xref: /freebsd/sys/kern/uipc_socket.c (revision b4af4f93c682e445bf159f0d1ec90b636296c946)
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_kern_tls.h"
111 #include "opt_sctp.h"
112 
113 #include <sys/param.h>
114 #include <sys/systm.h>
115 #include <sys/fcntl.h>
116 #include <sys/limits.h>
117 #include <sys/lock.h>
118 #include <sys/mac.h>
119 #include <sys/malloc.h>
120 #include <sys/mbuf.h>
121 #include <sys/mutex.h>
122 #include <sys/domain.h>
123 #include <sys/file.h>			/* for struct knote */
124 #include <sys/hhook.h>
125 #include <sys/kernel.h>
126 #include <sys/khelp.h>
127 #include <sys/ktls.h>
128 #include <sys/event.h>
129 #include <sys/eventhandler.h>
130 #include <sys/poll.h>
131 #include <sys/proc.h>
132 #include <sys/protosw.h>
133 #include <sys/sbuf.h>
134 #include <sys/socket.h>
135 #include <sys/socketvar.h>
136 #include <sys/resourcevar.h>
137 #include <net/route.h>
138 #include <sys/signalvar.h>
139 #include <sys/stat.h>
140 #include <sys/sx.h>
141 #include <sys/sysctl.h>
142 #include <sys/taskqueue.h>
143 #include <sys/uio.h>
144 #include <sys/un.h>
145 #include <sys/unpcb.h>
146 #include <sys/jail.h>
147 #include <sys/syslog.h>
148 #include <netinet/in.h>
149 #include <netinet/in_pcb.h>
150 #include <netinet/tcp.h>
151 
152 #include <net/vnet.h>
153 
154 #include <security/mac/mac_framework.h>
155 
156 #include <vm/uma.h>
157 
158 #ifdef COMPAT_FREEBSD32
159 #include <sys/mount.h>
160 #include <sys/sysent.h>
161 #include <compat/freebsd32/freebsd32.h>
162 #endif
163 
164 static int	soreceive_rcvoob(struct socket *so, struct uio *uio,
165 		    int flags);
166 static void	so_rdknl_lock(void *);
167 static void	so_rdknl_unlock(void *);
168 static void	so_rdknl_assert_locked(void *);
169 static void	so_rdknl_assert_unlocked(void *);
170 static void	so_wrknl_lock(void *);
171 static void	so_wrknl_unlock(void *);
172 static void	so_wrknl_assert_locked(void *);
173 static void	so_wrknl_assert_unlocked(void *);
174 
175 static void	filt_sordetach(struct knote *kn);
176 static int	filt_soread(struct knote *kn, long hint);
177 static void	filt_sowdetach(struct knote *kn);
178 static int	filt_sowrite(struct knote *kn, long hint);
179 static int	filt_soempty(struct knote *kn, long hint);
180 static int inline hhook_run_socket(struct socket *so, void *hctx, int32_t h_id);
181 fo_kqfilter_t	soo_kqfilter;
182 
183 static struct filterops soread_filtops = {
184 	.f_isfd = 1,
185 	.f_detach = filt_sordetach,
186 	.f_event = filt_soread,
187 };
188 static struct filterops sowrite_filtops = {
189 	.f_isfd = 1,
190 	.f_detach = filt_sowdetach,
191 	.f_event = filt_sowrite,
192 };
193 static struct filterops soempty_filtops = {
194 	.f_isfd = 1,
195 	.f_detach = filt_sowdetach,
196 	.f_event = filt_soempty,
197 };
198 
199 so_gen_t	so_gencnt;	/* generation count for sockets */
200 
201 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
202 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
203 
204 #define	VNET_SO_ASSERT(so)						\
205 	VNET_ASSERT(curvnet != NULL,					\
206 	    ("%s:%d curvnet is NULL, so=%p", __func__, __LINE__, (so)));
207 
208 VNET_DEFINE(struct hhook_head *, socket_hhh[HHOOK_SOCKET_LAST + 1]);
209 #define	V_socket_hhh		VNET(socket_hhh)
210 
211 /*
212  * Limit on the number of connections in the listen queue waiting
213  * for accept(2).
214  * NB: The original sysctl somaxconn is still available but hidden
215  * to prevent confusion about the actual purpose of this number.
216  */
217 static u_int somaxconn = SOMAXCONN;
218 
219 static int
220 sysctl_somaxconn(SYSCTL_HANDLER_ARGS)
221 {
222 	int error;
223 	int val;
224 
225 	val = somaxconn;
226 	error = sysctl_handle_int(oidp, &val, 0, req);
227 	if (error || !req->newptr )
228 		return (error);
229 
230 	/*
231 	 * The purpose of the UINT_MAX / 3 limit, is so that the formula
232 	 *   3 * so_qlimit / 2
233 	 * below, will not overflow.
234          */
235 
236 	if (val < 1 || val > UINT_MAX / 3)
237 		return (EINVAL);
238 
239 	somaxconn = val;
240 	return (0);
241 }
242 SYSCTL_PROC(_kern_ipc, OID_AUTO, soacceptqueue,
243     CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 0, sizeof(int),
244     sysctl_somaxconn, "I",
245     "Maximum listen socket pending connection accept queue size");
246 SYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn,
247     CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_SKIP | CTLFLAG_NEEDGIANT, 0,
248     sizeof(int), sysctl_somaxconn, "I",
249     "Maximum listen socket pending connection accept queue size (compat)");
250 
251 static int numopensockets;
252 SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD,
253     &numopensockets, 0, "Number of open sockets");
254 
255 /*
256  * accept_mtx locks down per-socket fields relating to accept queues.  See
257  * socketvar.h for an annotation of the protected fields of struct socket.
258  */
259 struct mtx accept_mtx;
260 MTX_SYSINIT(accept_mtx, &accept_mtx, "accept", MTX_DEF);
261 
262 /*
263  * so_global_mtx protects so_gencnt, numopensockets, and the per-socket
264  * so_gencnt field.
265  */
266 static struct mtx so_global_mtx;
267 MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF);
268 
269 /*
270  * General IPC sysctl name space, used by sockets and a variety of other IPC
271  * types.
272  */
273 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
274     "IPC");
275 
276 /*
277  * Initialize the socket subsystem and set up the socket
278  * memory allocator.
279  */
280 static uma_zone_t socket_zone;
281 int	maxsockets;
282 
283 static void
284 socket_zone_change(void *tag)
285 {
286 
287 	maxsockets = uma_zone_set_max(socket_zone, maxsockets);
288 }
289 
290 static void
291 socket_hhook_register(int subtype)
292 {
293 
294 	if (hhook_head_register(HHOOK_TYPE_SOCKET, subtype,
295 	    &V_socket_hhh[subtype],
296 	    HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
297 		printf("%s: WARNING: unable to register hook\n", __func__);
298 }
299 
300 static void
301 socket_hhook_deregister(int subtype)
302 {
303 
304 	if (hhook_head_deregister(V_socket_hhh[subtype]) != 0)
305 		printf("%s: WARNING: unable to deregister hook\n", __func__);
306 }
307 
308 static void
309 socket_init(void *tag)
310 {
311 
312 	socket_zone = uma_zcreate("socket", sizeof(struct socket), NULL, NULL,
313 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
314 	maxsockets = uma_zone_set_max(socket_zone, maxsockets);
315 	uma_zone_set_warning(socket_zone, "kern.ipc.maxsockets limit reached");
316 	EVENTHANDLER_REGISTER(maxsockets_change, socket_zone_change, NULL,
317 	    EVENTHANDLER_PRI_FIRST);
318 }
319 SYSINIT(socket, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, socket_init, NULL);
320 
321 static void
322 socket_vnet_init(const void *unused __unused)
323 {
324 	int i;
325 
326 	/* We expect a contiguous range */
327 	for (i = 0; i <= HHOOK_SOCKET_LAST; i++)
328 		socket_hhook_register(i);
329 }
330 VNET_SYSINIT(socket_vnet_init, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
331     socket_vnet_init, NULL);
332 
333 static void
334 socket_vnet_uninit(const void *unused __unused)
335 {
336 	int i;
337 
338 	for (i = 0; i <= HHOOK_SOCKET_LAST; i++)
339 		socket_hhook_deregister(i);
340 }
341 VNET_SYSUNINIT(socket_vnet_uninit, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
342     socket_vnet_uninit, NULL);
343 
344 /*
345  * Initialise maxsockets.  This SYSINIT must be run after
346  * tunable_mbinit().
347  */
348 static void
349 init_maxsockets(void *ignored)
350 {
351 
352 	TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
353 	maxsockets = imax(maxsockets, maxfiles);
354 }
355 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL);
356 
357 /*
358  * Sysctl to get and set the maximum global sockets limit.  Notify protocols
359  * of the change so that they can update their dependent limits as required.
360  */
361 static int
362 sysctl_maxsockets(SYSCTL_HANDLER_ARGS)
363 {
364 	int error, newmaxsockets;
365 
366 	newmaxsockets = maxsockets;
367 	error = sysctl_handle_int(oidp, &newmaxsockets, 0, req);
368 	if (error == 0 && req->newptr) {
369 		if (newmaxsockets > maxsockets &&
370 		    newmaxsockets <= maxfiles) {
371 			maxsockets = newmaxsockets;
372 			EVENTHANDLER_INVOKE(maxsockets_change);
373 		} else
374 			error = EINVAL;
375 	}
376 	return (error);
377 }
378 SYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets,
379     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, &maxsockets, 0,
380     sysctl_maxsockets, "IU",
381     "Maximum number of sockets available");
382 
383 /*
384  * Socket operation routines.  These routines are called by the routines in
385  * sys_socket.c or from a system process, and implement the semantics of
386  * socket operations by switching out to the protocol specific routines.
387  */
388 
389 /*
390  * Get a socket structure from our zone, and initialize it.  Note that it
391  * would probably be better to allocate socket and PCB at the same time, but
392  * I'm not convinced that all the protocols can be easily modified to do
393  * this.
394  *
395  * soalloc() returns a socket with a ref count of 0.
396  */
397 static struct socket *
398 soalloc(struct vnet *vnet)
399 {
400 	struct socket *so;
401 
402 	so = uma_zalloc(socket_zone, M_NOWAIT | M_ZERO);
403 	if (so == NULL)
404 		return (NULL);
405 #ifdef MAC
406 	if (mac_socket_init(so, M_NOWAIT) != 0) {
407 		uma_zfree(socket_zone, so);
408 		return (NULL);
409 	}
410 #endif
411 	if (khelp_init_osd(HELPER_CLASS_SOCKET, &so->osd)) {
412 		uma_zfree(socket_zone, so);
413 		return (NULL);
414 	}
415 
416 	/*
417 	 * The socket locking protocol allows to lock 2 sockets at a time,
418 	 * however, the first one must be a listening socket.  WITNESS lacks
419 	 * a feature to change class of an existing lock, so we use DUPOK.
420 	 */
421 	mtx_init(&so->so_lock, "socket", NULL, MTX_DEF | MTX_DUPOK);
422 	SOCKBUF_LOCK_INIT(&so->so_snd, "so_snd");
423 	SOCKBUF_LOCK_INIT(&so->so_rcv, "so_rcv");
424 	so->so_rcv.sb_sel = &so->so_rdsel;
425 	so->so_snd.sb_sel = &so->so_wrsel;
426 	sx_init(&so->so_snd.sb_sx, "so_snd_sx");
427 	sx_init(&so->so_rcv.sb_sx, "so_rcv_sx");
428 	TAILQ_INIT(&so->so_snd.sb_aiojobq);
429 	TAILQ_INIT(&so->so_rcv.sb_aiojobq);
430 	TASK_INIT(&so->so_snd.sb_aiotask, 0, soaio_snd, so);
431 	TASK_INIT(&so->so_rcv.sb_aiotask, 0, soaio_rcv, so);
432 #ifdef VIMAGE
433 	VNET_ASSERT(vnet != NULL, ("%s:%d vnet is NULL, so=%p",
434 	    __func__, __LINE__, so));
435 	so->so_vnet = vnet;
436 #endif
437 	/* We shouldn't need the so_global_mtx */
438 	if (hhook_run_socket(so, NULL, HHOOK_SOCKET_CREATE)) {
439 		/* Do we need more comprehensive error returns? */
440 		uma_zfree(socket_zone, so);
441 		return (NULL);
442 	}
443 	mtx_lock(&so_global_mtx);
444 	so->so_gencnt = ++so_gencnt;
445 	++numopensockets;
446 #ifdef VIMAGE
447 	vnet->vnet_sockcnt++;
448 #endif
449 	mtx_unlock(&so_global_mtx);
450 
451 	return (so);
452 }
453 
454 /*
455  * Free the storage associated with a socket at the socket layer, tear down
456  * locks, labels, etc.  All protocol state is assumed already to have been
457  * torn down (and possibly never set up) by the caller.
458  */
459 static void
460 sodealloc(struct socket *so)
461 {
462 
463 	KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count));
464 	KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL"));
465 
466 	mtx_lock(&so_global_mtx);
467 	so->so_gencnt = ++so_gencnt;
468 	--numopensockets;	/* Could be below, but faster here. */
469 #ifdef VIMAGE
470 	VNET_ASSERT(so->so_vnet != NULL, ("%s:%d so_vnet is NULL, so=%p",
471 	    __func__, __LINE__, so));
472 	so->so_vnet->vnet_sockcnt--;
473 #endif
474 	mtx_unlock(&so_global_mtx);
475 #ifdef MAC
476 	mac_socket_destroy(so);
477 #endif
478 	hhook_run_socket(so, NULL, HHOOK_SOCKET_CLOSE);
479 
480 	crfree(so->so_cred);
481 	khelp_destroy_osd(&so->osd);
482 	if (SOLISTENING(so)) {
483 		if (so->sol_accept_filter != NULL)
484 			accept_filt_setopt(so, NULL);
485 	} else {
486 		if (so->so_rcv.sb_hiwat)
487 			(void)chgsbsize(so->so_cred->cr_uidinfo,
488 			    &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY);
489 		if (so->so_snd.sb_hiwat)
490 			(void)chgsbsize(so->so_cred->cr_uidinfo,
491 			    &so->so_snd.sb_hiwat, 0, RLIM_INFINITY);
492 		sx_destroy(&so->so_snd.sb_sx);
493 		sx_destroy(&so->so_rcv.sb_sx);
494 		SOCKBUF_LOCK_DESTROY(&so->so_snd);
495 		SOCKBUF_LOCK_DESTROY(&so->so_rcv);
496 	}
497 	mtx_destroy(&so->so_lock);
498 	uma_zfree(socket_zone, so);
499 }
500 
501 /*
502  * socreate returns a socket with a ref count of 1.  The socket should be
503  * closed with soclose().
504  */
505 int
506 socreate(int dom, struct socket **aso, int type, int proto,
507     struct ucred *cred, struct thread *td)
508 {
509 	struct protosw *prp;
510 	struct socket *so;
511 	int error;
512 
513 	if (proto)
514 		prp = pffindproto(dom, proto, type);
515 	else
516 		prp = pffindtype(dom, type);
517 
518 	if (prp == NULL) {
519 		/* No support for domain. */
520 		if (pffinddomain(dom) == NULL)
521 			return (EAFNOSUPPORT);
522 		/* No support for socket type. */
523 		if (proto == 0 && type != 0)
524 			return (EPROTOTYPE);
525 		return (EPROTONOSUPPORT);
526 	}
527 	if (prp->pr_usrreqs->pru_attach == NULL ||
528 	    prp->pr_usrreqs->pru_attach == pru_attach_notsupp)
529 		return (EPROTONOSUPPORT);
530 
531 	if (prison_check_af(cred, prp->pr_domain->dom_family) != 0)
532 		return (EPROTONOSUPPORT);
533 
534 	if (prp->pr_type != type)
535 		return (EPROTOTYPE);
536 	so = soalloc(CRED_TO_VNET(cred));
537 	if (so == NULL)
538 		return (ENOBUFS);
539 
540 	so->so_type = type;
541 	so->so_cred = crhold(cred);
542 	if ((prp->pr_domain->dom_family == PF_INET) ||
543 	    (prp->pr_domain->dom_family == PF_INET6) ||
544 	    (prp->pr_domain->dom_family == PF_ROUTE))
545 		so->so_fibnum = td->td_proc->p_fibnum;
546 	else
547 		so->so_fibnum = 0;
548 	so->so_proto = prp;
549 #ifdef MAC
550 	mac_socket_create(cred, so);
551 #endif
552 	knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
553 	    so_rdknl_assert_locked, so_rdknl_assert_unlocked);
554 	knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
555 	    so_wrknl_assert_locked, so_wrknl_assert_unlocked);
556 	/*
557 	 * Auto-sizing of socket buffers is managed by the protocols and
558 	 * the appropriate flags must be set in the pru_attach function.
559 	 */
560 	CURVNET_SET(so->so_vnet);
561 	error = (*prp->pr_usrreqs->pru_attach)(so, proto, td);
562 	CURVNET_RESTORE();
563 	if (error) {
564 		sodealloc(so);
565 		return (error);
566 	}
567 	soref(so);
568 	*aso = so;
569 	return (0);
570 }
571 
572 #ifdef REGRESSION
573 static int regression_sonewconn_earlytest = 1;
574 SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW,
575     &regression_sonewconn_earlytest, 0, "Perform early sonewconn limit test");
576 #endif
577 
578 static struct timeval overinterval = { 60, 0 };
579 SYSCTL_TIMEVAL_SEC(_kern_ipc, OID_AUTO, sooverinterval, CTLFLAG_RW,
580     &overinterval,
581     "Delay in seconds between warnings for listen socket overflows");
582 
583 /*
584  * When an attempt at a new connection is noted on a socket which accepts
585  * connections, sonewconn is called.  If the connection is possible (subject
586  * to space constraints, etc.) then we allocate a new structure, properly
587  * linked into the data structure of the original socket, and return this.
588  * Connstatus may be 0, or SS_ISCONFIRMING, or SS_ISCONNECTED.
589  *
590  * Note: the ref count on the socket is 0 on return.
591  */
592 struct socket *
593 sonewconn(struct socket *head, int connstatus)
594 {
595 	struct sbuf descrsb;
596 	struct socket *so;
597 	int len, overcount;
598 	u_int qlen;
599 	const char localprefix[] = "local:";
600 	char descrbuf[SUNPATHLEN + sizeof(localprefix)];
601 #if defined(INET6)
602 	char addrbuf[INET6_ADDRSTRLEN];
603 #elif defined(INET)
604 	char addrbuf[INET_ADDRSTRLEN];
605 #endif
606 	bool dolog, over;
607 
608 	SOLISTEN_LOCK(head);
609 	over = (head->sol_qlen > 3 * head->sol_qlimit / 2);
610 #ifdef REGRESSION
611 	if (regression_sonewconn_earlytest && over) {
612 #else
613 	if (over) {
614 #endif
615 		head->sol_overcount++;
616 		dolog = !!ratecheck(&head->sol_lastover, &overinterval);
617 
618 		/*
619 		 * If we're going to log, copy the overflow count and queue
620 		 * length from the listen socket before dropping the lock.
621 		 * Also, reset the overflow count.
622 		 */
623 		if (dolog) {
624 			overcount = head->sol_overcount;
625 			head->sol_overcount = 0;
626 			qlen = head->sol_qlen;
627 		}
628 		SOLISTEN_UNLOCK(head);
629 
630 		if (dolog) {
631 			/*
632 			 * Try to print something descriptive about the
633 			 * socket for the error message.
