xref: /freebsd/sys/kern/uipc_socket.c (revision 1165fc9a526630487a1feb63daef65c5aee1a583)
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/capsicum.h>
116 #include <sys/fcntl.h>
117 #include <sys/limits.h>
118 #include <sys/lock.h>
119 #include <sys/mac.h>
120 #include <sys/malloc.h>
121 #include <sys/mbuf.h>
122 #include <sys/mutex.h>
123 #include <sys/domain.h>
124 #include <sys/file.h>			/* for struct knote */
125 #include <sys/hhook.h>
126 #include <sys/kernel.h>
127 #include <sys/khelp.h>
128 #include <sys/ktls.h>
129 #include <sys/event.h>
130 #include <sys/eventhandler.h>
131 #include <sys/poll.h>
132 #include <sys/proc.h>
133 #include <sys/protosw.h>
134 #include <sys/sbuf.h>
135 #include <sys/socket.h>
136 #include <sys/socketvar.h>
137 #include <sys/resourcevar.h>
138 #include <net/route.h>
139 #include <sys/signalvar.h>
140 #include <sys/stat.h>
141 #include <sys/sx.h>
142 #include <sys/sysctl.h>
143 #include <sys/taskqueue.h>
144 #include <sys/uio.h>
145 #include <sys/un.h>
146 #include <sys/unpcb.h>
147 #include <sys/jail.h>
148 #include <sys/syslog.h>
149 #include <netinet/in.h>
150 #include <netinet/in_pcb.h>
151 #include <netinet/tcp.h>
152 
153 #include <net/vnet.h>
154 
155 #include <security/mac/mac_framework.h>
156 
157 #include <vm/uma.h>
158 
159 #ifdef COMPAT_FREEBSD32
160 #include <sys/mount.h>
161 #include <sys/sysent.h>
162 #include <compat/freebsd32/freebsd32.h>
163 #endif
164 
165 static int	soreceive_rcvoob(struct socket *so, struct uio *uio,
166 		    int flags);
167 static void	so_rdknl_lock(void *);
168 static void	so_rdknl_unlock(void *);
169 static void	so_rdknl_assert_lock(void *, int);
170 static void	so_wrknl_lock(void *);
171 static void	so_wrknl_unlock(void *);
172 static void	so_wrknl_assert_lock(void *, int);
173 
174 static void	filt_sordetach(struct knote *kn);
175 static int	filt_soread(struct knote *kn, long hint);
176 static void	filt_sowdetach(struct knote *kn);
177 static int	filt_sowrite(struct knote *kn, long hint);
178 static int	filt_soempty(struct knote *kn, long hint);
179 static int inline hhook_run_socket(struct socket *so, void *hctx, int32_t h_id);
180 fo_kqfilter_t	soo_kqfilter;
181 
182 static struct filterops soread_filtops = {
183 	.f_isfd = 1,
184 	.f_detach = filt_sordetach,
185 	.f_event = filt_soread,
186 };
187 static struct filterops sowrite_filtops = {
188 	.f_isfd = 1,
189 	.f_detach = filt_sowdetach,
190 	.f_event = filt_sowrite,
191 };
192 static struct filterops soempty_filtops = {
193 	.f_isfd = 1,
194 	.f_detach = filt_sowdetach,
195 	.f_event = filt_soempty,
196 };
197 
198 so_gen_t	so_gencnt;	/* generation count for sockets */
199 
200 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
201 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
202 
203 #define	VNET_SO_ASSERT(so)						\
204 	VNET_ASSERT(curvnet != NULL,					\
205 	    ("%s:%d curvnet is NULL, so=%p", __func__, __LINE__, (so)));
206 
207 VNET_DEFINE(struct hhook_head *, socket_hhh[HHOOK_SOCKET_LAST + 1]);
208 #define	V_socket_hhh		VNET(socket_hhh)
209 
210 /*
211  * Limit on the number of connections in the listen queue waiting
212  * for accept(2).
213  * NB: The original sysctl somaxconn is still available but hidden
214  * to prevent confusion about the actual purpose of this number.
