != NULL)
3003 		pgsigio(&so->so_sigio, SIGURG, 0);
3004 	selwakeuppri(&so->so_rcv.sb_sel, PSOCK);
3005 }
3006 
3007 int
3008 sopoll(struct socket *so, int events, struct ucred *active_cred,
3009     struct thread *td)
3010 {
3011 
3012 	/*
3013 	 * We do not need to set or assert curvnet as long as everyone uses
3014 	 * sopoll_generic().
3015 	 */
3016 	return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred,
3017 	    td));
3018 }
3019 
3020 int
3021 sopoll_generic(struct socket *so, int events, struct ucred *active_cred,
3022     struct thread *td)
3023 {
3024 	int revents = 0;
3025 
3026 	SOCKBUF_LOCK(&so->so_snd);
3027 	SOCKBUF_LOCK(&so->so_rcv);
3028 	if (events & (POLLIN | POLLRDNORM))
3029 		if (soreadabledata(so))
3030 			revents |= events & (POLLIN | POLLRDNORM);
3031 
3032 	if (events & (POLLOUT | POLLWRNORM))
3033 		if (sowriteable(so))
3034 			revents |= events & (POLLOUT | POLLWRNORM);
3035 
3036 	if (events & (POLLPRI | POLLRDBAND))
3037 		if (so->so_oobmark || (so->so_rcv.sb_state & SBS_RCVATMARK))
3038 			revents |= events & (POLLPRI | POLLRDBAND);
3039 
3040 	if ((events & POLLINIGNEOF) == 0) {
3041 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3042 			revents |= events & (POLLIN | POLLRDNORM);
3043 			if (so->so_snd.sb_state & SBS_CANTSENDMORE)
3044 				revents |= POLLHUP;
3045 		}
3046 	}
3047 
3048 	if (revents == 0) {
3049 		if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
3050 			selrecord(td, &so->so_rcv.sb_sel);
3051 			so->so_rcv.sb_flags |= SB_SEL;
3052 		}
3053 
3054 		if (events & (POLLOUT | POLLWRNORM)) {
3055 			selrecord(td, &so->so_snd.sb_sel);
3056 			so->so_snd.sb_flags |= SB_SEL;
3057 		}
3058 	}
3059 
3060 	SOCKBUF_UNLOCK(&so->so_rcv);
3061 	SOCKBUF_UNLOCK(&so->so_snd);
3062 	return (revents);
3063 }
3064 
3065 int
3066 soo_kqfilter(struct file *fp, struct knote *kn)
3067 {
3068 	struct socket *so = kn->kn_fp->f_data;
3069 	struct sockbuf *sb;
3070 
3071 	switch (kn->kn_filter) {
3072 	case EVFILT_READ:
3073 		if (so->so_options & SO_ACCEPTCONN)
3074 			kn->kn_fop = &solisten_filtops;
3075 		else
3076 			kn->kn_fop = &soread_filtops;
3077 		sb = &so->so_rcv;
3078 		break;
3079 	case EVFILT_WRITE:
3080 		kn->kn_fop = &sowrite_filtops;
3081 		sb = &so->so_snd;
3082 		break;
3083 	default:
3084 		return (EINVAL);
3085 	}
3086 
3087 	SOCKBUF_LOCK(sb);
3088 	knlist_add(&sb->sb_sel.si_note, kn, 1);
3089 	sb->sb_flags |= SB_KNOTE;
3090 	SOCKBUF_UNLOCK(sb);
3091 	return (0);
3092 }
3093 
3094 /*
3095  * Some routines that return EOPNOTSUPP for entry points that are not
3096  * supported by a protocol.  Fill in as needed.
