xref: /freebsd/sys/kern/uipc_socket.c (revision 8522d140a568be6044aad4288042c72e8d3b72a7)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1988, 1990, 1993
5  *	The Regents of the University of California.
6  * Copyright (c) 2004 The FreeBSD Foundation
7  * Copyright (c) 2004-2008 Robert N. M. Watson
8  * All rights reserved.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)uipc_socket.c	8.3 (Berkeley) 4/15/94
35  */
36 
37 /*
38  * Comments on the socket life cycle:
39  *
40  * soalloc() sets of socket layer state for a socket, called only by
41  * socreate() and sonewconn().  Socket layer private.
42  *
43  * sodealloc() tears down socket layer state for a socket, called only by
44  * sofree() and sonewconn().  Socket layer private.
45  *
46  * pru_attach() associates protocol layer state with an allocated socket;
47  * called only once, may fail, aborting socket allocation.  This is called
48  * from socreate() and sonewconn().  Socket layer private.
49  *
50  * pru_detach() disassociates protocol layer state from an attached socket,
51  * and will be called exactly once for sockets in which pru_attach() has
52  * been successfully called.  If pru_attach() returned an error,
53  * pru_detach() will not be called.  Socket layer private.
54  *
55  * pru_abort() and pru_close() notify the protocol layer that the last
56  * consumer of a socket is starting to tear down the socket, and that the
57  * protocol should terminate the connection.  Historically, pru_abort() also
58  * detached protocol state from the socket state, but this is no longer the
59  * case.
60  *
61  * socreate() creates a socket and attaches protocol state.  This is a public
62  * interface that may be used by socket layer consumers to create new
63  * sockets.
64  *
65  * sonewconn() creates a socket and attaches protocol state.  This is a
66  * public interface  that may be used by protocols to create new sockets when
67  * a new connection is received and will be available for accept() on a
68  * listen socket.
69  *
70  * soclose() destroys a socket after possibly waiting for it to disconnect.
71  * This is a public interface that socket consumers should use to close and
72  * release a socket when done with it.
73  *
74  * soabort() destroys a socket without waiting for it to disconnect (used
75  * only for incoming connections that are already partially or fully
76  * connected).  This is used internally by the socket layer when clearing
77  * listen socket queues (due to overflow or close on the listen socket), but
78  * is also a public interface protocols may use to abort connections in
79  * their incomplete listen queues should they no longer be required.  Sockets
80  * placed in completed connection listen queues should not be aborted for
81  * reasons described in the comment above the soclose() implementation.  This
82  * is not a general purpose close routine, and except in the specific
83  * circumstances described here, should not be used.
84  *
85  * sofree() will free a socket and its protocol state if all references on
86  * the socket have been released, and is the public interface to attempt to
87  * free a socket when a reference is removed.  This is a socket layer private
88  * interface.
89  *
90  * NOTE: In addition to socreate() and soclose(), which provide a single
91  * socket reference to the consumer to be managed as required, there are two
92  * calls to explicitly manage socket references, soref(), and sorele().
93  * Currently, these are generally required only when transitioning a socket
94  * from a listen queue to a file descriptor, in order to prevent garbage
95  * collection of the socket at an untimely moment.  For a number of reasons,
96  * these interfaces are not preferred, and should be avoided.
97  *
98  * NOTE: With regard to VNETs the general rule is that callers do not set
99  * curvnet. Exceptions to this rule include soabort(), sodisconnect(),
100  * sofree() (and with that sorele(), sotryfree()), as well as sonewconn()
101  * and sorflush(), which are usually called from a pre-set VNET context.
102  * sopoll() currently does not need a VNET context to be set.
103  */
104 
105 #include <sys/cdefs.h>
106 __FBSDID("$FreeBSD$");
107 
108 #include "opt_inet.h"
109 #include "opt_inet6.h"
110 #include "opt_sctp.h"
111 
112 #include <sys/param.h>
113 #include <sys/systm.h>
114 #include <sys/fcntl.h>
115 #include <sys/limits.h>
116 #include <sys/lock.h>
117 #include <sys/mac.h>
118 #include <sys/malloc.h>
119 #include <sys/mbuf.h>
120 #include <sys/mutex.h>
121 #include <sys/domain.h>
122 #include <sys/file.h>			/* for struct knote */
123 #include <sys/hhook.h>
124 #include <sys/kernel.h>
125 #include <sys/khelp.h>
126 #include <sys/event.h>
127 #include <sys/eventhandler.h>
128 #include <sys/poll.h>
129 #include <sys/proc.h>
130 #include <sys/protosw.h>
131 #include <sys/socket.h>
132 #include <sys/socketvar.h>
133 #include <sys/resourcevar.h>
134 #include <net/route.h>
135 #include <sys/signalvar.h>
136 #include <sys/stat.h>
137 #include <sys/sx.h>
138 #include <sys/sysctl.h>
139 #include <sys/taskqueue.h>
140 #include <sys/uio.h>
141 #include <sys/jail.h>
142 #include <sys/syslog.h>
143 #include <netinet/in.h>
144 
145 #include <net/vnet.h>
146 
147 #include <security/mac/mac_framework.h>
148 
149 #include <vm/uma.h>
150 
151 #ifdef COMPAT_FREEBSD32
152 #include <sys/mount.h>
153 #include <sys/sysent.h>
154 #include <compat/freebsd32/freebsd32.h>
155 #endif
156 
157 static int	soreceive_rcvoob(struct socket *so, struct uio *uio,
158 		    int flags);
159 static void	so_rdknl_lock(void *);
160 static void	so_rdknl_unlock(void *);
161 static void	so_rdknl_assert_locked(void *);
162 static void	so_rdknl_assert_unlocked(void *);
163 static void	so_wrknl_lock(void *);
164 static void	so_wrknl_unlock(void *);
165 static void	so_wrknl_assert_locked(void *);
166 static void	so_wrknl_assert_unlocked(void *);
167 
168 static void	filt_sordetach(struct knote *kn);
169 static int	filt_soread(struct knote *kn, long hint);
170 static void	filt_sowdetach(struct knote *kn);
171 static int	filt_sowrite(struct knote *kn, long hint);
172 static int	filt_soempty(struct knote *kn, long hint);
173 static int inline hhook_run_socket(struct socket *so, void *hctx, int32_t h_id);
174 fo_kqfilter_t	soo_kqfilter;
175 
176 static struct filterops soread_filtops = {
177 	.f_isfd = 1,
178 	.f_detach = filt_sordetach,
179 	.f_event = filt_soread,
180 };
181 static struct filterops sowrite_filtops = {
182 	.f_isfd = 1,
183 	.f_detach = filt_sowdetach,
184 	.f_event = filt_sowrite,
185 };
186 static struct filterops soempty_filtops = {
187 	.f_isfd = 1,
188 	.f_detach = filt_sowdetach,
189 	.f_event = filt_soempty,
190 };
191 
192 so_gen_t	so_gencnt;	/* generation count for sockets */
193 
194 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
195 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
196 
197 #define	VNET_SO_ASSERT(so)						\
198 	VNET_ASSERT(curvnet != NULL,					\
199 	    ("%s:%d curvnet is NULL, so=%p", __func__, __LINE__, (so)));
200 
201 VNET_DEFINE(struct hhook_head *, socket_hhh[HHOOK_SOCKET_LAST + 1]);
202 #define	V_socket_hhh		VNET(socket_hhh)
203 
204 /*
205  * Limit on the number of connections in the listen queue waiting
206  * for accept(2).
207  * NB: The original sysctl somaxconn is still available but hidden
208  * to prevent confusion about the actual purpose of this number.
209  */
210 static u_int somaxconn = SOMAXCONN;
211 
212 static int
213 sysctl_somaxconn(SYSCTL_HANDLER_ARGS)
214 {
215 	int error;
216 	int val;
217 
218 	val = somaxconn;
219 	error = sysctl_handle_int(oidp, &val, 0, req);
220 	if (error || !req->newptr )
221 		return (error);
222 
223 	/*
224 	 * The purpose of the UINT_MAX / 3 limit, is so that the formula
225 	 *   3 * so_qlimit / 2
226 	 * below, will not overflow.
227          */
228 
229 	if (val < 1 || val > UINT_MAX / 3)
230 		return (EINVAL);
231 
232 	somaxconn = val;
233 	return (0);
234 }
235 SYSCTL_PROC(_kern_ipc, OID_AUTO, soacceptqueue, CTLTYPE_UINT | CTLFLAG_RW,
236     0, sizeof(int), sysctl_somaxconn, "I",
237     "Maximum listen socket pending connection accept queue size");
238 SYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn,
239     CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_SKIP,
240     0, sizeof(int), sysctl_somaxconn, "I",
241     "Maximum listen socket pending connection accept queue size (compat)");
242 
243 static int numopensockets;
244 SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD,
245     &numopensockets, 0, "Number of open sockets");
246 
247 /*
248  * accept_mtx locks down per-socket fields relating to accept queues.  See
249  * socketvar.h for an annotation of the protected fields of struct socket.
250  */
251 struct mtx accept_mtx;
252 MTX_SYSINIT(accept_mtx, &accept_mtx, "accept", MTX_DEF);
253 
254 /*
255  * so_global_mtx protects so_gencnt, numopensockets, and the per-socket
256  * so_gencnt field.
257  */
258 static struct mtx so_global_mtx;
259 MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF);
260 
261 /*
262  * General IPC sysctl name space, used by sockets and a variety of other IPC
263  * types.
264  */
265 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC");
266 
267 /*
268  * Initialize the socket subsystem and set up the socket
269  * memory allocator.
270  */
271 static uma_zone_t socket_zone;
272 int	maxsockets;
273 
274 static void
275 socket_zone_change(void *tag)
276 {
277 
278 	maxsockets = uma_zone_set_max(socket_zone, maxsockets);
279 }
280 
281 static void
282 socket_hhook_register(int subtype)
283 {
284 
285 	if (hhook_head_register(HHOOK_TYPE_SOCKET, subtype,
286 	    &V_socket_hhh[subtype],
287 	    HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
288 		printf("%s: WARNING: unable to register hook\n", __func__);
289 }
290 
291 static void
292 socket_hhook_deregister(int subtype)
293 {
294 
295 	if (hhook_head_deregister(V_socket_hhh[subtype]) != 0)
296 		printf("%s: WARNING: unable to deregister hook\n", __func__);
297 }
298 
299 static void
300 socket_init(void *tag)
301 {
302 
303 	socket_zone = uma_zcreate("socket", sizeof(struct socket), NULL, NULL,
304 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
305 	maxsockets = uma_zone_set_max(socket_zone, maxsockets);
306 	uma_zone_set_warning(socket_zone, "kern.ipc.maxsockets limit reached");
307 	EVENTHANDLER_REGISTER(maxsockets_change, socket_zone_change, NULL,
308 	    EVENTHANDLER_PRI_FIRST);
309 }
310 SYSINIT(socket, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, socket_init, NULL);
311 
312 static void
313 socket_vnet_init(const void *unused __unused)
314 {
315 	int i;
316 
317 	/* We expect a contiguous range */
318 	for (i = 0; i <= HHOOK_SOCKET_LAST; i++)
319 		socket_hhook_register(i);
320 }
321 VNET_SYSINIT(socket_vnet_init, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
322     socket_vnet_init, NULL);
323 
324 static void
325 socket_vnet_uninit(const void *unused __unused)
326 {
327 	int i;
328 
329 	for (i = 0; i <= HHOOK_SOCKET_LAST; i++)
330 		socket_hhook_deregister(i);
331 }
332 VNET_SYSUNINIT(socket_vnet_uninit, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
333     socket_vnet_uninit, NULL);
334 
335 /*
336  * Initialise maxsockets.  This SYSINIT must be run after
337  * tunable_mbinit().
338  */
339 static void
340 init_maxsockets(void *ignored)
341 {
342 
343 	TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
344 	maxsockets = imax(maxsockets, maxfiles);
345 }
346 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL);
347 
348 /*
349  * Sysctl to get and set the maximum global sockets limit.  Notify protocols
350  * of the change so that they can update their dependent limits as required.
351  */
352 static int
353 sysctl_maxsockets(SYSCTL_HANDLER_ARGS)
354 {
355 	int error, newmaxsockets;
356 
357 	newmaxsockets = maxsockets;
358 	error = sysctl_handle_int(oidp, &newmaxsockets, 0, req);
359 	if (error == 0 && req->newptr) {
360 		if (newmaxsockets > maxsockets &&
361 		    newmaxsockets <= maxfiles) {
362 			maxsockets = newmaxsockets;
363 			EVENTHANDLER_INVOKE(maxsockets_change);
364 		} else
365 			error = EINVAL;
366 	}
367 	return (error);
368 }
369 SYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets, CTLTYPE_INT|CTLFLAG_RW,
370     &maxsockets, 0, sysctl_maxsockets, "IU",
371     "Maximum number of sockets available");
372 
373 /*
374  * Socket operation routines.  These routines are called by the routines in
375  * sys_socket.c or from a system process, and implement the semantics of
376  * socket operations by switching out to the protocol specific routines.
377  */
378 
379 /*
380  * Get a socket structure from our zone, and initialize it.  Note that it
381  * would probably be better to allocate socket and PCB at the same time, but
382  * I'm not convinced that all the protocols can be easily modified to do
383  * this.
384  *
385  * soalloc() returns a socket with a ref count of 0.
386  */
387 static struct socket *
388 soalloc(struct vnet *vnet)
389 {
390 	struct socket *so;
391 
392 	so = uma_zalloc(socket_zone, M_NOWAIT | M_ZERO);
393 	if (so == NULL)
394 		return (NULL);
395 #ifdef MAC
396 	if (mac_socket_init(so, M_NOWAIT) != 0) {
397 		uma_zfree(socket_zone, so);
398 		return (NULL);
399 	}
400 #endif
401 	if (khelp_init_osd(HELPER_CLASS_SOCKET, &so->osd)) {
402 		uma_zfree(socket_zone, so);
403 		return (NULL);
404 	}
405 
406 	/*
407 	 * The socket locking protocol allows to lock 2 sockets at a time,
408 	 * however, the first one must be a listening socket.  WITNESS lacks
409 	 * a feature to change class of an existing lock, so we use DUPOK.
410 	 */
411 	mtx_init(&so->so_lock, "socket", NULL, MTX_DEF | MTX_DUPOK);
412 	SOCKBUF_LOCK_INIT(&so->so_snd, "so_snd");
413 	SOCKBUF_LOCK_INIT(&so->so_rcv, "so_rcv");
414 	so->so_rcv.sb_sel = &so->so_rdsel;
415 	so->so_snd.sb_sel = &so->so_wrsel;
416 	sx_init(&so->so_snd.sb_sx, "so_snd_sx");
417 	sx_init(&so->so_rcv.sb_sx, "so_rcv_sx");
418 	TAILQ_INIT(&so->so_snd.sb_aiojobq);
419 	TAILQ_INIT(&so->so_rcv.sb_aiojobq);
420 	TASK_INIT(&so->so_snd.sb_aiotask, 0, soaio_snd, so);
421 	TASK_INIT(&so->so_rcv.sb_aiotask, 0, soaio_rcv, so);
422 #ifdef VIMAGE
423 	VNET_ASSERT(vnet != NULL, ("%s:%d vnet is NULL, so=%p",
424 	    __func__, __LINE__, so));
425 	so->so_vnet = vnet;
426 #endif
427 	/* We shouldn't need the so_global_mtx */
428 	if (hhook_run_socket(so, NULL, HHOOK_SOCKET_CREATE)) {
429 		/* Do we need more comprehensive error returns? */
430 		uma_zfree(socket_zone, so);
431 		return (NULL);
432 	}
433 	mtx_lock(&so_global_mtx);
434 	so->so_gencnt = ++so_gencnt;
435 	++numopensockets;
436 #ifdef VIMAGE
437 	vnet->vnet_sockcnt++;
438 #endif
439 	mtx_unlock(&so_global_mtx);
440 
441 	return (so);
442 }
443 
444 /*
445  * Free the storage associated with a socket at the socket layer, tear down
446  * locks, labels, etc.  All protocol state is assumed already to have been
447  * torn down (and possibly never set up) by the caller.
448  */
449 static void
450 sodealloc(struct socket *so)
451 {
452 
453 	KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count));
454 	KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL"));
455 
456 	mtx_lock(&so_global_mtx);
457 	so->so_gencnt = ++so_gencnt;
458 	--numopensockets;	/* Could be below, but faster here. */
459 #ifdef VIMAGE
460 	VNET_ASSERT(so->so_vnet != NULL, ("%s:%d so_vnet is NULL, so=%p",
461 	    __func__, __LINE__, so));
462 	so->so_vnet->vnet_sockcnt--;
463 #endif
464 	mtx_unlock(&so_global_mtx);
465 #ifdef MAC
466 	mac_socket_destroy(so);
467 #endif
468 	hhook_run_socket(so, NULL, HHOOK_SOCKET_CLOSE);
469 
470 	crfree(so->so_cred);
471 	khelp_destroy_osd(&so->osd);
472 	if (SOLISTENING(so)) {
473 		if (so->sol_accept_filter != NULL)
474 			accept_filt_setopt(so, NULL);
475 	} else {
476 		if (so->so_rcv.sb_hiwat)
477 			(void)chgsbsize(so->so_cred->cr_uidinfo,
478 			    &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY);
479 		if (so->so_snd.sb_hiwat)
480 			(void)chgsbsize(so->so_cred->cr_uidinfo,
481 			    &so->so_snd.sb_hiwat, 0, RLIM_INFINITY);
482 		sx_destroy(&so->so_snd.sb_sx);
483 		sx_destroy(&so->so_rcv.sb_sx);
484 		SOCKBUF_LOCK_DESTROY(&so->so_snd);
485 		SOCKBUF_LOCK_DESTROY(&so->so_rcv);
486 	}
487 	mtx_destroy(&so->so_lock);
488 	uma_zfree(socket_zone, so);
489 }
490 
491 /*
492  * socreate returns a socket with a ref count of 1.  The socket should be
493  * closed with soclose().