215  */
216 static u_int somaxconn = SOMAXCONN;
217 
218 static int
219 sysctl_somaxconn(SYSCTL_HANDLER_ARGS)
220 {
221 	int error;
222 	int val;
223 
224 	val = somaxconn;
225 	error = sysctl_handle_int(oidp, &val, 0, req);
226 	if (error || !req->newptr )
227 		return (error);
228 
229 	/*
230 	 * The purpose of the UINT_MAX / 3 limit, is so that the formula
231 	 *   3 * so_qlimit / 2
232 	 * below, will not overflow.
233          */
234 
235 	if (val < 1 || val > UINT_MAX / 3)
236 		return (EINVAL);
237 
238 	somaxconn = val;
239 	return (0);
240 }
241 SYSCTL_PROC(_kern_ipc, OID_AUTO, soacceptqueue,
242     CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_MPSAFE, 0, sizeof(int),
243     sysctl_somaxconn, "I",
244     "Maximum listen socket pending connection accept queue size");
245 SYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn,
246     CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_SKIP | CTLFLAG_MPSAFE, 0,
247     sizeof(int), sysctl_somaxconn, "I",
248     "Maximum listen socket pending connection accept queue size (compat)");
249 
250 static int numopensockets;
251 SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD,
252     &numopensockets, 0, "Number of open sockets");
253 
254 /*
255  * accept_mtx locks down per-socket fields relating to accept queues.  See
256  * socketvar.h for an annotation of the protected fields of struct socket.
257  */
258 struct mtx accept_mtx;
259 MTX_SYSINIT(accept_mtx, &accept_mtx, "accept", MTX_DEF);
260 
261 /*
262  * so_global_mtx protects so_gencnt, numopensockets, and the per-socket
263  * so_gencnt field.
264  */
265 static struct mtx so_global_mtx;
266 MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF);
267 
268 /*
269  * General IPC sysctl name space, used by sockets and a variety of other IPC
270  * types.
271  */
272 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
273     "IPC");
274 
275 /*
276  * Initialize the socket subsystem and set up the socket
277  * memory allocator.
278  */
279 static uma_zone_t socket_zone;
280 int	maxsockets;
281 
282 static void
283 socket_zone_change(void *tag)
284 {
285 
286 	maxsockets = uma_zone_set_max(socket_zone, maxsockets);
287 }
288 
289 static void
290 socket_hhook_register(int subtype)
291 {
292 
293 	if (hhook_head_register(HHOOK_TYPE_SOCKET, subtype,
294 	    &V_socket_hhh[subtype],
295 	    HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
296 		printf("%s: WARNING: unable to register hook\n", __func__);
297 }
298 
299 static void
300 socket_hhook_deregister(int subtype)
301 {
302 
303 	if (hhook_head_deregister(V_socket_hhh[subtype]) != 0)
304 		printf("%s: WARNING: unable to deregister hook\n", __func__);
305 }
306 
307 static void
308 socket_init(void *tag)
309 {
310 
311 	socket_zone = uma_zcreate("socket", sizeof(struct socket), NULL, NULL,
312 	    NULL, NULL, UMA_ALIGN_PTR, 0);
313 	maxsockets = uma_zone_set_max(socket_zone, maxsockets);
314 	uma_zone_set_warning(socket_zone, "kern.ipc.maxsockets limit reached");
315 	EVENTHANDLER_REGISTER(maxsockets_change, socket_zone_change, NULL,
316 	    EVENTHANDLER_PRI_FIRST);
317 }
318 SYSINIT(socket, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, socket_init, NULL);
319 
320 static void
321 socket_vnet_init(const void *unused __unused)
322 {
323 	int i;
324 
325 	/* We expect a contiguous range */
326 	for (i = 0; i <= HHOOK_SOCKET_LAST; i++)
327 		socket_hhook_register(i);
328 }
329 VNET_SYSINIT(socket_vnet_init, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
330     socket_vnet_init, NULL);
331 
332 static void
333 socket_vnet_uninit(const void *unused __unused)
334 {
335 	int i;
336 
337 	for (i = 0; i <= HHOOK_SOCKET_LAST; i++)
338 		socket_hhook_deregister(i);
339 }
340 VNET_SYSUNINIT(socket_vnet_uninit, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
341     socket_vnet_uninit, NULL);
342 
343 /*
344  * Initialise maxsockets.  This SYSINIT must be run after
345  * tunable_mbinit().