3097  */
3098 int
3099 pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
3100 {
3101 
3102 	return EOPNOTSUPP;
3103 }
3104 
3105 int
3106 pru_aio_queue_notsupp(struct socket *so, struct kaiocb *job)
3107 {
3108 
3109 	return EOPNOTSUPP;
3110 }
3111 
3112 int
3113 pru_attach_notsupp(struct socket *so, int proto, struct thread *td)
3114 {
3115 
3116 	return EOPNOTSUPP;
3117 }
3118 
3119 int
3120 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3121 {
3122 
3123 	return EOPNOTSUPP;
3124 }
3125 
3126 int
3127 pru_bindat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3128     struct thread *td)
3129 {
3130 
3131 	return EOPNOTSUPP;
3132 }
3133 
3134 int
3135 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3136 {
3137 
3138 	return EOPNOTSUPP;
3139 }
3140 
3141 int
3142 pru_connectat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3143     struct thread *td)
3144 {
3145 
3146 	return EOPNOTSUPP;
3147 }
3148 
3149 int
3150 pru_connect2_notsupp(struct socket *so1, struct socket *so2)
3151 {
3152 
3153 	return EOPNOTSUPP;
3154 }
3155 
3156 int
3157 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
3158     struct ifnet *ifp, struct thread *td)
3159 {
3160 
3161 	return EOPNOTSUPP;
3162 }
3163 
3164 int
3165 pru_disconnect_notsupp(struct socket *so)
3166 {
3167 
3168 	return EOPNOTSUPP;
3169 }
3170 
3171 int
3172 pru_listen_notsupp(struct socket *so, int backlog, struct thread *td)
3173 {
3174 
3175 	return EOPNOTSUPP;
3176 }
3177 
3178 int
3179 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam)
3180 {
3181 
3182 	return EOPNOTSUPP;
3183 }
3184 
3185 int
3186 pru_rcvd_notsupp(struct socket *so, int flags)
3187 {
3188 
3189 	return EOPNOTSUPP;
3190 }
3191 
3192 int
3193 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
3194 {
3195 
3196 	return EOPNOTSUPP;
3197 }
3198 
3199 int
3200 pru_send_notsupp(struct socket *so, int flags, struct mbuf *m,
3201     struct sockaddr *addr, struct mbuf *control, struct thread *td)
3202 {
3203 
3204 	return EOPNOTSUPP;
3205 }
3206 
3207 int
3208 pru_ready_notsupp(struct socket *so, struct mbuf *m, int count)
3209 {
3210 
3211 	return (EOPNOTSUPP);
3212 }
3213 
3214 /*
3215  * This isn't really a ``null'' operation, but it's the default one and
3216  * doesn't do anything destructive.
3217  */
3218 int
3219 pru_sense_null(struct socket *so, struct stat *sb)
3220 {
3221 
3222 	sb->st_blksize = so->so_snd.sb_hiwat;
3223 	return 0;
3224 }
3225 
3226 int
3227 pru_shutdown_notsupp(struct socket *so)
3228 {
3229 
3230 	return EOPNOTSUPP;
3231 }
3232 
3233 int
3234 pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam)
3235 {
3236 
3237 	return EOPNOTSUPP;
3238 }
3239 
3240 int
3241 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio,
3242     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
3243 {
3244 
3245 	return EOPNOTSUPP;
3246 }
3247 
3248 int
3249 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr,
3250     struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
3251 {
3252 
3253 	return EOPNOTSUPP;
3254 }
3255 
3256 int
3257 pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred,
3258     struct thread *td)