494  */
495 int
496 socreate(int dom, struct socket **aso, int type, int proto,
497     struct ucred *cred, struct thread *td)
498 {
499 	struct protosw *prp;
500 	struct socket *so;
501 	int error;
502 
503 	if (proto)
504 		prp = pffindproto(dom, proto, type);
505 	else
506 		prp = pffindtype(dom, type);
507 
508 	if (prp == NULL) {
509 		/* No support for domain. */
510 		if (pffinddomain(dom) == NULL)
511 			return (EAFNOSUPPORT);
512 		/* No support for socket type. */
513 		if (proto == 0 && type != 0)
514 			return (EPROTOTYPE);
515 		return (EPROTONOSUPPORT);
516 	}
517 	if (prp->pr_usrreqs->pru_attach == NULL ||
518 	    prp->pr_usrreqs->pru_attach == pru_attach_notsupp)
519 		return (EPROTONOSUPPORT);
520 
521 	if (prison_check_af(cred, prp->pr_domain->dom_family) != 0)
522 		return (EPROTONOSUPPORT);
523 
524 	if (prp->pr_type != type)
525 		return (EPROTOTYPE);
526 	so = soalloc(CRED_TO_VNET(cred));
527 	if (so == NULL)
528 		return (ENOBUFS);
529 
530 	so->so_type = type;
531 	so->so_cred = crhold(cred);
532 	if ((prp->pr_domain->dom_family == PF_INET) ||
533 	    (prp->pr_domain->dom_family == PF_INET6) ||
534 	    (prp->pr_domain->dom_family == PF_ROUTE))
535 		so->so_fibnum = td->td_proc->p_fibnum;
536 	else
537 		so->so_fibnum = 0;
538 	so->so_proto = prp;
539 #ifdef MAC
540 	mac_socket_create(cred, so);
541 #endif
542 	knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
543 	    so_rdknl_assert_locked, so_rdknl_assert_unlocked);
544 	knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
545 	    so_wrknl_assert_locked, so_wrknl_assert_unlocked);
546 	/*
547 	 * Auto-sizing of socket buffers is managed by the protocols and
548 	 * the appropriate flags must be set in the pru_attach function.
549 	 */
550 	CURVNET_SET(so->so_vnet);
551 	error = (*prp->pr_usrreqs->pru_attach)(so, proto, td);
552 	CURVNET_RESTORE();
553 	if (error) {
554 		sodealloc(so);
555 		return (error);
556 	}
557 	soref(so);
558 	*aso = so;
559 	return (0);
560 }
561 
562 #ifdef REGRESSION
563 static int regression_sonewconn_earlytest = 1;
564 SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW,
565     &regression_sonewconn_earlytest, 0, "Perform early sonewconn limit test");
566 #endif
567 
568 /*
569  * When an attempt at a new connection is noted on a socket which accepts
570  * connections, sonewconn is called.  If the connection is possible (subject
571  * to space constraints, etc.) then we allocate a new structure, properly
572  * linked into the data structure of the original socket, and return this.
573  * Connstatus may be 0, or SS_ISCONFIRMING, or SS_ISCONNECTED.
574  *
575  * Note: the ref count on the socket is 0 on return.
576  */
577 struct socket *
578 sonewconn(struct socket *head, int connstatus)
579 {
580 	static struct timeval lastover;
581 	static struct timeval overinterval = { 60, 0 };
582 	static int overcount;
583 
584 	struct socket *so;
585 	u_int over;
586 
587 	SOLISTEN_LOCK(head);
588 	over = (head->sol_qlen > 3 * head->sol_qlimit / 2);
589 	SOLISTEN_UNLOCK(head);
590 #ifdef REGRESSION
591 	if (regression_sonewconn_earlytest && over) {
592 #else
593 	if (over) {
594 #endif
595 		overcount++;
596 
597 		if (ratecheck(&lastover, &overinterval)) {
598 			log(LOG_DEBUG, "%s: pcb %p: Listen queue overflow: "
599 			    "%i already in queue awaiting acceptance "
600 			    "(%d occurrences)\n",
601 			    __func__, head->so_pcb, head->sol_qlen, overcount);
602 
603 			overcount = 0;
604 		}
605 
606 		return (NULL);
607 	}
608 	VNET_ASSERT(head->so_vnet != NULL, ("%s: so %p vnet is NULL",
609 	    __func__, head));
610 	so = soalloc(head->so_vnet);
611 	if (so == NULL) {
612 		log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: "
613 		    "limit reached or out of memory\n",
614 		    __func__, head->so_pcb);
615 		return (NULL);
616 	}
617 	so->so_listen = head;
618 	so->so_type = head->so_type;
619 	so->so_linger = head->so_linger;
620 	so->so_state = head->so_state | SS_NOFDREF;
621 	so->so_fibnum = head->so_fibnum;
622 	so->so_proto = head->so_proto;
623 	so->so_cred = crhold(head->so_cred);
624 #ifdef MAC
625 	mac_socket_newconn(head, so);
626 #endif
627 	knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
628 	    so_rdknl_assert_locked, so_rdknl_assert_unlocked);
629 	knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
630 	    so_wrknl_assert_locked, so_wrknl_assert_unlocked);
631 	VNET_SO_ASSERT(head);
632 	if (soreserve(so, head->sol_sbsnd_hiwat, head->sol_sbrcv_hiwat)) {
633 		sodealloc(so);
634 		log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n",
635 		    __func__, head->so_pcb);
636 		return (NULL);
637 	}
638 	if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
639 		sodealloc(so);
640 		log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n",
641 		    __func__, head->so_pcb);
642 		return (NULL);
643 	}
644 	so->so_rcv.sb_lowat = head->sol_sbrcv_lowat;
645 	so->so_snd.sb_lowat = head->sol_sbsnd_lowat;
646 	so->so_rcv.sb_timeo = head->sol_sbrcv_timeo;
647 	so->so_snd.sb_timeo = head->sol_sbsnd_timeo;
648 	so->so_rcv.sb_flags |= head->sol_sbrcv_flags & SB_AUTOSIZE;
649 	so->so_snd.sb_flags |= head->sol_sbsnd_flags & SB_AUTOSIZE;
650 
651 	SOLISTEN_LOCK(head);
652 	if (head->sol_accept_filter != NULL)
653 		connstatus = 0;
654 	so->so_state |= connstatus;
655 	so->so_options = head->so_options & ~SO_ACCEPTCONN;
656 	soref(head); /* A socket on (in)complete queue refs head. */
657 	if (connstatus) {
658 		TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list);
659 		so->so_qstate = SQ_COMP;
660 		head->sol_qlen++;
661 		solisten_wakeup(head);	/* unlocks */
662 	} else {
663 		/*
664 		 * Keep removing sockets from the head until there's room for
665 		 * us to insert on the tail.  In pre-locking revisions, this
666 		 * was a simple if(), but as we could be racing with other
667 		 * threads and soabort() requires dropping locks, we must
668 		 * loop waiting for the condition to be true.
669 		 */
670 		while (head->sol_incqlen > head->sol_qlimit) {
671 			struct socket *sp;
672 
673 			sp = TAILQ_FIRST(&head->sol_incomp);
674 			TAILQ_REMOVE(&head->sol_incomp, sp, so_list);
675 			head->sol_incqlen--;
676 			SOCK_LOCK(sp);
677 			sp->so_qstate = SQ_NONE;
678 			sp->so_listen = NULL;
679 			SOCK_UNLOCK(sp);
680 			sorele(head);	/* does SOLISTEN_UNLOCK, head stays */
681 			soabort(sp);
682 			SOLISTEN_LOCK(head);
683 		}
684 		TAILQ_INSERT_TAIL(&head->sol_incomp, so, so_list);
685 		so->so_qstate = SQ_INCOMP;
686 		head->sol_incqlen++;
687 		SOLISTEN_UNLOCK(head);
688 	}
689 	return (so);
690 }
691 
692 #ifdef SCTP
693 /*
694  * Socket part of sctp_peeloff().  Detach a new socket from an
695  * association.  The new socket is returned with a reference.
696  */
697 struct socket *
698 sopeeloff(struct socket *head)
699 {
700 	struct socket *so;
701 
702 	VNET_ASSERT(head->so_vnet != NULL, ("%s:%d so_vnet is NULL, head=%p",
703 	    __func__, __LINE__, head));
704 	so = soalloc(head->so_vnet);
705 	if (so == NULL) {
706 		log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: "
707 		    "limit reached or out of memory\n",
708 		    __func__, head->so_pcb);
709 		return (NULL);
710 	}
711 	so->so_type = head->so_type;
712 	so->so_options = head->so_options;
713 	so->so_linger = head->so_linger;
714 	so->so_state = (head->so_state & SS_NBIO) | SS_ISCONNECTED;
715 	so->so_fibnum = head->so_fibnum;
716 	so->so_proto = head->so_proto;
717 	so->so_cred = crhold(head->so_cred);
718 #ifdef MAC
719 	mac_socket_newconn(head, so);
720 #endif
721 	knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
722 	    so_rdknl_assert_locked, so_rdknl_assert_unlocked);
723 	knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
724 	    so_wrknl_assert_locked, so_wrknl_assert_unlocked);
725 	VNET_SO_ASSERT(head);
726 	if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat)) {
727 		sodealloc(so);
728 		log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n",
729 		    __func__, head->so_pcb);
730 		return (NULL);
731 	}
732 	if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
733 		sodealloc(so);
734 		log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n",
735 		    __func__, head->so_pcb);
736 		return (NULL);
737 	}
738 	so->so_rcv.sb_lowat = head->so_rcv.sb_lowat;
739 	so->so_snd.sb_lowat = head->so_snd.sb_lowat;
740 	so->so_rcv.sb_timeo = head->so_rcv.sb_timeo;
741 	so->so_snd.sb_timeo = head->so_snd.sb_timeo;
742 	so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE;
743 	so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE;
744 
745 	soref(so);
746 
747 	return (so);
748 }
749 #endif	/* SCTP */
750 
751 int
752 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
753 {
754 	int error;
755 
756 	CURVNET_SET(so->so_vnet);
757 	error = (*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td);
758 	CURVNET_RESTORE();
759 	return (error);
760 }
761 
762 int
763 sobindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
764 {
765 	int error;
766 
767 	CURVNET_SET(so->so_vnet);
768 	error = (*so->so_proto->pr_usrreqs->pru_bindat)(fd, so, nam, td);
769 	CURVNET_RESTORE();
770 	return (error);
771 }
772 
773 /*
774  * solisten() transitions a socket from a non-listening state to a listening
775  * state, but can also be used to update the listen queue depth on an
776  * existing listen socket.  The protocol will call back into the sockets
777  * layer using solisten_proto_check() and solisten_proto() to check and set
778  * socket-layer listen state.  Call backs are used so that the protocol can
779  * acquire both protocol and socket layer locks in whatever order is required
780  * by the protocol.
781  *
782  * Protocol implementors are advised to hold the socket lock across the
783  * socket-layer test and set to avoid races at the socket layer.
784  */
785 int
786 solisten(struct socket *so, int backlog, struct thread *td)
787 {
788 	int error;
789 
790 	CURVNET_SET(so->so_vnet);
791 	error = (*so->so_proto->pr_usrreqs->pru_listen)(so, backlog, td);
792 	CURVNET_RESTORE();
793 	return (error);
794 }
795 
796 int
797 solisten_proto_check(struct socket *so)
798 {
799 
800 	SOCK_LOCK_ASSERT(so);
801 
802 	if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
803 	    SS_ISDISCONNECTING))
804 		return (EINVAL);
805 	return (0);
806 }
807 
808 void
809 solisten_proto(struct socket *so, int backlog)
810 {
811 	int sbrcv_lowat, sbsnd_lowat;
812 	u_int sbrcv_hiwat, sbsnd_hiwat;
813 	short sbrcv_flags, sbsnd_flags;
814 	sbintime_t sbrcv_timeo, sbsnd_timeo;
815 
816 	SOCK_LOCK_ASSERT(so);
817 
818 	if (SOLISTENING(so))
819 		goto listening;
820 
821 	/*
822 	 * Change this socket to listening state.
823 	 */
824 	sbrcv_lowat = so->so_rcv.sb_lowat;
825 	sbsnd_lowat = so->so_snd.sb_lowat;
826 	sbrcv_hiwat = so->so_rcv.sb_hiwat;
827 	sbsnd_hiwat = so->so_snd.sb_hiwat;
828 	sbrcv_flags = so->so_rcv.sb_flags;
829 	sbsnd_flags = so->so_snd.sb_flags;
830 	sbrcv_timeo = so->so_rcv.sb_timeo;
831 	sbsnd_timeo = so->so_snd.sb_timeo;
832 
833 	sbdestroy(&so->so_snd, so);
834 	sbdestroy(&so->so_rcv, so);
835 	sx_destroy(&so->so_snd.sb_sx);
836 	sx_destroy(&so->so_rcv.sb_sx);
837 	SOCKBUF_LOCK_DESTROY(&so->so_snd);
838 	SOCKBUF_LOCK_DESTROY(&so->so_rcv);
839 
840 #ifdef INVARIANTS
841 	bzero(&so->so_rcv,
842 	    sizeof(struct socket) - offsetof(struct socket, so_rcv));
843 #endif
844 
845 	so->sol_sbrcv_lowat = sbrcv_lowat;
846 	so->sol_sbsnd_lowat = sbsnd_lowat;
847 	so->sol_sbrcv_hiwat = sbrcv_hiwat;
848 	so->sol_sbsnd_hiwat = sbsnd_hiwat;
849 	so->sol_sbrcv_flags = sbrcv_flags;
850 	so->sol_sbsnd_flags = sbsnd_flags;
851 	so->sol_sbrcv_timeo = sbrcv_timeo;
852 	so->sol_sbsnd_timeo = sbsnd_timeo;
853 
854 	so->sol_qlen = so->sol_incqlen = 0;
855 	TAILQ_INIT(&so->sol_incomp);
856 	TAILQ_INIT(&so->sol_comp);
857 
858 	so->sol_accept_filter = NULL;
859 	so->sol_accept_filter_arg = NULL;
860 	so->sol_accept_filter_str = NULL;
861 
862 	so->sol_upcall = NULL;
863 	so->sol_upcallarg = NULL;
864 
865 	so->so_options |= SO_ACCEPTCONN;
866 
867 listening:
868 	if (backlog < 0 || backlog > somaxconn)
869 		backlog = somaxconn;
870 	so->sol_qlimit = backlog;
871 }
872 
873 /*
874  * Wakeup listeners/subsystems once we have a complete connection.
875  * Enters with lock, returns unlocked.
876  */
877 void
878 solisten_wakeup(struct socket *sol)
879 {
880 
881 	if (sol->sol_upcall != NULL)
882 		(void )sol->sol_upcall(sol, sol->sol_upcallarg, M_NOWAIT);
883 	else {
884 		selwakeuppri(&sol->so_rdsel, PSOCK);
885 		KNOTE_LOCKED(&sol->so_rdsel.si_note, 0);
886 	}
887 	SOLISTEN_UNLOCK(sol);
888 	wakeup_one(&sol->sol_comp);
889 }
890 
891 /*
892  * Return single connection off a listening socket queue.  Main consumer of
893  * the function is kern_accept4().  Some modules, that do their own accept
894  * management also use the function.
895  *
896  * Listening socket must be locked on entry and is returned unlocked on
897  * return.
898  * The flags argument is set of accept4(2) flags and ACCEPT4_INHERIT.
899  */
900 int
901 solisten_dequeue(struct socket *head, struct socket **ret, int flags)
902 {
903 	struct socket *so;
904 	int error;
905 
906 	SOLISTEN_LOCK_ASSERT(head);
907 
908 	while (!(head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp) &&
909 	    head->so_error == 0) {
910 		error = msleep(&head->sol_comp, &head->so_lock, PSOCK | PCATCH,
911 		    "accept", 0);
912 		if (error != 0) {
913 			SOLISTEN_UNLOCK(head);
914 			return (error);
915 		}
916 	}
917 	if (head->so_error) {
918 		error = head->so_error;
919 		head->so_error = 0;
920 	} else if ((head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp))
921 		error = EWOULDBLOCK;
922 	else
923 		error = 0;
924 	if (error) {
925 		SOLISTEN_UNLOCK(head);
926 		return (error);
927 	}
928 	so = TAILQ_FIRST(&head->sol_comp);
929 	SOCK_LOCK(so);
930 	KASSERT(so->so_qstate == SQ_COMP,
931 	    ("%s: so %p not SQ_COMP", __func__, so));
932 	soref(so);
933 	head->sol_qlen--;
934 	so->so_qstate = SQ_NONE;
935 	so->so_listen = NULL;
936 	TAILQ_REMOVE(&head->sol_comp, so, so_list);
937 	if (flags & ACCEPT4_INHERIT)
938 		so->so_state |= (head->so_state & SS_NBIO);
939 	else
940 		so->so_state |= (flags & SOCK_NONBLOCK) ? SS_NBIO : 0;
941 	SOCK_UNLOCK(so);
942 	sorele(head);
943 
944 	*ret = so;
945 	return (0);
946 }
947 
948 /*
949  * Evaluate the reference count and named references on a socket; if no
950  * references remain, free it.  This should be called whenever a reference is
951  * released, such as in sorele(), but also when named reference flags are
952  * cleared in socket or protocol code.
953  *
954  * sofree() will free the socket if:
955  *
956  * - There are no outstanding file descriptor references or related consumers
957  *   (so_count == 0).
958  *
959  * - The socket has been closed by user space, if ever open (SS_NOFDREF).
960  *
961  * - The protocol does not have an outstanding strong reference on the socket
962  *   (SS_PROTOREF).
963  *
964  * - The socket is not in a completed connection queue, so a process has been
965  *   notified that it is present.  If it is removed, the user process may
966  *   block in accept() despite select() saying the socket was ready.
967  */
968 void
969 sofree(struct socket *so)
970 {
971 	struct protosw *pr = so->so_proto;
972 
973 	SOCK_LOCK_ASSERT(so);
974 
975 	if ((so->so_state & SS_NOFDREF) == 0 || so->so_count != 0 ||
976 	    (so->so_state & SS_PROTOREF) || (so->so_qstate == SQ_COMP)) {
977 		SOCK_UNLOCK(so);
978 		return;
979 	}
980 
981 	if (!SOLISTENING(so) && so->so_qstate == SQ_INCOMP) {
982 		struct socket *sol;
983 
984 		sol = so->so_listen;
985 		KASSERT(sol, ("%s: so %p on incomp of NULL", __func__, so));
986 
987 		/*
988 		 * To solve race between close of a listening socket and
989 		 * a socket on its incomplete queue, we need to lock both.
990 		 * The order is first listening socket, then regular.
991 		 * Since we don't have SS_NOFDREF neither SS_PROTOREF, this
992 		 * function and the listening socket are the only pointers
993 		 * to so.  To preserve so and sol, we reference both and then
994 		 * relock.
995 		 * After relock the socket may not move to so_comp since it
996 		 * doesn't have PCB already, but it may be removed from
997 		 * so_incomp. If that happens, we share responsiblity on
998 		 * freeing the socket, but soclose() has already removed
999 		 * it from queue.