634 			 */
635 			sbuf_new(&descrsb, descrbuf, sizeof(descrbuf),
636 			    SBUF_FIXEDLEN);
637 			switch (head->so_proto->pr_domain->dom_family) {
638 #if defined(INET) || defined(INET6)
639 #ifdef INET
640 			case AF_INET:
641 #endif
642 #ifdef INET6
643 			case AF_INET6:
644 				if (head->so_proto->pr_domain->dom_family ==
645 				    AF_INET6 ||
646 				    (sotoinpcb(head)->inp_inc.inc_flags &
647 				    INC_ISIPV6)) {
648 					ip6_sprintf(addrbuf,
649 					    &sotoinpcb(head)->inp_inc.inc6_laddr);
650 					sbuf_printf(&descrsb, "[%s]", addrbuf);
651 				} else
652 #endif
653 				{
654 #ifdef INET
655 					inet_ntoa_r(
656 					    sotoinpcb(head)->inp_inc.inc_laddr,
657 					    addrbuf);
658 					sbuf_cat(&descrsb, addrbuf);
659 #endif
660 				}
661 				sbuf_printf(&descrsb, ":%hu (proto %u)",
662 				    ntohs(sotoinpcb(head)->inp_inc.inc_lport),
663 				    head->so_proto->pr_protocol);
664 				break;
665 #endif /* INET || INET6 */
666 			case AF_UNIX:
667 				sbuf_cat(&descrsb, localprefix);
668 				if (sotounpcb(head)->unp_addr != NULL)
669 					len =
670 					    sotounpcb(head)->unp_addr->sun_len -
671 					    offsetof(struct sockaddr_un,
672 					    sun_path);
673 				else
674 					len = 0;
675 				if (len > 0)
676 					sbuf_bcat(&descrsb,
677 					    sotounpcb(head)->unp_addr->sun_path,
678 					    len);
679 				else
680 					sbuf_cat(&descrsb, "(unknown)");
681 				break;
682 			}
683 
684 			/*
685 			 * If we can't print something more specific, at least
686 			 * print the domain name.
687 			 */
688 			if (sbuf_finish(&descrsb) != 0 ||
689 			    sbuf_len(&descrsb) <= 0) {
690 				sbuf_clear(&descrsb);
691 				sbuf_cat(&descrsb,
692 				    head->so_proto->pr_domain->dom_name ?:
693 				    "unknown");
694 				sbuf_finish(&descrsb);
695 			}
696 			KASSERT(sbuf_len(&descrsb) > 0,
697 			    ("%s: sbuf creation failed", __func__));
698 			log(LOG_DEBUG,
699 			    "%s: pcb %p (%s): Listen queue overflow: "
700 			    "%i already in queue awaiting acceptance "
701 			    "(%d occurrences)\n",
702 			    __func__, head->so_pcb, sbuf_data(&descrsb),
703 			    qlen, overcount);
704 			sbuf_delete(&descrsb);
705 
706 			overcount = 0;
707 		}
708 
709 		return (NULL);
710 	}
711 	SOLISTEN_UNLOCK(head);
712 	VNET_ASSERT(head->so_vnet != NULL, ("%s: so %p vnet is NULL",
713 	    __func__, head));
714 	so = soalloc(head->so_vnet);
715 	if (so == NULL) {
716 		log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: "
717 		    "limit reached or out of memory\n",
718 		    __func__, head->so_pcb);
719 		return (NULL);
720 	}
721 	so->so_listen = head;
722 	so->so_type = head->so_type;
723 	so->so_linger = head->so_linger;
724 	so->so_state = head->so_state | SS_NOFDREF;
725 	so->so_fibnum = head->so_fibnum;
726 	so->so_proto = head->so_proto;
727 	so->so_cred = crhold(head->so_cred);
728 #ifdef MAC
729 	mac_socket_newconn(head, so);
730 #endif
731 	knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
732 	    so_rdknl_assert_locked, so_rdknl_assert_unlocked);
733 	knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
734 	    so_wrknl_assert_locked, so_wrknl_assert_unlocked);
735 	VNET_SO_ASSERT(head);
736 	if (soreserve(so, head->sol_sbsnd_hiwat, head->sol_sbrcv_hiwat)) {
737 		sodealloc(so);
738 		log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n",
739 		    __func__, head->so_pcb);
740 		return (NULL);
741 	}
742 	if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
743 		sodealloc(so);
744 		log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n",
745 		    __func__, head->so_pcb);
746 		return (NULL);
747 	}
748 	so->so_rcv.sb_lowat = head->sol_sbrcv_lowat;
749 	so->so_snd.sb_lowat = head->sol_sbsnd_lowat;
750 	so->so_rcv.sb_timeo = head->sol_sbrcv_timeo;
751 	so->so_snd.sb_timeo = head->sol_sbsnd_timeo;
752 	so->so_rcv.sb_flags |= head->sol_sbrcv_flags & SB_AUTOSIZE;
753 	so->so_snd.sb_flags |= head->sol_sbsnd_flags & SB_AUTOSIZE;
754 
755 	SOLISTEN_LOCK(head);
756 	if (head->sol_accept_filter != NULL)
757 		connstatus = 0;
758 	so->so_state |= connstatus;
759 	so->so_options = head->so_options & ~SO_ACCEPTCONN;
760 	soref(head); /* A socket on (in)complete queue refs head. */
761 	if (connstatus) {
762 		TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list);
763 		so->so_qstate = SQ_COMP;
764 		head->sol_qlen++;
765 		solisten_wakeup(head);	/* unlocks */
766 	} else {
767 		/*
768 		 * Keep removing sockets from the head until there's room for
769 		 * us to insert on the tail.  In pre-locking revisions, this
770 		 * was a simple if(), but as we could be racing with other
771 		 * threads and soabort() requires dropping locks, we must
772 		 * loop waiting for the condition to be true.
773 		 */
774 		while (head->sol_incqlen > head->sol_qlimit) {
775 			struct socket *sp;
776 
777 			sp = TAILQ_FIRST(&head->sol_incomp);
778 			TAILQ_REMOVE(&head->sol_incomp, sp, so_list);
779 			head->sol_incqlen--;
780 			SOCK_LOCK(sp);
781 			sp->so_qstate = SQ_NONE;
782 			sp->so_listen = NULL;
783 			SOCK_UNLOCK(sp);
784 			sorele(head);	/* does SOLISTEN_UNLOCK, head stays */
785 			soabort(sp);
786 			SOLISTEN_LOCK(head);
787 		}
788 		TAILQ_INSERT_TAIL(&head->sol_incomp, so, so_list);
789 		so->so_qstate = SQ_INCOMP;
790 		head->sol_incqlen++;
791 		SOLISTEN_UNLOCK(head);
792 	}
793 	return (so);
794 }
795 
796 #if defined(SCTP) || defined(SCTP_SUPPORT)
797 /*
798  * Socket part of sctp_peeloff().  Detach a new socket from an
799  * association.  The new socket is returned with a reference.
800  */
801 struct socket *
802 sopeeloff(struct socket *head)
803 {
804 	struct socket *so;
805 
806 	VNET_ASSERT(head->so_vnet != NULL, ("%s:%d so_vnet is NULL, head=%p",
807 	    __func__, __LINE__, head));
808 	so = soalloc(head->so_vnet);
809 	if (so == NULL) {
810 		log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: "
811 		    "limit reached or out of memory\n",
812 		    __func__, head->so_pcb);
813 		return (NULL);
814 	}
815 	so->so_type = head->so_type;
816 	so->so_options = head->so_options;
817 	so->so_linger = head->so_linger;
818 	so->so_state = (head->so_state & SS_NBIO) | SS_ISCONNECTED;
819 	so->so_fibnum = head->so_fibnum;
820 	so->so_proto = head->so_proto;
821 	so->so_cred = crhold(head->so_cred);
822 #ifdef MAC
823 	mac_socket_newconn(head, so);
824 #endif
825 	knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
826 	    so_rdknl_assert_locked, so_rdknl_assert_unlocked);
827 	knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
828 	    so_wrknl_assert_locked, so_wrknl_assert_unlocked);
829 	VNET_SO_ASSERT(head);
830 	if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat)) {
831 		sodealloc(so);
832 		log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n",
833 		    __func__, head->so_pcb);
834 		return (NULL);
835 	}
836 	if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
837 		sodealloc(so);
838 		log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n",
839 		    __func__, head->so_pcb);
840 		return (NULL);
841 	}
842 	so->so_rcv.sb_lowat = head->so_rcv.sb_lowat;
843 	so->so_snd.sb_lowat = head->so_snd.sb_lowat;
844 	so->so_rcv.sb_timeo = head->so_rcv.sb_timeo;
845 	so->so_snd.sb_timeo = head->so_snd.sb_timeo;
846 	so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE;
847 	so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE;
848 
849 	soref(so);
850 
851 	return (so);
852 }
853 #endif	/* SCTP */
854 
855 int
856 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
857 {
858 	int error;
859 
860 	CURVNET_SET(so->so_vnet);
861 	error = (*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td);
862 	CURVNET_RESTORE();
863 	return (error);
864 }
865 
866 int
867 sobindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
868 {
869 	int error;
870 
871 	CURVNET_SET(so->so_vnet);
872 	error = (*so->so_proto->pr_usrreqs->pru_bindat)(fd, so, nam, td);
873 	CURVNET_RESTORE();
874 	return (error);
875 }
876 
877 /*
878  * solisten() transitions a socket from a non-listening state to a listening
879  * state, but can also be used to update the listen queue depth on an
880  * existing listen socket.  The protocol will call back into the sockets
881  * layer using solisten_proto_check() and solisten_proto() to check and set
882  * socket-layer listen state.  Call backs are used so that the protocol can
883  * acquire both protocol and socket layer locks in whatever order is required
884  * by the protocol.
885  *
886  * Protocol implementors are advised to hold the socket lock across the
887  * socket-layer test and set to avoid races at the socket layer.
888  */
889 int
890 solisten(struct socket *so, int backlog, struct thread *td)
891 {
892 	int error;
893 
894 	CURVNET_SET(so->so_vnet);
895 	error = (*so->so_proto->pr_usrreqs->pru_listen)(so, backlog, td);
896 	CURVNET_RESTORE();
897 	return (error);
898 }
899 
900 int
901 solisten_proto_check(struct socket *so)
902 {
903 
904 	SOCK_LOCK_ASSERT(so);
905 
906 	if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
907 	    SS_ISDISCONNECTING))
908 		return (EINVAL);
909 	return (0);
910 }
911 
912 void
913 solisten_proto(struct socket *so, int backlog)
914 {
915 	int sbrcv_lowat, sbsnd_lowat;
916 	u_int sbrcv_hiwat, sbsnd_hiwat;
917 	short sbrcv_flags, sbsnd_flags;
918 	sbintime_t sbrcv_timeo, sbsnd_timeo;
919 
920 	SOCK_LOCK_ASSERT(so);
921 
922 	if (SOLISTENING(so))
923 		goto listening;
924 
925 	/*
926 	 * Change this socket to listening state.
927 	 */
928 	sbrcv_lowat = so->so_rcv.sb_lowat;
929 	sbsnd_lowat = so->so_snd.sb_lowat;
930 	sbrcv_hiwat = so->so_rcv.sb_hiwat;
931 	sbsnd_hiwat = so->so_snd.sb_hiwat;
932 	sbrcv_flags = so->so_rcv.sb_flags;
933 	sbsnd_flags = so->so_snd.sb_flags;
934 	sbrcv_timeo = so->so_rcv.sb_timeo;
935 	sbsnd_timeo = so->so_snd.sb_timeo;
936 
937 	sbdestroy(&so->so_snd, so);
938 	sbdestroy(&so->so_rcv, so);
939 	sx_destroy(&so->so_snd.sb_sx);
940 	sx_destroy(&so->so_rcv.sb_sx);
941 	SOCKBUF_LOCK_DESTROY(&so->so_snd);
942 	SOCKBUF_LOCK_DESTROY(&so->so_rcv);
943 
944 #ifdef INVARIANTS
945 	bzero(&so->so_rcv,
946 	    sizeof(struct socket) - offsetof(struct socket, so_rcv));
947 #endif
948 
949 	so->sol_sbrcv_lowat = sbrcv_lowat;
950 	so->sol_sbsnd_lowat = sbsnd_lowat;
951 	so->sol_sbrcv_hiwat = sbrcv_hiwat;
952 	so->sol_sbsnd_hiwat = sbsnd_hiwat;
953 	so->sol_sbrcv_flags = sbrcv_flags;
954 	so->sol_sbsnd_flags = sbsnd_flags;
955 	so->sol_sbrcv_timeo = sbrcv_timeo;
956 	so->sol_sbsnd_timeo = sbsnd_timeo;
957 
958 	so->sol_qlen = so->sol_incqlen = 0;
959 	TAILQ_INIT(&so->sol_incomp);
960 	TAILQ_INIT(&so->sol_comp);
961 
962 	so->sol_accept_filter = NULL;
963 	so->sol_accept_filter_arg = NULL;
964 	so->sol_accept_filter_str = NULL;
965 
966 	so->sol_upcall = NULL;
967 	so->sol_upcallarg = NULL;
968 
969 	so->so_options |= SO_ACCEPTCONN;
970 
971 listening:
972 	if (backlog < 0 || backlog > somaxconn)
973 		backlog = somaxconn;
974 	so->sol_qlimit = backlog;
975 }
976 
977 /*
978  * Wakeup listeners/subsystems once we have a complete connection.
979  * Enters with lock, returns unlocked.
980  */
981 void
982 solisten_wakeup(struct socket *sol)
983 {
984 
985 	if (sol->sol_upcall != NULL)
986 		(void )sol->sol_upcall(sol, sol->sol_upcallarg, M_NOWAIT);
987 	else {
988 		selwakeuppri(&sol->so_rdsel, PSOCK);
989 		KNOTE_LOCKED(&sol->so_rdsel.si_note, 0);
990 	}
991 	SOLISTEN_UNLOCK(sol);
992 	wakeup_one(&sol->sol_comp);
993 	if ((sol->so_state & SS_ASYNC) && sol->so_sigio != NULL)
994 		pgsigio(&sol->so_sigio, SIGIO, 0);
995 }
996 
997 /*
998  * Return single connection off a listening socket queue.  Main consumer of
999  * the function is kern_accept4().  Some modules, that do their own accept
1000  * management also use the function.
1001  *
1002  * Listening socket must be locked on entry and is returned unlocked on
1003  * return.
1004  * The flags argument is set of accept4(2) flags and ACCEPT4_INHERIT.
1005  */
1006 int
1007 solisten_dequeue(struct socket *head, struct socket **ret, int flags)
1008 {
1009 	struct socket *so;
1010 	int error;
1011 
1012 	SOLISTEN_LOCK_ASSERT(head);
1013 
1014 	while (!(head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp) &&
1015 	    head->so_error == 0) {
1016 		error = msleep(&head->sol_comp, &head->so_lock, PSOCK | PCATCH,
1017 		    "accept", 0);
1018 		if (error != 0) {
1019 			SOLISTEN_UNLOCK(head);
1020 			return (error);
1021 		}
1022 	}
1023 	if (head->so_error) {
1024 		error = head->so_error;
1025 		head->so_error = 0;
1026 	} else if ((head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp))
1027 		error = EWOULDBLOCK;
1028 	else
1029 		error = 0;
1030 	if (error) {
1031 		SOLISTEN_UNLOCK(head);
1032 		return (error);
1033 	}
1034 	so = TAILQ_FIRST(&head->sol_comp);
1035 	SOCK_LOCK(so);
1036 	KASSERT(so->so_qstate == SQ_COMP,
1037 	    ("%s: so %p not SQ_COMP", __func__, so));
1038 	soref(so);
1039 	head->sol_qlen--;
1040 	so->so_qstate = SQ_NONE;
1041 	so->so_listen = NULL;
1042 	TAILQ_REMOVE(&head->sol_comp, so, so_list);
1043 	if (flags & ACCEPT4_INHERIT)
1044 		so->so_state |= (head->so_state & SS_NBIO);
1045 	else
1046 		so->so_state |= (flags & SOCK_NONBLOCK) ? SS_NBIO : 0;
1047 	SOCK_UNLOCK(so);
1048 	sorele(head);
1049 
1050 	*ret = so;
1051 	return (0);
1052 }
1053 
1054 /*
1055  * Evaluate the reference count and named references on a socket; if no
1056  * references remain, free it.  This should be called whenever a reference is
1057  * released, such as in sorele(), but also when named reference flags are
1058  * cleared in socket or protocol code.
1059  *
1060  * sofree() will free the socket if:
1061  *
1062  * - There are no outstanding file descriptor references or related consumers
1063  *   (so_count == 0).
1064  *
1065  * - The socket has been closed by user space, if ever open (SS_NOFDREF).
1066  *
1067  * - The protocol does not have an outstanding strong reference on the socket
1068  *   (SS_PROTOREF).
1069  *
1070  * - The socket is not in a completed connection queue, so a process has been
1071  *   notified that it is present.  If it is removed, the user process may
1072  *   block in accept() despite select() saying the socket was ready.
1073  */
1074 void
1075 sofree(struct socket *so)
1076 {
1077 	struct protosw *pr = so->so_proto;
1078 
1079 	SOCK_LOCK_ASSERT(so);
1080 
1081 	if ((so->so_state & SS_NOFDREF) == 0 || so->so_count != 0 ||
1082 	    (so->so_state & SS_PROTOREF) || (so->so_qstate == SQ_COMP)) {
1083 		SOCK_UNLOCK(so);
1084 		return;
1085 	}
1086 
1087 	if (!SOLISTENING(so) && so->so_qstate == SQ_INCOMP) {
1088 		struct socket *sol;
1089 
1090 		sol = so->so_listen;
1091 		KASSERT(sol, ("%s: so %p on incomp of NULL", __func__, so));
1092 
1093 		/*
1094 		 * To solve race between close of a listening socket and
1095 		 * a socket on its incomplete queue, we need to lock both.
1096 		 * The order is first listening socket, then regular.
1097 		 * Since we don't have SS_NOFDREF neither SS_PROTOREF, this
1098 		 * function and the listening socket are the only pointers
1099 		 * to so.  To preserve so and sol, we reference both and then
1100 		 * relock.
1101 		 * After relock the socket may not move to so_comp since it
1102 		 * doesn't have PCB already, but it may be removed from
1103 		 * so_incomp. If that happens, we share responsiblity on
1104 		 * freeing the socket, but soclose() has already removed
1105 		 * it from queue.
1106 		 */
1107 		soref(sol);
1108 		soref(so);
1109 		SOCK_UNLOCK(so);
1110 		SOLISTEN_LOCK(sol);
1111 		SOCK_LOCK(so);
1112 		if (so->so_qstate == SQ_INCOMP) {
1113 			KASSERT(so->so_listen == sol,
1114 			    ("%s: so %p migrated out of sol %p",
1115 			    __func__, so, sol));
1116 			TAILQ_REMOVE(&sol->sol_incomp, so, so_list);
1117 			sol->sol_incqlen--;
1118 			/* This is guarenteed not to be the last. */
1119 			refcount_release(&sol->so_count);
1120 			so->so_qstate = SQ_NONE;
1121 			so->so_listen = NULL;
1122 		} else
1123 			KASSERT(so->so_listen == NULL,
1124 			    ("%s: so %p not on (in)comp with so_listen",
1125 			    __func__, so));
1126 		sorele(sol);
1127 		KASSERT(so->so_count == 1,
1128 		    ("%s: so %p count %u", __func__, so, so->so_count));
1129 		so->so_count = 0;
1130 	}
1131 	if (SOLISTENING(so))
1132 		so->so_error = ECONNABORTED;
1133 	SOCK_UNLOCK(so);
1134 
1135 	if (so->so_dtor != NULL)
1136 		so->so_dtor(so);
1137 
1138 	VNET_SO_ASSERT(so);
1139 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
1140 		(*pr->pr_domain->dom_dispose)(so);
1141 	if (pr->pr_usrreqs->pru_detach != NULL)
1142 		(*pr->pr_usrreqs->pru_detach)(so);
1143 
1144 	/*
1145 	 * From this point on, we assume that no other references to this
1146 	 * socket exist anywhere else in the stack.  Therefore, no locks need
1147 	 * to be acquired or held.