3259 {
3260 
3261 	return EOPNOTSUPP;
3262 }
3263 
3264 static void
3265 filt_sordetach(struct knote *kn)
3266 {
3267 	struct socket *so = kn->kn_fp->f_data;
3268 
3269 	SOCKBUF_LOCK(&so->so_rcv);
3270 	knlist_remove(&so->so_rcv.sb_sel.si_note, kn, 1);
3271 	if (knlist_empty(&so->so_rcv.sb_sel.si_note))
3272 		so->so_rcv.sb_flags &= ~SB_KNOTE;
3273 	SOCKBUF_UNLOCK(&so->so_rcv);
3274 }
3275 
3276 /*ARGSUSED*/
3277 static int
3278 filt_soread(struct knote *kn, long hint)
3279 {
3280 	struct socket *so;
3281 
3282 	so = kn->kn_fp->f_data;
3283 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
3284 
3285 	kn->kn_data = sbavail(&so->so_rcv) - so->so_rcv.sb_ctl;
3286 	if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3287 		kn->kn_flags |= EV_EOF;
3288 		kn->kn_fflags = so->so_error;
3289 		return (1);
3290 	} else if (so->so_error)	/* temporary udp error */
3291 		return (1);
3292 
3293 	if (kn->kn_sfflags & NOTE_LOWAT) {
3294 		if (kn->kn_data >= kn->kn_sdata)
3295 			return 1;
3296 	} else {
3297 		if (sbavail(&so->so_rcv) >= so->so_rcv.sb_lowat)
3298 			return 1;
3299 	}
3300 
3301 	/* This hook returning non-zero indicates an event, not error */
3302 	return (hhook_run_socket(so, NULL, HHOOK_FILT_SOREAD));
3303 }
3304 
3305 static void
3306 filt_sowdetach(struct knote *kn)
3307 {
3308 	struct socket *so = kn->kn_fp->f_data;
3309 
3310 	SOCKBUF_LOCK(&so->so_snd);
3311 	knlist_remove(&so->so_snd.sb_sel.si_note, kn, 1);
3312 	if (knlist_empty(&so->so_snd.sb_sel.si_note))
3313 		so->so_snd.sb_flags &= ~SB_KNOTE;
3314 	SOCKBUF_UNLOCK(&so->so_snd);
3315 }
3316 
3317 /*ARGSUSED*/
3318 static int
3319 filt_sowrite(struct knote *kn, long hint)
3320 {
3321 	struct socket *so;
3322 
3323 	so = kn->kn_fp->f_data;
3324 	SOCKBUF_LOCK_ASSERT(&so->so_snd);
3325 	kn->kn_data = sbspace(&so->so_snd);
3326 
3327 	hhook_run_socket(so, kn, HHOOK_FILT_SOWRITE);
3328 
3329 	if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
3330 		kn->kn_flags |= EV_EOF;
3331 		kn->kn_fflags = so->so_error;
3332 		return (1);
3333 	} else if (so->so_error)	/* temporary udp error */
3334 		return (1);
3335 	else if (((so->so_state & SS_ISCONNECTED) == 0) &&
3336 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
3337 		return (0);
3338 	else if (kn->kn_sfflags & NOTE_LOWAT)
3339 		return (kn->kn_data >= kn->kn_sdata);
3340 	else
3341 		return (kn->kn_data >= so->so_snd.sb_lowat);
3342 }
3343 
3344 /*ARGSUSED*/
3345 static int
3346 filt_solisten(struct knote *kn, long hint)
3347 {
3348 	struct socket *so = kn->kn_fp->f_data;
3349 
3350 	kn->kn_data = so->so_qlen;
3351 	return (!TAILQ_EMPTY(&so->so_comp));
3352 }
3353 
3354 int
3355 socheckuid(struct socket *so, uid_t uid)
3356 {
3357 
3358 	if (so == NULL)
3359 		return (EPERM);
3360 	if (so->so_cred->cr_uid != uid)
3361 		return (EPERM);
3362 	return (0);
3363 }
3364 
3365 /*
3366  * These functions are used by protocols to notify the socket layer (and its
3367  * consumers) of state changes in the sockets driven by protocol-side events.
3368  */
3369 
3370 /*
3371  * Procedures to manipulate state flags of socket and do appropriate wakeups.