1000 		 */
1001 		soref(sol);
1002 		soref(so);
1003 		SOCK_UNLOCK(so);
1004 		SOLISTEN_LOCK(sol);
1005 		SOCK_LOCK(so);
1006 		if (so->so_qstate == SQ_INCOMP) {
1007 			KASSERT(so->so_listen == sol,
1008 			    ("%s: so %p migrated out of sol %p",
1009 			    __func__, so, sol));
1010 			TAILQ_REMOVE(&sol->sol_incomp, so, so_list);
1011 			sol->sol_incqlen--;
1012 			/* This is guarenteed not to be the last. */
1013 			refcount_release(&sol->so_count);
1014 			so->so_qstate = SQ_NONE;
1015 			so->so_listen = NULL;
1016 		} else
1017 			KASSERT(so->so_listen == NULL,
1018 			    ("%s: so %p not on (in)comp with so_listen",
1019 			    __func__, so));
1020 		sorele(sol);
1021 		KASSERT(so->so_count == 1,
1022 		    ("%s: so %p count %u", __func__, so, so->so_count));
1023 		so->so_count = 0;
1024 	}
1025 	if (SOLISTENING(so))
1026 		so->so_error = ECONNABORTED;
1027 	SOCK_UNLOCK(so);
1028 
1029 	if (so->so_dtor != NULL)
1030 		so->so_dtor(so);
1031 
1032 	VNET_SO_ASSERT(so);
1033 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
1034 		(*pr->pr_domain->dom_dispose)(so);
1035 	if (pr->pr_usrreqs->pru_detach != NULL)
1036 		(*pr->pr_usrreqs->pru_detach)(so);
1037 
1038 	/*
1039 	 * From this point on, we assume that no other references to this
1040 	 * socket exist anywhere else in the stack.  Therefore, no locks need
1041 	 * to be acquired or held.
1042 	 *
1043 	 * We used to do a lot of socket buffer and socket locking here, as
1044 	 * well as invoke sorflush() and perform wakeups.  The direct call to
1045 	 * dom_dispose() and sbrelease_internal() are an inlining of what was
1046 	 * necessary from sorflush().
1047 	 *
1048 	 * Notice that the socket buffer and kqueue state are torn down
1049 	 * before calling pru_detach.  This means that protocols shold not
1050 	 * assume they can perform socket wakeups, etc, in their detach code.
1051 	 */
1052 	if (!SOLISTENING(so)) {
1053 		sbdestroy(&so->so_snd, so);
1054 		sbdestroy(&so->so_rcv, so);
1055 	}
1056 	seldrain(&so->so_rdsel);
1057 	seldrain(&so->so_wrsel);
1058 	knlist_destroy(&so->so_rdsel.si_note);
1059 	knlist_destroy(&so->so_wrsel.si_note);
1060 	sodealloc(so);
1061 }
1062 
1063 /*
1064  * Close a socket on last file table reference removal.  Initiate disconnect
1065  * if connected.  Free socket when disconnect complete.
1066  *
1067  * This function will sorele() the socket.  Note that soclose() may be called
1068  * prior to the ref count reaching zero.  The actual socket structure will
1069  * not be freed until the ref count reaches zero.
1070  */
1071 int
1072 soclose(struct socket *so)
1073 {
1074 	struct accept_queue lqueue;
1075 	bool listening;
1076 	int error = 0;
1077 
1078 	KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter"));
1079 
1080 	CURVNET_SET(so->so_vnet);
1081 	funsetown(&so->so_sigio);
1082 	if (so->so_state & SS_ISCONNECTED) {
1083 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
1084 			error = sodisconnect(so);
1085 			if (error) {
1086 				if (error == ENOTCONN)
1087 					error = 0;
1088 				goto drop;
1089 			}
1090 		}
1091 		if (so->so_options & SO_LINGER) {
1092 			if ((so->so_state & SS_ISDISCONNECTING) &&
1093 			    (so->so_state & SS_NBIO))
1094 				goto drop;
1095 			while (so->so_state & SS_ISCONNECTED) {
1096 				error = tsleep(&so->so_timeo,
1097 				    PSOCK | PCATCH, "soclos",
1098 				    so->so_linger * hz);
1099 				if (error)
1100 					break;
1101 			}
1102 		}
1103 	}
1104 
1105 drop:
1106 	if (so->so_proto->pr_usrreqs->pru_close != NULL)
1107 		(*so->so_proto->pr_usrreqs->pru_close)(so);
1108 
1109 	SOCK_LOCK(so);
1110 	if ((listening = (so->so_options & SO_ACCEPTCONN))) {
1111 		struct socket *sp;
1112 
1113 		TAILQ_INIT(&lqueue);
1114 		TAILQ_SWAP(&lqueue, &so->sol_incomp, socket, so_list);
1115 		TAILQ_CONCAT(&lqueue, &so->sol_comp, so_list);
1116 
1117 		so->sol_qlen = so->sol_incqlen = 0;
1118 
1119 		TAILQ_FOREACH(sp, &lqueue, so_list) {
1120 			SOCK_LOCK(sp);
1121 			sp->so_qstate = SQ_NONE;
1122 			sp->so_listen = NULL;
1123 			SOCK_UNLOCK(sp);
1124 			/* Guaranteed not to be the last. */
1125 			refcount_release(&so->so_count);
1126 		}
1127 	}
1128 	KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF"));
1129 	so->so_state |= SS_NOFDREF;
1130 	sorele(so);
1131 	if (listening) {
1132 		struct socket *sp;
1133 
1134 		TAILQ_FOREACH(sp, &lqueue, so_list) {
1135 			SOCK_LOCK(sp);
1136 			if (sp->so_count == 0) {
1137 				SOCK_UNLOCK(sp);
1138 				soabort(sp);
1139 			} else
1140 				/* sp is now in sofree() */
1141 				SOCK_UNLOCK(sp);
1142 		}
1143 	}
1144 	CURVNET_RESTORE();
1145 	return (error);
1146 }
1147 
1148 /*
1149  * soabort() is used to abruptly tear down a connection, such as when a
1150  * resource limit is reached (listen queue depth exceeded), or if a listen
1151  * socket is closed while there are sockets waiting to be accepted.
1152  *
1153  * This interface is tricky, because it is called on an unreferenced socket,
1154  * and must be called only by a thread that has actually removed the socket
1155  * from the listen queue it was on, or races with other threads are risked.
1156  *
1157  * This interface will call into the protocol code, so must not be called
1158  * with any socket locks held.  Protocols do call it while holding their own
1159  * recursible protocol mutexes, but this is something that should be subject
1160  * to review in the future.
1161  */
1162 void
1163 soabort(struct socket *so)
1164 {
1165 
1166 	/*
1167 	 * In as much as is possible, assert that no references to this
1168 	 * socket are held.  This is not quite the same as asserting that the
1169 	 * current thread is responsible for arranging for no references, but
1170 	 * is as close as we can get for now.
1171 	 */
1172 	KASSERT(so->so_count == 0, ("soabort: so_count"));
1173 	KASSERT((so->so_state & SS_PROTOREF) == 0, ("soabort: SS_PROTOREF"));
1174 	KASSERT(so->so_state & SS_NOFDREF, ("soabort: !SS_NOFDREF"));
1175 	KASSERT(so->so_qstate == SQ_NONE, ("soabort: !SQ_NONE"));
1176 	VNET_SO_ASSERT(so);
1177 
1178 	if (so->so_proto->pr_usrreqs->pru_abort != NULL)
1179 		(*so->so_proto->pr_usrreqs->pru_abort)(so);
1180 	SOCK_LOCK(so);
1181 	sofree(so);
1182 }
1183 
1184 int
1185 soaccept(struct socket *so, struct sockaddr **nam)
1186 {
1187 	int error;
1188 
1189 	SOCK_LOCK(so);
1190 	KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF"));
1191 	so->so_state &= ~SS_NOFDREF;
1192 	SOCK_UNLOCK(so);
1193 
1194 	CURVNET_SET(so->so_vnet);
1195 	error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam);
1196 	CURVNET_RESTORE();
1197 	return (error);
1198 }
1199 
1200 int
1201 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
1202 {
1203 
1204 	return (soconnectat(AT_FDCWD, so, nam, td));
1205 }
1206 
1207 int
1208 soconnectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
1209 {
1210 	int error;
1211 
1212 	if (so->so_options & SO_ACCEPTCONN)
1213 		return (EOPNOTSUPP);
1214 
1215 	CURVNET_SET(so->so_vnet);
1216 	/*
1217 	 * If protocol is connection-based, can only connect once.
1218 	 * Otherwise, if connected, try to disconnect first.  This allows
1219 	 * user to disconnect by connecting to, e.g., a null address.
1220 	 */
1221 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
1222 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
1223 	    (error = sodisconnect(so)))) {
1224 		error = EISCONN;
1225 	} else {
1226 		/*
1227 		 * Prevent accumulated error from previous connection from
1228 		 * biting us.
1229 		 */
1230 		so->so_error = 0;
1231 		if (fd == AT_FDCWD) {
1232 			error = (*so->so_proto->pr_usrreqs->pru_connect)(so,
1233 			    nam, td);
1234 		} else {
1235 			error = (*so->so_proto->pr_usrreqs->pru_connectat)(fd,
1236 			    so, nam, td);
1237 		}
1238 	}
1239 	CURVNET_RESTORE();
1240 
1241 	return (error);
1242 }
1243 
1244 int
1245 soconnect2(struct socket *so1, struct socket *so2)
1246 {
1247 	int error;
1248 
1249 	CURVNET_SET(so1->so_vnet);
1250 	error = (*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2);
1251 	CURVNET_RESTORE();
1252 	return (error);
1253 }
1254 
1255 int
1256 sodisconnect(struct socket *so)
1257 {
1258 	int error;
1259 
1260 	if ((so->so_state & SS_ISCONNECTED) == 0)
1261 		return (ENOTCONN);
1262 	if (so->so_state & SS_ISDISCONNECTING)
1263 		return (EALREADY);
1264 	VNET_SO_ASSERT(so);
1265 	error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so);
1266 	return (error);
1267 }
1268 
1269 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? 0 : SBL_WAIT)
1270 
1271 int
1272 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio,
1273     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1274 {
1275 	long space;
1276 	ssize_t resid;
1277 	int clen = 0, error, dontroute;
1278 
1279 	KASSERT(so->so_type == SOCK_DGRAM, ("sosend_dgram: !SOCK_DGRAM"));
1280 	KASSERT(so->so_proto->pr_flags & PR_ATOMIC,
1281 	    ("sosend_dgram: !PR_ATOMIC"));
1282 
1283 	if (uio != NULL)
1284 		resid = uio->uio_resid;
1285 	else
1286 		resid = top->m_pkthdr.len;
1287 	/*
1288 	 * In theory resid should be unsigned.  However, space must be
1289 	 * signed, as it might be less than 0 if we over-committed, and we
1290 	 * must use a signed comparison of space and resid.  On the other
1291 	 * hand, a negative resid causes us to loop sending 0-length
1292 	 * segments to the protocol.
1293 	 */
1294 	if (resid < 0) {
1295 		error = EINVAL;
1296 		goto out;
1297 	}
1298 
1299 	dontroute =
1300 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0;
1301 	if (td != NULL)
1302 		td->td_ru.ru_msgsnd++;
1303 	if (control != NULL)
1304 		clen = control->m_len;
1305 
1306 	SOCKBUF_LOCK(&so->so_snd);
1307 	if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1308 		SOCKBUF_UNLOCK(&so->so_snd);
1309 		error = EPIPE;
1310 		goto out;
1311 	}
1312 	if (so->so_error) {
1313 		error = so->so_error;
1314 		so->so_error = 0;
1315 		SOCKBUF_UNLOCK(&so->so_snd);
1316 		goto out;
1317 	}
1318 	if ((so->so_state & SS_ISCONNECTED) == 0) {
1319 		/*
1320 		 * `sendto' and `sendmsg' is allowed on a connection-based
1321 		 * socket if it supports implied connect.  Return ENOTCONN if
1322 		 * not connected and no address is supplied.
1323 		 */
1324 		if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1325 		    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1326 			if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1327 			    !(resid == 0 && clen != 0)) {
1328 				SOCKBUF_UNLOCK(&so->so_snd);
1329 				error = ENOTCONN;
1330 				goto out;
1331 			}
1332 		} else if (addr == NULL) {
1333 			if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1334 				error = ENOTCONN;
1335 			else
1336 				error = EDESTADDRREQ;
1337 			SOCKBUF_UNLOCK(&so->so_snd);
1338 			goto out;
1339 		}
1340 	}
1341 
1342 	/*
1343 	 * Do we need MSG_OOB support in SOCK_DGRAM?  Signs here may be a
1344 	 * problem and need fixing.
1345 	 */
1346 	space = sbspace(&so->so_snd);
1347 	if (flags & MSG_OOB)
1348 		space += 1024;
1349 	space -= clen;
1350 	SOCKBUF_UNLOCK(&so->so_snd);
1351 	if (resid > space) {
1352 		error = EMSGSIZE;
1353 		goto out;
1354 	}
1355 	if (uio == NULL) {
1356 		resid = 0;
1357 		if (flags & MSG_EOR)
1358 			top->m_flags |= M_EOR;
1359 	} else {
1360 		/*
1361 		 * Copy the data from userland into a mbuf chain.
1362 		 * If no data is to be copied in, a single empty mbuf
1363 		 * is returned.
1364 		 */
1365 		top = m_uiotombuf(uio, M_WAITOK, space, max_hdr,
1366 		    (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0)));
1367 		if (top == NULL) {
1368 			error = EFAULT;	/* only possible error */
1369 			goto out;
1370 		}
1371 		space -= resid - uio->uio_resid;
1372 		resid = uio->uio_resid;
1373 	}
1374 	KASSERT(resid == 0, ("sosend_dgram: resid != 0"));
1375 	/*
1376 	 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock
1377 	 * than with.
1378 	 */
1379 	if (dontroute) {
1380 		SOCK_LOCK(so);
1381 		so->so_options |= SO_DONTROUTE;
1382 		SOCK_UNLOCK(so);
1383 	}
1384 	/*
1385 	 * XXX all the SBS_CANTSENDMORE checks previously done could be out
1386 	 * of date.  We could have received a reset packet in an interrupt or
1387 	 * maybe we slept while doing page faults in uiomove() etc.  We could
1388 	 * probably recheck again inside the locking protection here, but
1389 	 * there are probably other places that this also happens.  We must
1390 	 * rethink this.
1391 	 */
1392 	VNET_SO_ASSERT(so);
1393 	error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1394 	    (flags & MSG_OOB) ? PRUS_OOB :
1395 	/*
1396 	 * If the user set MSG_EOF, the protocol understands this flag and
1397 	 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND.
1398 	 */
1399 	    ((flags & MSG_EOF) &&
1400 	     (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1401 	     (resid <= 0)) ?
1402 		PRUS_EOF :
1403 		/* If there is more to send set PRUS_MORETOCOME */
1404 		(flags & MSG_MORETOCOME) ||
1405 		(resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1406 		top, addr, control, td);
1407 	if (dontroute) {
1408 		SOCK_LOCK(so);
1409 		so->so_options &= ~SO_DONTROUTE;
1410 		SOCK_UNLOCK(so);
1411 	}
1412 	clen = 0;
1413 	control = NULL;
1414 	top = NULL;
1415 out:
1416 	if (top != NULL)
1417 		m_freem(top);
1418 	if (control != NULL)
1419 		m_freem(control);
1420 	return (error);
1421 }
1422 
1423 /*
1424  * Send on a socket.  If send must go all at once and message is larger than
1425  * send buffering, then hard error.  Lock against other senders.  If must go
1426  * all at once and not enough room now, then inform user that this would
1427  * block and do nothing.  Otherwise, if nonblocking, send as much as
1428  * possible.  The data to be sent is described by "uio" if nonzero, otherwise
1429  * by the mbuf chain "top" (which must be null if uio is not).  Data provided
1430  * in mbuf chain must be small enough to send all at once.
1431  *
1432  * Returns nonzero on error, timeout or signal; callers must check for short
1433  * counts if EINTR/ERESTART are returned.  Data and control buffers are freed
1434  * on return.
1435  */
1436 int
1437 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio,
1438     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1439 {
1440 	long space;
1441 	ssize_t resid;
1442 	int clen = 0, error, dontroute;
1443 	int atomic = sosendallatonce(so) || top;
1444 
1445 	if (uio != NULL)
1446 		resid = uio->uio_resid;
1447 	else
1448 		resid = top->m_pkthdr.len;
1449 	/*
1450 	 * In theory resid should be unsigned.  However, space must be
1451 	 * signed, as it might be less than 0 if we over-committed, and we
1452 	 * must use a signed comparison of space and resid.  On the other
1453 	 * hand, a negative resid causes us to loop sending 0-length
1454 	 * segments to the protocol.
1455 	 *
1456 	 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
1457 	 * type sockets since that's an error.
1458 	 */
1459 	if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
1460 		error = EINVAL;
1461 		goto out;
1462 	}
1463 
1464 	dontroute =
1465 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
1466 	    (so->so_proto->pr_flags & PR_ATOMIC);
1467 	if (td != NULL)
1468 		td->td_ru.ru_msgsnd++;
1469 	if (control != NULL)
1470 		clen = control->m_len;
1471 
1472 	error = sblock(&so->so_snd, SBLOCKWAIT(flags));
1473 	if (error)
1474 		goto out;
1475 
1476 restart:
1477 	do {
1478 		SOCKBUF_LOCK(&so->so_snd);
1479 		if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1480 			SOCKBUF_UNLOCK(&so->so_snd);
1481 			error = EPIPE;
1482 			goto release;
1483 		}
1484 		if (so->so_error) {
1485 			error = so->so_error;
1486 			so->so_error = 0;
1487 			SOCKBUF_UNLOCK(&so->so_snd);
1488 			goto release;
1489 		}
1490 		if ((so->so_state & SS_ISCONNECTED) == 0) {
1491 			/*
1492 			 * `sendto' and `sendmsg' is allowed on a connection-
1493 			 * based socket if it supports implied connect.
1494 			 * Return ENOTCONN if not connected and no address is
1495 			 * supplied.