1148 	 *
1149 	 * We used to do a lot of socket buffer and socket locking here, as
1150 	 * well as invoke sorflush() and perform wakeups.  The direct call to
1151 	 * dom_dispose() and sbdestroy() are an inlining of what was
1152 	 * necessary from sorflush().
1153 	 *
1154 	 * Notice that the socket buffer and kqueue state are torn down
1155 	 * before calling pru_detach.  This means that protocols shold not
1156 	 * assume they can perform socket wakeups, etc, in their detach code.
1157 	 */
1158 	if (!SOLISTENING(so)) {
1159 		sbdestroy(&so->so_snd, so);
1160 		sbdestroy(&so->so_rcv, so);
1161 	}
1162 	seldrain(&so->so_rdsel);
1163 	seldrain(&so->so_wrsel);
1164 	knlist_destroy(&so->so_rdsel.si_note);
1165 	knlist_destroy(&so->so_wrsel.si_note);
1166 	sodealloc(so);
1167 }
1168 
1169 /*
1170  * Close a socket on last file table reference removal.  Initiate disconnect
1171  * if connected.  Free socket when disconnect complete.
1172  *
1173  * This function will sorele() the socket.  Note that soclose() may be called
1174  * prior to the ref count reaching zero.  The actual socket structure will
1175  * not be freed until the ref count reaches zero.
1176  */
1177 int
1178 soclose(struct socket *so)
1179 {
1180 	struct accept_queue lqueue;
1181 	bool listening;
1182 	int error = 0;
1183 
1184 	KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter"));
1185 
1186 	CURVNET_SET(so->so_vnet);
1187 	funsetown(&so->so_sigio);
1188 	if (so->so_state & SS_ISCONNECTED) {
1189 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
1190 			error = sodisconnect(so);
1191 			if (error) {
1192 				if (error == ENOTCONN)
1193 					error = 0;
1194 				goto drop;
1195 			}
1196 		}
1197 		if (so->so_options & SO_LINGER) {
1198 			if ((so->so_state & SS_ISDISCONNECTING) &&
1199 			    (so->so_state & SS_NBIO))
1200 				goto drop;
1201 			while (so->so_state & SS_ISCONNECTED) {
1202 				error = tsleep(&so->so_timeo,
1203 				    PSOCK | PCATCH, "soclos",
1204 				    so->so_linger * hz);
1205 				if (error)
1206 					break;
1207 			}
1208 		}
1209 	}
1210 
1211 drop:
1212 	if (so->so_proto->pr_usrreqs->pru_close != NULL)
1213 		(*so->so_proto->pr_usrreqs->pru_close)(so);
1214 
1215 	SOCK_LOCK(so);
1216 	if ((listening = (so->so_options & SO_ACCEPTCONN))) {
1217 		struct socket *sp;
1218 
1219 		TAILQ_INIT(&lqueue);
1220 		TAILQ_SWAP(&lqueue, &so->sol_incomp, socket, so_list);
1221 		TAILQ_CONCAT(&lqueue, &so->sol_comp, so_list);
1222 
1223 		so->sol_qlen = so->sol_incqlen = 0;
1224 
1225 		TAILQ_FOREACH(sp, &lqueue, so_list) {
1226 			SOCK_LOCK(sp);
1227 			sp->so_qstate = SQ_NONE;
1228 			sp->so_listen = NULL;
1229 			SOCK_UNLOCK(sp);
1230 			/* Guaranteed not to be the last. */
1231 			refcount_release(&so->so_count);
1232 		}
1233 	}
1234 	KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF"));
1235 	so->so_state |= SS_NOFDREF;
1236 	sorele(so);
1237 	if (listening) {
1238 		struct socket *sp, *tsp;
1239 
1240 		TAILQ_FOREACH_SAFE(sp, &lqueue, so_list, tsp) {
1241 			SOCK_LOCK(sp);
1242 			if (sp->so_count == 0) {
1243 				SOCK_UNLOCK(sp);
1244 				soabort(sp);
1245 			} else
1246 				/* sp is now in sofree() */
1247 				SOCK_UNLOCK(sp);
1248 		}
1249 	}
1250 	CURVNET_RESTORE();
1251 	return (error);
1252 }
1253 
1254 /*
1255  * soabort() is used to abruptly tear down a connection, such as when a
1256  * resource limit is reached (listen queue depth exceeded), or if a listen
1257  * socket is closed while there are sockets waiting to be accepted.
1258  *
1259  * This interface is tricky, because it is called on an unreferenced socket,
1260  * and must be called only by a thread that has actually removed the socket
1261  * from the listen queue it was on, or races with other threads are risked.
1262  *
1263  * This interface will call into the protocol code, so must not be called
1264  * with any socket locks held.  Protocols do call it while holding their own
1265  * recursible protocol mutexes, but this is something that should be subject
1266  * to review in the future.
1267  */
1268 void
1269 soabort(struct socket *so)
1270 {
1271 
1272 	/*
1273 	 * In as much as is possible, assert that no references to this
1274 	 * socket are held.  This is not quite the same as asserting that the
1275 	 * current thread is responsible for arranging for no references, but
1276 	 * is as close as we can get for now.
1277 	 */
1278 	KASSERT(so->so_count == 0, ("soabort: so_count"));
1279 	KASSERT((so->so_state & SS_PROTOREF) == 0, ("soabort: SS_PROTOREF"));
1280 	KASSERT(so->so_state & SS_NOFDREF, ("soabort: !SS_NOFDREF"));
1281 	VNET_SO_ASSERT(so);
1282 
1283 	if (so->so_proto->pr_usrreqs->pru_abort != NULL)
1284 		(*so->so_proto->pr_usrreqs->pru_abort)(so);
1285 	SOCK_LOCK(so);
1286 	sofree(so);
1287 }
1288 
1289 int
1290 soaccept(struct socket *so, struct sockaddr **nam)
1291 {
1292 	int error;
1293 
1294 	SOCK_LOCK(so);
1295 	KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF"));
1296 	so->so_state &= ~SS_NOFDREF;
1297 	SOCK_UNLOCK(so);
1298 
1299 	CURVNET_SET(so->so_vnet);
1300 	error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam);
1301 	CURVNET_RESTORE();
1302 	return (error);
1303 }
1304 
1305 int
1306 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
1307 {
1308 
1309 	return (soconnectat(AT_FDCWD, so, nam, td));
1310 }
1311 
1312 int
1313 soconnectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
1314 {
1315 	int error;
1316 
1317 	if (so->so_options & SO_ACCEPTCONN)
1318 		return (EOPNOTSUPP);
1319 
1320 	CURVNET_SET(so->so_vnet);
1321 	/*
1322 	 * If protocol is connection-based, can only connect once.
1323 	 * Otherwise, if connected, try to disconnect first.  This allows
1324 	 * user to disconnect by connecting to, e.g., a null address.
1325 	 */
1326 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
1327 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
1328 	    (error = sodisconnect(so)))) {
1329 		error = EISCONN;
1330 	} else {
1331 		/*
1332 		 * Prevent accumulated error from previous connection from
1333 		 * biting us.
1334 		 */
1335 		so->so_error = 0;
1336 		if (fd == AT_FDCWD) {
1337 			error = (*so->so_proto->pr_usrreqs->pru_connect)(so,
1338 			    nam, td);
1339 		} else {
1340 			error = (*so->so_proto->pr_usrreqs->pru_connectat)(fd,
1341 			    so, nam, td);
1342 		}
1343 	}
1344 	CURVNET_RESTORE();
1345 
1346 	return (error);
1347 }
1348 
1349 int
1350 soconnect2(struct socket *so1, struct socket *so2)
1351 {
1352 	int error;
1353 
1354 	CURVNET_SET(so1->so_vnet);
1355 	error = (*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2);
1356 	CURVNET_RESTORE();
1357 	return (error);
1358 }
1359 
1360 int
1361 sodisconnect(struct socket *so)
1362 {
1363 	int error;
1364 
1365 	if ((so->so_state & SS_ISCONNECTED) == 0)
1366 		return (ENOTCONN);
1367 	if (so->so_state & SS_ISDISCONNECTING)
1368 		return (EALREADY);
1369 	VNET_SO_ASSERT(so);
1370 	error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so);
1371 	return (error);
1372 }
1373 
1374 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? 0 : SBL_WAIT)
1375 
1376 int
1377 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio,
1378     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1379 {
1380 	long space;
1381 	ssize_t resid;
1382 	int clen = 0, error, dontroute;
1383 
1384 	KASSERT(so->so_type == SOCK_DGRAM, ("sosend_dgram: !SOCK_DGRAM"));
1385 	KASSERT(so->so_proto->pr_flags & PR_ATOMIC,
1386 	    ("sosend_dgram: !PR_ATOMIC"));
1387 
1388 	if (uio != NULL)
1389 		resid = uio->uio_resid;
1390 	else
1391 		resid = top->m_pkthdr.len;
1392 	/*
1393 	 * In theory resid should be unsigned.  However, space must be
1394 	 * signed, as it might be less than 0 if we over-committed, and we
1395 	 * must use a signed comparison of space and resid.  On the other
1396 	 * hand, a negative resid causes us to loop sending 0-length
1397 	 * segments to the protocol.
1398 	 */
1399 	if (resid < 0) {
1400 		error = EINVAL;
1401 		goto out;
1402 	}
1403 
1404 	dontroute =
1405 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0;
1406 	if (td != NULL)
1407 		td->td_ru.ru_msgsnd++;
1408 	if (control != NULL)
1409 		clen = control->m_len;
1410 
1411 	SOCKBUF_LOCK(&so->so_snd);
1412 	if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1413 		SOCKBUF_UNLOCK(&so->so_snd);
1414 		error = EPIPE;
1415 		goto out;
1416 	}
1417 	if (so->so_error) {
1418 		error = so->so_error;
1419 		so->so_error = 0;
1420 		SOCKBUF_UNLOCK(&so->so_snd);
1421 		goto out;
1422 	}
1423 	if ((so->so_state & SS_ISCONNECTED) == 0) {
1424 		/*
1425 		 * `sendto' and `sendmsg' is allowed on a connection-based
1426 		 * socket if it supports implied connect.  Return ENOTCONN if
1427 		 * not connected and no address is supplied.
1428 		 */
1429 		if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1430 		    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1431 			if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1432 			    !(resid == 0 && clen != 0)) {
1433 				SOCKBUF_UNLOCK(&so->so_snd);
1434 				error = ENOTCONN;
1435 				goto out;
1436 			}
1437 		} else if (addr == NULL) {
1438 			if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1439 				error = ENOTCONN;
1440 			else
1441 				error = EDESTADDRREQ;
1442 			SOCKBUF_UNLOCK(&so->so_snd);
1443 			goto out;
1444 		}
1445 	}
1446 
1447 	/*
1448 	 * Do we need MSG_OOB support in SOCK_DGRAM?  Signs here may be a
1449 	 * problem and need fixing.
1450 	 */
1451 	space = sbspace(&so->so_snd);
1452 	if (flags & MSG_OOB)
1453 		space += 1024;
1454 	space -= clen;
1455 	SOCKBUF_UNLOCK(&so->so_snd);
1456 	if (resid > space) {
1457 		error = EMSGSIZE;
1458 		goto out;
1459 	}
1460 	if (uio == NULL) {
1461 		resid = 0;
1462 		if (flags & MSG_EOR)
1463 			top->m_flags |= M_EOR;
1464 	} else {
1465 		/*
1466 		 * Copy the data from userland into a mbuf chain.
1467 		 * If no data is to be copied in, a single empty mbuf
1468 		 * is returned.
1469 		 */
1470 		top = m_uiotombuf(uio, M_WAITOK, space, max_hdr,
1471 		    (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0)));
1472 		if (top == NULL) {
1473 			error = EFAULT;	/* only possible error */
1474 			goto out;
1475 		}
1476 		space -= resid - uio->uio_resid;
1477 		resid = uio->uio_resid;
1478 	}
1479 	KASSERT(resid == 0, ("sosend_dgram: resid != 0"));
1480 	/*
1481 	 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock
1482 	 * than with.
1483 	 */
1484 	if (dontroute) {
1485 		SOCK_LOCK(so);
1486 		so->so_options |= SO_DONTROUTE;
1487 		SOCK_UNLOCK(so);
1488 	}
1489 	/*
1490 	 * XXX all the SBS_CANTSENDMORE checks previously done could be out
1491 	 * of date.  We could have received a reset packet in an interrupt or
1492 	 * maybe we slept while doing page faults in uiomove() etc.  We could
1493 	 * probably recheck again inside the locking protection here, but
1494 	 * there are probably other places that this also happens.  We must
1495 	 * rethink this.
1496 	 */
1497 	VNET_SO_ASSERT(so);
1498 	error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1499 	    (flags & MSG_OOB) ? PRUS_OOB :
1500 	/*
1501 	 * If the user set MSG_EOF, the protocol understands this flag and
1502 	 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND.
1503 	 */
1504 	    ((flags & MSG_EOF) &&
1505 	     (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1506 	     (resid <= 0)) ?
1507 		PRUS_EOF :
1508 		/* If there is more to send set PRUS_MORETOCOME */
1509 		(flags & MSG_MORETOCOME) ||
1510 		(resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1511 		top, addr, control, td);
1512 	if (dontroute) {
1513 		SOCK_LOCK(so);
1514 		so->so_options &= ~SO_DONTROUTE;
1515 		SOCK_UNLOCK(so);
1516 	}
1517 	clen = 0;
1518 	control = NULL;
1519 	top = NULL;
1520 out:
1521 	if (top != NULL)
1522 		m_freem(top);
1523 	if (control != NULL)
1524 		m_freem(control);
1525 	return (error);
1526 }
1527 
1528 /*
1529  * Send on a socket.  If send must go all at once and message is larger than
1530  * send buffering, then hard error.  Lock against other senders.  If must go
1531  * all at once and not enough room now, then inform user that this would
1532  * block and do nothing.  Otherwise, if nonblocking, send as much as
1533  * possible.  The data to be sent is described by "uio" if nonzero, otherwise
1534  * by the mbuf chain "top" (which must be null if uio is not).  Data provided
1535  * in mbuf chain must be small enough to send all at once.
1536  *
1537  * Returns nonzero on error, timeout or signal; callers must check for short
1538  * counts if EINTR/ERESTART are returned.  Data and control buffers are freed
1539  * on return.
1540  */
1541 int
1542 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio,
1543     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1544 {
1545 	long space;
1546 	ssize_t resid;
1547 	int clen = 0, error, dontroute;
1548 	int atomic = sosendallatonce(so) || top;
1549 	int pru_flag;
1550 #ifdef KERN_TLS
1551 	struct ktls_session *tls;
1552 	int tls_enq_cnt, tls_pruflag;
1553 	uint8_t tls_rtype;
1554 
1555 	tls = NULL;
1556 	tls_rtype = TLS_RLTYPE_APP;
1557 #endif
1558 	if (uio != NULL)
1559 		resid = uio->uio_resid;
1560 	else if ((top->m_flags & M_PKTHDR) != 0)
1561 		resid = top->m_pkthdr.len;
1562 	else
1563 		resid = m_length(top, NULL);
1564 	/*
1565 	 * In theory resid should be unsigned.  However, space must be
1566 	 * signed, as it might be less than 0 if we over-committed, and we
1567 	 * must use a signed comparison of space and resid.  On the other
1568 	 * hand, a negative resid causes us to loop sending 0-length
1569 	 * segments to the protocol.
1570 	 *
1571 	 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
1572 	 * type sockets since that's an error.
1573 	 */
1574 	if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
1575 		error = EINVAL;
1576 		goto out;
1577 	}
1578 
1579 	dontroute =
1580 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
1581 	    (so->so_proto->pr_flags & PR_ATOMIC);
1582 	if (td != NULL)
1583 		td->td_ru.ru_msgsnd++;
1584 	if (control != NULL)
1585 		clen = control->m_len;
1586 
1587 	error = sblock(&so->so_snd, SBLOCKWAIT(flags));
1588 	if (error)
1589 		goto out;
1590 
1591 #ifdef KERN_TLS
1592 	tls_pruflag = 0;
1593 	tls = ktls_hold(so->so_snd.sb_tls_info);
1594 	if (tls != NULL) {
1595 		if (tls->mode == TCP_TLS_MODE_SW)
1596 			tls_pruflag = PRUS_NOTREADY;
1597 
1598 		if (control != NULL) {
1599 			struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1600 
1601 			if (clen >= sizeof(*cm) &&
1602 			    cm->cmsg_type == TLS_SET_RECORD_TYPE) {
1603 				tls_rtype = *((uint8_t *)CMSG_DATA(cm));
1604 				clen = 0;
1605 				m_freem(control);
1606 				control = NULL;
1607 				atomic = 1;
1608 			}
1609 		}
1610 	}
1611 #endif
1612 
1613 restart:
1614 	do {
1615 		SOCKBUF_LOCK(&so->so_snd);
1616 		if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1617 			SOCKBUF_UNLOCK(&so->so_snd);
1618 			error = EPIPE;
1619 			goto release;
1620 		}
1621 		if (so->so_error) {
1622 			error = so->so_error;
1623 			so->so_error = 0;
1624 			SOCKBUF_UNLOCK(&so->so_snd);
1625 			goto release;
1626 		}
1627 		if ((so->so_state & SS_ISCONNECTED) == 0) {
1628 			/*
1629 			 * `sendto' and `sendmsg' is allowed on a connection-
1630 			 * based socket if it supports implied connect.
1631 			 * Return ENOTCONN if not connected and no address is
1632 			 * supplied.