346  */
347 static void
348 init_maxsockets(void *ignored)
349 {
350 
351 	TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
352 	maxsockets = imax(maxsockets, maxfiles);
353 }
354 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL);
355 
356 /*
357  * Sysctl to get and set the maximum global sockets limit.  Notify protocols
358  * of the change so that they can update their dependent limits as required.
359  */
360 static int
361 sysctl_maxsockets(SYSCTL_HANDLER_ARGS)
362 {
363 	int error, newmaxsockets;
364 
365 	newmaxsockets = maxsockets;
366 	error = sysctl_handle_int(oidp, &newmaxsockets, 0, req);
367 	if (error == 0 && req->newptr && newmaxsockets != maxsockets) {
368 		if (newmaxsockets > maxsockets &&
369 		    newmaxsockets <= maxfiles) {
370 			maxsockets = newmaxsockets;
371 			EVENTHANDLER_INVOKE(maxsockets_change);
372 		} else
373 			error = EINVAL;
374 	}
375 	return (error);
376 }
377 SYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets,
378     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE, &maxsockets, 0,
379     sysctl_maxsockets, "IU",
380     "Maximum number of sockets available");
381 
382 /*
383  * Socket operation routines.  These routines are called by the routines in
384  * sys_socket.c or from a system process, and implement the semantics of
385  * socket operations by switching out to the protocol specific routines.
386  */
387 
388 /*
389  * Get a socket structure from our zone, and initialize it.  Note that it
390  * would probably be better to allocate socket and PCB at the same time, but
391  * I'm not convinced that all the protocols can be easily modified to do
392  * this.
393  *
394  * soalloc() returns a socket with a ref count of 0.
395  */
396 static struct socket *
397 soalloc(struct vnet *vnet)
398 {
399 	struct socket *so;
400 
401 	so = uma_zalloc(socket_zone, M_NOWAIT | M_ZERO);
402 	if (so == NULL)
403 		return (NULL);
404 #ifdef MAC
405 	if (mac_socket_init(so, M_NOWAIT) != 0) {
406 		uma_zfree(socket_zone, so);
407 		return (NULL);
408 	}
409 #endif
410 	if (khelp_init_osd(HELPER_CLASS_SOCKET, &so->osd)) {
411 		uma_zfree(socket_zone, so);
412 		return (NULL);
413 	}
414 
415 	/*
416 	 * The socket locking protocol allows to lock 2 sockets at a time,
417 	 * however, the first one must be a listening socket.  WITNESS lacks
418 	 * a feature to change class of an existing lock, so we use DUPOK.
419 	 */
420 	mtx_init(&so->so_lock, "socket", NULL, MTX_DEF | MTX_DUPOK);
421 	mtx_init(&so->so_snd_mtx, "so_snd", NULL, MTX_DEF);
422 	mtx_init(&so->so_rcv_mtx, "so_rcv", NULL, MTX_DEF);
423 	so->so_rcv.sb_sel = &so->so_rdsel;
424 	so->so_snd.sb_sel = &so->so_wrsel;
425 	sx_init(&so->so_snd_sx, "so_snd_sx");
426 	sx_init(&so->so_rcv_sx, "so_rcv_sx");
427 	TAILQ_INIT(&so->so_snd.sb_aiojobq);
428 	TAILQ_INIT(&so->so_rcv.sb_aiojobq);
429 	TASK_INIT(&so->so_snd.sb_aiotask, 0, soaio_snd, so);
430 	TASK_INIT(&so->so_rcv.sb_aiotask, 0, soaio_rcv, so);
431 #ifdef VIMAGE
432 	VNET_ASSERT(vnet != NULL, ("%s:%d vnet is NULL, so=%p",
433 	    __func__, __LINE__, so));
434 	so->so_vnet = vnet;
435 #endif
436 	/* We shouldn't need the so_global_mtx */
437 	if (hhook_run_socket(so, NULL, HHOOK_SOCKET_CREATE)) {
438 		/* Do we need more comprehensive error returns? */
439 		uma_zfree(socket_zone, so);
440 		return (NULL);
441 	}
442 	mtx_lock(&so_global_mtx);
443 	so->so_gencnt = ++so_gencnt;
444 	++numopensockets;
445 #ifdef VIMAGE
446 	vnet->vnet_sockcnt++;
447 #endif
448 	mtx_unlock(&so_global_mtx);
449 
450 	return (so);
451 }
452 
453 /*
454  * Free the storage associated with a socket at the socket layer, tear down
455  * locks, labels, etc.  All protocol state is assumed already to have been
456  * torn down (and possibly never set up) by the caller.