1496 			 */
1497 			if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1498 			    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1499 				if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1500 				    !(resid == 0 && clen != 0)) {
1501 					SOCKBUF_UNLOCK(&so->so_snd);
1502 					error = ENOTCONN;
1503 					goto release;
1504 				}
1505 			} else if (addr == NULL) {
1506 				SOCKBUF_UNLOCK(&so->so_snd);
1507 				if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1508 					error = ENOTCONN;
1509 				else
1510 					error = EDESTADDRREQ;
1511 				goto release;
1512 			}
1513 		}
1514 		space = sbspace(&so->so_snd);
1515 		if (flags & MSG_OOB)
1516 			space += 1024;
1517 		if ((atomic && resid > so->so_snd.sb_hiwat) ||
1518 		    clen > so->so_snd.sb_hiwat) {
1519 			SOCKBUF_UNLOCK(&so->so_snd);
1520 			error = EMSGSIZE;
1521 			goto release;
1522 		}
1523 		if (space < resid + clen &&
1524 		    (atomic || space < so->so_snd.sb_lowat || space < clen)) {
1525 			if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO)) {
1526 				SOCKBUF_UNLOCK(&so->so_snd);
1527 				error = EWOULDBLOCK;
1528 				goto release;
1529 			}
1530 			error = sbwait(&so->so_snd);
1531 			SOCKBUF_UNLOCK(&so->so_snd);
1532 			if (error)
1533 				goto release;
1534 			goto restart;
1535 		}
1536 		SOCKBUF_UNLOCK(&so->so_snd);
1537 		space -= clen;
1538 		do {
1539 			if (uio == NULL) {
1540 				resid = 0;
1541 				if (flags & MSG_EOR)
1542 					top->m_flags |= M_EOR;
1543 			} else {
1544 				/*
1545 				 * Copy the data from userland into a mbuf
1546 				 * chain.  If resid is 0, which can happen
1547 				 * only if we have control to send, then
1548 				 * a single empty mbuf is returned.  This
1549 				 * is a workaround to prevent protocol send
1550 				 * methods to panic.
1551 				 */
1552 				top = m_uiotombuf(uio, M_WAITOK, space,
1553 				    (atomic ? max_hdr : 0),
1554 				    (atomic ? M_PKTHDR : 0) |
1555 				    ((flags & MSG_EOR) ? M_EOR : 0));
1556 				if (top == NULL) {
1557 					error = EFAULT; /* only possible error */
1558 					goto release;
1559 				}
1560 				space -= resid - uio->uio_resid;
1561 				resid = uio->uio_resid;
1562 			}
1563 			if (dontroute) {
1564 				SOCK_LOCK(so);
1565 				so->so_options |= SO_DONTROUTE;
1566 				SOCK_UNLOCK(so);
1567 			}
1568 			/*
1569 			 * XXX all the SBS_CANTSENDMORE checks previously
1570 			 * done could be out of date.  We could have received
1571 			 * a reset packet in an interrupt or maybe we slept
1572 			 * while doing page faults in uiomove() etc.  We
1573 			 * could probably recheck again inside the locking
1574 			 * protection here, but there are probably other
1575 			 * places that this also happens.  We must rethink
1576 			 * this.
1577 			 */
1578 			VNET_SO_ASSERT(so);
1579 			error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1580 			    (flags & MSG_OOB) ? PRUS_OOB :
1581 			/*
1582 			 * If the user set MSG_EOF, the protocol understands
1583 			 * this flag and nothing left to send then use
1584 			 * PRU_SEND_EOF instead of PRU_SEND.
1585 			 */
1586 			    ((flags & MSG_EOF) &&
1587 			     (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1588 			     (resid <= 0)) ?
1589 				PRUS_EOF :
1590 			/* If there is more to send set PRUS_MORETOCOME. */
1591 			    (flags & MSG_MORETOCOME) ||
1592 			    (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1593 			    top, addr, control, td);
1594 			if (dontroute) {
1595 				SOCK_LOCK(so);
1596 				so->so_options &= ~SO_DONTROUTE;
1597 				SOCK_UNLOCK(so);
1598 			}
1599 			clen = 0;
1600 			control = NULL;
1601 			top = NULL;
1602 			if (error)
1603 				goto release;
1604 		} while (resid && space > 0);
1605 	} while (resid);
1606 
1607 release:
1608 	sbunlock(&so->so_snd);
1609 out:
1610 	if (top != NULL)
1611 		m_freem(top);
1612 	if (control != NULL)
1613 		m_freem(control);
1614 	return (error);
1615 }
1616 
1617 int
1618 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
1619     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1620 {
1621 	int error;
1622 
1623 	CURVNET_SET(so->so_vnet);
1624 	if (!SOLISTENING(so))
1625 		error = so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio,
1626 		    top, control, flags, td);
1627 	else {
1628 		m_freem(top);
1629 		m_freem(control);
1630 		error = ENOTCONN;
1631 	}
1632 	CURVNET_RESTORE();
1633 	return (error);
1634 }
1635 
1636 /*
1637  * The part of soreceive() that implements reading non-inline out-of-band
1638  * data from a socket.  For more complete comments, see soreceive(), from
1639  * which this code originated.
1640  *
1641  * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
1642  * unable to return an mbuf chain to the caller.
1643  */
1644 static int
1645 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags)
1646 {
1647 	struct protosw *pr = so->so_proto;
1648 	struct mbuf *m;
1649 	int error;
1650 
1651 	KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
1652 	VNET_SO_ASSERT(so);
1653 
1654 	m = m_get(M_WAITOK, MT_DATA);
1655 	error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK);
1656 	if (error)
1657 		goto bad;
1658 	do {
1659 		error = uiomove(mtod(m, void *),
1660 		    (int) min(uio->uio_resid, m->m_len), uio);
1661 		m = m_free(m);
1662 	} while (uio->uio_resid && error == 0 && m);
1663 bad:
1664 	if (m != NULL)
1665 		m_freem(m);
1666 	return (error);
1667 }
1668 
1669 /*
1670  * Following replacement or removal of the first mbuf on the first mbuf chain
1671  * of a socket buffer, push necessary state changes back into the socket
1672  * buffer so that other consumers see the values consistently.  'nextrecord'
1673  * is the callers locally stored value of the original value of
1674  * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes.
1675  * NOTE: 'nextrecord' may be NULL.
1676  */
1677 static __inline void
1678 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord)
1679 {
1680 
1681 	SOCKBUF_LOCK_ASSERT(sb);
1682 	/*
1683 	 * First, update for the new value of nextrecord.  If necessary, make
1684 	 * it the first record.
1685 	 */
1686 	if (sb->sb_mb != NULL)
1687 		sb->sb_mb->m_nextpkt = nextrecord;
1688 	else
1689 		sb->sb_mb = nextrecord;
1690 
1691 	/*
1692 	 * Now update any dependent socket buffer fields to reflect the new
1693 	 * state.  This is an expanded inline of SB_EMPTY_FIXUP(), with the
1694 	 * addition of a second clause that takes care of the case where
1695 	 * sb_mb has been updated, but remains the last record.
1696 	 */
1697 	if (sb->sb_mb == NULL) {
1698 		sb->sb_mbtail = NULL;
1699 		sb->sb_lastrecord = NULL;
1700 	} else if (sb->sb_mb->m_nextpkt == NULL)
1701 		sb->sb_lastrecord = sb->sb_mb;
1702 }
1703 
1704 /*
1705  * Implement receive operations on a socket.  We depend on the way that
1706  * records are added to the sockbuf by sbappend.  In particular, each record
1707  * (mbufs linked through m_next) must begin with an address if the protocol
1708  * so specifies, followed by an optional mbuf or mbufs containing ancillary
1709  * data, and then zero or more mbufs of data.  In order to allow parallelism
1710  * between network receive and copying to user space, as well as avoid
1711  * sleeping with a mutex held, we release the socket buffer mutex during the
1712  * user space copy.  Although the sockbuf is locked, new data may still be
1713  * appended, and thus we must maintain consistency of the sockbuf during that
1714  * time.
1715  *
1716  * The caller may receive the data as a single mbuf chain by supplying an
1717  * mbuf **mp0 for use in returning the chain.  The uio is then used only for
1718  * the count in uio_resid.
1719  */
1720 int
1721 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio,
1722     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1723 {
1724 	struct mbuf *m, **mp;
1725 	int flags, error, offset;
1726 	ssize_t len;
1727 	struct protosw *pr = so->so_proto;
1728 	struct mbuf *nextrecord;
1729 	int moff, type = 0;
1730 	ssize_t orig_resid = uio->uio_resid;
1731 
1732 	mp = mp0;
1733 	if (psa != NULL)
1734 		*psa = NULL;
1735 	if (controlp != NULL)
1736 		*controlp = NULL;
1737 	if (flagsp != NULL)
1738 		flags = *flagsp &~ MSG_EOR;
1739 	else
1740 		flags = 0;
1741 	if (flags & MSG_OOB)
1742 		return (soreceive_rcvoob(so, uio, flags));
1743 	if (mp != NULL)
1744 		*mp = NULL;
1745 	if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING)
1746 	    && uio->uio_resid) {
1747 		VNET_SO_ASSERT(so);
1748 		(*pr->pr_usrreqs->pru_rcvd)(so, 0);
1749 	}
1750 
1751 	error = sblock(&so->so_rcv, SBLOCKWAIT(flags));
1752 	if (error)
1753 		return (error);
1754 
1755 restart:
1756 	SOCKBUF_LOCK(&so->so_rcv);
1757 	m = so->so_rcv.sb_mb;
1758 	/*
1759 	 * If we have less data than requested, block awaiting more (subject
1760 	 * to any timeout) if:
1761 	 *   1. the current count is less than the low water mark, or
1762 	 *   2. MSG_DONTWAIT is not set
1763 	 */
1764 	if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1765 	    sbavail(&so->so_rcv) < uio->uio_resid) &&
1766 	    sbavail(&so->so_rcv) < so->so_rcv.sb_lowat &&
1767 	    m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
1768 		KASSERT(m != NULL || !sbavail(&so->so_rcv),
1769 		    ("receive: m == %p sbavail == %u",
1770 		    m, sbavail(&so->so_rcv)));
1771 		if (so->so_error) {
1772 			if (m != NULL)
1773 				goto dontblock;
1774 			error = so->so_error;
1775 			if ((flags & MSG_PEEK) == 0)
1776 				so->so_error = 0;
1777 			SOCKBUF_UNLOCK(&so->so_rcv);
1778 			goto release;
1779 		}
1780 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1781 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1782 			if (m == NULL) {
1783 				SOCKBUF_UNLOCK(&so->so_rcv);
1784 				goto release;
1785 			} else
1786 				goto dontblock;
1787 		}
1788 		for (; m != NULL; m = m->m_next)
1789 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
1790 				m = so->so_rcv.sb_mb;
1791 				goto dontblock;
1792 			}
1793 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1794 		    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1795 			SOCKBUF_UNLOCK(&so->so_rcv);
1796 			error = ENOTCONN;
1797 			goto release;
1798 		}
1799 		if (uio->uio_resid == 0) {
1800 			SOCKBUF_UNLOCK(&so->so_rcv);
1801 			goto release;
1802 		}
1803 		if ((so->so_state & SS_NBIO) ||
1804 		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
1805 			SOCKBUF_UNLOCK(&so->so_rcv);
1806 			error = EWOULDBLOCK;
1807 			goto release;
1808 		}
1809 		SBLASTRECORDCHK(&so->so_rcv);
1810 		SBLASTMBUFCHK(&so->so_rcv);
1811 		error = sbwait(&so->so_rcv);
1812 		SOCKBUF_UNLOCK(&so->so_rcv);
1813 		if (error)
1814 			goto release;
1815 		goto restart;
1816 	}
1817 dontblock:
1818 	/*
1819 	 * From this point onward, we maintain 'nextrecord' as a cache of the
1820 	 * pointer to the next record in the socket buffer.  We must keep the
1821 	 * various socket buffer pointers and local stack versions of the
1822 	 * pointers in sync, pushing out modifications before dropping the
1823 	 * socket buffer mutex, and re-reading them when picking it up.
1824 	 *
1825 	 * Otherwise, we will race with the network stack appending new data
1826 	 * or records onto the socket buffer by using inconsistent/stale
1827 	 * versions of the field, possibly resulting in socket buffer
1828 	 * corruption.
1829 	 *
1830 	 * By holding the high-level sblock(), we prevent simultaneous
1831 	 * readers from pulling off the front of the socket buffer.
1832 	 */
1833 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1834 	if (uio->uio_td)
1835 		uio->uio_td->td_ru.ru_msgrcv++;
1836 	KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb"));
1837 	SBLASTRECORDCHK(&so->so_rcv);
1838 	SBLASTMBUFCHK(&so->so_rcv);
1839 	nextrecord = m->m_nextpkt;
1840 	if (pr->pr_flags & PR_ADDR) {
1841 		KASSERT(m->m_type == MT_SONAME,
1842 		    ("m->m_type == %d", m->m_type));
1843 		orig_resid = 0;
1844 		if (psa != NULL)
1845 			*psa = sodupsockaddr(mtod(m, struct sockaddr *),
1846 			    M_NOWAIT);
1847 		if (flags & MSG_PEEK) {
1848 			m = m->m_next;
1849 		} else {
1850 			sbfree(&so->so_rcv, m);
1851 			so->so_rcv.sb_mb = m_free(m);
1852 			m = so->so_rcv.sb_mb;
1853 			sockbuf_pushsync(&so->so_rcv, nextrecord);
1854 		}
1855 	}
1856 
1857 	/*
1858 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
1859 	 * in the first mbuf chain on the socket buffer.  If MSG_PEEK, we
1860 	 * just copy the data; if !MSG_PEEK, we call into the protocol to
1861 	 * perform externalization (or freeing if controlp == NULL).
1862 	 */
1863 	if (m != NULL && m->m_type == MT_CONTROL) {
1864 		struct mbuf *cm = NULL, *cmn;
1865 		struct mbuf **cme = &cm;
1866 
1867 		do {
1868 			if (flags & MSG_PEEK) {
1869 				if (controlp != NULL) {
1870 					*controlp = m_copym(m, 0, m->m_len,
1871 					    M_NOWAIT);
1872 					controlp = &(*controlp)->m_next;
1873 				}
1874 				m = m->m_next;
1875 			} else {
1876 				sbfree(&so->so_rcv, m);
1877 				so->so_rcv.sb_mb = m->m_next;
1878 				m->m_next = NULL;
1879 				*cme = m;
1880 				cme = &(*cme)->m_next;
1881 				m = so->so_rcv.sb_mb;
1882 			}
1883 		} while (m != NULL && m->m_type == MT_CONTROL);
1884 		if ((flags & MSG_PEEK) == 0)
1885 			sockbuf_pushsync(&so->so_rcv, nextrecord);
1886 		while (cm != NULL) {
1887 			cmn = cm->m_next;
1888 			cm->m_next = NULL;
1889 			if (pr->pr_domain->dom_externalize != NULL) {
1890 				SOCKBUF_UNLOCK(&so->so_rcv);
1891 				VNET_SO_ASSERT(so);
1892 				error = (*pr->pr_domain->dom_externalize)
1893 				    (cm, controlp, flags);
1894 				SOCKBUF_LOCK(&so->so_rcv);
1895 			} else if (controlp != NULL)
1896 				*controlp = cm;
1897 			else
1898 				m_freem(cm);
1899 			if (controlp != NULL) {
1900 				orig_resid = 0;
1901 				while (*controlp != NULL)
1902 					controlp = &(*controlp)->m_next;
1903 			}
1904 			cm = cmn;
1905 		}
1906 		if (m != NULL)
1907 			nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1908 		else
1909 			nextrecord = so->so_rcv.sb_mb;
1910 		orig_resid = 0;
1911 	}
1912 	if (m != NULL) {
1913 		if ((flags & MSG_PEEK) == 0) {
1914 			KASSERT(m->m_nextpkt == nextrecord,
1915 			    ("soreceive: post-control, nextrecord !sync"));
1916 			if (nextrecord == NULL) {
1917 				KASSERT(so->so_rcv.sb_mb == m,
1918 				    ("soreceive: post-control, sb_mb!=m"));
1919 				KASSERT(so->so_rcv.sb_lastrecord == m,
1920 				    ("soreceive: post-control, lastrecord!=m"));
1921 			}
1922 		}
1923 		type = m->m_type;
1924 		if (type == MT_OOBDATA)
1925 			flags |= MSG_OOB;
1926 	} else {
1927 		if ((flags & MSG_PEEK) == 0) {
1928 			KASSERT(so->so_rcv.sb_mb == nextrecord,
1929 			    ("soreceive: sb_mb != nextrecord"));
1930 			if (so->so_rcv.sb_mb == NULL) {
1931 				KASSERT(so->so_rcv.sb_lastrecord == NULL,
1932 				    ("soreceive: sb_lastercord != NULL"));
1933 			}
1934 		}
1935 	}
1936 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1937 	SBLASTRECORDCHK(&so->so_rcv);
1938 	SBLASTMBUFCHK(&so->so_rcv);
1939 
1940 	/*
1941 	 * Now continue to read any data mbufs off of the head of the socket
1942 	 * buffer until the read request is satisfied.  Note that 'type' is
1943 	 * used to store the type of any mbuf reads that have happened so far
1944 	 * such that soreceive() can stop reading if the type changes, which
1945 	 * causes soreceive() to return only one of regular data and inline
1946 	 * out-of-band data in a single socket receive operation.
1947 	 */
1948 	moff = 0;
1949 	offset = 0;
1950 	while (m != NULL && !(m->m_flags & M_NOTAVAIL) && uio->uio_resid > 0
1951 	    && error == 0) {
1952 		/*
1953 		 * If the type of mbuf has changed since the last mbuf
1954 		 * examined ('type'), end the receive operation.
1955 		 */
1956 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1957 		if (m->m_type == MT_OOBDATA || m->m_type == MT_CONTROL) {
1958 			if (type != m->m_type)
1959 				break;
1960 		} else if (type == MT_OOBDATA)
1961 			break;
1962 		else
1963 		    KASSERT(m->m_type == MT_DATA,
1964 			("m->m_type == %d", m->m_type));
1965 		so->so_rcv.sb_state &= ~SBS_RCVATMARK;
1966 		len = uio->uio_resid;
1967 		if (so->so_oobmark && len > so->so_oobmark - offset)
1968 			len = so->so_oobmark - offset;
1969 		if (len > m->m_len - moff)
1970 			len = m->m_len - moff;
1971 		/*
1972 		 * If mp is set, just pass back the mbufs.  Otherwise copy
1973 		 * them out via the uio, then free.  Sockbuf must be
1974 		 * consistent here (points to current mbuf, it points to next
1975 		 * record) when we drop priority; we must note any additions
1976 		 * to the sockbuf when we block interrupts again.
1977 		 */
1978 		if (mp == NULL) {
1979 			SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1980 			SBLASTRECORDCHK(&so->so_rcv);
1981 			SBLASTMBUFCHK(&so->so_rcv);
1982 			SOCKBUF_UNLOCK(&so->so_rcv);
1983 			error = uiomove(mtod(m, char *) + moff, (int)len, uio);
1984 			SOCKBUF_LOCK(&so->so_rcv);
1985 			if (error) {
1986 				/*
1987 				 * The MT_SONAME mbuf has already been removed
1988 				 * from the record, so it is necessary to
1989 				 * remove the data mbufs, if any, to preserve
1990 				 * the invariant in the case of PR_ADDR that
1991 				 * requires MT_SONAME mbufs at the head of
1992 				 * each record.