1633 			 */
1634 			if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1635 			    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1636 				if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1637 				    !(resid == 0 && clen != 0)) {
1638 					SOCKBUF_UNLOCK(&so->so_snd);
1639 					error = ENOTCONN;
1640 					goto release;
1641 				}
1642 			} else if (addr == NULL) {
1643 				SOCKBUF_UNLOCK(&so->so_snd);
1644 				if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1645 					error = ENOTCONN;
1646 				else
1647 					error = EDESTADDRREQ;
1648 				goto release;
1649 			}
1650 		}
1651 		space = sbspace(&so->so_snd);
1652 		if (flags & MSG_OOB)
1653 			space += 1024;
1654 		if ((atomic && resid > so->so_snd.sb_hiwat) ||
1655 		    clen > so->so_snd.sb_hiwat) {
1656 			SOCKBUF_UNLOCK(&so->so_snd);
1657 			error = EMSGSIZE;
1658 			goto release;
1659 		}
1660 		if (space < resid + clen &&
1661 		    (atomic || space < so->so_snd.sb_lowat || space < clen)) {
1662 			if ((so->so_state & SS_NBIO) ||
1663 			    (flags & (MSG_NBIO | MSG_DONTWAIT)) != 0) {
1664 				SOCKBUF_UNLOCK(&so->so_snd);
1665 				error = EWOULDBLOCK;
1666 				goto release;
1667 			}
1668 			error = sbwait(&so->so_snd);
1669 			SOCKBUF_UNLOCK(&so->so_snd);
1670 			if (error)
1671 				goto release;
1672 			goto restart;
1673 		}
1674 		SOCKBUF_UNLOCK(&so->so_snd);
1675 		space -= clen;
1676 		do {
1677 			if (uio == NULL) {
1678 				resid = 0;
1679 				if (flags & MSG_EOR)
1680 					top->m_flags |= M_EOR;
1681 #ifdef KERN_TLS
1682 				if (tls != NULL) {
1683 					ktls_frame(top, tls, &tls_enq_cnt,
1684 					    tls_rtype);
1685 					tls_rtype = TLS_RLTYPE_APP;
1686 				}
1687 #endif
1688 			} else {
1689 				/*
1690 				 * Copy the data from userland into a mbuf
1691 				 * chain.  If resid is 0, which can happen
1692 				 * only if we have control to send, then
1693 				 * a single empty mbuf is returned.  This
1694 				 * is a workaround to prevent protocol send
1695 				 * methods to panic.
1696 				 */
1697 #ifdef KERN_TLS
1698 				if (tls != NULL) {
1699 					top = m_uiotombuf(uio, M_WAITOK, space,
1700 					    tls->params.max_frame_len,
1701 					    M_EXTPG |
1702 					    ((flags & MSG_EOR) ? M_EOR : 0));
1703 					if (top != NULL) {
1704 						ktls_frame(top, tls,
1705 						    &tls_enq_cnt, tls_rtype);
1706 					}
1707 					tls_rtype = TLS_RLTYPE_APP;
1708 				} else
1709 #endif
1710 					top = m_uiotombuf(uio, M_WAITOK, space,
1711 					    (atomic ? max_hdr : 0),
1712 					    (atomic ? M_PKTHDR : 0) |
1713 					    ((flags & MSG_EOR) ? M_EOR : 0));
1714 				if (top == NULL) {
1715 					error = EFAULT; /* only possible error */
1716 					goto release;
1717 				}
1718 				space -= resid - uio->uio_resid;
1719 				resid = uio->uio_resid;
1720 			}
1721 			if (dontroute) {
1722 				SOCK_LOCK(so);
1723 				so->so_options |= SO_DONTROUTE;
1724 				SOCK_UNLOCK(so);
1725 			}
1726 			/*
1727 			 * XXX all the SBS_CANTSENDMORE checks previously
1728 			 * done could be out of date.  We could have received
1729 			 * a reset packet in an interrupt or maybe we slept
1730 			 * while doing page faults in uiomove() etc.  We
1731 			 * could probably recheck again inside the locking
1732 			 * protection here, but there are probably other
1733 			 * places that this also happens.  We must rethink
1734 			 * this.
1735 			 */
1736 			VNET_SO_ASSERT(so);
1737 
1738 			pru_flag = (flags & MSG_OOB) ? PRUS_OOB :
1739 			/*
1740 			 * If the user set MSG_EOF, the protocol understands
1741 			 * this flag and nothing left to send then use
1742 			 * PRU_SEND_EOF instead of PRU_SEND.
1743 			 */
1744 			    ((flags & MSG_EOF) &&
1745 			     (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1746 			     (resid <= 0)) ?
1747 				PRUS_EOF :
1748 			/* If there is more to send set PRUS_MORETOCOME. */
1749 			    (flags & MSG_MORETOCOME) ||
1750 			    (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0;
1751 
1752 #ifdef KERN_TLS
1753 			pru_flag |= tls_pruflag;
1754 #endif
1755 
1756 			error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1757 			    pru_flag, top, addr, control, td);
1758 
1759 			if (dontroute) {
1760 				SOCK_LOCK(so);
1761 				so->so_options &= ~SO_DONTROUTE;
1762 				SOCK_UNLOCK(so);
1763 			}
1764 
1765 #ifdef KERN_TLS
1766 			if (tls != NULL && tls->mode == TCP_TLS_MODE_SW) {
1767 				/*
1768 				 * Note that error is intentionally
1769 				 * ignored.
1770 				 *
1771 				 * Like sendfile(), we rely on the
1772 				 * completion routine (pru_ready())
1773 				 * to free the mbufs in the event that
1774 				 * pru_send() encountered an error and
1775 				 * did not append them to the sockbuf.
1776 				 */
1777 				soref(so);
1778 				ktls_enqueue(top, so, tls_enq_cnt);
1779 			}
1780 #endif
1781 			clen = 0;
1782 			control = NULL;
1783 			top = NULL;
1784 			if (error)
1785 				goto release;
1786 		} while (resid && space > 0);
1787 	} while (resid);
1788 
1789 release:
1790 	sbunlock(&so->so_snd);
1791 out:
1792 #ifdef KERN_TLS
1793 	if (tls != NULL)
1794 		ktls_free(tls);
1795 #endif
1796 	if (top != NULL)
1797 		m_freem(top);
1798 	if (control != NULL)
1799 		m_freem(control);
1800 	return (error);
1801 }
1802 
1803 int
1804 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
1805     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1806 {
1807 	int error;
1808 
1809 	CURVNET_SET(so->so_vnet);
1810 	if (!SOLISTENING(so))
1811 		error = so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio,
1812 		    top, control, flags, td);
1813 	else {
1814 		m_freem(top);
1815 		m_freem(control);
1816 		error = ENOTCONN;
1817 	}
1818 	CURVNET_RESTORE();
1819 	return (error);
1820 }
1821 
1822 /*
1823  * The part of soreceive() that implements reading non-inline out-of-band
1824  * data from a socket.  For more complete comments, see soreceive(), from
1825  * which this code originated.
1826  *
1827  * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
1828  * unable to return an mbuf chain to the caller.
1829  */
1830 static int
1831 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags)
1832 {
1833 	struct protosw *pr = so->so_proto;
1834 	struct mbuf *m;
1835 	int error;
1836 
1837 	KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
1838 	VNET_SO_ASSERT(so);
1839 
1840 	m = m_get(M_WAITOK, MT_DATA);
1841 	error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK);
1842 	if (error)
1843 		goto bad;
1844 	do {
1845 		error = uiomove(mtod(m, void *),
1846 		    (int) min(uio->uio_resid, m->m_len), uio);
1847 		m = m_free(m);
1848 	} while (uio->uio_resid && error == 0 && m);
1849 bad:
1850 	if (m != NULL)
1851 		m_freem(m);
1852 	return (error);
1853 }
1854 
1855 /*
1856  * Following replacement or removal of the first mbuf on the first mbuf chain
1857  * of a socket buffer, push necessary state changes back into the socket
1858  * buffer so that other consumers see the values consistently.  'nextrecord'
1859  * is the callers locally stored value of the original value of
1860  * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes.
1861  * NOTE: 'nextrecord' may be NULL.
1862  */
1863 static __inline void
1864 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord)
1865 {
1866 
1867 	SOCKBUF_LOCK_ASSERT(sb);
1868 	/*
1869 	 * First, update for the new value of nextrecord.  If necessary, make
1870 	 * it the first record.
1871 	 */
1872 	if (sb->sb_mb != NULL)
1873 		sb->sb_mb->m_nextpkt = nextrecord;
1874 	else
1875 		sb->sb_mb = nextrecord;
1876 
1877 	/*
1878 	 * Now update any dependent socket buffer fields to reflect the new
1879 	 * state.  This is an expanded inline of SB_EMPTY_FIXUP(), with the
1880 	 * addition of a second clause that takes care of the case where
1881 	 * sb_mb has been updated, but remains the last record.
1882 	 */
1883 	if (sb->sb_mb == NULL) {
1884 		sb->sb_mbtail = NULL;
1885 		sb->sb_lastrecord = NULL;
1886 	} else if (sb->sb_mb->m_nextpkt == NULL)
1887 		sb->sb_lastrecord = sb->sb_mb;
1888 }
1889 
1890 /*
1891  * Implement receive operations on a socket.  We depend on the way that
1892  * records are added to the sockbuf by sbappend.  In particular, each record
1893  * (mbufs linked through m_next) must begin with an address if the protocol
1894  * so specifies, followed by an optional mbuf or mbufs containing ancillary
1895  * data, and then zero or more mbufs of data.  In order to allow parallelism
1896  * between network receive and copying to user space, as well as avoid
1897  * sleeping with a mutex held, we release the socket buffer mutex during the
1898  * user space copy.  Although the sockbuf is locked, new data may still be
1899  * appended, and thus we must maintain consistency of the sockbuf during that
1900  * time.
1901  *
1902  * The caller may receive the data as a single mbuf chain by supplying an
1903  * mbuf **mp0 for use in returning the chain.  The uio is then used only for
1904  * the count in uio_resid.
1905  */
1906 int
1907 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio,
1908     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1909 {
1910 	struct mbuf *m, **mp;
1911 	int flags, error, offset;
1912 	ssize_t len;
1913 	struct protosw *pr = so->so_proto;
1914 	struct mbuf *nextrecord;
1915 	int moff, type = 0;
1916 	ssize_t orig_resid = uio->uio_resid;
1917 
1918 	mp = mp0;
1919 	if (psa != NULL)
1920 		*psa = NULL;
1921 	if (controlp != NULL)
1922 		*controlp = NULL;
1923 	if (flagsp != NULL)
1924 		flags = *flagsp &~ MSG_EOR;
1925 	else
1926 		flags = 0;
1927 	if (flags & MSG_OOB)
1928 		return (soreceive_rcvoob(so, uio, flags));
1929 	if (mp != NULL)
1930 		*mp = NULL;
1931 	if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING)
1932 	    && uio->uio_resid) {
1933 		VNET_SO_ASSERT(so);
1934 		(*pr->pr_usrreqs->pru_rcvd)(so, 0);
1935 	}
1936 
1937 	error = sblock(&so->so_rcv, SBLOCKWAIT(flags));
1938 	if (error)
1939 		return (error);
1940 
1941 restart:
1942 	SOCKBUF_LOCK(&so->so_rcv);
1943 	m = so->so_rcv.sb_mb;
1944 	/*
1945 	 * If we have less data than requested, block awaiting more (subject
1946 	 * to any timeout) if:
1947 	 *   1. the current count is less than the low water mark, or
1948 	 *   2. MSG_DONTWAIT is not set
1949 	 */
1950 	if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1951 	    sbavail(&so->so_rcv) < uio->uio_resid) &&
1952 	    sbavail(&so->so_rcv) < so->so_rcv.sb_lowat &&
1953 	    m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
1954 		KASSERT(m != NULL || !sbavail(&so->so_rcv),
1955 		    ("receive: m == %p sbavail == %u",
1956 		    m, sbavail(&so->so_rcv)));
1957 		if (so->so_error) {
1958 			if (m != NULL)
1959 				goto dontblock;
1960 			error = so->so_error;
1961 			if ((flags & MSG_PEEK) == 0)
1962 				so->so_error = 0;
1963 			SOCKBUF_UNLOCK(&so->so_rcv);
1964 			goto release;
1965 		}
1966 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1967 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1968 			if (m != NULL)
1969 				goto dontblock;
1970 #ifdef KERN_TLS
1971 			else if (so->so_rcv.sb_tlsdcc == 0 &&
1972 			    so->so_rcv.sb_tlscc == 0) {
1973 #else
1974 			else {
1975 #endif
1976 				SOCKBUF_UNLOCK(&so->so_rcv);
1977 				goto release;
1978 			}
1979 		}
1980 		for (; m != NULL; m = m->m_next)
1981 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
1982 				m = so->so_rcv.sb_mb;
1983 				goto dontblock;
1984 			}
1985 		if ((so->so_state & (SS_ISCONNECTING | SS_ISCONNECTED |
1986 		    SS_ISDISCONNECTING | SS_ISDISCONNECTED)) == 0 &&
1987 		    (so->so_proto->pr_flags & PR_CONNREQUIRED) != 0) {
1988 			SOCKBUF_UNLOCK(&so->so_rcv);
1989 			error = ENOTCONN;
1990 			goto release;
1991 		}
1992 		if (uio->uio_resid == 0) {
1993 			SOCKBUF_UNLOCK(&so->so_rcv);
1994 			goto release;
1995 		}
1996 		if ((so->so_state & SS_NBIO) ||
1997 		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
1998 			SOCKBUF_UNLOCK(&so->so_rcv);
1999 			error = EWOULDBLOCK;
2000 			goto release;
2001 		}
2002 		SBLASTRECORDCHK(&so->so_rcv);
2003 		SBLASTMBUFCHK(&so->so_rcv);
2004 		error = sbwait(&so->so_rcv);
2005 		SOCKBUF_UNLOCK(&so->so_rcv);
2006 		if (error)
2007 			goto release;
2008 		goto restart;
2009 	}
2010 dontblock:
2011 	/*
2012 	 * From this point onward, we maintain 'nextrecord' as a cache of the
2013 	 * pointer to the next record in the socket buffer.  We must keep the
2014 	 * various socket buffer pointers and local stack versions of the
2015 	 * pointers in sync, pushing out modifications before dropping the
2016 	 * socket buffer mutex, and re-reading them when picking it up.
2017 	 *
2018 	 * Otherwise, we will race with the network stack appending new data
2019 	 * or records onto the socket buffer by using inconsistent/stale
2020 	 * versions of the field, possibly resulting in socket buffer
2021 	 * corruption.
2022 	 *
2023 	 * By holding the high-level sblock(), we prevent simultaneous
2024 	 * readers from pulling off the front of the socket buffer.
2025 	 */
2026 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2027 	if (uio->uio_td)
2028 		uio->uio_td->td_ru.ru_msgrcv++;
2029 	KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb"));
2030 	SBLASTRECORDCHK(&so->so_rcv);
2031 	SBLASTMBUFCHK(&so->so_rcv);
2032 	nextrecord = m->m_nextpkt;
2033 	if (pr->pr_flags & PR_ADDR) {
2034 		KASSERT(m->m_type == MT_SONAME,
2035 		    ("m->m_type == %d", m->m_type));
2036 		orig_resid = 0;
2037 		if (psa != NULL)
2038 			*psa = sodupsockaddr(mtod(m, struct sockaddr *),
2039 			    M_NOWAIT);
2040 		if (flags & MSG_PEEK) {
2041 			m = m->m_next;
2042 		} else {
2043 			sbfree(&so->so_rcv, m);
2044 			so->so_rcv.sb_mb = m_free(m);
2045 			m = so->so_rcv.sb_mb;
2046 			sockbuf_pushsync(&so->so_rcv, nextrecord);
2047 		}
2048 	}
2049 
2050 	/*
2051 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
2052 	 * in the first mbuf chain on the socket buffer.  If MSG_PEEK, we
2053 	 * just copy the data; if !MSG_PEEK, we call into the protocol to
2054 	 * perform externalization (or freeing if controlp == NULL).
2055 	 */
2056 	if (m != NULL && m->m_type == MT_CONTROL) {
2057 		struct mbuf *cm = NULL, *cmn;
2058 		struct mbuf **cme = &cm;
2059 
2060 		do {
2061 			if (flags & MSG_PEEK) {
2062 				if (controlp != NULL) {
2063 					*controlp = m_copym(m, 0, m->m_len,
2064 					    M_NOWAIT);
2065 					controlp = &(*controlp)->m_next;
2066 				}
2067 				m = m->m_next;
2068 			} else {
2069 				sbfree(&so->so_rcv, m);
2070 				so->so_rcv.sb_mb = m->m_next;
2071 				m->m_next = NULL;
2072 				*cme = m;
2073 				cme = &(*cme)->m_next;
2074 				m = so->so_rcv.sb_mb;
2075 			}
2076 		} while (m != NULL && m->m_type == MT_CONTROL);
2077 		if ((flags & MSG_PEEK) == 0)
2078 			sockbuf_pushsync(&so->so_rcv, nextrecord);
2079 		while (cm != NULL) {
2080 			cmn = cm->m_next;
2081 			cm->m_next = NULL;
2082 			if (pr->pr_domain->dom_externalize != NULL) {
2083 				SOCKBUF_UNLOCK(&so->so_rcv);
2084 				VNET_SO_ASSERT(so);
2085 				error = (*pr->pr_domain->dom_externalize)
2086 				    (cm, controlp, flags);
2087 				SOCKBUF_LOCK(&so->so_rcv);
2088 			} else if (controlp != NULL)
2089 				*controlp = cm;
2090 			else
2091 				m_freem(cm);
2092 			if (controlp != NULL) {
2093 				orig_resid = 0;
2094 				while (*controlp != NULL)
2095 					controlp = &(*controlp)->m_next;
2096 			}
2097 			cm = cmn;
2098 		}
2099 		if (m != NULL)
2100 			nextrecord = so->so_rcv.sb_mb->m_nextpkt;
2101 		else
2102 			nextrecord = so->so_rcv.sb_mb;
2103 		orig_resid = 0;
2104 	}
2105 	if (m != NULL) {
2106 		if ((flags & MSG_PEEK) == 0) {
2107 			KASSERT(m->m_nextpkt == nextrecord,
2108 			    ("soreceive: post-control, nextrecord !sync"));
2109 			if (nextrecord == NULL) {
2110 				KASSERT(so->so_rcv.sb_mb == m,
2111 				    ("soreceive: post-control, sb_mb!=m"));
2112 				KASSERT(so->so_rcv.sb_lastrecord == m,
2113 				    ("soreceive: post-control, lastrecord!=m"));
2114 			}
2115 		}
2116 		type = m->m_type;
2117 		if (type == MT_OOBDATA)
2118 			flags |= MSG_OOB;
2119 	} else {
2120 		if ((flags & MSG_PEEK) == 0) {
2121 			KASSERT(so->so_rcv.sb_mb == nextrecord,
2122 			    ("soreceive: sb_mb != nextrecord"));
2123 			if (so->so_rcv.sb_mb == NULL) {
2124 				KASSERT(so->so_rcv.sb_lastrecord == NULL,
2125 				    ("soreceive: sb_lastercord != NULL"));
2126 			}
2127 		}
2128 	}
2129 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2130 	SBLASTRECORDCHK(&so->so_rcv);
2131 	SBLASTMBUFCHK(&so->so_rcv);
2132 
2133 	/*
2134 	 * Now continue to read any data mbufs off of the head of the socket
2135 	 * buffer until the read request is satisfied.  Note that 'type' is
2136 	 * used to store the type of any mbuf reads that have happened so far
2137 	 * such that soreceive() can stop reading if the type changes, which
2138 	 * causes soreceive() to return only one of regular data and inline
2139 	 * out-of-band data in a single socket receive operation.
2140 	 */
2141 	moff = 0;
2142 	offset = 0;
2143 	while (m != NULL && !(m->m_flags & M_NOTAVAIL) && uio->uio_resid > 0
2144 	    && error == 0) {
2145 		/*
2146 		 * If the type of mbuf has changed since the last mbuf
2147 		 * examined ('type'), end the receive operation.