3372  *
3373  * Normal sequence from the active (originating) side is that
3374  * soisconnecting() is called during processing of connect() call, resulting
3375  * in an eventual call to soisconnected() if/when the connection is
3376  * established.  When the connection is torn down soisdisconnecting() is
3377  * called during processing of disconnect() call, and soisdisconnected() is
3378  * called when the connection to the peer is totally severed.  The semantics
3379  * of these routines are such that connectionless protocols can call
3380  * soisconnected() and soisdisconnected() only, bypassing the in-progress
3381  * calls when setting up a ``connection'' takes no time.
3382  *
3383  * From the passive side, a socket is created with two queues of sockets:
3384  * so_incomp for connections in progress and so_comp for connections already
3385  * made and awaiting user acceptance.  As a protocol is preparing incoming
3386  * connections, it creates a socket structure queued on so_incomp by calling
3387  * sonewconn().  When the connection is established, soisconnected() is
3388  * called, and transfers the socket structure to so_comp, making it available
3389  * to accept().
3390  *
3391  * If a socket is closed with sockets on either so_incomp or so_comp, these
3392  * sockets are dropped.
3393  *
3394  * If higher-level protocols are implemented in the kernel, the wakeups done
3395  * here will sometimes cause software-interrupt process scheduling.
3396  */
3397 void
3398 soisconnecting(struct socket *so)
3399 {
3400 
3401 	SOCK_LOCK(so);
3402 	so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
3403 	so->so_state |= SS_ISCONNECTING;
3404 	SOCK_UNLOCK(so);
3405 }
3406 
3407 void
3408 soisconnected(struct socket *so)
3409 {
3410 	struct socket *head;
3411 	int ret;
3412 
3413 restart:
3414 	ACCEPT_LOCK();
3415 	SOCK_LOCK(so);
3416 	so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
3417 	so->so_state |= SS_ISCONNECTED;
3418 	head = so->so_head;
3419 	if (head != NULL && (so->so_qstate & SQ_INCOMP)) {
3420 		if ((so->so_options & SO_ACCEPTFILTER) == 0) {
3421 			SOCK_UNLOCK(so);
3422 			TAILQ_REMOVE(&head->so_incomp, so, so_list);
3423 			head->so_incqlen--;
3424 			so->so_qstate &= ~SQ_INCOMP;
3425 			TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
3426 			head->so_qlen++;
3427 			so->so_qstate |= SQ_COMP;
3428 			ACCEPT_UNLOCK();
3429 			sorwakeup(head);
3430 			wakeup_one(&head->so_timeo);
3431 		} else {
3432 			ACCEPT_UNLOCK();
3433 			soupcall_set(so, SO_RCV,
3434 			    head->so_accf->so_accept_filter->accf_callback,
3435 			    head->so_accf->so_accept_filter_arg);
3436 			so->so_options &= ~SO_ACCEPTFILTER;
3437 			ret = head->so_accf->so_accept_filter->accf_callback(so,
3438 			    head->so_accf->so_accept_filter_arg, M_NOWAIT);
3439 			if (ret == SU_ISCONNECTED)
3440 				soupcall_clear(so, SO_RCV);
3441 			SOCK_UNLOCK(so);
3442 			if (ret == SU_ISCONNECTED)
3443 				goto restart;
3444 		}
3445 		return;
3446 	}
3447 	SOCK_UNLOCK(so);
3448 	ACCEPT_UNLOCK();
3449 	wakeup(&so->so_timeo);
3450 	sorwakeup(so);
3451 	sowwakeup(so);
3452 }
3453 
3454 void
3455 soisdisconnecting(struct socket *so)
3456 {
3457 
3458 	/*
3459 	 * Note: This code assumes that SOCK_LOCK(so) and
3460 	 * SOCKBUF_LOCK(&so->so_rcv) are the same.