1993 				 */
1994 				if (pr->pr_flags & PR_ATOMIC &&
1995 				    ((flags & MSG_PEEK) == 0))
1996 					(void)sbdroprecord_locked(&so->so_rcv);
1997 				SOCKBUF_UNLOCK(&so->so_rcv);
1998 				goto release;
1999 			}
2000 		} else
2001 			uio->uio_resid -= len;
2002 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2003 		if (len == m->m_len - moff) {
2004 			if (m->m_flags & M_EOR)
2005 				flags |= MSG_EOR;
2006 			if (flags & MSG_PEEK) {
2007 				m = m->m_next;
2008 				moff = 0;
2009 			} else {
2010 				nextrecord = m->m_nextpkt;
2011 				sbfree(&so->so_rcv, m);
2012 				if (mp != NULL) {
2013 					m->m_nextpkt = NULL;
2014 					*mp = m;
2015 					mp = &m->m_next;
2016 					so->so_rcv.sb_mb = m = m->m_next;
2017 					*mp = NULL;
2018 				} else {
2019 					so->so_rcv.sb_mb = m_free(m);
2020 					m = so->so_rcv.sb_mb;
2021 				}
2022 				sockbuf_pushsync(&so->so_rcv, nextrecord);
2023 				SBLASTRECORDCHK(&so->so_rcv);
2024 				SBLASTMBUFCHK(&so->so_rcv);
2025 			}
2026 		} else {
2027 			if (flags & MSG_PEEK)
2028 				moff += len;
2029 			else {
2030 				if (mp != NULL) {
2031 					if (flags & MSG_DONTWAIT) {
2032 						*mp = m_copym(m, 0, len,
2033 						    M_NOWAIT);
2034 						if (*mp == NULL) {
2035 							/*
2036 							 * m_copym() couldn't
2037 							 * allocate an mbuf.
2038 							 * Adjust uio_resid back
2039 							 * (it was adjusted
2040 							 * down by len bytes,
2041 							 * which we didn't end
2042 							 * up "copying" over).
2043 							 */
2044 							uio->uio_resid += len;
2045 							break;
2046 						}
2047 					} else {
2048 						SOCKBUF_UNLOCK(&so->so_rcv);
2049 						*mp = m_copym(m, 0, len,
2050 						    M_WAITOK);
2051 						SOCKBUF_LOCK(&so->so_rcv);
2052 					}
2053 				}
2054 				sbcut_locked(&so->so_rcv, len);
2055 			}
2056 		}
2057 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2058 		if (so->so_oobmark) {
2059 			if ((flags & MSG_PEEK) == 0) {
2060 				so->so_oobmark -= len;
2061 				if (so->so_oobmark == 0) {
2062 					so->so_rcv.sb_state |= SBS_RCVATMARK;
2063 					break;
2064 				}
2065 			} else {
2066 				offset += len;
2067 				if (offset == so->so_oobmark)
2068 					break;
2069 			}
2070 		}
2071 		if (flags & MSG_EOR)
2072 			break;
2073 		/*
2074 		 * If the MSG_WAITALL flag is set (for non-atomic socket), we
2075 		 * must not quit until "uio->uio_resid == 0" or an error
2076 		 * termination.  If a signal/timeout occurs, return with a
2077 		 * short count but without error.  Keep sockbuf locked
2078 		 * against other readers.
2079 		 */
2080 		while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
2081 		    !sosendallatonce(so) && nextrecord == NULL) {
2082 			SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2083 			if (so->so_error ||
2084 			    so->so_rcv.sb_state & SBS_CANTRCVMORE)
2085 				break;
2086 			/*
2087 			 * Notify the protocol that some data has been
2088 			 * drained before blocking.
2089 			 */
2090 			if (pr->pr_flags & PR_WANTRCVD) {
2091 				SOCKBUF_UNLOCK(&so->so_rcv);
2092 				VNET_SO_ASSERT(so);
2093 				(*pr->pr_usrreqs->pru_rcvd)(so, flags);
2094 				SOCKBUF_LOCK(&so->so_rcv);
2095 			}
2096 			SBLASTRECORDCHK(&so->so_rcv);
2097 			SBLASTMBUFCHK(&so->so_rcv);
2098 			/*
2099 			 * We could receive some data while was notifying
2100 			 * the protocol. Skip blocking in this case.
2101 			 */
2102 			if (so->so_rcv.sb_mb == NULL) {
2103 				error = sbwait(&so->so_rcv);
2104 				if (error) {
2105 					SOCKBUF_UNLOCK(&so->so_rcv);
2106 					goto release;
2107 				}
2108 			}
2109 			m = so->so_rcv.sb_mb;
2110 			if (m != NULL)
2111 				nextrecord = m->m_nextpkt;
2112 		}
2113 	}
2114 
2115 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2116 	if (m != NULL && pr->pr_flags & PR_ATOMIC) {
2117 		flags |= MSG_TRUNC;
2118 		if ((flags & MSG_PEEK) == 0)
2119 			(void) sbdroprecord_locked(&so->so_rcv);
2120 	}
2121 	if ((flags & MSG_PEEK) == 0) {
2122 		if (m == NULL) {
2123 			/*
2124 			 * First part is an inline SB_EMPTY_FIXUP().  Second
2125 			 * part makes sure sb_lastrecord is up-to-date if
2126 			 * there is still data in the socket buffer.
2127 			 */
2128 			so->so_rcv.sb_mb = nextrecord;
2129 			if (so->so_rcv.sb_mb == NULL) {
2130 				so->so_rcv.sb_mbtail = NULL;
2131 				so->so_rcv.sb_lastrecord = NULL;
2132 			} else if (nextrecord->m_nextpkt == NULL)
2133 				so->so_rcv.sb_lastrecord = nextrecord;
2134 		}
2135 		SBLASTRECORDCHK(&so->so_rcv);
2136 		SBLASTMBUFCHK(&so->so_rcv);
2137 		/*
2138 		 * If soreceive() is being done from the socket callback,
2139 		 * then don't need to generate ACK to peer to update window,
2140 		 * since ACK will be generated on return to TCP.
2141 		 */
2142 		if (!(flags & MSG_SOCALLBCK) &&
2143 		    (pr->pr_flags & PR_WANTRCVD)) {
2144 			SOCKBUF_UNLOCK(&so->so_rcv);
2145 			VNET_SO_ASSERT(so);
2146 			(*pr->pr_usrreqs->pru_rcvd)(so, flags);
2147 			SOCKBUF_LOCK(&so->so_rcv);
2148 		}
2149 	}
2150 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2151 	if (orig_resid == uio->uio_resid && orig_resid &&
2152 	    (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
2153 		SOCKBUF_UNLOCK(&so->so_rcv);
2154 		goto restart;
2155 	}
2156 	SOCKBUF_UNLOCK(&so->so_rcv);
2157 
2158 	if (flagsp != NULL)
2159 		*flagsp |= flags;
2160 release:
2161 	sbunlock(&so->so_rcv);
2162 	return (error);
2163 }
2164 
2165 /*
2166  * Optimized version of soreceive() for stream (TCP) sockets.
2167  */
2168 int
2169 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio,
2170     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2171 {
2172 	int len = 0, error = 0, flags, oresid;
2173 	struct sockbuf *sb;
2174 	struct mbuf *m, *n = NULL;
2175 
2176 	/* We only do stream sockets. */
2177 	if (so->so_type != SOCK_STREAM)
2178 		return (EINVAL);
2179 	if (psa != NULL)
2180 		*psa = NULL;
2181 	if (flagsp != NULL)
2182 		flags = *flagsp &~ MSG_EOR;
2183 	else
2184 		flags = 0;
2185 	if (controlp != NULL)
2186 		*controlp = NULL;
2187 	if (flags & MSG_OOB)
2188 		return (soreceive_rcvoob(so, uio, flags));
2189 	if (mp0 != NULL)
2190 		*mp0 = NULL;
2191 
2192 	sb = &so->so_rcv;
2193 
2194 	/* Prevent other readers from entering the socket. */
2195 	error = sblock(sb, SBLOCKWAIT(flags));
2196 	if (error)
2197 		goto out;
2198 	SOCKBUF_LOCK(sb);
2199 
2200 	/* Easy one, no space to copyout anything. */
2201 	if (uio->uio_resid == 0) {
2202 		error = EINVAL;
2203 		goto out;
2204 	}
2205 	oresid = uio->uio_resid;
2206 
2207 	/* We will never ever get anything unless we are or were connected. */
2208 	if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) {
2209 		error = ENOTCONN;
2210 		goto out;
2211 	}
2212 
2213 restart:
2214 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2215 
2216 	/* Abort if socket has reported problems. */
2217 	if (so->so_error) {
2218 		if (sbavail(sb) > 0)
2219 			goto deliver;
2220 		if (oresid > uio->uio_resid)
2221 			goto out;
2222 		error = so->so_error;
2223 		if (!(flags & MSG_PEEK))
2224 			so->so_error = 0;
2225 		goto out;
2226 	}
2227 
2228 	/* Door is closed.  Deliver what is left, if any. */
2229 	if (sb->sb_state & SBS_CANTRCVMORE) {
2230 		if (sbavail(sb) > 0)
2231 			goto deliver;
2232 		else
2233 			goto out;
2234 	}
2235 
2236 	/* Socket buffer is empty and we shall not block. */
2237 	if (sbavail(sb) == 0 &&
2238 	    ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) {
2239 		error = EAGAIN;
2240 		goto out;
2241 	}
2242 
2243 	/* Socket buffer got some data that we shall deliver now. */
2244 	if (sbavail(sb) > 0 && !(flags & MSG_WAITALL) &&
2245 	    ((so->so_state & SS_NBIO) ||
2246 	     (flags & (MSG_DONTWAIT|MSG_NBIO)) ||
2247 	     sbavail(sb) >= sb->sb_lowat ||
2248 	     sbavail(sb) >= uio->uio_resid ||
2249 	     sbavail(sb) >= sb->sb_hiwat) ) {
2250 		goto deliver;
2251 	}
2252 
2253 	/* On MSG_WAITALL we must wait until all data or error arrives. */
2254 	if ((flags & MSG_WAITALL) &&
2255 	    (sbavail(sb) >= uio->uio_resid || sbavail(sb) >= sb->sb_hiwat))
2256 		goto deliver;
2257 
2258 	/*
2259 	 * Wait and block until (more) data comes in.
2260 	 * NB: Drops the sockbuf lock during wait.
2261 	 */
2262 	error = sbwait(sb);
2263 	if (error)
2264 		goto out;
2265 	goto restart;
2266 
2267 deliver:
2268 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2269 	KASSERT(sbavail(sb) > 0, ("%s: sockbuf empty", __func__));
2270 	KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__));
2271 
2272 	/* Statistics. */
2273 	if (uio->uio_td)
2274 		uio->uio_td->td_ru.ru_msgrcv++;
2275 
2276 	/* Fill uio until full or current end of socket buffer is reached. */
2277 	len = min(uio->uio_resid, sbavail(sb));
2278 	if (mp0 != NULL) {
2279 		/* Dequeue as many mbufs as possible. */
2280 		if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) {
2281 			if (*mp0 == NULL)
2282 				*mp0 = sb->sb_mb;
2283 			else
2284 				m_cat(*mp0, sb->sb_mb);
2285 			for (m = sb->sb_mb;
2286 			     m != NULL && m->m_len <= len;
2287 			     m = m->m_next) {
2288 				KASSERT(!(m->m_flags & M_NOTAVAIL),
2289 				    ("%s: m %p not available", __func__, m));
2290 				len -= m->m_len;
2291 				uio->uio_resid -= m->m_len;
2292 				sbfree(sb, m);
2293 				n = m;
2294 			}
2295 			n->m_next = NULL;
2296 			sb->sb_mb = m;
2297 			sb->sb_lastrecord = sb->sb_mb;
2298 			if (sb->sb_mb == NULL)
2299 				SB_EMPTY_FIXUP(sb);
2300 		}
2301 		/* Copy the remainder. */
2302 		if (len > 0) {
2303 			KASSERT(sb->sb_mb != NULL,
2304 			    ("%s: len > 0 && sb->sb_mb empty", __func__));
2305 
2306 			m = m_copym(sb->sb_mb, 0, len, M_NOWAIT);
2307 			if (m == NULL)
2308 				len = 0;	/* Don't flush data from sockbuf. */
2309 			else
2310 				uio->uio_resid -= len;
2311 			if (*mp0 != NULL)
2312 				m_cat(*mp0, m);
2313 			else
2314 				*mp0 = m;
2315 			if (*mp0 == NULL) {
2316 				error = ENOBUFS;
2317 				goto out;
2318 			}
2319 		}
2320 	} else {
2321 		/* NB: Must unlock socket buffer as uiomove may sleep. */
2322 		SOCKBUF_UNLOCK(sb);
2323 		error = m_mbuftouio(uio, sb->sb_mb, len);
2324 		SOCKBUF_LOCK(sb);
2325 		if (error)
2326 			goto out;
2327 	}
2328 	SBLASTRECORDCHK(sb);
2329 	SBLASTMBUFCHK(sb);
2330 
2331 	/*
2332 	 * Remove the delivered data from the socket buffer unless we
2333 	 * were only peeking.
2334 	 */
2335 	if (!(flags & MSG_PEEK)) {
2336 		if (len > 0)
2337 			sbdrop_locked(sb, len);
2338 
2339 		/* Notify protocol that we drained some data. */
2340 		if ((so->so_proto->pr_flags & PR_WANTRCVD) &&
2341 		    (((flags & MSG_WAITALL) && uio->uio_resid > 0) ||
2342 		     !(flags & MSG_SOCALLBCK))) {
2343 			SOCKBUF_UNLOCK(sb);
2344 			VNET_SO_ASSERT(so);
2345 			(*so->so_proto->pr_usrreqs->pru_rcvd)(so, flags);
2346 			SOCKBUF_LOCK(sb);
2347 		}
2348 	}
2349 
2350 	/*
2351 	 * For MSG_WAITALL we may have to loop again and wait for
2352 	 * more data to come in.
2353 	 */
2354 	if ((flags & MSG_WAITALL) && uio->uio_resid > 0)
2355 		goto restart;
2356 out:
2357 	SOCKBUF_LOCK_ASSERT(sb);
2358 	SBLASTRECORDCHK(sb);
2359 	SBLASTMBUFCHK(sb);
2360 	SOCKBUF_UNLOCK(sb);
2361 	sbunlock(sb);
2362 	return (error);
2363 }
2364 
2365 /*
2366  * Optimized version of soreceive() for simple datagram cases from userspace.
2367  * Unlike in the stream case, we're able to drop a datagram if copyout()
2368  * fails, and because we handle datagrams atomically, we don't need to use a
2369  * sleep lock to prevent I/O interlacing.
2370  */
2371 int
2372 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio,
2373     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2374 {
2375 	struct mbuf *m, *m2;
2376 	int flags, error;
2377 	ssize_t len;
2378 	struct protosw *pr = so->so_proto;
2379 	struct mbuf *nextrecord;
2380 
2381 	if (psa != NULL)
2382 		*psa = NULL;
2383 	if (controlp != NULL)
2384 		*controlp = NULL;
2385 	if (flagsp != NULL)
2386 		flags = *flagsp &~ MSG_EOR;
2387 	else
2388 		flags = 0;
2389 
2390 	/*
2391 	 * For any complicated cases, fall back to the full
2392 	 * soreceive_generic().
2393 	 */
2394 	if (mp0 != NULL || (flags & MSG_PEEK) || (flags & MSG_OOB))
2395 		return (soreceive_generic(so, psa, uio, mp0, controlp,
2396 		    flagsp));
2397 
2398 	/*
2399 	 * Enforce restrictions on use.
2400 	 */
2401 	KASSERT((pr->pr_flags & PR_WANTRCVD) == 0,
2402 	    ("soreceive_dgram: wantrcvd"));
2403 	KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic"));
2404 	KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0,
2405 	    ("soreceive_dgram: SBS_RCVATMARK"));
2406 	KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0,
2407 	    ("soreceive_dgram: P_CONNREQUIRED"));
2408 
2409 	/*
2410 	 * Loop blocking while waiting for a datagram.
2411 	 */
2412 	SOCKBUF_LOCK(&so->so_rcv);
2413 	while ((m = so->so_rcv.sb_mb) == NULL) {
2414 		KASSERT(sbavail(&so->so_rcv) == 0,
2415 		    ("soreceive_dgram: sb_mb NULL but sbavail %u",
2416 		    sbavail(&so->so_rcv)));
2417 		if (so->so_error) {
2418 			error = so->so_error;
2419 			so->so_error = 0;
2420 			SOCKBUF_UNLOCK(&so->so_rcv);
2421 			return (error);
2422 		}
2423 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE ||
2424 		    uio->uio_resid == 0) {
2425 			SOCKBUF_UNLOCK(&so->so_rcv);
2426 			return (0);
2427 		}
2428 		if ((so->so_state & SS_NBIO) ||
2429 		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2430 			SOCKBUF_UNLOCK(&so->so_rcv);
2431 			return (EWOULDBLOCK);
2432 		}
2433 		SBLASTRECORDCHK(&so->so_rcv);
2434 		SBLASTMBUFCHK(&so->so_rcv);
2435 		error = sbwait(&so->so_rcv);
2436 		if (error) {
2437 			SOCKBUF_UNLOCK(&so->so_rcv);
2438 			return (error);
2439 		}
2440 	}
2441 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2442 
2443 	if (uio->uio_td)
2444 		uio->uio_td->td_ru.ru_msgrcv++;
2445 	SBLASTRECORDCHK(&so->so_rcv);
2446 	SBLASTMBUFCHK(&so->so_rcv);
2447 	nextrecord = m->m_nextpkt;
2448 	if (nextrecord == NULL) {
2449 		KASSERT(so->so_rcv.sb_lastrecord == m,
2450 		    ("soreceive_dgram: lastrecord != m"));
2451 	}
2452 
2453 	KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord,
2454 	    ("soreceive_dgram: m_nextpkt != nextrecord"));
2455 
2456 	/*
2457 	 * Pull 'm' and its chain off the front of the packet queue.
2458 	 */
2459 	so->so_rcv.sb_mb = NULL;
2460 	sockbuf_pushsync(&so->so_rcv, nextrecord);
2461 
2462 	/*
2463 	 * Walk 'm's chain and free that many bytes from the socket buffer.
2464 	 */
2465 	for (m2 = m; m2 != NULL; m2 = m2->m_next)
2466 		sbfree(&so->so_rcv, m2);
2467 
2468 	/*
2469 	 * Do a few last checks before we let go of the lock.