2148 		 */
2149 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2150 		if (m->m_type == MT_OOBDATA || m->m_type == MT_CONTROL) {
2151 			if (type != m->m_type)
2152 				break;
2153 		} else if (type == MT_OOBDATA)
2154 			break;
2155 		else
2156 		    KASSERT(m->m_type == MT_DATA,
2157 			("m->m_type == %d", m->m_type));
2158 		so->so_rcv.sb_state &= ~SBS_RCVATMARK;
2159 		len = uio->uio_resid;
2160 		if (so->so_oobmark && len > so->so_oobmark - offset)
2161 			len = so->so_oobmark - offset;
2162 		if (len > m->m_len - moff)
2163 			len = m->m_len - moff;
2164 		/*
2165 		 * If mp is set, just pass back the mbufs.  Otherwise copy
2166 		 * them out via the uio, then free.  Sockbuf must be
2167 		 * consistent here (points to current mbuf, it points to next
2168 		 * record) when we drop priority; we must note any additions
2169 		 * to the sockbuf when we block interrupts again.
2170 		 */
2171 		if (mp == NULL) {
2172 			SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2173 			SBLASTRECORDCHK(&so->so_rcv);
2174 			SBLASTMBUFCHK(&so->so_rcv);
2175 			SOCKBUF_UNLOCK(&so->so_rcv);
2176 			if ((m->m_flags & M_EXTPG) != 0)
2177 				error = m_unmappedtouio(m, moff, uio, (int)len);
2178 			else
2179 				error = uiomove(mtod(m, char *) + moff,
2180 				    (int)len, uio);
2181 			SOCKBUF_LOCK(&so->so_rcv);
2182 			if (error) {
2183 				/*
2184 				 * The MT_SONAME mbuf has already been removed
2185 				 * from the record, so it is necessary to
2186 				 * remove the data mbufs, if any, to preserve
2187 				 * the invariant in the case of PR_ADDR that
2188 				 * requires MT_SONAME mbufs at the head of
2189 				 * each record.
2190 				 */
2191 				if (pr->pr_flags & PR_ATOMIC &&
2192 				    ((flags & MSG_PEEK) == 0))
2193 					(void)sbdroprecord_locked(&so->so_rcv);
2194 				SOCKBUF_UNLOCK(&so->so_rcv);
2195 				goto release;
2196 			}
2197 		} else
2198 			uio->uio_resid -= len;
2199 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2200 		if (len == m->m_len - moff) {
2201 			if (m->m_flags & M_EOR)
2202 				flags |= MSG_EOR;
2203 			if (flags & MSG_PEEK) {
2204 				m = m->m_next;
2205 				moff = 0;
2206 			} else {
2207 				nextrecord = m->m_nextpkt;
2208 				sbfree(&so->so_rcv, m);
2209 				if (mp != NULL) {
2210 					m->m_nextpkt = NULL;
2211 					*mp = m;
2212 					mp = &m->m_next;
2213 					so->so_rcv.sb_mb = m = m->m_next;
2214 					*mp = NULL;
2215 				} else {
2216 					so->so_rcv.sb_mb = m_free(m);
2217 					m = so->so_rcv.sb_mb;
2218 				}
2219 				sockbuf_pushsync(&so->so_rcv, nextrecord);
2220 				SBLASTRECORDCHK(&so->so_rcv);
2221 				SBLASTMBUFCHK(&so->so_rcv);
2222 			}
2223 		} else {
2224 			if (flags & MSG_PEEK)
2225 				moff += len;
2226 			else {
2227 				if (mp != NULL) {
2228 					if (flags & MSG_DONTWAIT) {
2229 						*mp = m_copym(m, 0, len,
2230 						    M_NOWAIT);
2231 						if (*mp == NULL) {
2232 							/*
2233 							 * m_copym() couldn't
2234 							 * allocate an mbuf.
2235 							 * Adjust uio_resid back
2236 							 * (it was adjusted
2237 							 * down by len bytes,
2238 							 * which we didn't end
2239 							 * up "copying" over).
2240 							 */
2241 							uio->uio_resid += len;
2242 							break;
2243 						}
2244 					} else {
2245 						SOCKBUF_UNLOCK(&so->so_rcv);
2246 						*mp = m_copym(m, 0, len,
2247 						    M_WAITOK);
2248 						SOCKBUF_LOCK(&so->so_rcv);
2249 					}
2250 				}
2251 				sbcut_locked(&so->so_rcv, len);
2252 			}
2253 		}
2254 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2255 		if (so->so_oobmark) {
2256 			if ((flags & MSG_PEEK) == 0) {
2257 				so->so_oobmark -= len;
2258 				if (so->so_oobmark == 0) {
2259 					so->so_rcv.sb_state |= SBS_RCVATMARK;
2260 					break;
2261 				}
2262 			} else {
2263 				offset += len;
2264 				if (offset == so->so_oobmark)
2265 					break;
2266 			}
2267 		}
2268 		if (flags & MSG_EOR)
2269 			break;
2270 		/*
2271 		 * If the MSG_WAITALL flag is set (for non-atomic socket), we
2272 		 * must not quit until "uio->uio_resid == 0" or an error
2273 		 * termination.  If a signal/timeout occurs, return with a
2274 		 * short count but without error.  Keep sockbuf locked
2275 		 * against other readers.
2276 		 */
2277 		while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
2278 		    !sosendallatonce(so) && nextrecord == NULL) {
2279 			SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2280 			if (so->so_error ||
2281 			    so->so_rcv.sb_state & SBS_CANTRCVMORE)
2282 				break;
2283 			/*
2284 			 * Notify the protocol that some data has been
2285 			 * drained before blocking.
2286 			 */
2287 			if (pr->pr_flags & PR_WANTRCVD) {
2288 				SOCKBUF_UNLOCK(&so->so_rcv);
2289 				VNET_SO_ASSERT(so);
2290 				(*pr->pr_usrreqs->pru_rcvd)(so, flags);
2291 				SOCKBUF_LOCK(&so->so_rcv);
2292 			}
2293 			SBLASTRECORDCHK(&so->so_rcv);
2294 			SBLASTMBUFCHK(&so->so_rcv);
2295 			/*
2296 			 * We could receive some data while was notifying
2297 			 * the protocol. Skip blocking in this case.
2298 			 */
2299 			if (so->so_rcv.sb_mb == NULL) {
2300 				error = sbwait(&so->so_rcv);
2301 				if (error) {
2302 					SOCKBUF_UNLOCK(&so->so_rcv);
2303 					goto release;
2304 				}
2305 			}
2306 			m = so->so_rcv.sb_mb;
2307 			if (m != NULL)
2308 				nextrecord = m->m_nextpkt;
2309 		}
2310 	}
2311 
2312 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2313 	if (m != NULL && pr->pr_flags & PR_ATOMIC) {
2314 		flags |= MSG_TRUNC;
2315 		if ((flags & MSG_PEEK) == 0)
2316 			(void) sbdroprecord_locked(&so->so_rcv);
2317 	}
2318 	if ((flags & MSG_PEEK) == 0) {
2319 		if (m == NULL) {
2320 			/*
2321 			 * First part is an inline SB_EMPTY_FIXUP().  Second
2322 			 * part makes sure sb_lastrecord is up-to-date if
2323 			 * there is still data in the socket buffer.
2324 			 */
2325 			so->so_rcv.sb_mb = nextrecord;
2326 			if (so->so_rcv.sb_mb == NULL) {
2327 				so->so_rcv.sb_mbtail = NULL;
2328 				so->so_rcv.sb_lastrecord = NULL;
2329 			} else if (nextrecord->m_nextpkt == NULL)
2330 				so->so_rcv.sb_lastrecord = nextrecord;
2331 		}
2332 		SBLASTRECORDCHK(&so->so_rcv);
2333 		SBLASTMBUFCHK(&so->so_rcv);
2334 		/*
2335 		 * If soreceive() is being done from the socket callback,
2336 		 * then don't need to generate ACK to peer to update window,
2337 		 * since ACK will be generated on return to TCP.
2338 		 */
2339 		if (!(flags & MSG_SOCALLBCK) &&
2340 		    (pr->pr_flags & PR_WANTRCVD)) {
2341 			SOCKBUF_UNLOCK(&so->so_rcv);
2342 			VNET_SO_ASSERT(so);
2343 			(*pr->pr_usrreqs->pru_rcvd)(so, flags);
2344 			SOCKBUF_LOCK(&so->so_rcv);
2345 		}
2346 	}
2347 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2348 	if (orig_resid == uio->uio_resid && orig_resid &&
2349 	    (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
2350 		SOCKBUF_UNLOCK(&so->so_rcv);
2351 		goto restart;
2352 	}
2353 	SOCKBUF_UNLOCK(&so->so_rcv);
2354 
2355 	if (flagsp != NULL)
2356 		*flagsp |= flags;
2357 release:
2358 	sbunlock(&so->so_rcv);
2359 	return (error);
2360 }
2361 
2362 /*
2363  * Optimized version of soreceive() for stream (TCP) sockets.
2364  */
2365 int
2366 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio,
2367     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2368 {
2369 	int len = 0, error = 0, flags, oresid;
2370 	struct sockbuf *sb;
2371 	struct mbuf *m, *n = NULL;
2372 
2373 	/* We only do stream sockets. */
2374 	if (so->so_type != SOCK_STREAM)
2375 		return (EINVAL);
2376 	if (psa != NULL)
2377 		*psa = NULL;
2378 	if (flagsp != NULL)
2379 		flags = *flagsp &~ MSG_EOR;
2380 	else
2381 		flags = 0;
2382 	if (controlp != NULL)
2383 		*controlp = NULL;
2384 	if (flags & MSG_OOB)
2385 		return (soreceive_rcvoob(so, uio, flags));
2386 	if (mp0 != NULL)
2387 		*mp0 = NULL;
2388 
2389 	sb = &so->so_rcv;
2390 
2391 #ifdef KERN_TLS
2392 	/*
2393 	 * KTLS store TLS records as records with a control message to
2394 	 * describe the framing.
2395 	 *
2396 	 * We check once here before acquiring locks to optimize the
2397 	 * common case.
2398 	 */
2399 	if (sb->sb_tls_info != NULL)
2400 		return (soreceive_generic(so, psa, uio, mp0, controlp,
2401 		    flagsp));
2402 #endif
2403 
2404 	/* Prevent other readers from entering the socket. */
2405 	error = sblock(sb, SBLOCKWAIT(flags));
2406 	if (error)
2407 		return (error);
2408 	SOCKBUF_LOCK(sb);
2409 
2410 #ifdef KERN_TLS
2411 	if (sb->sb_tls_info != NULL) {
2412 		SOCKBUF_UNLOCK(sb);
2413 		sbunlock(sb);
2414 		return (soreceive_generic(so, psa, uio, mp0, controlp,
2415 		    flagsp));
2416 	}
2417 #endif
2418 
2419 	/* Easy one, no space to copyout anything. */
2420 	if (uio->uio_resid == 0) {
2421 		error = EINVAL;
2422 		goto out;
2423 	}
2424 	oresid = uio->uio_resid;
2425 
2426 	/* We will never ever get anything unless we are or were connected. */
2427 	if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) {
2428 		error = ENOTCONN;
2429 		goto out;
2430 	}
2431 
2432 restart:
2433 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2434 
2435 	/* Abort if socket has reported problems. */
2436 	if (so->so_error) {
2437 		if (sbavail(sb) > 0)
2438 			goto deliver;
2439 		if (oresid > uio->uio_resid)
2440 			goto out;
2441 		error = so->so_error;
2442 		if (!(flags & MSG_PEEK))
2443 			so->so_error = 0;
2444 		goto out;
2445 	}
2446 
2447 	/* Door is closed.  Deliver what is left, if any. */
2448 	if (sb->sb_state & SBS_CANTRCVMORE) {
2449 		if (sbavail(sb) > 0)
2450 			goto deliver;
2451 		else
2452 			goto out;
2453 	}
2454 
2455 	/* Socket buffer is empty and we shall not block. */
2456 	if (sbavail(sb) == 0 &&
2457 	    ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) {
2458 		error = EAGAIN;
2459 		goto out;
2460 	}
2461 
2462 	/* Socket buffer got some data that we shall deliver now. */
2463 	if (sbavail(sb) > 0 && !(flags & MSG_WAITALL) &&
2464 	    ((so->so_state & SS_NBIO) ||
2465 	     (flags & (MSG_DONTWAIT|MSG_NBIO)) ||
2466 	     sbavail(sb) >= sb->sb_lowat ||
2467 	     sbavail(sb) >= uio->uio_resid ||
2468 	     sbavail(sb) >= sb->sb_hiwat) ) {
2469 		goto deliver;
2470 	}
2471 
2472 	/* On MSG_WAITALL we must wait until all data or error arrives. */
2473 	if ((flags & MSG_WAITALL) &&
2474 	    (sbavail(sb) >= uio->uio_resid || sbavail(sb) >= sb->sb_hiwat))
2475 		goto deliver;
2476 
2477 	/*
2478 	 * Wait and block until (more) data comes in.
2479 	 * NB: Drops the sockbuf lock during wait.
2480 	 */
2481 	error = sbwait(sb);
2482 	if (error)
2483 		goto out;
2484 	goto restart;
2485 
2486 deliver:
2487 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2488 	KASSERT(sbavail(sb) > 0, ("%s: sockbuf empty", __func__));
2489 	KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__));
2490 
2491 	/* Statistics. */
2492 	if (uio->uio_td)
2493 		uio->uio_td->td_ru.ru_msgrcv++;
2494 
2495 	/* Fill uio until full or current end of socket buffer is reached. */
2496 	len = min(uio->uio_resid, sbavail(sb));
2497 	if (mp0 != NULL) {
2498 		/* Dequeue as many mbufs as possible. */
2499 		if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) {
2500 			if (*mp0 == NULL)
2501 				*mp0 = sb->sb_mb;
2502 			else
2503 				m_cat(*mp0, sb->sb_mb);
2504 			for (m = sb->sb_mb;
2505 			     m != NULL && m->m_len <= len;
2506 			     m = m->m_next) {
2507 				KASSERT(!(m->m_flags & M_NOTAVAIL),
2508 				    ("%s: m %p not available", __func__, m));
2509 				len -= m->m_len;
2510 				uio->uio_resid -= m->m_len;
2511 				sbfree(sb, m);
2512 				n = m;
2513 			}
2514 			n->m_next = NULL;
2515 			sb->sb_mb = m;
2516 			sb->sb_lastrecord = sb->sb_mb;
2517 			if (sb->sb_mb == NULL)
2518 				SB_EMPTY_FIXUP(sb);
2519 		}
2520 		/* Copy the remainder. */
2521 		if (len > 0) {
2522 			KASSERT(sb->sb_mb != NULL,
2523 			    ("%s: len > 0 && sb->sb_mb empty", __func__));
2524 
2525 			m = m_copym(sb->sb_mb, 0, len, M_NOWAIT);
2526 			if (m == NULL)
2527 				len = 0;	/* Don't flush data from sockbuf. */
2528 			else
2529 				uio->uio_resid -= len;
2530 			if (*mp0 != NULL)
2531 				m_cat(*mp0, m);
2532 			else
2533 				*mp0 = m;
2534 			if (*mp0 == NULL) {
2535 				error = ENOBUFS;
2536 				goto out;
2537 			}
2538 		}
2539 	} else {
2540 		/* NB: Must unlock socket buffer as uiomove may sleep. */
2541 		SOCKBUF_UNLOCK(sb);
2542 		error = m_mbuftouio(uio, sb->sb_mb, len);
2543 		SOCKBUF_LOCK(sb);
2544 		if (error)
2545 			goto out;
2546 	}
2547 	SBLASTRECORDCHK(sb);
2548 	SBLASTMBUFCHK(sb);
2549 
2550 	/*
2551 	 * Remove the delivered data from the socket buffer unless we
2552 	 * were only peeking.
2553 	 */
2554 	if (!(flags & MSG_PEEK)) {
2555 		if (len > 0)
2556 			sbdrop_locked(sb, len);
2557 
2558 		/* Notify protocol that we drained some data. */
2559 		if ((so->so_proto->pr_flags & PR_WANTRCVD) &&
2560 		    (((flags & MSG_WAITALL) && uio->uio_resid > 0) ||
2561 		     !(flags & MSG_SOCALLBCK))) {
2562 			SOCKBUF_UNLOCK(sb);
2563 			VNET_SO_ASSERT(so);
2564 			(*so->so_proto->pr_usrreqs->pru_rcvd)(so, flags);
2565 			SOCKBUF_LOCK(sb);
2566 		}
2567 	}
2568 
2569 	/*
2570 	 * For MSG_WAITALL we may have to loop again and wait for
2571 	 * more data to come in.
2572 	 */
2573 	if ((flags & MSG_WAITALL) && uio->uio_resid > 0)
2574 		goto restart;
2575 out:
2576 	SOCKBUF_LOCK_ASSERT(sb);
2577 	SBLASTRECORDCHK(sb);
2578 	SBLASTMBUFCHK(sb);
2579 	SOCKBUF_UNLOCK(sb);
2580 	sbunlock(sb);
2581 	return (error);
2582 }
2583 
2584 /*
2585  * Optimized version of soreceive() for simple datagram cases from userspace.
2586  * Unlike in the stream case, we're able to drop a datagram if copyout()
2587  * fails, and because we handle datagrams atomically, we don't need to use a
2588  * sleep lock to prevent I/O interlacing.
2589  */
2590 int
2591 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio,
2592     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2593 {
2594 	struct mbuf *m, *m2;
2595 	int flags, error;
2596 	ssize_t len;
2597 	struct protosw *pr = so->so_proto;
2598 	struct mbuf *nextrecord;
2599 
2600 	if (psa != NULL)
2601 		*psa = NULL;
2602 	if (controlp != NULL)
2603 		*controlp = NULL;
2604 	if (flagsp != NULL)
2605 		flags = *flagsp &~ MSG_EOR;
2606 	else
2607 		flags = 0;
2608 
2609 	/*
2610 	 * For any complicated cases, fall back to the full
2611 	 * soreceive_generic().
2612 	 */
2613 	if (mp0 != NULL || (flags & MSG_PEEK) || (flags & MSG_OOB))
2614 		return (soreceive_generic(so, psa, uio, mp0, controlp,
2615 		    flagsp));
2616 
2617 	/*
2618 	 * Enforce restrictions on use.
2619 	 */
2620 	KASSERT((pr->pr_flags & PR_WANTRCVD) == 0,
2621 	    ("soreceive_dgram: wantrcvd"));
2622 	KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic"));
2623 	KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0,
2624 	    ("soreceive_dgram: SBS_RCVATMARK"));
2625 	KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0,
2626 	    ("soreceive_dgram: P_CONNREQUIRED"));
2627 
2628 	/*
2629 	 * Loop blocking while waiting for a datagram.
2630 	 */
2631 	SOCKBUF_LOCK(&so->so_rcv);
2632 	while ((m = so->so_rcv.sb_mb) == NULL) {
2633 		KASSERT(sbavail(&so->so_rcv) == 0,
2634 		    ("soreceive_dgram: sb_mb NULL but sbavail %u",
2635 		    sbavail(&so->so_rcv)));
2636 		if (so->so_error) {
2637 			error = so->so_error;
2638 			so->so_error = 0;
2639 			SOCKBUF_UNLOCK(&so->so_rcv);
2640 			return (error);
2641 		}
2642 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE ||
2643 		    uio->uio_resid == 0) {
2644 			SOCKBUF_UNLOCK(&so->so_rcv);
2645 			return (0);
2646 		}
2647 		if ((so->so_state & SS_NBIO) ||
2648 		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2649 			SOCKBUF_UNLOCK(&so->so_rcv);
2650 			return (EWOULDBLOCK);
2651 		}
2652 		SBLASTRECORDCHK(&so->so_rcv);
2653 		SBLASTMBUFCHK(&so->so_rcv);
2654 		error = sbwait(&so->so_rcv);
2655 		if (error) {
2656 			SOCKBUF_UNLOCK(&so->so_rcv);
2657 			return (error);
2658 		}
2659 	}
2660 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2661 
2662 	if (uio->uio_td)
2663 		uio->uio_td->td_ru.ru_msgrcv++;
2664 	SBLASTRECORDCHK(&so->so_rcv);
2665 	SBLASTMBUFCHK(&so->so_rcv);
2666 	nextrecord = m->m_nextpkt;
2667 	if (nextrecord == NULL) {
2668 		KASSERT(so->so_rcv.sb_lastrecord == m,
2669 		    ("soreceive_dgram: lastrecord != m"));
2670 	}
2671 
2672 	KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord,
2673 	    ("soreceive_dgram: m_nextpkt != nextrecord"));
2674 
2675 	/*
2676 	 * Pull 'm' and its chain off the front of the packet queue.