3461 	 */
3462 	SOCKBUF_LOCK(&so->so_rcv);
3463 	so->so_state &= ~SS_ISCONNECTING;
3464 	so->so_state |= SS_ISDISCONNECTING;
3465 	socantrcvmore_locked(so);
3466 	SOCKBUF_LOCK(&so->so_snd);
3467 	socantsendmore_locked(so);
3468 	wakeup(&so->so_timeo);
3469 }
3470 
3471 void
3472 soisdisconnected(struct socket *so)
3473 {
3474 
3475 	/*
3476 	 * Note: This code assumes that SOCK_LOCK(so) and
3477 	 * SOCKBUF_LOCK(&so->so_rcv) are the same.
3478 	 */
3479 	SOCKBUF_LOCK(&so->so_rcv);
3480 	so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
3481 	so->so_state |= SS_ISDISCONNECTED;
3482 	socantrcvmore_locked(so);
3483 	SOCKBUF_LOCK(&so->so_snd);
3484 	sbdrop_locked(&so->so_snd, sbused(&so->so_snd));
3485 	socantsendmore_locked(so);
3486 	wakeup(&so->so_timeo);
3487 }
3488 
3489 /*
3490  * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
3491  */
3492 struct sockaddr *
3493 sodupsockaddr(const struct sockaddr *sa, int mflags)
3494 {
3495 	struct sockaddr *sa2;
3496 
3497 	sa2 = malloc(sa->sa_len, M_SONAME, mflags);
3498 	if (sa2)
3499 		bcopy(sa, sa2, sa->sa_len);
3500 	return sa2;
3501 }
3502 
3503 /*
3504  * Register per-socket buffer upcalls.
3505  */
3506 void
3507 soupcall_set(struct socket *so, int which,
3508     int (*func)(struct socket *, void *, int), void *arg)
3509 {
3510 	struct sockbuf *sb;
3511 
3512 	switch (which) {
3513 	case SO_RCV:
3514 		sb = &so->so_rcv;
3515 		break;
3516 	case SO_SND:
3517 		sb = &so->so_snd;
3518 		break;
3519 	default:
3520 		panic("soupcall_set: bad which");
3521 	}
3522 	SOCKBUF_LOCK_ASSERT(sb);
3523 #if 0
3524 	/* XXX: accf_http actually wants to do this on purpose. */
3525 	KASSERT(sb->sb_upcall == NULL, ("soupcall_set: overwriting upcall"));
3526 #endif
3527 	sb->sb_upcall = func;
3528 	sb->sb_upcallarg = arg;
3529 	sb->sb_flags |= SB_UPCALL;
3530 }
3531 
3532 void
3533 soupcall_clear(struct socket *so, int which)
3534 {
3535 	struct sockbuf *sb;
3536 
3537 	switch (which) {
3538 	case SO_RCV:
3539 		sb = &so->so_rcv;
3540 		break;
3541 	case SO_SND:
3542 		sb = &so->so_snd;
3543 		break;
3544 	default:
3545 		panic("soupcall_clear: bad which");
3546 	}
3547 	SOCKBUF_LOCK_ASSERT(sb);
3548 	KASSERT(sb->sb_upcall != NULL, ("soupcall_clear: no upcall to clear"));
3549 	sb->sb_upcall = NULL;
3550 	sb->sb_upcallarg = NULL;
3551 	sb->sb_flags &= ~SB_UPCALL;
3552 }
3553 
3554 /*
3555  * Create an external-format (``xsocket'') structure using the information in
3556  * the kernel-format socket structure pointed to by so.  This is done to
3557  * reduce the spew of irrelevant information over this interface, to isolate
3558  * user code from changes in the kernel structure, and potentially to provide
3559  * information-hiding if we decide that some of this information should be
3560  * hidden from users.