457  */
458 static void
459 sodealloc(struct socket *so)
460 {
461 
462 	KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count));
463 	KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL"));
464 
465 	mtx_lock(&so_global_mtx);
466 	so->so_gencnt = ++so_gencnt;
467 	--numopensockets;	/* Could be below, but faster here. */
468 #ifdef VIMAGE
469 	VNET_ASSERT(so->so_vnet != NULL, ("%s:%d so_vnet is NULL, so=%p",
470 	    __func__, __LINE__, so));
471 	so->so_vnet->vnet_sockcnt--;
472 #endif
473 	mtx_unlock(&so_global_mtx);
474 #ifdef MAC
475 	mac_socket_destroy(so);
476 #endif
477 	hhook_run_socket(so, NULL, HHOOK_SOCKET_CLOSE);
478 
479 	khelp_destroy_osd(&so->osd);
480 	if (SOLISTENING(so)) {
481 		if (so->sol_accept_filter != NULL)
482 			accept_filt_setopt(so, NULL);
483 	} else {
484 		if (so->so_rcv.sb_hiwat)
485 			(void)chgsbsize(so->so_cred->cr_uidinfo,
486 			    &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY);
487 		if (so->so_snd.sb_hiwat)
488 			(void)chgsbsize(so->so_cred->cr_uidinfo,
489 			    &so->so_snd.sb_hiwat, 0, RLIM_INFINITY);
490 		sx_destroy(&so->so_snd_sx);
491 		sx_destroy(&so->so_rcv_sx);
492 		mtx_destroy(&so->so_snd_mtx);
493 		mtx_destroy(&so->so_rcv_mtx);
494 	}
495 	crfree(so->so_cred);
496 	mtx_destroy(&so->so_lock);
497 	uma_zfree(socket_zone, so);
498 }
499 
500 /*
501  * socreate returns a socket with a ref count of 1 and a file descriptor
502  * reference.  The socket should be closed with soclose().
503  */
504 int
505 socreate(int dom, struct socket **aso, int type, int proto,
506     struct ucred *cred, struct thread *td)
507 {
508 	struct protosw *prp;
509 	struct socket *so;
510 	int error;
511 
512 	if (proto)
513 		prp = pffindproto(dom, proto, type);
514 	else
515 		prp = pffindtype(dom, type);
516 
517 	if (prp == NULL) {
518 		/* No support for domain. */
519 		if (pffinddomain(dom) == NULL)
520 			return (EAFNOSUPPORT);
521 		/* No support for socket type. */
522 		if (proto == 0 && type != 0)
523 			return (EPROTOTYPE);
524 		return (EPROTONOSUPPORT);
525 	}
526 	if (prp->pr_usrreqs->pru_attach == NULL ||
527 	    prp->pr_usrreqs->pru_attach == pru_attach_notsupp)
528 		return (EPROTONOSUPPORT);
529 
530 	if (IN_CAPABILITY_MODE(td) && (prp->pr_flags & PR_CAPATTACH) == 0)
531 		return (ECAPMODE);
532 
533 	if (prison_check_af(cred, prp->pr_domain->dom_family) != 0)
534 		return (EPROTONOSUPPORT);
535 
536 	if (prp->pr_type != type)
537 		return (EPROTOTYPE);
538 	so = soalloc(CRED_TO_VNET(cred));
539 	if (so == NULL)
540 		return (ENOBUFS);
541 
542 	so->so_type = type;
543 	so->so_cred = crhold(cred);
544 	if ((prp->pr_domain->dom_family == PF_INET) ||
545 	    (prp->pr_domain->dom_family == PF_INET6) ||
546 	    (prp->pr_domain->dom_family == PF_ROUTE))
547 		so->so_fibnum = td->td_proc->p_fibnum;
548 	else
549 		so->so_fibnum = 0;
550 	so->so_proto = prp;
551 #ifdef MAC
552 	mac_socket_create(cred, so);
553 #endif
554 	knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
555 	    so_rdknl_assert_lock);
556 	knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
557 	    so_wrknl_assert_lock);
558 	if ((prp->pr_flags & PR_SOCKBUF) == 0) {
559 		so->so_snd.sb_mtx = &so->so_snd_mtx;
560 		so->so_rcv.sb_mtx = &so->so_rcv_mtx;
561 	}
562 	/*
563 	 * Auto-sizing of socket buffers is managed by the protocols and
564 	 * the appropriate flags must be set in the pru_attach function.