2677 	 */
2678 	so->so_rcv.sb_mb = NULL;
2679 	sockbuf_pushsync(&so->so_rcv, nextrecord);
2680 
2681 	/*
2682 	 * Walk 'm's chain and free that many bytes from the socket buffer.
2683 	 */
2684 	for (m2 = m; m2 != NULL; m2 = m2->m_next)
2685 		sbfree(&so->so_rcv, m2);
2686 
2687 	/*
2688 	 * Do a few last checks before we let go of the lock.
2689 	 */
2690 	SBLASTRECORDCHK(&so->so_rcv);
2691 	SBLASTMBUFCHK(&so->so_rcv);
2692 	SOCKBUF_UNLOCK(&so->so_rcv);
2693 
2694 	if (pr->pr_flags & PR_ADDR) {
2695 		KASSERT(m->m_type == MT_SONAME,
2696 		    ("m->m_type == %d", m->m_type));
2697 		if (psa != NULL)
2698 			*psa = sodupsockaddr(mtod(m, struct sockaddr *),
2699 			    M_NOWAIT);
2700 		m = m_free(m);
2701 	}
2702 	if (m == NULL) {
2703 		/* XXXRW: Can this happen? */
2704 		return (0);
2705 	}
2706 
2707 	/*
2708 	 * Packet to copyout() is now in 'm' and it is disconnected from the
2709 	 * queue.
2710 	 *
2711 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
2712 	 * in the first mbuf chain on the socket buffer.  We call into the
2713 	 * protocol to perform externalization (or freeing if controlp ==
2714 	 * NULL). In some cases there can be only MT_CONTROL mbufs without
2715 	 * MT_DATA mbufs.
2716 	 */
2717 	if (m->m_type == MT_CONTROL) {
2718 		struct mbuf *cm = NULL, *cmn;
2719 		struct mbuf **cme = &cm;
2720 
2721 		do {
2722 			m2 = m->m_next;
2723 			m->m_next = NULL;
2724 			*cme = m;
2725 			cme = &(*cme)->m_next;
2726 			m = m2;
2727 		} while (m != NULL && m->m_type == MT_CONTROL);
2728 		while (cm != NULL) {
2729 			cmn = cm->m_next;
2730 			cm->m_next = NULL;
2731 			if (pr->pr_domain->dom_externalize != NULL) {
2732 				error = (*pr->pr_domain->dom_externalize)
2733 				    (cm, controlp, flags);
2734 			} else if (controlp != NULL)
2735 				*controlp = cm;
2736 			else
2737 				m_freem(cm);
2738 			if (controlp != NULL) {
2739 				while (*controlp != NULL)
2740 					controlp = &(*controlp)->m_next;
2741 			}
2742 			cm = cmn;
2743 		}
2744 	}
2745 	KASSERT(m == NULL || m->m_type == MT_DATA,
2746 	    ("soreceive_dgram: !data"));
2747 	while (m != NULL && uio->uio_resid > 0) {
2748 		len = uio->uio_resid;
2749 		if (len > m->m_len)
2750 			len = m->m_len;
2751 		error = uiomove(mtod(m, char *), (int)len, uio);
2752 		if (error) {
2753 			m_freem(m);
2754 			return (error);
2755 		}
2756 		if (len == m->m_len)
2757 			m = m_free(m);
2758 		else {
2759 			m->m_data += len;
2760 			m->m_len -= len;
2761 		}
2762 	}
2763 	if (m != NULL) {
2764 		flags |= MSG_TRUNC;
2765 		m_freem(m);
2766 	}
2767 	if (flagsp != NULL)
2768 		*flagsp |= flags;
2769 	return (0);
2770 }
2771 
2772 int
2773 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
2774     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2775 {
2776 	int error;
2777 
2778 	CURVNET_SET(so->so_vnet);
2779 	if (!SOLISTENING(so))
2780 		error = (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio,
2781 		    mp0, controlp, flagsp));
2782 	else
2783 		error = ENOTCONN;
2784 	CURVNET_RESTORE();
2785 	return (error);
2786 }
2787 
2788 int
2789 soshutdown(struct socket *so, int how)
2790 {
2791 	struct protosw *pr = so->so_proto;
2792 	int error, soerror_enotconn;
2793 
2794 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
2795 		return (EINVAL);
2796 
2797 	soerror_enotconn = 0;
2798 	if ((so->so_state &
2799 	    (SS_ISCONNECTED | SS_ISCONNECTING | SS_ISDISCONNECTING)) == 0) {
2800 		/*
2801 		 * POSIX mandates us to return ENOTCONN when shutdown(2) is
2802 		 * invoked on a datagram sockets, however historically we would
2803 		 * actually tear socket down. This is known to be leveraged by
2804 		 * some applications to unblock process waiting in recvXXX(2)
2805 		 * by other process that it shares that socket with. Try to meet
2806 		 * both backward-compatibility and POSIX requirements by forcing
2807 		 * ENOTCONN but still asking protocol to perform pru_shutdown().
2808 		 */
2809 		if (so->so_type != SOCK_DGRAM && !SOLISTENING(so))
2810 			return (ENOTCONN);
2811 		soerror_enotconn = 1;
2812 	}
2813 
2814 	if (SOLISTENING(so)) {
2815 		if (how != SHUT_WR) {
2816 			SOLISTEN_LOCK(so);
2817 			so->so_error = ECONNABORTED;
2818 			solisten_wakeup(so);	/* unlocks so */
2819 		}
2820 		goto done;
2821 	}
2822 
2823 	CURVNET_SET(so->so_vnet);
2824 	if (pr->pr_usrreqs->pru_flush != NULL)
2825 		(*pr->pr_usrreqs->pru_flush)(so, how);
2826 	if (how != SHUT_WR)
2827 		sorflush(so);
2828 	if (how != SHUT_RD) {
2829 		error = (*pr->pr_usrreqs->pru_shutdown)(so);
2830 		wakeup(&so->so_timeo);
2831 		CURVNET_RESTORE();
2832 		return ((error == 0 && soerror_enotconn) ? ENOTCONN : error);
2833 	}
2834 	wakeup(&so->so_timeo);
2835 	CURVNET_RESTORE();
2836 
2837 done:
2838 	return (soerror_enotconn ? ENOTCONN : 0);
2839 }
2840 
2841 void
2842 sorflush(struct socket *so)
2843 {
2844 	struct sockbuf *sb = &so->so_rcv;
2845 	struct protosw *pr = so->so_proto;
2846 	struct socket aso;
2847 
2848 	VNET_SO_ASSERT(so);
2849 
2850 	/*
2851 	 * In order to avoid calling dom_dispose with the socket buffer mutex
2852 	 * held, and in order to generally avoid holding the lock for a long
2853 	 * time, we make a copy of the socket buffer and clear the original
2854 	 * (except locks, state).  The new socket buffer copy won't have
2855 	 * initialized locks so we can only call routines that won't use or
2856 	 * assert those locks.
2857 	 *
2858 	 * Dislodge threads currently blocked in receive and wait to acquire
2859 	 * a lock against other simultaneous readers before clearing the
2860 	 * socket buffer.  Don't let our acquire be interrupted by a signal
2861 	 * despite any existing socket disposition on interruptable waiting.
2862 	 */
2863 	socantrcvmore(so);
2864 	(void) sblock(sb, SBL_WAIT | SBL_NOINTR);
2865 
2866 	/*
2867 	 * Invalidate/clear most of the sockbuf structure, but leave selinfo
2868 	 * and mutex data unchanged.
2869 	 */
2870 	SOCKBUF_LOCK(sb);
2871 	bzero(&aso, sizeof(aso));
2872 	aso.so_pcb = so->so_pcb;
2873 	bcopy(&sb->sb_startzero, &aso.so_rcv.sb_startzero,
2874 	    sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2875 	bzero(&sb->sb_startzero,
2876 	    sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2877 	SOCKBUF_UNLOCK(sb);
2878 	sbunlock(sb);
2879 
2880 	/*
2881 	 * Dispose of special rights and flush the copied socket.  Don't call
2882 	 * any unsafe routines (that rely on locks being initialized) on aso.
2883 	 */
2884 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
2885 		(*pr->pr_domain->dom_dispose)(&aso);
2886 	sbrelease_internal(&aso.so_rcv, so);
2887 }
2888 
2889 /*
2890  * Wrapper for Socket established helper hook.
2891  * Parameters: socket, context of the hook point, hook id.
2892  */
2893 static int inline
2894 hhook_run_socket(struct socket *so, void *hctx, int32_t h_id)
2895 {
2896 	struct socket_hhook_data hhook_data = {
2897 		.so = so,
2898 		.hctx = hctx,
2899 		.m = NULL,
2900 		.status = 0
2901 	};
2902 
2903 	CURVNET_SET(so->so_vnet);
2904 	HHOOKS_RUN_IF(V_socket_hhh[h_id], &hhook_data, &so->osd);
2905 	CURVNET_RESTORE();
2906 
2907 	/* Ugly but needed, since hhooks return void for now */
2908 	return (hhook_data.status);
2909 }
2910 
2911 /*
2912  * Perhaps this routine, and sooptcopyout(), below, ought to come in an
2913  * additional variant to handle the case where the option value needs to be
2914  * some kind of integer, but not a specific size.  In addition to their use
2915  * here, these functions are also called by the protocol-level pr_ctloutput()
2916  * routines.
2917  */
2918 int
2919 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2920 {
2921 	size_t	valsize;
2922 
2923 	/*
2924 	 * If the user gives us more than we wanted, we ignore it, but if we
2925 	 * don't get the minimum length the caller wants, we return EINVAL.
2926 	 * On success, sopt->sopt_valsize is set to however much we actually
2927 	 * retrieved.
2928 	 */
2929 	if ((valsize = sopt->sopt_valsize) < minlen)
2930 		return EINVAL;
2931 	if (valsize > len)
2932 		sopt->sopt_valsize = valsize = len;
2933 
2934 	if (sopt->sopt_td != NULL)
2935 		return (copyin(sopt->sopt_val, buf, valsize));
2936 
2937 	bcopy(sopt->sopt_val, buf, valsize);
2938 	return (0);
2939 }
2940 
2941 /*
2942  * Kernel version of setsockopt(2).
2943  *
2944  * XXX: optlen is size_t, not socklen_t
2945  */
2946 int
2947 so_setsockopt(struct socket *so, int level, int optname, void *optval,
2948     size_t optlen)
2949 {
2950 	struct sockopt sopt;
2951 
2952 	sopt.sopt_level = level;
2953 	sopt.sopt_name = optname;
2954 	sopt.sopt_dir = SOPT_SET;
2955 	sopt.sopt_val = optval;
2956 	sopt.sopt_valsize = optlen;
2957 	sopt.sopt_td = NULL;
2958 	return (sosetopt(so, &sopt));
2959 }
2960 
2961 int
2962 sosetopt(struct socket *so, struct sockopt *sopt)
2963 {
2964 	int	error, optval;
2965 	struct	linger l;
2966 	struct	timeval tv;
2967 	sbintime_t val;
2968 	uint32_t val32;
2969 #ifdef MAC
2970 	struct mac extmac;
2971 #endif
2972 
2973 	CURVNET_SET(so->so_vnet);
2974 	error = 0;
2975 	if (sopt->sopt_level != SOL_SOCKET) {
2976 		if (so->so_proto->pr_ctloutput != NULL)
2977 			error = (*so->so_proto->pr_ctloutput)(so, sopt);
2978 		else
2979 			error = ENOPROTOOPT;
2980 	} else {
2981 		switch (sopt->sopt_name) {
2982 		case SO_ACCEPTFILTER:
2983 			error = accept_filt_setopt(so, sopt);
2984 			if (error)
2985 				goto bad;
2986 			break;
2987 
2988 		case SO_LINGER:
2989 			error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
2990 			if (error)
2991 				goto bad;
2992 			if (l.l_linger < 0 ||
2993 			    l.l_linger > USHRT_MAX ||
2994 			    l.l_linger > (INT_MAX / hz)) {
2995 				error = EDOM;
2996 				goto bad;
2997 			}
2998 			SOCK_LOCK(so);
2999 			so->so_linger = l.l_linger;
3000 			if (l.l_onoff)
3001 				so->so_options |= SO_LINGER;
3002 			else
3003 				so->so_options &= ~SO_LINGER;
3004 			SOCK_UNLOCK(so);
3005 			break;
3006 
3007 		case SO_DEBUG:
3008 		case SO_KEEPALIVE:
3009 		case SO_DONTROUTE:
3010 		case SO_USELOOPBACK:
3011 		case SO_BROADCAST:
3012 		case SO_REUSEADDR:
3013 		case SO_REUSEPORT:
3014 		case SO_REUSEPORT_LB:
3015 		case SO_OOBINLINE:
3016 		case SO_TIMESTAMP:
3017 		case SO_BINTIME:
3018 		case SO_NOSIGPIPE:
3019 		case SO_NO_DDP:
3020 		case SO_NO_OFFLOAD:
3021 			error = sooptcopyin(sopt, &optval, sizeof optval,
3022 			    sizeof optval);
3023 			if (error)
3024 				goto bad;
3025 			SOCK_LOCK(so);
3026 			if (optval)
3027 				so->so_options |= sopt->sopt_name;
3028 			else
3029 				so->so_options &= ~sopt->sopt_name;
3030 			SOCK_UNLOCK(so);
3031 			break;
3032 
3033 		case SO_SETFIB:
3034 			error = sooptcopyin(sopt, &optval, sizeof optval,
3035 			    sizeof optval);
3036 			if (error)
3037 				goto bad;
3038 
3039 			if (optval < 0 || optval >= rt_numfibs) {
3040 				error = EINVAL;
3041 				goto bad;
3042 			}
3043 			if (((so->so_proto->pr_domain->dom_family == PF_INET) ||
3044 			   (so->so_proto->pr_domain->dom_family == PF_INET6) ||
3045 			   (so->so_proto->pr_domain->dom_family == PF_ROUTE)))
3046 				so->so_fibnum = optval;
3047 			else
3048 				so->so_fibnum = 0;
3049 			break;
3050 
3051 		case SO_USER_COOKIE:
3052 			error = sooptcopyin(sopt, &val32, sizeof val32,
3053 			    sizeof val32);
3054 			if (error)
3055 				goto bad;
3056 			so->so_user_cookie = val32;
3057 			break;
3058 
3059 		case SO_SNDBUF:
3060 		case SO_RCVBUF:
3061 		case SO_SNDLOWAT:
3062 		case SO_RCVLOWAT:
3063 			error = sooptcopyin(sopt, &optval, sizeof optval,
3064 			    sizeof optval);
3065 			if (error)
3066 				goto bad;
3067 
3068 			/*
3069 			 * Values < 1 make no sense for any of these options,
3070 			 * so disallow them.
3071 			 */
3072 			if (optval < 1) {
3073 				error = EINVAL;
3074 				goto bad;
3075 			}
3076 
3077 			error = sbsetopt(so, sopt->sopt_name, optval);
3078 			break;
3079 
3080 		case SO_SNDTIMEO:
3081 		case SO_RCVTIMEO:
3082 #ifdef COMPAT_FREEBSD32
3083 			if (SV_CURPROC_FLAG(SV_ILP32)) {
3084 				struct timeval32 tv32;
3085 
3086 				error = sooptcopyin(sopt, &tv32, sizeof tv32,
3087 				    sizeof tv32);
3088 				CP(tv32, tv, tv_sec);
3089 				CP(tv32, tv, tv_usec);
3090 			} else
3091 #endif
3092 				error = sooptcopyin(sopt, &tv, sizeof tv,
3093 				    sizeof tv);
3094 			if (error)
3095 				goto bad;
3096 			if (tv.tv_sec < 0 || tv.tv_usec < 0 ||
3097 			    tv.tv_usec >= 1000000) {
3098 				error = EDOM;
3099 				goto bad;
3100 			}
3101 			if (tv.tv_sec > INT32_MAX)
3102 				val = SBT_MAX;
3103 			else
3104 				val = tvtosbt(tv);
3105 			switch (sopt->sopt_name) {
3106 			case SO_SNDTIMEO:
3107 				so->so_snd.sb_timeo = val;
3108 				break;
3109 			case SO_RCVTIMEO:
3110 				so->so_rcv.sb_timeo = val;
3111 				break;
3112 			}
3113 			break;
3114 
3115 		case SO_LABEL:
3116 #ifdef MAC
3117 			error = sooptcopyin(sopt, &extmac, sizeof extmac,
3118 			    sizeof extmac);
3119 			if (error)
3120 				goto bad;
3121 			error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
3122 			    so, &extmac);
3123 #else
3124 			error = EOPNOTSUPP;
3125 #endif
3126 			break;
3127 
3128 		case SO_TS_CLOCK:
3129 			error = sooptcopyin(sopt, &optval, sizeof optval,
3130 			    sizeof optval);
3131 			if (error)
3132 				goto bad;
3133 			if (optval < 0 || optval > SO_TS_CLOCK_MAX) {
3134 				error = EINVAL;
3135 				goto bad;
3136 			}
3137 			so->so_ts_clock = optval;
3138 			break;
3139 
3140 		case SO_MAX_PACING_RATE:
3141 			error = sooptcopyin(sopt, &val32, sizeof(val32),
3142 			    sizeof(val32));
3143 			if (error)
3144 				goto bad;
3145 			so->so_max_pacing_rate = val32;
3146 			break;
3147 
3148 		default:
3149 			if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
3150 				error = hhook_run_socket(so, sopt,
3151 				    HHOOK_SOCKET_OPT);
3152 			else
3153 				error = ENOPROTOOPT;
3154 			break;
3155 		}
3156 		if (error == 0 && so->so_proto->pr_ctloutput != NULL)
3157 			(void)(*so->so_proto->pr_ctloutput)(so, sopt);
3158 	}
3159 bad:
3160 	CURVNET_RESTORE();
3161 	return (error);
3162 }
3163 
3164 /*
3165  * Helper routine for getsockopt.
3166  */
3167 int
3168 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
3169 {
3170 	int	error;
3171 	size_t	valsize;
3172 
3173 	error = 0;
3174 
3175 	/*
3176 	 * Documented get behavior is that we always return a value, possibly
3177 	 * truncated to fit in the user's buffer.  Traditional behavior is
3178 	 * that we always tell the user precisely how much we copied, rather
3179 	 * than something useful like the total amount we had available for
3180 	 * her.  Note that this interface is not idempotent; the entire
3181 	 * answer must be generated ahead of time.