2470 	 */
2471 	SBLASTRECORDCHK(&so->so_rcv);
2472 	SBLASTMBUFCHK(&so->so_rcv);
2473 	SOCKBUF_UNLOCK(&so->so_rcv);
2474 
2475 	if (pr->pr_flags & PR_ADDR) {
2476 		KASSERT(m->m_type == MT_SONAME,
2477 		    ("m->m_type == %d", m->m_type));
2478 		if (psa != NULL)
2479 			*psa = sodupsockaddr(mtod(m, struct sockaddr *),
2480 			    M_NOWAIT);
2481 		m = m_free(m);
2482 	}
2483 	if (m == NULL) {
2484 		/* XXXRW: Can this happen? */
2485 		return (0);
2486 	}
2487 
2488 	/*
2489 	 * Packet to copyout() is now in 'm' and it is disconnected from the
2490 	 * queue.
2491 	 *
2492 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
2493 	 * in the first mbuf chain on the socket buffer.  We call into the
2494 	 * protocol to perform externalization (or freeing if controlp ==
2495 	 * NULL). In some cases there can be only MT_CONTROL mbufs without
2496 	 * MT_DATA mbufs.
2497 	 */
2498 	if (m->m_type == MT_CONTROL) {
2499 		struct mbuf *cm = NULL, *cmn;
2500 		struct mbuf **cme = &cm;
2501 
2502 		do {
2503 			m2 = m->m_next;
2504 			m->m_next = NULL;
2505 			*cme = m;
2506 			cme = &(*cme)->m_next;
2507 			m = m2;
2508 		} while (m != NULL && m->m_type == MT_CONTROL);
2509 		while (cm != NULL) {
2510 			cmn = cm->m_next;
2511 			cm->m_next = NULL;
2512 			if (pr->pr_domain->dom_externalize != NULL) {
2513 				error = (*pr->pr_domain->dom_externalize)
2514 				    (cm, controlp, flags);
2515 			} else if (controlp != NULL)
2516 				*controlp = cm;
2517 			else
2518 				m_freem(cm);
2519 			if (controlp != NULL) {
2520 				while (*controlp != NULL)
2521 					controlp = &(*controlp)->m_next;
2522 			}
2523 			cm = cmn;
2524 		}
2525 	}
2526 	KASSERT(m == NULL || m->m_type == MT_DATA,
2527 	    ("soreceive_dgram: !data"));
2528 	while (m != NULL && uio->uio_resid > 0) {
2529 		len = uio->uio_resid;
2530 		if (len > m->m_len)
2531 			len = m->m_len;
2532 		error = uiomove(mtod(m, char *), (int)len, uio);
2533 		if (error) {
2534 			m_freem(m);
2535 			return (error);
2536 		}
2537 		if (len == m->m_len)
2538 			m = m_free(m);
2539 		else {
2540 			m->m_data += len;
2541 			m->m_len -= len;
2542 		}
2543 	}
2544 	if (m != NULL) {
2545 		flags |= MSG_TRUNC;
2546 		m_freem(m);
2547 	}
2548 	if (flagsp != NULL)
2549 		*flagsp |= flags;
2550 	return (0);
2551 }
2552 
2553 int
2554 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
2555     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2556 {
2557 	int error;
2558 
2559 	CURVNET_SET(so->so_vnet);
2560 	if (!SOLISTENING(so))
2561 		error = (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio,
2562 		    mp0, controlp, flagsp));
2563 	else
2564 		error = ENOTCONN;
2565 	CURVNET_RESTORE();
2566 	return (error);
2567 }
2568 
2569 int
2570 soshutdown(struct socket *so, int how)
2571 {
2572 	struct protosw *pr = so->so_proto;
2573 	int error, soerror_enotconn;
2574 
2575 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
2576 		return (EINVAL);
2577 
2578 	soerror_enotconn = 0;
2579 	if ((so->so_state &
2580 	    (SS_ISCONNECTED | SS_ISCONNECTING | SS_ISDISCONNECTING)) == 0) {
2581 		/*
2582 		 * POSIX mandates us to return ENOTCONN when shutdown(2) is
2583 		 * invoked on a datagram sockets, however historically we would
2584 		 * actually tear socket down. This is known to be leveraged by
2585 		 * some applications to unblock process waiting in recvXXX(2)
2586 		 * by other process that it shares that socket with. Try to meet
2587 		 * both backward-compatibility and POSIX requirements by forcing
2588 		 * ENOTCONN but still asking protocol to perform pru_shutdown().
2589 		 */
2590 		if (so->so_type != SOCK_DGRAM && !SOLISTENING(so))
2591 			return (ENOTCONN);
2592 		soerror_enotconn = 1;
2593 	}
2594 
2595 	if (SOLISTENING(so)) {
2596 		if (how != SHUT_WR) {
2597 			SOLISTEN_LOCK(so);
2598 			so->so_error = ECONNABORTED;
2599 			solisten_wakeup(so);	/* unlocks so */
2600 		}
2601 		goto done;
2602 	}
2603 
2604 	CURVNET_SET(so->so_vnet);
2605 	if (pr->pr_usrreqs->pru_flush != NULL)
2606 		(*pr->pr_usrreqs->pru_flush)(so, how);
2607 	if (how != SHUT_WR)
2608 		sorflush(so);
2609 	if (how != SHUT_RD) {
2610 		error = (*pr->pr_usrreqs->pru_shutdown)(so);
2611 		wakeup(&so->so_timeo);
2612 		CURVNET_RESTORE();
2613 		return ((error == 0 && soerror_enotconn) ? ENOTCONN : error);
2614 	}
2615 	wakeup(&so->so_timeo);
2616 	CURVNET_RESTORE();
2617 
2618 done:
2619 	return (soerror_enotconn ? ENOTCONN : 0);
2620 }
2621 
2622 void
2623 sorflush(struct socket *so)
2624 {
2625 	struct sockbuf *sb = &so->so_rcv;
2626 	struct protosw *pr = so->so_proto;
2627 	struct socket aso;
2628 
2629 	VNET_SO_ASSERT(so);
2630 
2631 	/*
2632 	 * In order to avoid calling dom_dispose with the socket buffer mutex
2633 	 * held, and in order to generally avoid holding the lock for a long
2634 	 * time, we make a copy of the socket buffer and clear the original
2635 	 * (except locks, state).  The new socket buffer copy won't have
2636 	 * initialized locks so we can only call routines that won't use or
2637 	 * assert those locks.
2638 	 *
2639 	 * Dislodge threads currently blocked in receive and wait to acquire
2640 	 * a lock against other simultaneous readers before clearing the
2641 	 * socket buffer.  Don't let our acquire be interrupted by a signal
2642 	 * despite any existing socket disposition on interruptable waiting.
2643 	 */
2644 	socantrcvmore(so);
2645 	(void) sblock(sb, SBL_WAIT | SBL_NOINTR);
2646 
2647 	/*
2648 	 * Invalidate/clear most of the sockbuf structure, but leave selinfo
2649 	 * and mutex data unchanged.
2650 	 */
2651 	SOCKBUF_LOCK(sb);
2652 	bzero(&aso, sizeof(aso));
2653 	aso.so_pcb = so->so_pcb;
2654 	bcopy(&sb->sb_startzero, &aso.so_rcv.sb_startzero,
2655 	    sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2656 	bzero(&sb->sb_startzero,
2657 	    sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2658 	SOCKBUF_UNLOCK(sb);
2659 	sbunlock(sb);
2660 
2661 	/*
2662 	 * Dispose of special rights and flush the copied socket.  Don't call
2663 	 * any unsafe routines (that rely on locks being initialized) on aso.
2664 	 */
2665 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
2666 		(*pr->pr_domain->dom_dispose)(&aso);
2667 	sbrelease_internal(&aso.so_rcv, so);
2668 }
2669 
2670 /*
2671  * Wrapper for Socket established helper hook.
2672  * Parameters: socket, context of the hook point, hook id.
2673  */
2674 static int inline
2675 hhook_run_socket(struct socket *so, void *hctx, int32_t h_id)
2676 {
2677 	struct socket_hhook_data hhook_data = {
2678 		.so = so,
2679 		.hctx = hctx,
2680 		.m = NULL,
2681 		.status = 0
2682 	};
2683 
2684 	CURVNET_SET(so->so_vnet);
2685 	HHOOKS_RUN_IF(V_socket_hhh[h_id], &hhook_data, &so->osd);
2686 	CURVNET_RESTORE();
2687 
2688 	/* Ugly but needed, since hhooks return void for now */
2689 	return (hhook_data.status);
2690 }
2691 
2692 /*
2693  * Perhaps this routine, and sooptcopyout(), below, ought to come in an
2694  * additional variant to handle the case where the option value needs to be
2695  * some kind of integer, but not a specific size.  In addition to their use
2696  * here, these functions are also called by the protocol-level pr_ctloutput()
2697  * routines.
2698  */
2699 int
2700 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2701 {
2702 	size_t	valsize;
2703 
2704 	/*
2705 	 * If the user gives us more than we wanted, we ignore it, but if we
2706 	 * don't get the minimum length the caller wants, we return EINVAL.
2707 	 * On success, sopt->sopt_valsize is set to however much we actually
2708 	 * retrieved.
2709 	 */
2710 	if ((valsize = sopt->sopt_valsize) < minlen)
2711 		return EINVAL;
2712 	if (valsize > len)
2713 		sopt->sopt_valsize = valsize = len;
2714 
2715 	if (sopt->sopt_td != NULL)
2716 		return (copyin(sopt->sopt_val, buf, valsize));
2717 
2718 	bcopy(sopt->sopt_val, buf, valsize);
2719 	return (0);
2720 }
2721 
2722 /*
2723  * Kernel version of setsockopt(2).
2724  *
2725  * XXX: optlen is size_t, not socklen_t
2726  */
2727 int
2728 so_setsockopt(struct socket *so, int level, int optname, void *optval,
2729     size_t optlen)
2730 {
2731 	struct sockopt sopt;
2732 
2733 	sopt.sopt_level = level;
2734 	sopt.sopt_name = optname;
2735 	sopt.sopt_dir = SOPT_SET;
2736 	sopt.sopt_val = optval;
2737 	sopt.sopt_valsize = optlen;
2738 	sopt.sopt_td = NULL;
2739 	return (sosetopt(so, &sopt));
2740 }
2741 
2742 int
2743 sosetopt(struct socket *so, struct sockopt *sopt)
2744 {
2745 	int	error, optval;
2746 	struct	linger l;
2747 	struct	timeval tv;
2748 	sbintime_t val;
2749 	uint32_t val32;
2750 #ifdef MAC
2751 	struct mac extmac;
2752 #endif
2753 
2754 	CURVNET_SET(so->so_vnet);
2755 	error = 0;
2756 	if (sopt->sopt_level != SOL_SOCKET) {
2757 		if (so->so_proto->pr_ctloutput != NULL) {
2758 			error = (*so->so_proto->pr_ctloutput)(so, sopt);
2759 			CURVNET_RESTORE();
2760 			return (error);
2761 		}
2762 		error = ENOPROTOOPT;
2763 	} else {
2764 		switch (sopt->sopt_name) {
2765 		case SO_ACCEPTFILTER:
2766 			error = accept_filt_setopt(so, sopt);
2767 			if (error)
2768 				goto bad;
2769 			break;
2770 
2771 		case SO_LINGER:
2772 			error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
2773 			if (error)
2774 				goto bad;
2775 
2776 			SOCK_LOCK(so);
2777 			so->so_linger = l.l_linger;
2778 			if (l.l_onoff)
2779 				so->so_options |= SO_LINGER;
2780 			else
2781 				so->so_options &= ~SO_LINGER;
2782 			SOCK_UNLOCK(so);
2783 			break;
2784 
2785 		case SO_DEBUG:
2786 		case SO_KEEPALIVE:
2787 		case SO_DONTROUTE:
2788 		case SO_USELOOPBACK:
2789 		case SO_BROADCAST:
2790 		case SO_REUSEADDR:
2791 		case SO_REUSEPORT:
2792 		case SO_REUSEPORT_LB:
2793 		case SO_OOBINLINE:
2794 		case SO_TIMESTAMP:
2795 		case SO_BINTIME:
2796 		case SO_NOSIGPIPE:
2797 		case SO_NO_DDP:
2798 		case SO_NO_OFFLOAD:
2799 			error = sooptcopyin(sopt, &optval, sizeof optval,
2800 			    sizeof optval);
2801 			if (error)
2802 				goto bad;
2803 			SOCK_LOCK(so);
2804 			if (optval)
2805 				so->so_options |= sopt->sopt_name;
2806 			else
2807 				so->so_options &= ~sopt->sopt_name;
2808 			SOCK_UNLOCK(so);
2809 			break;
2810 
2811 		case SO_SETFIB:
2812 			error = sooptcopyin(sopt, &optval, sizeof optval,
2813 			    sizeof optval);
2814 			if (error)
2815 				goto bad;
2816 
2817 			if (optval < 0 || optval >= rt_numfibs) {
2818 				error = EINVAL;
2819 				goto bad;
2820 			}
2821 			if (((so->so_proto->pr_domain->dom_family == PF_INET) ||
2822 			   (so->so_proto->pr_domain->dom_family == PF_INET6) ||
2823 			   (so->so_proto->pr_domain->dom_family == PF_ROUTE)))
2824 				so->so_fibnum = optval;
2825 			else
2826 				so->so_fibnum = 0;
2827 			break;
2828 
2829 		case SO_USER_COOKIE:
2830 			error = sooptcopyin(sopt, &val32, sizeof val32,
2831 			    sizeof val32);
2832 			if (error)
2833 				goto bad;
2834 			so->so_user_cookie = val32;
2835 			break;
2836 
2837 		case SO_SNDBUF:
2838 		case SO_RCVBUF:
2839 		case SO_SNDLOWAT:
2840 		case SO_RCVLOWAT:
2841 			error = sooptcopyin(sopt, &optval, sizeof optval,
2842 			    sizeof optval);
2843 			if (error)
2844 				goto bad;
2845 
2846 			/*
2847 			 * Values < 1 make no sense for any of these options,
2848 			 * so disallow them.
2849 			 */
2850 			if (optval < 1) {
2851 				error = EINVAL;
2852 				goto bad;
2853 			}
2854 
2855 			error = sbsetopt(so, sopt->sopt_name, optval);
2856 			break;
2857 
2858 		case SO_SNDTIMEO:
2859 		case SO_RCVTIMEO:
2860 #ifdef COMPAT_FREEBSD32
2861 			if (SV_CURPROC_FLAG(SV_ILP32)) {
2862 				struct timeval32 tv32;
2863 
2864 				error = sooptcopyin(sopt, &tv32, sizeof tv32,
2865 				    sizeof tv32);
2866 				CP(tv32, tv, tv_sec);
2867 				CP(tv32, tv, tv_usec);
2868 			} else
2869 #endif
2870 				error = sooptcopyin(sopt, &tv, sizeof tv,
2871 				    sizeof tv);
2872 			if (error)
2873 				goto bad;
2874 			if (tv.tv_sec < 0 || tv.tv_usec < 0 ||
2875 			    tv.tv_usec >= 1000000) {
2876 				error = EDOM;
2877 				goto bad;
2878 			}
2879 			if (tv.tv_sec > INT32_MAX)
2880 				val = SBT_MAX;
2881 			else
2882 				val = tvtosbt(tv);
2883 			switch (sopt->sopt_name) {
2884 			case SO_SNDTIMEO:
2885 				so->so_snd.sb_timeo = val;
2886 				break;
2887 			case SO_RCVTIMEO:
2888 				so->so_rcv.sb_timeo = val;
2889 				break;
2890 			}
2891 			break;
2892 
2893 		case SO_LABEL:
2894 #ifdef MAC
2895 			error = sooptcopyin(sopt, &extmac, sizeof extmac,
2896 			    sizeof extmac);
2897 			if (error)
2898 				goto bad;
2899 			error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
2900 			    so, &extmac);
2901 #else
2902 			error = EOPNOTSUPP;
2903 #endif
2904 			break;
2905 
2906 		case SO_TS_CLOCK:
2907 			error = sooptcopyin(sopt, &optval, sizeof optval,
2908 			    sizeof optval);
2909 			if (error)
2910 				goto bad;
2911 			if (optval < 0 || optval > SO_TS_CLOCK_MAX) {
2912 				error = EINVAL;
2913 				goto bad;
2914 			}
2915 			so->so_ts_clock = optval;
2916 			break;
2917 
2918 		case SO_MAX_PACING_RATE:
2919 			error = sooptcopyin(sopt, &val32, sizeof(val32),
2920 			    sizeof(val32));
2921 			if (error)
2922 				goto bad;
2923 			so->so_max_pacing_rate = val32;
2924 			break;
2925 
2926 		default:
2927 			if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
2928 				error = hhook_run_socket(so, sopt,
2929 				    HHOOK_SOCKET_OPT);
2930 			else
2931 				error = ENOPROTOOPT;
2932 			break;
2933 		}
2934 		if (error == 0 && so->so_proto->pr_ctloutput != NULL)
2935 			(void)(*so->so_proto->pr_ctloutput)(so, sopt);
2936 	}
2937 bad:
2938 	CURVNET_RESTORE();
2939 	return (error);
2940 }
2941 
2942 /*
2943  * Helper routine for getsockopt.
2944  */
2945 int
2946 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
2947 {
2948 	int	error;
2949 	size_t	valsize;
2950 
2951 	error = 0;
2952 
2953 	/*
2954 	 * Documented get behavior is that we always return a value, possibly
2955 	 * truncated to fit in the user's buffer.  Traditional behavior is
2956 	 * that we always tell the user precisely how much we copied, rather
2957 	 * than something useful like the total amount we had available for
2958 	 * her.  Note that this interface is not idempotent; the entire
2959 	 * answer must be generated ahead of time.