565 	 */
566 	CURVNET_SET(so->so_vnet);
567 	error = (*prp->pr_usrreqs->pru_attach)(so, proto, td);
568 	CURVNET_RESTORE();
569 	if (error) {
570 		sodealloc(so);
571 		return (error);
572 	}
573 	soref(so);
574 	*aso = so;
575 	return (0);
576 }
577 
578 #ifdef REGRESSION
579 static int regression_sonewconn_earlytest = 1;
580 SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW,
581     &regression_sonewconn_earlytest, 0, "Perform early sonewconn limit test");
582 #endif
583 
584 static struct timeval overinterval = { 60, 0 };
585 SYSCTL_TIMEVAL_SEC(_kern_ipc, OID_AUTO, sooverinterval, CTLFLAG_RW,
586     &overinterval,
587     "Delay in seconds between warnings for listen socket overflows");
588 
589 /*
590  * When an attempt at a new connection is noted on a socket which accepts
591  * connections, sonewconn is called.  If the connection is possible (subject
592  * to space constraints, etc.) then we allocate a new structure, properly
593  * linked into the data structure of the original socket, and return this.
594  * Connstatus may be 0, or SS_ISCONFIRMING, or SS_ISCONNECTED.
595  *
596  * Note: the ref count on the socket is 0 on return.
597  */
598 struct socket *
599 sonewconn(struct socket *head, int connstatus)
600 {
601 	struct sbuf descrsb;
602 	struct socket *so;
603 	int len, overcount;
604 	u_int qlen;
605 	const char localprefix[] = "local:";
606 	char descrbuf[SUNPATHLEN + sizeof(localprefix)];
607 #if defined(INET6)
608 	char addrbuf[INET6_ADDRSTRLEN];
609 #elif defined(INET)
610 	char addrbuf[INET_ADDRSTRLEN];
611 #endif
612 	bool dolog, over;
613 
614 	SOLISTEN_LOCK(head);
615 	over = (head->sol_qlen > 3 * head->sol_qlimit / 2);
616 #ifdef REGRESSION
617 	if (regression_sonewconn_earlytest && over) {
618 #else
619 	if (over) {
620 #endif
621 		head->sol_overcount++;
622 		dolog = !!ratecheck(&head->sol_lastover, &overinterval);
623 
624 		/*
625 		 * If we're going to log, copy the overflow count and queue
626 		 * length from the listen socket before dropping the lock.
627 		 * Also, reset the overflow count.
628 		 */
629 		if (dolog) {
630 			overcount = head->sol_overcount;
631 			head->sol_overcount = 0;
632 			qlen = head->sol_qlen;
633 		}
634 		SOLISTEN_UNLOCK(head);
635 
636 		if (dolog) {
637 			/*
638 			 * Try to print something descriptive about the
639 			 * socket for the error message.