3182 	 */
3183 	valsize = min(len, sopt->sopt_valsize);
3184 	sopt->sopt_valsize = valsize;
3185 	if (sopt->sopt_val != NULL) {
3186 		if (sopt->sopt_td != NULL)
3187 			error = copyout(buf, sopt->sopt_val, valsize);
3188 		else
3189 			bcopy(buf, sopt->sopt_val, valsize);
3190 	}
3191 	return (error);
3192 }
3193 
3194 int
3195 sogetopt(struct socket *so, struct sockopt *sopt)
3196 {
3197 	int	error, optval;
3198 	struct	linger l;
3199 	struct	timeval tv;
3200 #ifdef MAC
3201 	struct mac extmac;
3202 #endif
3203 
3204 	CURVNET_SET(so->so_vnet);
3205 	error = 0;
3206 	if (sopt->sopt_level != SOL_SOCKET) {
3207 		if (so->so_proto->pr_ctloutput != NULL)
3208 			error = (*so->so_proto->pr_ctloutput)(so, sopt);
3209 		else
3210 			error = ENOPROTOOPT;
3211 		CURVNET_RESTORE();
3212 		return (error);
3213 	} else {
3214 		switch (sopt->sopt_name) {
3215 		case SO_ACCEPTFILTER:
3216 			error = accept_filt_getopt(so, sopt);
3217 			break;
3218 
3219 		case SO_LINGER:
3220 			SOCK_LOCK(so);
3221 			l.l_onoff = so->so_options & SO_LINGER;
3222 			l.l_linger = so->so_linger;
3223 			SOCK_UNLOCK(so);
3224 			error = sooptcopyout(sopt, &l, sizeof l);
3225 			break;
3226 
3227 		case SO_USELOOPBACK:
3228 		case SO_DONTROUTE:
3229 		case SO_DEBUG:
3230 		case SO_KEEPALIVE:
3231 		case SO_REUSEADDR:
3232 		case SO_REUSEPORT:
3233 		case SO_REUSEPORT_LB:
3234 		case SO_BROADCAST:
3235 		case SO_OOBINLINE:
3236 		case SO_ACCEPTCONN:
3237 		case SO_TIMESTAMP:
3238 		case SO_BINTIME:
3239 		case SO_NOSIGPIPE:
3240 		case SO_NO_DDP:
3241 		case SO_NO_OFFLOAD:
3242 			optval = so->so_options & sopt->sopt_name;
3243 integer:
3244 			error = sooptcopyout(sopt, &optval, sizeof optval);
3245 			break;
3246 
3247 		case SO_DOMAIN:
3248 			optval = so->so_proto->pr_domain->dom_family;
3249 			goto integer;
3250 
3251 		case SO_TYPE:
3252 			optval = so->so_type;
3253 			goto integer;
3254 
3255 		case SO_PROTOCOL:
3256 			optval = so->so_proto->pr_protocol;
3257 			goto integer;
3258 
3259 		case SO_ERROR:
3260 			SOCK_LOCK(so);
3261 			optval = so->so_error;
3262 			so->so_error = 0;
3263 			SOCK_UNLOCK(so);
3264 			goto integer;
3265 
3266 		case SO_SNDBUF:
3267 			optval = SOLISTENING(so) ? so->sol_sbsnd_hiwat :
3268 			    so->so_snd.sb_hiwat;
3269 			goto integer;
3270 
3271 		case SO_RCVBUF:
3272 			optval = SOLISTENING(so) ? so->sol_sbrcv_hiwat :
3273 			    so->so_rcv.sb_hiwat;
3274 			goto integer;
3275 
3276 		case SO_SNDLOWAT:
3277 			optval = SOLISTENING(so) ? so->sol_sbsnd_lowat :
3278 			    so->so_snd.sb_lowat;
3279 			goto integer;
3280 
3281 		case SO_RCVLOWAT:
3282 			optval = SOLISTENING(so) ? so->sol_sbrcv_lowat :
3283 			    so->so_rcv.sb_lowat;
3284 			goto integer;
3285 
3286 		case SO_SNDTIMEO:
3287 		case SO_RCVTIMEO:
3288 			tv = sbttotv(sopt->sopt_name == SO_SNDTIMEO ?
3289 			    so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
3290 #ifdef COMPAT_FREEBSD32
3291 			if (SV_CURPROC_FLAG(SV_ILP32)) {
3292 				struct timeval32 tv32;
3293 
3294 				CP(tv, tv32, tv_sec);
3295 				CP(tv, tv32, tv_usec);
3296 				error = sooptcopyout(sopt, &tv32, sizeof tv32);
3297 			} else
3298 #endif
3299 				error = sooptcopyout(sopt, &tv, sizeof tv);
3300 			break;
3301 
3302 		case SO_LABEL:
3303 #ifdef MAC
3304 			error = sooptcopyin(sopt, &extmac, sizeof(extmac),
3305 			    sizeof(extmac));
3306 			if (error)
3307 				goto bad;
3308 			error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
3309 			    so, &extmac);
3310 			if (error)
3311 				goto bad;
3312 			error = sooptcopyout(sopt, &extmac, sizeof extmac);
3313 #else
3314 			error = EOPNOTSUPP;
3315 #endif
3316 			break;
3317 
3318 		case SO_PEERLABEL:
3319 #ifdef MAC
3320 			error = sooptcopyin(sopt, &extmac, sizeof(extmac),
3321 			    sizeof(extmac));
3322 			if (error)
3323 				goto bad;
3324 			error = mac_getsockopt_peerlabel(
3325 			    sopt->sopt_td->td_ucred, so, &extmac);
3326 			if (error)
3327 				goto bad;
3328 			error = sooptcopyout(sopt, &extmac, sizeof extmac);
3329 #else
3330 			error = EOPNOTSUPP;
3331 #endif
3332 			break;
3333 
3334 		case SO_LISTENQLIMIT:
3335 			optval = SOLISTENING(so) ? so->sol_qlimit : 0;
3336 			goto integer;
3337 
3338 		case SO_LISTENQLEN:
3339 			optval = SOLISTENING(so) ? so->sol_qlen : 0;
3340 			goto integer;
3341 
3342 		case SO_LISTENINCQLEN:
3343 			optval = SOLISTENING(so) ? so->sol_incqlen : 0;
3344 			goto integer;
3345 
3346 		case SO_TS_CLOCK:
3347 			optval = so->so_ts_clock;
3348 			goto integer;
3349 
3350 		case SO_MAX_PACING_RATE:
3351 			optval = so->so_max_pacing_rate;
3352 			goto integer;
3353 
3354 		default:
3355 			if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
3356 				error = hhook_run_socket(so, sopt,
3357 				    HHOOK_SOCKET_OPT);
3358 			else
3359 				error = ENOPROTOOPT;
3360 			break;
3361 		}
3362 	}
3363 #ifdef MAC
3364 bad:
3365 #endif
3366 	CURVNET_RESTORE();
3367 	return (error);
3368 }
3369 
3370 int
3371 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
3372 {
3373 	struct mbuf *m, *m_prev;
3374 	int sopt_size = sopt->sopt_valsize;
3375 
3376 	MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
3377 	if (m == NULL)
3378 		return ENOBUFS;
3379 	if (sopt_size > MLEN) {
3380 		MCLGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT);
3381 		if ((m->m_flags & M_EXT) == 0) {
3382 			m_free(m);
3383 			return ENOBUFS;
3384 		}
3385 		m->m_len = min(MCLBYTES, sopt_size);
3386 	} else {
3387 		m->m_len = min(MLEN, sopt_size);
3388 	}
3389 	sopt_size -= m->m_len;
3390 	*mp = m;
3391 	m_prev = m;
3392 
3393 	while (sopt_size) {
3394 		MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
3395 		if (m == NULL) {
3396 			m_freem(*mp);
3397 			return ENOBUFS;
3398 		}
3399 		if (sopt_size > MLEN) {
3400 			MCLGET(m, sopt->sopt_td != NULL ? M_WAITOK :
3401 			    M_NOWAIT);
3402 			if ((m->m_flags & M_EXT) == 0) {
3403 				m_freem(m);
3404 				m_freem(*mp);
3405 				return ENOBUFS;
3406 			}
3407 			m->m_len = min(MCLBYTES, sopt_size);
3408 		} else {
3409 			m->m_len = min(MLEN, sopt_size);
3410 		}
3411 		sopt_size -= m->m_len;
3412 		m_prev->m_next = m;
3413 		m_prev = m;
3414 	}
3415 	return (0);
3416 }
3417 
3418 int
3419 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
3420 {
3421 	struct mbuf *m0 = m;
3422 
3423 	if (sopt->sopt_val == NULL)
3424 		return (0);
3425 	while (m != NULL && sopt->sopt_valsize >= m->m_len) {
3426 		if (sopt->sopt_td != NULL) {
3427 			int error;
3428 
3429 			error = copyin(sopt->sopt_val, mtod(m, char *),
3430 			    m->m_len);
3431 			if (error != 0) {
3432 				m_freem(m0);
3433 				return(error);
3434 			}
3435 		} else
3436 			bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
3437 		sopt->sopt_valsize -= m->m_len;
3438 		sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
3439 		m = m->m_next;
3440 	}
3441 	if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
3442 		panic("ip6_sooptmcopyin");
3443 	return (0);
3444 }
3445 
3446 int
3447 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
3448 {
3449 	struct mbuf *m0 = m;
3450 	size_t valsize = 0;
3451 
3452 	if (sopt->sopt_val == NULL)
3453 		return (0);
3454 	while (m != NULL && sopt->sopt_valsize >= m->m_len) {
3455 		if (sopt->sopt_td != NULL) {
3456 			int error;
3457 
3458 			error = copyout(mtod(m, char *), sopt->sopt_val,
3459 			    m->m_len);
3460 			if (error != 0) {
3461 				m_freem(m0);
3462 				return(error);
3463 			}
3464 		} else
3465 			bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
3466 		sopt->sopt_valsize -= m->m_len;
3467 		sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
3468 		valsize += m->m_len;
3469 		m = m->m_next;
3470 	}
3471 	if (m != NULL) {
3472 		/* enough soopt buffer should be given from user-land */
3473 		m_freem(m0);
3474 		return(EINVAL);
3475 	}
3476 	sopt->sopt_valsize = valsize;
3477 	return (0);
3478 }
3479 
3480 /*
3481  * sohasoutofband(): protocol notifies socket layer of the arrival of new
3482  * out-of-band data, which will then notify socket consumers.
3483  */
3484 void
3485 sohasoutofband(struct socket *so)
3486 {
3487 
3488 	if (so->so_sigio != NULL)
3489 		pgsigio(&so->so_sigio, SIGURG, 0);
3490 	selwakeuppri(&so->so_rdsel, PSOCK);
3491 }
3492 
3493 int
3494 sopoll(struct socket *so, int events, struct ucred *active_cred,
3495     struct thread *td)
3496 {
3497 
3498 	/*
3499 	 * We do not need to set or assert curvnet as long as everyone uses
3500 	 * sopoll_generic().
3501 	 */
3502 	return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred,
3503 	    td));
3504 }
3505 
3506 int
3507 sopoll_generic(struct socket *so, int events, struct ucred *active_cred,
3508     struct thread *td)
3509 {
3510 	int revents;
3511 
3512 	SOCK_LOCK(so);
3513 	if (SOLISTENING(so)) {
3514 		if (!(events & (POLLIN | POLLRDNORM)))
3515 			revents = 0;
3516 		else if (!TAILQ_EMPTY(&so->sol_comp))
3517 			revents = events & (POLLIN | POLLRDNORM);
3518 		else if ((events & POLLINIGNEOF) == 0 && so->so_error)
3519 			revents = (events & (POLLIN | POLLRDNORM)) | POLLHUP;
3520 		else {
3521 			selrecord(td, &so->so_rdsel);
3522 			revents = 0;
3523 		}
3524 	} else {
3525 		revents = 0;
3526 		SOCKBUF_LOCK(&so->so_snd);
3527 		SOCKBUF_LOCK(&so->so_rcv);
3528 		if (events & (POLLIN | POLLRDNORM))
3529 			if (soreadabledata(so))
3530 				revents |= events & (POLLIN | POLLRDNORM);
3531 		if (events & (POLLOUT | POLLWRNORM))
3532 			if (sowriteable(so))
3533 				revents |= events & (POLLOUT | POLLWRNORM);
3534 		if (events & (POLLPRI | POLLRDBAND))
3535 			if (so->so_oobmark ||
3536 			    (so->so_rcv.sb_state & SBS_RCVATMARK))
3537 				revents |= events & (POLLPRI | POLLRDBAND);
3538 		if ((events & POLLINIGNEOF) == 0) {
3539 			if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3540 				revents |= events & (POLLIN | POLLRDNORM);
3541 				if (so->so_snd.sb_state & SBS_CANTSENDMORE)
3542 					revents |= POLLHUP;
3543 			}
3544 		}
3545 		if (revents == 0) {
3546 			if (events &
3547 			    (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
3548 				selrecord(td, &so->so_rdsel);
3549 				so->so_rcv.sb_flags |= SB_SEL;
3550 			}
3551 			if (events & (POLLOUT | POLLWRNORM)) {
3552 				selrecord(td, &so->so_wrsel);
3553 				so->so_snd.sb_flags |= SB_SEL;
3554 			}
3555 		}
3556 		SOCKBUF_UNLOCK(&so->so_rcv);
3557 		SOCKBUF_UNLOCK(&so->so_snd);
3558 	}
3559 	SOCK_UNLOCK(so);
3560 	return (revents);
3561 }
3562 
3563 int
3564 soo_kqfilter(struct file *fp, struct knote *kn)
3565 {
3566 	struct socket *so = kn->kn_fp->f_data;
3567 	struct sockbuf *sb;
3568 	struct knlist *knl;
3569 
3570 	switch (kn->kn_filter) {
3571 	case EVFILT_READ:
3572 		kn->kn_fop = &soread_filtops;
3573 		knl = &so->so_rdsel.si_note;
3574 		sb = &so->so_rcv;
3575 		break;
3576 	case EVFILT_WRITE:
3577 		kn->kn_fop = &sowrite_filtops;
3578 		knl = &so->so_wrsel.si_note;
3579 		sb = &so->so_snd;
3580 		break;
3581 	case EVFILT_EMPTY:
3582 		kn->kn_fop = &soempty_filtops;
3583 		knl = &so->so_wrsel.si_note;
3584 		sb = &so->so_snd;
3585 		break;
3586 	default:
3587 		return (EINVAL);
3588 	}
3589 
3590 	SOCK_LOCK(so);
3591 	if (SOLISTENING(so)) {
3592 		knlist_add(knl, kn, 1);
3593 	} else {
3594 		SOCKBUF_LOCK(sb);
3595 		knlist_add(knl, kn, 1);
3596 		sb->sb_flags |= SB_KNOTE;
3597 		SOCKBUF_UNLOCK(sb);
3598 	}
3599 	SOCK_UNLOCK(so);
3600 	return (0);
3601 }
3602 
3603 /*
3604  * Some routines that return EOPNOTSUPP for entry points that are not
3605  * supported by a protocol.  Fill in as needed.
3606  */
3607 int
3608 pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
3609 {
3610 
3611 	return EOPNOTSUPP;
3612 }
3613 
3614 int
3615 pru_aio_queue_notsupp(struct socket *so, struct kaiocb *job)
3616 {
3617 
3618 	return EOPNOTSUPP;
3619 }
3620 
3621 int
3622 pru_attach_notsupp(struct socket *so, int proto, struct thread *td)
3623 {
3624 
3625 	return EOPNOTSUPP;
3626 }
3627 
3628 int
3629 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3630 {
3631 
3632 	return EOPNOTSUPP;
3633 }
3634 
3635 int
3636 pru_bindat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3637     struct thread *td)
3638 {
3639 
3640 	return EOPNOTSUPP;
3641 }
3642 
3643 int
3644 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3645 {
3646 
3647 	return EOPNOTSUPP;
3648 }
3649 
3650 int
3651 pru_connectat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3652     struct thread *td)
3653 {
3654 
3655 	return EOPNOTSUPP;
3656 }
3657 
3658 int
3659 pru_connect2_notsupp(struct socket *so1, struct socket *so2)
3660 {
3661 
3662 	return EOPNOTSUPP;
3663 }
3664 
3665 int
3666 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
3667     struct ifnet *ifp, struct thread *td)
3668 {
3669 
3670 	return EOPNOTSUPP;
3671 }
3672 
3673 int
3674 pru_disconnect_notsupp(struct socket *so)
3675 {
3676 
3677 	return EOPNOTSUPP;
3678 }
3679 
3680 int
3681 pru_listen_notsupp(struct socket *so, int backlog, struct thread *td)
3682 {
3683 
3684 	return EOPNOTSUPP;
3685 }
3686 
3687 int
3688 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam)
3689 {
3690 
3691 	return EOPNOTSUPP;
3692 }
3693 
3694 int
3695 pru_rcvd_notsupp(struct socket *so, int flags)
3696 {
3697 
3698 	return EOPNOTSUPP;
3699 }
3700 
3701 int
3702 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
3703 {
3704 
3705 	return EOPNOTSUPP;
3706 }
3707 
3708 int
3709 pru_send_notsupp(struct socket *so, int flags, struct mbuf *m,
3710     struct sockaddr *addr, struct mbuf *control, struct thread *td)
3711 {
3712 
3713 	return EOPNOTSUPP;
3714 }
3715 
3716 int
3717 pru_ready_notsupp(struct socket *so, struct mbuf *m, int count)
3718 {
3719 
3720 	return (EOPNOTSUPP);
3721 }
3722 
3723 /*
3724  * This isn't really a ``null'' operation, but it's the default one and
3725  * doesn't do anything destructive.