2960 	 */
2961 	valsize = min(len, sopt->sopt_valsize);
2962 	sopt->sopt_valsize = valsize;
2963 	if (sopt->sopt_val != NULL) {
2964 		if (sopt->sopt_td != NULL)
2965 			error = copyout(buf, sopt->sopt_val, valsize);
2966 		else
2967 			bcopy(buf, sopt->sopt_val, valsize);
2968 	}
2969 	return (error);
2970 }
2971 
2972 int
2973 sogetopt(struct socket *so, struct sockopt *sopt)
2974 {
2975 	int	error, optval;
2976 	struct	linger l;
2977 	struct	timeval tv;
2978 #ifdef MAC
2979 	struct mac extmac;
2980 #endif
2981 
2982 	CURVNET_SET(so->so_vnet);
2983 	error = 0;
2984 	if (sopt->sopt_level != SOL_SOCKET) {
2985 		if (so->so_proto->pr_ctloutput != NULL)
2986 			error = (*so->so_proto->pr_ctloutput)(so, sopt);
2987 		else
2988 			error = ENOPROTOOPT;
2989 		CURVNET_RESTORE();
2990 		return (error);
2991 	} else {
2992 		switch (sopt->sopt_name) {
2993 		case SO_ACCEPTFILTER:
2994 			error = accept_filt_getopt(so, sopt);
2995 			break;
2996 
2997 		case SO_LINGER:
2998 			SOCK_LOCK(so);
2999 			l.l_onoff = so->so_options & SO_LINGER;
3000 			l.l_linger = so->so_linger;
3001 			SOCK_UNLOCK(so);
3002 			error = sooptcopyout(sopt, &l, sizeof l);
3003 			break;
3004 
3005 		case SO_USELOOPBACK:
3006 		case SO_DONTROUTE:
3007 		case SO_DEBUG:
3008 		case SO_KEEPALIVE:
3009 		case SO_REUSEADDR:
3010 		case SO_REUSEPORT:
3011 		case SO_REUSEPORT_LB:
3012 		case SO_BROADCAST:
3013 		case SO_OOBINLINE:
3014 		case SO_ACCEPTCONN:
3015 		case SO_TIMESTAMP:
3016 		case SO_BINTIME:
3017 		case SO_NOSIGPIPE:
3018 			optval = so->so_options & sopt->sopt_name;
3019 integer:
3020 			error = sooptcopyout(sopt, &optval, sizeof optval);
3021 			break;
3022 
3023 		case SO_DOMAIN:
3024 			optval = so->so_proto->pr_domain->dom_family;
3025 			goto integer;
3026 
3027 		case SO_TYPE:
3028 			optval = so->so_type;
3029 			goto integer;
3030 
3031 		case SO_PROTOCOL:
3032 			optval = so->so_proto->pr_protocol;
3033 			goto integer;
3034 
3035 		case SO_ERROR:
3036 			SOCK_LOCK(so);
3037 			optval = so->so_error;
3038 			so->so_error = 0;
3039 			SOCK_UNLOCK(so);
3040 			goto integer;
3041 
3042 		case SO_SNDBUF:
3043 			optval = SOLISTENING(so) ? so->sol_sbsnd_hiwat :
3044 			    so->so_snd.sb_hiwat;
3045 			goto integer;
3046 
3047 		case SO_RCVBUF:
3048 			optval = SOLISTENING(so) ? so->sol_sbrcv_hiwat :
3049 			    so->so_rcv.sb_hiwat;
3050 			goto integer;
3051 
3052 		case SO_SNDLOWAT:
3053 			optval = SOLISTENING(so) ? so->sol_sbsnd_lowat :
3054 			    so->so_snd.sb_lowat;
3055 			goto integer;
3056 
3057 		case SO_RCVLOWAT:
3058 			optval = SOLISTENING(so) ? so->sol_sbrcv_lowat :
3059 			    so->so_rcv.sb_lowat;
3060 			goto integer;
3061 
3062 		case SO_SNDTIMEO:
3063 		case SO_RCVTIMEO:
3064 			tv = sbttotv(sopt->sopt_name == SO_SNDTIMEO ?
3065 			    so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
3066 #ifdef COMPAT_FREEBSD32
3067 			if (SV_CURPROC_FLAG(SV_ILP32)) {
3068 				struct timeval32 tv32;
3069 
3070 				CP(tv, tv32, tv_sec);
3071 				CP(tv, tv32, tv_usec);
3072 				error = sooptcopyout(sopt, &tv32, sizeof tv32);
3073 			} else
3074 #endif
3075 				error = sooptcopyout(sopt, &tv, sizeof tv);
3076 			break;
3077 
3078 		case SO_LABEL:
3079 #ifdef MAC
3080 			error = sooptcopyin(sopt, &extmac, sizeof(extmac),
3081 			    sizeof(extmac));
3082 			if (error)
3083 				goto bad;
3084 			error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
3085 			    so, &extmac);
3086 			if (error)
3087 				goto bad;
3088 			error = sooptcopyout(sopt, &extmac, sizeof extmac);
3089 #else
3090 			error = EOPNOTSUPP;
3091 #endif
3092 			break;
3093 
3094 		case SO_PEERLABEL:
3095 #ifdef MAC
3096 			error = sooptcopyin(sopt, &extmac, sizeof(extmac),
3097 			    sizeof(extmac));
3098 			if (error)
3099 				goto bad;
3100 			error = mac_getsockopt_peerlabel(
3101 			    sopt->sopt_td->td_ucred, so, &extmac);
3102 			if (error)
3103 				goto bad;
3104 			error = sooptcopyout(sopt, &extmac, sizeof extmac);
3105 #else
3106 			error = EOPNOTSUPP;
3107 #endif
3108 			break;
3109 
3110 		case SO_LISTENQLIMIT:
3111 			optval = SOLISTENING(so) ? so->sol_qlimit : 0;
3112 			goto integer;
3113 
3114 		case SO_LISTENQLEN:
3115 			optval = SOLISTENING(so) ? so->sol_qlen : 0;
3116 			goto integer;
3117 
3118 		case SO_LISTENINCQLEN:
3119 			optval = SOLISTENING(so) ? so->sol_incqlen : 0;
3120 			goto integer;
3121 
3122 		case SO_TS_CLOCK:
3123 			optval = so->so_ts_clock;
3124 			goto integer;
3125 
3126 		case SO_MAX_PACING_RATE:
3127 			optval = so->so_max_pacing_rate;
3128 			goto integer;
3129 
3130 		default:
3131 			if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
3132 				error = hhook_run_socket(so, sopt,
3133 				    HHOOK_SOCKET_OPT);
3134 			else
3135 				error = ENOPROTOOPT;
3136 			break;
3137 		}
3138 	}
3139 #ifdef MAC
3140 bad:
3141 #endif
3142 	CURVNET_RESTORE();
3143 	return (error);
3144 }
3145 
3146 int
3147 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
3148 {
3149 	struct mbuf *m, *m_prev;
3150 	int sopt_size = sopt->sopt_valsize;
3151 
3152 	MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
3153 	if (m == NULL)
3154 		return ENOBUFS;
3155 	if (sopt_size > MLEN) {
3156 		MCLGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT);
3157 		if ((m->m_flags & M_EXT) == 0) {
3158 			m_free(m);
3159 			return ENOBUFS;
3160 		}
3161 		m->m_len = min(MCLBYTES, sopt_size);
3162 	} else {
3163 		m->m_len = min(MLEN, sopt_size);
3164 	}
3165 	sopt_size -= m->m_len;
3166 	*mp = m;
3167 	m_prev = m;
3168 
3169 	while (sopt_size) {
3170 		MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
3171 		if (m == NULL) {
3172 			m_freem(*mp);
3173 			return ENOBUFS;
3174 		}
3175 		if (sopt_size > MLEN) {
3176 			MCLGET(m, sopt->sopt_td != NULL ? M_WAITOK :
3177 			    M_NOWAIT);
3178 			if ((m->m_flags & M_EXT) == 0) {
3179 				m_freem(m);
3180 				m_freem(*mp);
3181 				return ENOBUFS;
3182 			}
3183 			m->m_len = min(MCLBYTES, sopt_size);
3184 		} else {
3185 			m->m_len = min(MLEN, sopt_size);
3186 		}
3187 		sopt_size -= m->m_len;
3188 		m_prev->m_next = m;
3189 		m_prev = m;
3190 	}
3191 	return (0);
3192 }
3193 
3194 int
3195 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
3196 {
3197 	struct mbuf *m0 = m;
3198 
3199 	if (sopt->sopt_val == NULL)
3200 		return (0);
3201 	while (m != NULL && sopt->sopt_valsize >= m->m_len) {
3202 		if (sopt->sopt_td != NULL) {
3203 			int error;
3204 
3205 			error = copyin(sopt->sopt_val, mtod(m, char *),
3206 			    m->m_len);
3207 			if (error != 0) {
3208 				m_freem(m0);
3209 				return(error);
3210 			}
3211 		} else
3212 			bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
3213 		sopt->sopt_valsize -= m->m_len;
3214 		sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
3215 		m = m->m_next;
3216 	}
3217 	if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
3218 		panic("ip6_sooptmcopyin");
3219 	return (0);
3220 }
3221 
3222 int
3223 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
3224 {
3225 	struct mbuf *m0 = m;
3226 	size_t valsize = 0;
3227 
3228 	if (sopt->sopt_val == NULL)
3229 		return (0);
3230 	while (m != NULL && sopt->sopt_valsize >= m->m_len) {
3231 		if (sopt->sopt_td != NULL) {
3232 			int error;
3233 
3234 			error = copyout(mtod(m, char *), sopt->sopt_val,
3235 			    m->m_len);
3236 			if (error != 0) {
3237 				m_freem(m0);
3238 				return(error);
3239 			}
3240 		} else
3241 			bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
3242 		sopt->sopt_valsize -= m->m_len;
3243 		sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
3244 		valsize += m->m_len;
3245 		m = m->m_next;
3246 	}
3247 	if (m != NULL) {
3248 		/* enough soopt buffer should be given from user-land */
3249 		m_freem(m0);
3250 		return(EINVAL);
3251 	}
3252 	sopt->sopt_valsize = valsize;
3253 	return (0);
3254 }
3255 
3256 /*
3257  * sohasoutofband(): protocol notifies socket layer of the arrival of new
3258  * out-of-band data, which will then notify socket consumers.
3259  */
3260 void
3261 sohasoutofband(struct socket *so)
3262 {
3263 
3264 	if (so->so_sigio != NULL)
3265 		pgsigio(&so->so_sigio, SIGURG, 0);
3266 	selwakeuppri(&so->so_rdsel, PSOCK);
3267 }
3268 
3269 int
3270 sopoll(struct socket *so, int events, struct ucred *active_cred,
3271     struct thread *td)
3272 {
3273 
3274 	/*
3275 	 * We do not need to set or assert curvnet as long as everyone uses
3276 	 * sopoll_generic().
3277 	 */
3278 	return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred,
3279 	    td));
3280 }
3281 
3282 int
3283 sopoll_generic(struct socket *so, int events, struct ucred *active_cred,
3284     struct thread *td)
3285 {
3286 	int revents;
3287 
3288 	SOCK_LOCK(so);
3289 	if (SOLISTENING(so)) {
3290 		if (!(events & (POLLIN | POLLRDNORM)))
3291 			revents = 0;
3292 		else if (!TAILQ_EMPTY(&so->sol_comp))
3293 			revents = events & (POLLIN | POLLRDNORM);
3294 		else if ((events & POLLINIGNEOF) == 0 && so->so_error)
3295 			revents = (events & (POLLIN | POLLRDNORM)) | POLLHUP;
3296 		else {
3297 			selrecord(td, &so->so_rdsel);
3298 			revents = 0;
3299 		}
3300 	} else {
3301 		revents = 0;
3302 		SOCKBUF_LOCK(&so->so_snd);
3303 		SOCKBUF_LOCK(&so->so_rcv);
3304 		if (events & (POLLIN | POLLRDNORM))
3305 			if (soreadabledata(so))
3306 				revents |= events & (POLLIN | POLLRDNORM);
3307 		if (events & (POLLOUT | POLLWRNORM))
3308 			if (sowriteable(so))
3309 				revents |= events & (POLLOUT | POLLWRNORM);
3310 		if (events & (POLLPRI | POLLRDBAND))
3311 			if (so->so_oobmark ||
3312 			    (so->so_rcv.sb_state & SBS_RCVATMARK))
3313 				revents |= events & (POLLPRI | POLLRDBAND);
3314 		if ((events & POLLINIGNEOF) == 0) {
3315 			if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3316 				revents |= events & (POLLIN | POLLRDNORM);
3317 				if (so->so_snd.sb_state & SBS_CANTSENDMORE)
3318 					revents |= POLLHUP;
3319 			}
3320 		}
3321 		if (revents == 0) {
3322 			if (events &
3323 			    (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
3324 				selrecord(td, &so->so_rdsel);
3325 				so->so_rcv.sb_flags |= SB_SEL;
3326 			}
3327 			if (events & (POLLOUT | POLLWRNORM)) {
3328 				selrecord(td, &so->so_wrsel);
3329 				so->so_snd.sb_flags |= SB_SEL;
3330 			}
3331 		}
3332 		SOCKBUF_UNLOCK(&so->so_rcv);
3333 		SOCKBUF_UNLOCK(&so->so_snd);
3334 	}
3335 	SOCK_UNLOCK(so);
3336 	return (revents);
3337 }
3338 
3339 int
3340 soo_kqfilter(struct file *fp, struct knote *kn)
3341 {
3342 	struct socket *so = kn->kn_fp->f_data;
3343 	struct sockbuf *sb;
3344 	struct knlist *knl;
3345 
3346 	switch (kn->kn_filter) {
3347 	case EVFILT_READ:
3348 		kn->kn_fop = &soread_filtops;
3349 		knl = &so->so_rdsel.si_note;
3350 		sb = &so->so_rcv;
3351 		break;
3352 	case EVFILT_WRITE:
3353 		kn->kn_fop = &sowrite_filtops;
3354 		knl = &so->so_wrsel.si_note;
3355 		sb = &so->so_snd;
3356 		break;
3357 	case EVFILT_EMPTY:
3358 		kn->kn_fop = &soempty_filtops;
3359 		knl = &so->so_wrsel.si_note;
3360 		sb = &so->so_snd;
3361 		break;
3362 	default:
3363 		return (EINVAL);
3364 	}
3365 
3366 	SOCK_LOCK(so);
3367 	if (SOLISTENING(so)) {
3368 		knlist_add(knl, kn, 1);
3369 	} else {
3370 		SOCKBUF_LOCK(sb);
3371 		knlist_add(knl, kn, 1);
3372 		sb->sb_flags |= SB_KNOTE;
3373 		SOCKBUF_UNLOCK(sb);
3374 	}
3375 	SOCK_UNLOCK(so);
3376 	return (0);
3377 }
3378 
3379 /*
3380  * Some routines that return EOPNOTSUPP for entry points that are not
3381  * supported by a protocol.  Fill in as needed.
3382  */
3383 int
3384 pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
3385 {
3386 
3387 	return EOPNOTSUPP;
3388 }
3389 
3390 int
3391 pru_aio_queue_notsupp(struct socket *so, struct kaiocb *job)
3392 {
3393 
3394 	return EOPNOTSUPP;
3395 }
3396 
3397 int
3398 pru_attach_notsupp(struct socket *so, int proto, struct thread *td)
3399 {
3400 
3401 	return EOPNOTSUPP;
3402 }
3403 
3404 int
3405 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3406 {
3407 
3408 	return EOPNOTSUPP;
3409 }
3410 
3411 int
3412 pru_bindat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3413     struct thread *td)
3414 {
3415 
3416 	return EOPNOTSUPP;
3417 }
3418 
3419 int
3420 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3421 {
3422 
3423 	return EOPNOTSUPP;
3424 }
3425 
3426 int
3427 pru_connectat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3428     struct thread *td)
3429 {
3430 
3431 	return EOPNOTSUPP;
3432 }
3433 
3434 int
3435 pru_connect2_notsupp(struct socket *so1, struct socket *so2)
3436 {
3437 
3438 	return EOPNOTSUPP;
3439 }
3440 
3441 int
3442 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
3443     struct ifnet *ifp, struct thread *td)
3444 {
3445 
3446 	return EOPNOTSUPP;
3447 }
3448 
3449 int
3450 pru_disconnect_notsupp(struct socket *so)
3451 {
3452 
3453 	return EOPNOTSUPP;
3454 }
3455 
3456 int
3457 pru_listen_notsupp(struct socket *so, int backlog, struct thread *td)
3458 {
3459 
3460 	return EOPNOTSUPP;
3461 }
3462 
3463 int
3464 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam)
3465 {
3466 
3467 	return EOPNOTSUPP;
3468 }
3469 
3470 int
3471 pru_rcvd_notsupp(struct socket *so, int flags)
3472 {
3473 
3474 	return EOPNOTSUPP;
3475 }
3476 
3477 int
3478 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
3479 {
3480 
3481 	return EOPNOTSUPP;
3482 }
3483 
3484 int
3485 pru_send_notsupp(struct socket *so, int flags, struct mbuf *m,
3486     struct sockaddr *addr, struct mbuf *control, struct thread *td)
3487 {
3488 
3489 	return EOPNOTSUPP;
3490 }
3491 
3492 int
3493 pru_ready_notsupp(struct socket *so, struct mbuf *m, int count)
3494 {
3495 
3496 	return (EOPNOTSUPP);
3497 }
3498 
3499 /*
3500  * This isn't really a ``null'' operation, but it's the default one and
3501  * doesn't do anything destructive.