640 			 */
641 			sbuf_new(&descrsb, descrbuf, sizeof(descrbuf),
642 			    SBUF_FIXEDLEN);
643 			switch (head->so_proto->pr_domain->dom_family) {
644 #if defined(INET) || defined(INET6)
645 #ifdef INET
646 			case AF_INET:
647 #endif
648 #ifdef INET6
649 			case AF_INET6:
650 				if (head->so_proto->pr_domain->dom_family ==
651 				    AF_INET6 ||
652 				    (sotoinpcb(head)->inp_inc.inc_flags &
653 				    INC_ISIPV6)) {
654 					ip6_sprintf(addrbuf,
655 					    &sotoinpcb(head)->inp_inc.inc6_laddr);
656 					sbuf_printf(&descrsb, "[%s]", addrbuf);
657 				} else
658 #endif
659 				{
660 #ifdef INET
661 					inet_ntoa_r(
662 					    sotoinpcb(head)->inp_inc.inc_laddr,
663 					    addrbuf);
664 					sbuf_cat(&descrsb, addrbuf);
665 #endif
666 				}
667 				sbuf_printf(&descrsb, ":%hu (proto %u)",
668 				    ntohs(sotoinpcb(head)->inp_inc.inc_lport),
669 				    head->so_proto->pr_protocol);
670 				break;
671 #endif /* INET || INET6 */
672 			case AF_UNIX:
673 				sbuf_cat(&descrsb, localprefix);
674 				if (sotounpcb(head)->unp_addr != NULL)
675 					len =
676 					    sotounpcb(head)->unp_addr->sun_len -
677 					    offsetof(struct sockaddr_un,
678 					    sun_path);
679 				else
680 					len = 0;
681 				if (len > 0)
682 					sbuf_bcat(&descrsb,
683 					    sotounpcb(head)->unp_addr->sun_path,
684 					    len);
685 				else
686 					sbuf_cat(&descrsb, "(unknown)");
687 				break;
688 			}
689 
690 			/*
691 			 * If we can't print something more specific, at least
692 			 * print the domain name.
693 			 */
694 			if (sbuf_finish(&descrsb) != 0 ||
695 			    sbuf_len(&descrsb) <= 0) {
696 				sbuf_clear(&descrsb);
697 				sbuf_cat(&descrsb,
698 				    head->so_proto->pr_domain->dom_name ?:
699 				    "unknown");
700 				sbuf_finish(&descrsb);
701 			}
702 			KASSERT(sbuf_len(&descrsb) > 0,
703 			    ("%s: sbuf creation failed", __func__));
704 			if (head->so_cred == 0) {
705 				log(LOG_DEBUG,
706 			    	"%s: pcb %p (%s): Listen queue overflow: "
707 			    	"%i already in queue awaiting acceptance "
708 			    	"(%d occurrences)\n",
709 			    	__func__, head->so_pcb, sbuf_data(&descrsb),
710 			    	qlen, overcount);
711 			} else {
712 				log(LOG_DEBUG, "%s: pcb %p (%s): Listen queue overflow: "
713 				    "%i already in queue awaiting acceptance "
714 				    "(%d occurrences), euid %d, rgid %d, jail %s\n",
715 				    __func__, head->so_pcb, sbuf_data(&descrsb),
716 				    qlen, overcount,
717 				    head->so_cred->cr_uid, head->so_cred->cr_rgid,
718 				    head->so_cred->cr_prison ?
719 					head->so_cred->cr_prison->pr_name :
720 					"not_jailed");
721 			}
722 			sbuf_delete(&descrsb);
723 
724 			overcount = 0;
725 		}
726 
727 		return (NULL);
728 	}
729 	SOLISTEN_UNLOCK(head);
730 	VNET_ASSERT(head->so_vnet != NULL, ("%s: so %p vnet is NULL",
731 	    __func__, head));
732 	so = soalloc(head->so_vnet);
733 	if (so == NULL) {
734 		log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: "
735 		    "limit reached or out of memory\n",
736 		    __func__, head->so_pcb);
737 		return (NULL);
738 	}
739 	so->so_listen = head;
740 	so->so_type = head->so_type;
741 	so->so_options = head->so_options & ~SO_ACCEPTCONN;
742 	so->so_linger = head->so_linger;
743 	so->so_state = head->so_state;
744 	so->so_fibnum = head->so_fibnum;
745 	so->so_proto = head->so_proto;
746 	so->so_cred = crhold(head->so_cred);
747 #ifdef MAC
748 	mac_socket_newconn(head, so);
749 #endif
750 	knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
751 	    so_rdknl_assert_lock);
752 	knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
753 	    so_wrknl_assert_lock);
754 	VNET_SO_ASSERT(head);
755 	if (soreserve(so, head->sol_sbsnd_hiwat, head->sol_sbrcv_hiwat)) {
756 		sodealloc(so);
757 		log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n",