3726  */
3727 int
3728 pru_sense_null(struct socket *so, struct stat *sb)
3729 {
3730 
3731 	sb->st_blksize = so->so_snd.sb_hiwat;
3732 	return 0;
3733 }
3734 
3735 int
3736 pru_shutdown_notsupp(struct socket *so)
3737 {
3738 
3739 	return EOPNOTSUPP;
3740 }
3741 
3742 int
3743 pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam)
3744 {
3745 
3746 	return EOPNOTSUPP;
3747 }
3748 
3749 int
3750 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio,
3751     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
3752 {
3753 
3754 	return EOPNOTSUPP;
3755 }
3756 
3757 int
3758 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr,
3759     struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
3760 {
3761 
3762 	return EOPNOTSUPP;
3763 }
3764 
3765 int
3766 pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred,
3767     struct thread *td)
3768 {
3769 
3770 	return EOPNOTSUPP;
3771 }
3772 
3773 static void
3774 filt_sordetach(struct knote *kn)
3775 {
3776 	struct socket *so = kn->kn_fp->f_data;
3777 
3778 	so_rdknl_lock(so);
3779 	knlist_remove(&so->so_rdsel.si_note, kn, 1);
3780 	if (!SOLISTENING(so) && knlist_empty(&so->so_rdsel.si_note))
3781 		so->so_rcv.sb_flags &= ~SB_KNOTE;
3782 	so_rdknl_unlock(so);
3783 }
3784 
3785 /*ARGSUSED*/
3786 static int
3787 filt_soread(struct knote *kn, long hint)
3788 {
3789 	struct socket *so;
3790 
3791 	so = kn->kn_fp->f_data;
3792 
3793 	if (SOLISTENING(so)) {
3794 		SOCK_LOCK_ASSERT(so);
3795 		kn->kn_data = so->sol_qlen;
3796 		if (so->so_error) {
3797 			kn->kn_flags |= EV_EOF;
3798 			kn->kn_fflags = so->so_error;
3799 			return (1);
3800 		}
3801 		return (!TAILQ_EMPTY(&so->sol_comp));
3802 	}
3803 
3804 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
3805 
3806 	kn->kn_data = sbavail(&so->so_rcv) - so->so_rcv.sb_ctl;
3807 	if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3808 		kn->kn_flags |= EV_EOF;
3809 		kn->kn_fflags = so->so_error;
3810 		return (1);
3811 	} else if (so->so_error)	/* temporary udp error */
3812 		return (1);
3813 
3814 	if (kn->kn_sfflags & NOTE_LOWAT) {
3815 		if (kn->kn_data >= kn->kn_sdata)
3816 			return (1);
3817 	} else if (sbavail(&so->so_rcv) >= so->so_rcv.sb_lowat)
3818 		return (1);
3819 
3820 	/* This hook returning non-zero indicates an event, not error */
3821 	return (hhook_run_socket(so, NULL, HHOOK_FILT_SOREAD));
3822 }
3823 
3824 static void
3825 filt_sowdetach(struct knote *kn)
3826 {
3827 	struct socket *so = kn->kn_fp->f_data;
3828 
3829 	so_wrknl_lock(so);
3830 	knlist_remove(&so->so_wrsel.si_note, kn, 1);
3831 	if (!SOLISTENING(so) && knlist_empty(&so->so_wrsel.si_note))
3832 		so->so_snd.sb_flags &= ~SB_KNOTE;
3833 	so_wrknl_unlock(so);
3834 }
3835 
3836 /*ARGSUSED*/
3837 static int
3838 filt_sowrite(struct knote *kn, long hint)
3839 {
3840 	struct socket *so;
3841 
3842 	so = kn->kn_fp->f_data;
3843 
3844 	if (SOLISTENING(so))
3845 		return (0);
3846 
3847 	SOCKBUF_LOCK_ASSERT(&so->so_snd);
3848 	kn->kn_data = sbspace(&so->so_snd);
3849 
3850 	hhook_run_socket(so, kn, HHOOK_FILT_SOWRITE);
3851 
3852 	if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
3853 		kn->kn_flags |= EV_EOF;
3854 		kn->kn_fflags = so->so_error;
3855 		return (1);
3856 	} else if (so->so_error)	/* temporary udp error */
3857 		return (1);
3858 	else if (((so->so_state & SS_ISCONNECTED) == 0) &&
3859 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
3860 		return (0);
3861 	else if (kn->kn_sfflags & NOTE_LOWAT)
3862 		return (kn->kn_data >= kn->kn_sdata);
3863 	else
3864 		return (kn->kn_data >= so->so_snd.sb_lowat);
3865 }
3866 
3867 static int
3868 filt_soempty(struct knote *kn, long hint)
3869 {
3870 	struct socket *so;
3871 
3872 	so = kn->kn_fp->f_data;
3873 
3874 	if (SOLISTENING(so))
3875 		return (1);
3876 
3877 	SOCKBUF_LOCK_ASSERT(&so->so_snd);
3878 	kn->kn_data = sbused(&so->so_snd);
3879 
3880 	if (kn->kn_data == 0)
3881 		return (1);
3882 	else
3883 		return (0);
3884 }
3885 
3886 int
3887 socheckuid(struct socket *so, uid_t uid)
3888 {
3889 
3890 	if (so == NULL)
3891 		return (EPERM);
3892 	if (so->so_cred->cr_uid != uid)
3893 		return (EPERM);
3894 	return (0);
3895 }
3896 
3897 /*
3898  * These functions are used by protocols to notify the socket layer (and its
3899  * consumers) of state changes in the sockets driven by protocol-side events.
3900  */
3901 
3902 /*
3903  * Procedures to manipulate state flags of socket and do appropriate wakeups.
3904  *
3905  * Normal sequence from the active (originating) side is that
3906  * soisconnecting() is called during processing of connect() call, resulting
3907  * in an eventual call to soisconnected() if/when the connection is
3908  * established.  When the connection is torn down soisdisconnecting() is
3909  * called during processing of disconnect() call, and soisdisconnected() is
3910  * called when the connection to the peer is totally severed.  The semantics
3911  * of these routines are such that connectionless protocols can call
3912  * soisconnected() and soisdisconnected() only, bypassing the in-progress
3913  * calls when setting up a ``connection'' takes no time.
3914  *
3915  * From the passive side, a socket is created with two queues of sockets:
3916  * so_incomp for connections in progress and so_comp for connections already
3917  * made and awaiting user acceptance.  As a protocol is preparing incoming
3918  * connections, it creates a socket structure queued on so_incomp by calling
3919  * sonewconn().  When the connection is established, soisconnected() is
3920  * called, and transfers the socket structure to so_comp, making it available
3921  * to accept().
3922  *
3923  * If a socket is closed with sockets on either so_incomp or so_comp, these
3924  * sockets are dropped.
3925  *
3926  * If higher-level protocols are implemented in the kernel, the wakeups done
3927  * here will sometimes cause software-interrupt process scheduling.
3928  */
3929 void
3930 soisconnecting(struct socket *so)
3931 {
3932 
3933 	SOCK_LOCK(so);
3934 	so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
3935 	so->so_state |= SS_ISCONNECTING;
3936 	SOCK_UNLOCK(so);
3937 }
3938 
3939 void
3940 soisconnected(struct socket *so)
3941 {
3942 
3943 	SOCK_LOCK(so);
3944 	so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
3945 	so->so_state |= SS_ISCONNECTED;
3946 
3947 	if (so->so_qstate == SQ_INCOMP) {
3948 		struct socket *head = so->so_listen;
3949 		int ret;
3950 
3951 		KASSERT(head, ("%s: so %p on incomp of NULL", __func__, so));
3952 		/*
3953 		 * Promoting a socket from incomplete queue to complete, we
3954 		 * need to go through reverse order of locking.  We first do
3955 		 * trylock, and if that doesn't succeed, we go the hard way
3956 		 * leaving a reference and rechecking consistency after proper
3957 		 * locking.
3958 		 */
3959 		if (__predict_false(SOLISTEN_TRYLOCK(head) == 0)) {
3960 			soref(head);
3961 			SOCK_UNLOCK(so);
3962 			SOLISTEN_LOCK(head);
3963 			SOCK_LOCK(so);
3964 			if (__predict_false(head != so->so_listen)) {
3965 				/*
3966 				 * The socket went off the listen queue,
3967 				 * should be lost race to close(2) of sol.
3968 				 * The socket is about to soabort().
3969 				 */
3970 				SOCK_UNLOCK(so);
3971 				sorele(head);
3972 				return;
3973 			}
3974 			/* Not the last one, as so holds a ref. */
3975 			refcount_release(&head->so_count);
3976 		}
3977 again:
3978 		if ((so->so_options & SO_ACCEPTFILTER) == 0) {
3979 			TAILQ_REMOVE(&head->sol_incomp, so, so_list);
3980 			head->sol_incqlen--;
3981 			TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list);
3982 			head->sol_qlen++;
3983 			so->so_qstate = SQ_COMP;
3984 			SOCK_UNLOCK(so);
3985 			solisten_wakeup(head);	/* unlocks */
3986 		} else {
3987 			SOCKBUF_LOCK(&so->so_rcv);
3988 			soupcall_set(so, SO_RCV,
3989 			    head->sol_accept_filter->accf_callback,
3990 			    head->sol_accept_filter_arg);
3991 			so->so_options &= ~SO_ACCEPTFILTER;
3992 			ret = head->sol_accept_filter->accf_callback(so,
3993 			    head->sol_accept_filter_arg, M_NOWAIT);
3994 			if (ret == SU_ISCONNECTED) {
3995 				soupcall_clear(so, SO_RCV);
3996 				SOCKBUF_UNLOCK(&so->so_rcv);
3997 				goto again;
3998 			}
3999 			SOCKBUF_UNLOCK(&so->so_rcv);
4000 			SOCK_UNLOCK(so);
4001 			SOLISTEN_UNLOCK(head);
4002 		}
4003 		return;
4004 	}
4005 	SOCK_UNLOCK(so);
4006 	wakeup(&so->so_timeo);
4007 	sorwakeup(so);
4008 	sowwakeup(so);
4009 }
4010 
4011 void
4012 soisdisconnecting(struct socket *so)
4013 {
4014 
4015 	SOCK_LOCK(so);
4016 	so->so_state &= ~SS_ISCONNECTING;
4017 	so->so_state |= SS_ISDISCONNECTING;
4018 
4019 	if (!SOLISTENING(so)) {
4020 		SOCKBUF_LOCK(&so->so_rcv);
4021 		socantrcvmore_locked(so);
4022 		SOCKBUF_LOCK(&so->so_snd);
4023 		socantsendmore_locked(so);
4024 	}
4025 	SOCK_UNLOCK(so);
4026 	wakeup(&so->so_timeo);
4027 }
4028 
4029 void
4030 soisdisconnected(struct socket *so)
4031 {
4032 
4033 	SOCK_LOCK(so);
4034 
4035 	/*
4036 	 * There is at least one reader of so_state that does not
4037 	 * acquire socket lock, namely soreceive_generic().  Ensure
4038 	 * that it never sees all flags that track connection status
4039 	 * cleared, by ordering the update with a barrier semantic of
4040 	 * our release thread fence.
4041 	 */
4042 	so->so_state |= SS_ISDISCONNECTED;
4043 	atomic_thread_fence_rel();
4044 	so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
4045 
4046 	if (!SOLISTENING(so)) {
4047 		SOCK_UNLOCK(so);
4048 		SOCKBUF_LOCK(&so->so_rcv);
4049 		socantrcvmore_locked(so);
4050 		SOCKBUF_LOCK(&so->so_snd);
4051 		sbdrop_locked(&so->so_snd, sbused(&so->so_snd));
4052 		socantsendmore_locked(so);
4053 	} else
4054 		SOCK_UNLOCK(so);
4055 	wakeup(&so->so_timeo);
4056 }
4057 
4058 /*
4059  * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
4060  */
4061 struct sockaddr *
4062 sodupsockaddr(const struct sockaddr *sa, int mflags)
4063 {
4064 	struct sockaddr *sa2;
4065 
4066 	sa2 = malloc(sa->sa_len, M_SONAME, mflags);
4067 	if (sa2)
4068 		bcopy(sa, sa2, sa->sa_len);
4069 	return sa2;
4070 }
4071 
4072 /*
4073  * Register per-socket destructor.
4074  */
4075 void
4076 sodtor_set(struct socket *so, so_dtor_t *func)
4077 {
4078 
4079 	SOCK_LOCK_ASSERT(so);
4080 	so->so_dtor = func;
4081 }
4082 
4083 /*
4084  * Register per-socket buffer upcalls.
4085  */
4086 void
4087 soupcall_set(struct socket *so, int which, so_upcall_t func, void *arg)
4088 {
4089 	struct sockbuf *sb;
4090 
4091 	KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so));
4092 
4093 	switch (which) {
4094 	case SO_RCV:
4095 		sb = &so->so_rcv;
4096 		break;
4097 	case SO_SND:
4098 		sb = &so->so_snd;
4099 		break;
4100 	default:
4101 		panic("soupcall_set: bad which");
4102 	}
4103 	SOCKBUF_LOCK_ASSERT(sb);
4104 	sb->sb_upcall = func;
4105 	sb->sb_upcallarg = arg;
4106 	sb->sb_flags |= SB_UPCALL;
4107 }
4108 
4109 void
4110 soupcall_clear(struct socket *so, int which)
4111 {
4112 	struct sockbuf *sb;
4113 
4114 	KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so));
4115 
4116 	switch (which) {
4117 	case SO_RCV:
4118 		sb = &so->so_rcv;
4119 		break;
4120 	case SO_SND:
4121 		sb = &so->so_snd;
4122 		break;
4123 	default:
4124 		panic("soupcall_clear: bad which");
4125 	}
4126 	SOCKBUF_LOCK_ASSERT(sb);
4127 	KASSERT(sb->sb_upcall != NULL,
4128 	    ("%s: so %p no upcall to clear", __func__, so));
4129 	sb->sb_upcall = NULL;
4130 	sb->sb_upcallarg = NULL;
4131 	sb->sb_flags &= ~SB_UPCALL;
4132 }
4133 
4134 void
4135 solisten_upcall_set(struct socket *so, so_upcall_t func, void *arg)
4136 {
4137 
4138 	SOLISTEN_LOCK_ASSERT(so);
4139 	so->sol_upcall = func;
4140 	so->sol_upcallarg = arg;
4141 }
4142 
4143 static void
4144 so_rdknl_lock(void *arg)
4145 {
4146 	struct socket *so = arg;
4147 
4148 	if (SOLISTENING(so))
4149 		SOCK_LOCK(so);
4150 	else
4151 		SOCKBUF_LOCK(&so->so_rcv);
4152 }
4153 
4154 static void
4155 so_rdknl_unlock(void *arg)
4156 {
4157 	struct socket *so = arg;
4158 
4159 	if (SOLISTENING(so))
4160 		SOCK_UNLOCK(so);
4161 	else
4162 		SOCKBUF_UNLOCK(&so->so_rcv);
4163 }
4164 
4165 static void
4166 so_rdknl_assert_locked(void *arg)
4167 {
4168 	struct socket *so = arg;
4169 
4170 	if (SOLISTENING(so))
4171 		SOCK_LOCK_ASSERT(so);
4172 	else
4173 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
4174 }
4175 
4176 static void
4177 so_rdknl_assert_unlocked(void *arg)
4178 {
4179 	struct socket *so = arg;
4180 
4181 	if (SOLISTENING(so))
4182 		SOCK_UNLOCK_ASSERT(so);
4183 	else
4184 		SOCKBUF_UNLOCK_ASSERT(&so->so_rcv);
4185 }
4186 
4187 static void
4188 so_wrknl_lock(void *arg)
4189 {
4190 	struct socket *so = arg;
4191 
4192 	if (SOLISTENING(so))
4193 		SOCK_LOCK(so);
4194 	else
4195 		SOCKBUF_LOCK(&so->so_snd);
4196 }
4197 
4198 static void
4199 so_wrknl_unlock(void *arg)
4200 {
4201 	struct socket *so = arg;
4202 
4203 	if (SOLISTENING(so))
4204 		SOCK_UNLOCK(so);
4205 	else
4206 		SOCKBUF_UNLOCK(&so->so_snd);
4207 }
4208 
4209 static void
4210 so_wrknl_assert_locked(void *arg)
4211 {
4212 	struct socket *so = arg;
4213 
4214 	if (SOLISTENING(so))
4215 		SOCK_LOCK_ASSERT(so);
4216 	else
4217 		SOCKBUF_LOCK_ASSERT(&so->so_snd);
4218 }
4219 
4220 static void
4221 so_wrknl_assert_unlocked(void *arg)
4222 {
4223 	struct socket *so = arg;
4224 
4225 	if (SOLISTENING(so))
4226 		SOCK_UNLOCK_ASSERT(so);
4227 	else
4228 		SOCKBUF_UNLOCK_ASSERT(&so->so_snd);
4229 }
4230 
4231 /*
4232  * Create an external-format (``xsocket'') structure using the information in
4233  * the kernel-format socket structure pointed to by so.  This is done to
4234  * reduce the spew of irrelevant information over this interface, to isolate
4235  * user code from changes in the kernel structure, and potentially to provide
4236  * information-hiding if we decide that some of this information should be
4237  * hidden from users.
4238  */
4239 void
4240 sotoxsocket(struct socket *so, struct xsocket *xso)
4241 {
4242 
4243 	bzero(xso, sizeof(*xso));
4244 	xso->xso_len = sizeof *xso;
4245 	xso->xso_so = (uintptr_t)so;
4246 	xso->so_type = so->so_type;
4247 	xso->so_options = so->so_options;
4248 	xso->so_linger = so->so_linger;
4249 	xso->so_state = so->so_state;
4250 	xso->so_pcb = (uintptr_t)so->so_pcb;
4251 	xso->xso_protocol = so->so_proto->pr_protocol;
4252 	xso->xso_family = so->so_proto->pr_domain->dom_family;
4253 	xso->so_timeo = so->so_timeo;
4254 	xso->so_error = so->so_error;
4255 	xso->so_uid = so->so_cred->cr_uid;
4256 	xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
4257 	if (SOLISTENING(so)) {
4258 		xso->so_qlen = so->sol_qlen;
4259 		xso->so_incqlen = so->sol_incqlen;
4260 		xso->so_qlimit = so->sol_qlimit;
4261 		xso->so_oobmark = 0;
4262 	} else {
4263 		xso->so_state |= so->so_qstate;
4264 		xso->so_qlen = xso->so_incqlen = xso->so_qlimit = 0;
4265 		xso->so_oobmark = so->so_oobmark;
4266 		sbtoxsockbuf(&so->so_snd, &xso->so_snd);
4267 		sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
4268 	}
4269 }
4270 
4271 struct sockbuf *
4272 so_sockbuf_rcv(struct socket *so)
4273 {
4274 
4275 	return (&so->so_rcv);
4276 }
4277 
4278 struct sockbuf *
4279 so_sockbuf_snd(struct socket *so)
4280 {
4281 
4282 	return (&so->so_snd);
4283 }
4284 
4285 int
4286 so_state_get(const struct socket *so)
4287 {
4288 
4289 	return (so->so_state);
4290 }
4291 
4292 void
4293 so_state_set(struct socket *so, int val)
4294 {
4295 
4296 	so->so_state = val;
4297 }
4298 
4299 int
4300 so_options_get(const struct socket *so)
4301 {
4302 
4303 	return (so->so_options);
4304 }
4305 
4306 void
4307 so_options_set(struct socket *so, int val)
4308 {
4309 
4310 	so->so_options = val;
4311 }
4312 
4313 int
4314 so_error_get(const struct socket *so)
4315 {
4316 
4317 	return (so->so_error);
4318 }
4319 
4320 void
4321 so_error_set(struct socket *so, int val)
4322 {
4323 
4324 	so->so_error = val;
4325 }
4326 
4327 int
4328 so_linger_get(const struct socket *so)
4329 {
4330 
4331 	return (so->so_linger);
4332 }
4333 
4334 void
4335 so_linger_set(struct socket *so, int val)
4336 {
4337 
4338 	KASSERT(val >= 0 && val <= USHRT_MAX && val <= (INT_MAX / hz),
4339 	    ("%s: val %d out of range", __func__, val));
4340 
4341 	so->so_linger = val;
4342 }
4343 
4344 struct protosw *
4345 so_protosw_get(const struct socket *so)
4346 {
4347 
4348 	return (so->so_proto);
4349 }
4350 
4351 void
4352 so_protosw_set(struct socket *so, struct protosw *val)
4353 {
4354 
4355 	so->so_proto = val;
4356 }
4357 
4358 void
4359 so_sorwakeup(struct socket *so)
4360 {
4361 
4362 	sorwakeup(so);
4363 }
4364 
4365 void
4366 so_sowwakeup(struct socket *so)
4367 {
4368 
4369 	sowwakeup(so);
4370 }
4371 
4372 void
4373 so_sorwakeup_locked(struct socket *so)
4374 {
4375 
4376 	sorwakeup_locked(so);
4377 }
4378 
4379 void
4380 so_sowwakeup_locked(struct socket *so)
4381 {
4382 
4383 	sowwakeup_locked(so);
4384 }
4385 
4386 void
4387 so_lock(struct socket *so)
4388 {
4389 
4390 	SOCK_LOCK(so);
4391 }
4392 
4393 void
4394 so_unlock(struct socket *so)
4395 {
4396 
4397 	SOCK_UNLOCK(so);
4398 }
4399