3502  */
3503 int
3504 pru_sense_null(struct socket *so, struct stat *sb)
3505 {
3506 
3507 	sb->st_blksize = so->so_snd.sb_hiwat;
3508 	return 0;
3509 }
3510 
3511 int
3512 pru_shutdown_notsupp(struct socket *so)
3513 {
3514 
3515 	return EOPNOTSUPP;
3516 }
3517 
3518 int
3519 pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam)
3520 {
3521 
3522 	return EOPNOTSUPP;
3523 }
3524 
3525 int
3526 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio,
3527     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
3528 {
3529 
3530 	return EOPNOTSUPP;
3531 }
3532 
3533 int
3534 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr,
3535     struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
3536 {
3537 
3538 	return EOPNOTSUPP;
3539 }
3540 
3541 int
3542 pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred,
3543     struct thread *td)
3544 {
3545 
3546 	return EOPNOTSUPP;
3547 }
3548 
3549 static void
3550 filt_sordetach(struct knote *kn)
3551 {
3552 	struct socket *so = kn->kn_fp->f_data;
3553 
3554 	so_rdknl_lock(so);
3555 	knlist_remove(&so->so_rdsel.si_note, kn, 1);
3556 	if (!SOLISTENING(so) && knlist_empty(&so->so_rdsel.si_note))
3557 		so->so_rcv.sb_flags &= ~SB_KNOTE;
3558 	so_rdknl_unlock(so);
3559 }
3560 
3561 /*ARGSUSED*/
3562 static int
3563 filt_soread(struct knote *kn, long hint)
3564 {
3565 	struct socket *so;
3566 
3567 	so = kn->kn_fp->f_data;
3568 
3569 	if (SOLISTENING(so)) {
3570 		SOCK_LOCK_ASSERT(so);
3571 		kn->kn_data = so->sol_qlen;
3572 		if (so->so_error) {
3573 			kn->kn_flags |= EV_EOF;
3574 			kn->kn_fflags = so->so_error;
3575 			return (1);
3576 		}
3577 		return (!TAILQ_EMPTY(&so->sol_comp));
3578 	}
3579 
3580 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
3581 
3582 	kn->kn_data = sbavail(&so->so_rcv) - so->so_rcv.sb_ctl;
3583 	if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3584 		kn->kn_flags |= EV_EOF;
3585 		kn->kn_fflags = so->so_error;
3586 		return (1);
3587 	} else if (so->so_error)	/* temporary udp error */
3588 		return (1);
3589 
3590 	if (kn->kn_sfflags & NOTE_LOWAT) {
3591 		if (kn->kn_data >= kn->kn_sdata)
3592 			return (1);
3593 	} else if (sbavail(&so->so_rcv) >= so->so_rcv.sb_lowat)
3594 		return (1);
3595 
3596 	/* This hook returning non-zero indicates an event, not error */
3597 	return (hhook_run_socket(so, NULL, HHOOK_FILT_SOREAD));
3598 }
3599 
3600 static void
3601 filt_sowdetach(struct knote *kn)
3602 {
3603 	struct socket *so = kn->kn_fp->f_data;
3604 
3605 	so_wrknl_lock(so);
3606 	knlist_remove(&so->so_wrsel.si_note, kn, 1);
3607 	if (!SOLISTENING(so) && knlist_empty(&so->so_wrsel.si_note))
3608 		so->so_snd.sb_flags &= ~SB_KNOTE;
3609 	so_wrknl_unlock(so);
3610 }
3611 
3612 /*ARGSUSED*/
3613 static int
3614 filt_sowrite(struct knote *kn, long hint)
3615 {
3616 	struct socket *so;
3617 
3618 	so = kn->kn_fp->f_data;
3619 
3620 	if (SOLISTENING(so))
3621 		return (0);
3622 
3623 	SOCKBUF_LOCK_ASSERT(&so->so_snd);
3624 	kn->kn_data = sbspace(&so->so_snd);
3625 
3626 	hhook_run_socket(so, kn, HHOOK_FILT_SOWRITE);
3627 
3628 	if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
3629 		kn->kn_flags |= EV_EOF;
3630 		kn->kn_fflags = so->so_error;
3631 		return (1);
3632 	} else if (so->so_error)	/* temporary udp error */
3633 		return (1);
3634 	else if (((so->so_state & SS_ISCONNECTED) == 0) &&
3635 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
3636 		return (0);
3637 	else if (kn->kn_sfflags & NOTE_LOWAT)
3638 		return (kn->kn_data >= kn->kn_sdata);
3639 	else
3640 		return (kn->kn_data >= so->so_snd.sb_lowat);
3641 }
3642 
3643 static int
3644 filt_soempty(struct knote *kn, long hint)
3645 {
3646 	struct socket *so;
3647 
3648 	so = kn->kn_fp->f_data;
3649 
3650 	if (SOLISTENING(so))
3651 		return (1);
3652 
3653 	SOCKBUF_LOCK_ASSERT(&so->so_snd);
3654 	kn->kn_data = sbused(&so->so_snd);
3655 
3656 	if (kn->kn_data == 0)
3657 		return (1);
3658 	else
3659 		return (0);
3660 }
3661 
3662 int
3663 socheckuid(struct socket *so, uid_t uid)
3664 {
3665 
3666 	if (so == NULL)
3667 		return (EPERM);
3668 	if (so->so_cred->cr_uid != uid)
3669 		return (EPERM);
3670 	return (0);
3671 }
3672 
3673 /*
3674  * These functions are used by protocols to notify the socket layer (and its
3675  * consumers) of state changes in the sockets driven by protocol-side events.
3676  */
3677 
3678 /*
3679  * Procedures to manipulate state flags of socket and do appropriate wakeups.
3680  *
3681  * Normal sequence from the active (originating) side is that
3682  * soisconnecting() is called during processing of connect() call, resulting
3683  * in an eventual call to soisconnected() if/when the connection is
3684  * established.  When the connection is torn down soisdisconnecting() is
3685  * called during processing of disconnect() call, and soisdisconnected() is
3686  * called when the connection to the peer is totally severed.  The semantics
3687  * of these routines are such that connectionless protocols can call
3688  * soisconnected() and soisdisconnected() only, bypassing the in-progress
3689  * calls when setting up a ``connection'' takes no time.
3690  *
3691  * From the passive side, a socket is created with two queues of sockets:
3692  * so_incomp for connections in progress and so_comp for connections already
3693  * made and awaiting user acceptance.  As a protocol is preparing incoming
3694  * connections, it creates a socket structure queued on so_incomp by calling
3695  * sonewconn().  When the connection is established, soisconnected() is
3696  * called, and transfers the socket structure to so_comp, making it available
3697  * to accept().
3698  *
3699  * If a socket is closed with sockets on either so_incomp or so_comp, these
3700  * sockets are dropped.
3701  *
3702  * If higher-level protocols are implemented in the kernel, the wakeups done
3703  * here will sometimes cause software-interrupt process scheduling.
3704  */
3705 void
3706 soisconnecting(struct socket *so)
3707 {
3708 
3709 	SOCK_LOCK(so);
3710 	so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
3711 	so->so_state |= SS_ISCONNECTING;
3712 	SOCK_UNLOCK(so);
3713 }
3714 
3715 void
3716 soisconnected(struct socket *so)
3717 {
3718 
3719 	SOCK_LOCK(so);
3720 	so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
3721 	so->so_state |= SS_ISCONNECTED;
3722 
3723 	if (so->so_qstate == SQ_INCOMP) {
3724 		struct socket *head = so->so_listen;
3725 		int ret;
3726 
3727 		KASSERT(head, ("%s: so %p on incomp of NULL", __func__, so));
3728 		/*
3729 		 * Promoting a socket from incomplete queue to complete, we
3730 		 * need to go through reverse order of locking.  We first do
3731 		 * trylock, and if that doesn't succeed, we go the hard way
3732 		 * leaving a reference and rechecking consistency after proper
3733 		 * locking.
3734 		 */
3735 		if (__predict_false(SOLISTEN_TRYLOCK(head) == 0)) {
3736 			soref(head);
3737 			SOCK_UNLOCK(so);
3738 			SOLISTEN_LOCK(head);
3739 			SOCK_LOCK(so);
3740 			if (__predict_false(head != so->so_listen)) {
3741 				/*
3742 				 * The socket went off the listen queue,
3743 				 * should be lost race to close(2) of sol.
3744 				 * The socket is about to soabort().
3745 				 */
3746 				SOCK_UNLOCK(so);
3747 				sorele(head);
3748 				return;
3749 			}
3750 			/* Not the last one, as so holds a ref. */
3751 			refcount_release(&head->so_count);
3752 		}
3753 again:
3754 		if ((so->so_options & SO_ACCEPTFILTER) == 0) {
3755 			TAILQ_REMOVE(&head->sol_incomp, so, so_list);
3756 			head->sol_incqlen--;
3757 			TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list);
3758 			head->sol_qlen++;
3759 			so->so_qstate = SQ_COMP;
3760 			SOCK_UNLOCK(so);
3761 			solisten_wakeup(head);	/* unlocks */
3762 		} else {
3763 			SOCKBUF_LOCK(&so->so_rcv);
3764 			soupcall_set(so, SO_RCV,
3765 			    head->sol_accept_filter->accf_callback,
3766 			    head->sol_accept_filter_arg);
3767 			so->so_options &= ~SO_ACCEPTFILTER;
3768 			ret = head->sol_accept_filter->accf_callback(so,
3769 			    head->sol_accept_filter_arg, M_NOWAIT);
3770 			if (ret == SU_ISCONNECTED) {
3771 				soupcall_clear(so, SO_RCV);
3772 				SOCKBUF_UNLOCK(&so->so_rcv);
3773 				goto again;
3774 			}
3775 			SOCKBUF_UNLOCK(&so->so_rcv);
3776 			SOCK_UNLOCK(so);
3777 			SOLISTEN_UNLOCK(head);
3778 		}
3779 		return;
3780 	}
3781 	SOCK_UNLOCK(so);
3782 	wakeup(&so->so_timeo);
3783 	sorwakeup(so);
3784 	sowwakeup(so);
3785 }
3786 
3787 void
3788 soisdisconnecting(struct socket *so)
3789 {
3790 
3791 	SOCK_LOCK(so);
3792 	so->so_state &= ~SS_ISCONNECTING;
3793 	so->so_state |= SS_ISDISCONNECTING;
3794 
3795 	if (!SOLISTENING(so)) {
3796 		SOCKBUF_LOCK(&so->so_rcv);
3797 		socantrcvmore_locked(so);
3798 		SOCKBUF_LOCK(&so->so_snd);
3799 		socantsendmore_locked(so);
3800 	}
3801 	SOCK_UNLOCK(so);
3802 	wakeup(&so->so_timeo);
3803 }
3804 
3805 void
3806 soisdisconnected(struct socket *so)
3807 {
3808 
3809 	SOCK_LOCK(so);
3810 	so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
3811 	so->so_state |= SS_ISDISCONNECTED;
3812 
3813 	if (!SOLISTENING(so)) {
3814 		SOCK_UNLOCK(so);
3815 		SOCKBUF_LOCK(&so->so_rcv);
3816 		socantrcvmore_locked(so);
3817 		SOCKBUF_LOCK(&so->so_snd);
3818 		sbdrop_locked(&so->so_snd, sbused(&so->so_snd));
3819 		socantsendmore_locked(so);
3820 	} else
3821 		SOCK_UNLOCK(so);
3822 	wakeup(&so->so_timeo);
3823 }
3824 
3825 /*
3826  * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
3827  */
3828 struct sockaddr *
3829 sodupsockaddr(const struct sockaddr *sa, int mflags)
3830 {
3831 	struct sockaddr *sa2;
3832 
3833 	sa2 = malloc(sa->sa_len, M_SONAME, mflags);
3834 	if (sa2)
3835 		bcopy(sa, sa2, sa->sa_len);
3836 	return sa2;
3837 }
3838 
3839 /*
3840  * Register per-socket destructor.
3841  */
3842 void
3843 sodtor_set(struct socket *so, so_dtor_t *func)
3844 {
3845 
3846 	SOCK_LOCK_ASSERT(so);
3847 	so->so_dtor = func;
3848 }
3849 
3850 /*
3851  * Register per-socket buffer upcalls.
3852  */
3853 void
3854 soupcall_set(struct socket *so, int which, so_upcall_t func, void *arg)
3855 {
3856 	struct sockbuf *sb;
3857 
3858 	KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so));
3859 
3860 	switch (which) {
3861 	case SO_RCV:
3862 		sb = &so->so_rcv;
3863 		break;
3864 	case SO_SND:
3865 		sb = &so->so_snd;
3866 		break;
3867 	default:
3868 		panic("soupcall_set: bad which");
3869 	}
3870 	SOCKBUF_LOCK_ASSERT(sb);
3871 	sb->sb_upcall = func;
3872 	sb->sb_upcallarg = arg;
3873 	sb->sb_flags |= SB_UPCALL;
3874 }
3875 
3876 void
3877 soupcall_clear(struct socket *so, int which)
3878 {
3879 	struct sockbuf *sb;
3880 
3881 	KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so));
3882 
3883 	switch (which) {
3884 	case SO_RCV:
3885 		sb = &so->so_rcv;
3886 		break;
3887 	case SO_SND:
3888 		sb = &so->so_snd;
3889 		break;
3890 	default:
3891 		panic("soupcall_clear: bad which");
3892 	}
3893 	SOCKBUF_LOCK_ASSERT(sb);
3894 	KASSERT(sb->sb_upcall != NULL,
3895 	    ("%s: so %p no upcall to clear", __func__, so));
3896 	sb->sb_upcall = NULL;
3897 	sb->sb_upcallarg = NULL;
3898 	sb->sb_flags &= ~SB_UPCALL;
3899 }
3900 
3901 void
3902 solisten_upcall_set(struct socket *so, so_upcall_t func, void *arg)
3903 {
3904 
3905 	SOLISTEN_LOCK_ASSERT(so);
3906 	so->sol_upcall = func;
3907 	so->sol_upcallarg = arg;
3908 }
3909 
3910 static void
3911 so_rdknl_lock(void *arg)
3912 {
3913 	struct socket *so = arg;
3914 
3915 	if (SOLISTENING(so))
3916 		SOCK_LOCK(so);
3917 	else
3918 		SOCKBUF_LOCK(&so->so_rcv);
3919 }
3920 
3921 static void
3922 so_rdknl_unlock(void *arg)
3923 {
3924 	struct socket *so = arg;
3925 
3926 	if (SOLISTENING(so))
3927 		SOCK_UNLOCK(so);
3928 	else
3929 		SOCKBUF_UNLOCK(&so->so_rcv);
3930 }
3931 
3932 static void
3933 so_rdknl_assert_locked(void *arg)
3934 {
3935 	struct socket *so = arg;
3936 
3937 	if (SOLISTENING(so))
3938 		SOCK_LOCK_ASSERT(so);
3939 	else
3940 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
3941 }
3942 
3943 static void
3944 so_rdknl_assert_unlocked(void *arg)
3945 {
3946 	struct socket *so = arg;
3947 
3948 	if (SOLISTENING(so))
3949 		SOCK_UNLOCK_ASSERT(so);
3950 	else
3951 		SOCKBUF_UNLOCK_ASSERT(&so->so_rcv);
3952 }
3953 
3954 static void
3955 so_wrknl_lock(void *arg)
3956 {
3957 	struct socket *so = arg;
3958 
3959 	if (SOLISTENING(so))
3960 		SOCK_LOCK(so);
3961 	else
3962 		SOCKBUF_LOCK(&so->so_snd);
3963 }
3964 
3965 static void
3966 so_wrknl_unlock(void *arg)
3967 {
3968 	struct socket *so = arg;
3969 
3970 	if (SOLISTENING(so))
3971 		SOCK_UNLOCK(so);
3972 	else
3973 		SOCKBUF_UNLOCK(&so->so_snd);
3974 }
3975 
3976 static void
3977 so_wrknl_assert_locked(void *arg)
3978 {
3979 	struct socket *so = arg;
3980 
3981 	if (SOLISTENING(so))
3982 		SOCK_LOCK_ASSERT(so);
3983 	else
3984 		SOCKBUF_LOCK_ASSERT(&so->so_snd);
3985 }
3986 
3987 static void
3988 so_wrknl_assert_unlocked(void *arg)
3989 {
3990 	struct socket *so = arg;
3991 
3992 	if (SOLISTENING(so))
3993 		SOCK_UNLOCK_ASSERT(so);
3994 	else
3995 		SOCKBUF_UNLOCK_ASSERT(&so->so_snd);
3996 }
3997 
3998 /*
3999  * Create an external-format (``xsocket'') structure using the information in
4000  * the kernel-format socket structure pointed to by so.  This is done to
4001  * reduce the spew of irrelevant information over this interface, to isolate
4002  * user code from changes in the kernel structure, and potentially to provide
4003  * information-hiding if we decide that some of this information should be
4004  * hidden from users.
4005  */
4006 void
4007 sotoxsocket(struct socket *so, struct xsocket *xso)
4008 {
4009 
4010 	xso->xso_len = sizeof *xso;
4011 	xso->xso_so = (uintptr_t)so;
4012 	xso->so_type = so->so_type;
4013 	xso->so_options = so->so_options;
4014 	xso->so_linger = so->so_linger;
4015 	xso->so_state = so->so_state;
4016 	xso->so_pcb = (uintptr_t)so->so_pcb;
4017 	xso->xso_protocol = so->so_proto->pr_protocol;
4018 	xso->xso_family = so->so_proto->pr_domain->dom_family;
4019 	xso->so_timeo = so->so_timeo;
4020 	xso->so_error = so->so_error;
4021 	xso->so_uid = so->so_cred->cr_uid;
4022 	xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
4023 	if (SOLISTENING(so)) {
4024 		xso->so_qlen = so->sol_qlen;
4025 		xso->so_incqlen = so->sol_incqlen;
4026 		xso->so_qlimit = so->sol_qlimit;
4027 		xso->so_oobmark = 0;
4028 		bzero(&xso->so_snd, sizeof(xso->so_snd));
4029 		bzero(&xso->so_rcv, sizeof(xso->so_rcv));
4030 	} else {
4031 		xso->so_state |= so->so_qstate;
4032 		xso->so_qlen = xso->so_incqlen = xso->so_qlimit = 0;
4033 		xso->so_oobmark = so->so_oobmark;
4034 		sbtoxsockbuf(&so->so_snd, &xso->so_snd);
4035 		sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
4036 	}
4037 }
4038 
4039 struct sockbuf *
4040 so_sockbuf_rcv(struct socket *so)
4041 {
4042 
4043 	return (&so->so_rcv);
4044 }
4045 
4046 struct sockbuf *
4047 so_sockbuf_snd(struct socket *so)
4048 {
4049 
4050 	return (&so->so_snd);
4051 }
4052 
4053 int
4054 so_state_get(const struct socket *so)
4055 {
4056 
4057 	return (so->so_state);
4058 }
4059 
4060 void
4061 so_state_set(struct socket *so, int val)
4062 {
4063 
4064 	so->so_state = val;
4065 }
4066 
4067 int
4068 so_options_get(const struct socket *so)
4069 {
4070 
4071 	return (so->so_options);
4072 }
4073 
4074 void
4075 so_options_set(struct socket *so, int val)
4076 {
4077 
4078 	so->so_options = val;
4079 }
4080 
4081 int
4082 so_error_get(const struct socket *so)
4083 {
4084 
4085 	return (so->so_error);
4086 }
4087 
4088 void
4089 so_error_set(struct socket *so, int val)
4090 {
4091 
4092 	so->so_error = val;
4093 }
4094 
4095 int
4096 so_linger_get(const struct socket *so)
4097 {
4098 
4099 	return (so->so_linger);
4100 }
4101 
4102 void
4103 so_linger_set(struct socket *so, int val)
4104 {
4105 
4106 	so->so_linger = val;
4107 }
4108 
4109 struct protosw *
4110 so_protosw_get(const struct socket *so)
4111 {
4112 
4113 	return (so->so_proto);
4114 }
4115 
4116 void
4117 so_protosw_set(struct socket *so, struct protosw *val)
4118 {
4119 
4120 	so->so_proto = val;
4121 }
4122 
4123 void
4124 so_sorwakeup(struct socket *so)
4125 {
4126 
4127 	sorwakeup(so);
4128 }
4129 
4130 void
4131 so_sowwakeup(struct socket *so)
4132 {
4133 
4134 	sowwakeup(so);
4135 }
4136 
4137 void
4138 so_sorwakeup_locked(struct socket *so)
4139 {
4140 
4141 	sorwakeup_locked(so);
4142 }
4143 
4144 void
4145 so_sowwakeup_locked(struct socket *so)
4146 {
4147 
4148 	sowwakeup_locked(so);
4149 }
4150 
4151 void
4152 so_lock(struct socket *so)
4153 {
4154 
4155 	SOCK_LOCK(so);
4156 }
4157 
4158 void
4159 so_unlock(struct socket *so)
4160 {
4161 
4162 	SOCK_UNLOCK(so);
4163 }
4164