3561  */
3562 void
3563 sotoxsocket(struct socket *so, struct xsocket *xso)
3564 {
3565 
3566 	xso->xso_len = sizeof *xso;
3567 	xso->xso_so = so;
3568 	xso->so_type = so->so_type;
3569 	xso->so_options = so->so_options;
3570 	xso->so_linger = so->so_linger;
3571 	xso->so_state = so->so_state;
3572 	xso->so_pcb = so->so_pcb;
3573 	xso->xso_protocol = so->so_proto->pr_protocol;
3574 	xso->xso_family = so->so_proto->pr_domain->dom_family;
3575 	xso->so_qlen = so->so_qlen;
3576 	xso->so_incqlen = so->so_incqlen;
3577 	xso->so_qlimit = so->so_qlimit;
3578 	xso->so_timeo = so->so_timeo;
3579 	xso->so_error = so->so_error;
3580 	xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
3581 	xso->so_oobmark = so->so_oobmark;
3582 	sbtoxsockbuf(&so->so_snd, &xso->so_snd);
3583 	sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
3584 	xso->so_uid = so->so_cred->cr_uid;
3585 }
3586 
3587 
3588 /*
3589  * Socket accessor functions to provide external consumers with
3590  * a safe interface to socket state
3591  *
3592  */
3593 
3594 void
3595 so_listeners_apply_all(struct socket *so, void (*func)(struct socket *, void *),
3596     void *arg)
3597 {
3598 
3599 	TAILQ_FOREACH(so, &so->so_comp, so_list)
3600 		func(so, arg);
3601 }
3602 
3603 struct sockbuf *
3604 so_sockbuf_rcv(struct socket *so)
3605 {
3606 
3607 	return (&so->so_rcv);
3608 }
3609 
3610 struct sockbuf *
3611 so_sockbuf_snd(struct socket *so)
3612 {
3613 
3614 	return (&so->so_snd);
3615 }
3616 
3617 int
3618 so_state_get(const struct socket *so)
3619 {
3620 
3621 	return (so->so_state);
3622 }
3623 
3624 void
3625 so_state_set(struct socket *so, int val)
3626 {
3627 
3628 	so->so_state = val;
3629 }
3630 
3631 int
3632 so_options_get(const struct socket *so)
3633 {
3634 
3635 	return (so->so_options);
3636 }
3637 
3638 void
3639 so_options_set(struct socket *so, int val)
3640 {
3641 
3642 	so->so_options = val;
3643 }
3644 
3645 int
3646 so_error_get(const struct socket *so)
3647 {
3648 
3649 	return (so->so_error);
3650 }
3651 
3652 void
3653 so_error_set(struct socket *so, int val)
3654 {
3655 
3656 	so->so_error = val;
3657 }
3658 
3659 int
3660 so_linger_get(const struct socket *so)
3661 {
3662 
3663 	return (so->so_linger);
3664 }
3665 
3666 void
3667 so_linger_set(struct socket *so, int val)
3668 {
3669 
3670 	so->so_linger = val;
3671 }
3672 
3673 struct protosw *
3674 so_protosw_get(const struct socket *so)
3675 {
3676 
3677 	return (so->so_proto);
3678 }
3679 
3680 void
3681 so_protosw_set(struct socket *so, struct protosw *val)
3682 {
3683 
3684 	so->so_proto = val;
3685 }
3686 
3687 void
3688 so_sorwakeup(struct socket *so)
3689 {
3690 
3691 	sorwakeup(so);
3692 }
3693 
3694 void
3695 so_sowwakeup(struct socket *so)
3696 {
3697 
3698 	sowwakeup(so);
3699 }
3700 
3701 void
3702 so_sorwakeup_locked(struct socket *so)
3703 {
3704 
3705 	sorwakeup_locked(so);
3706 }
3707 
3708 void
3709 so_sowwakeup_locked(struct socket *so)
3710 {
3711 
3712 	sowwakeup_locked(so);
3713 }
3714 
3715 void
3716 so_lock(struct socket *so)
3717 {
3718 
3719 	SOCK_LOCK(so);
3720 }
3721 
3722 void
3723 so_unlock(struct socket *so)
3724 {
3725 
3726 	SOCK_UNLOCK(so);
3727 }
3728