758 		    __func__, head->so_pcb);
759 		return (NULL);
760 	}
761 	if ((so->so_proto->pr_flags & PR_SOCKBUF) == 0) {
762 		so->so_snd.sb_mtx = &so->so_snd_mtx;
763 		so->so_rcv.sb_mtx = &so->so_rcv_mtx;
764 	}
765 	/* Socket starts with a reference from the listen queue. */
766 	refcount_init(&so->so_count, 1);
767 	if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
768 		MPASS(refcount_release(&so->so_count));
769 		sodealloc(so);
770 		log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n",
771 		    __func__, head->so_pcb);
772 		return (NULL);
773 	}
774 	so->so_rcv.sb_lowat = head->sol_sbrcv_lowat;
775 	so->so_snd.sb_lowat = head->sol_sbsnd_lowat;
776 	so->so_rcv.sb_timeo = head->sol_sbrcv_timeo;
777 	so->so_snd.sb_timeo = head->sol_sbsnd_timeo;
778 	so->so_rcv.sb_flags |= head->sol_sbrcv_flags & SB_AUTOSIZE;
779 	so->so_snd.sb_flags |= head->sol_sbsnd_flags & SB_AUTOSIZE;
780 
781 	SOLISTEN_LOCK(head);
782 	if (head->sol_accept_filter != NULL)
783 		connstatus = 0;
784 	so->so_state |= connstatus;
785 	soref(head); /* A socket on (in)complete queue refs head. */
786 	if (connstatus) {
787 		TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list);
788 		so->so_qstate = SQ_COMP;
789 		head->sol_qlen++;
790 		solisten_wakeup(head);	/* unlocks */
791 	} else {
792 		/*
793 		 * Keep removing sockets from the head until there's room for
794 		 * us to insert on the tail.  In pre-locking revisions, this
795 		 * was a simple if(), but as we could be racing with other
796 		 * threads and soabort() requires dropping locks, we must
797 		 * loop waiting for the condition to be true.
798 		 */
799 		while (head->sol_incqlen > head->sol_qlimit) {
800 			struct socket *sp;
801 
802 			sp = TAILQ_FIRST(&head->sol_incomp);
803 			TAILQ_REMOVE(&head->sol_incomp, sp, so_list);
804 			head->sol_incqlen--;
805 			SOCK_LOCK(sp);
806 			sp->so_qstate = SQ_NONE;
807 			sp->so_listen = NULL;
808 			SOCK_UNLOCK(sp);
809 			sorele_locked(head);	/* does SOLISTEN_UNLOCK, head stays */
810 			soabort(sp);
811 			SOLISTEN_LOCK(head);
812 		}
813 		TAILQ_INSERT_TAIL(&head->sol_incomp, so, so_list);
814 		so->so_qstate = SQ_INCOMP;
815 		head->sol_incqlen++;
816 		SOLISTEN_UNLOCK(head);
817 	}
818 	return (so);
819 }
820 
821 #if defined(SCTP) || defined(SCTP_SUPPORT)
822 /*
823  * Socket part of sctp_peeloff().  Detach a new socket from an
824  * association.  The new socket is returned with a reference.
825  */
826 struct socket *
827 sopeeloff(struct socket *head)
828 {
829 	struct socket *so;
830 
831 	VNET_ASSERT(head->so_vnet != NULL, ("%s:%d so_vnet is NULL, head=%p",
832 	    __func__, __LINE__, head));
833 	so = soalloc(head->so_vnet);
834 	if (so == NULL) {
835 		log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: "
836 		    "limit reached or out of memory\n",
837 		    __func__, head->so_pcb);
838 		return (NULL);
839 	}
840 	so->so_type = head->so_type;
841 	so->so_options = head->so_options;
842 	so->so_linger = head->so_linger;
843 	so->so_state = (head->so_state & SS_NBIO) | SS_ISCONNECTED;
844 	so->so_fibnum = head->so_fibnum;
845 	so->so_proto = head->so_proto;
846 	so->so_cred = crhold(head->so_cred);
847 #ifdef MAC
848 	mac_socket_newconn(head, so);
849 #endif
850 	knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
851 	    so_rdknl_assert_lock);
852 	knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
853 	    so_wrknl_assert_lock);
854 	VNET_SO_ASSERT(head);
855 	if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat)) {
856 		sodealloc(so);
857 		log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n",
858 		    __func__, head->so_pcb);
859 		return (NULL);
860 	}
861 	if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
862 		sodealloc(so);
863 		log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n",
864 		    __func__, head->so_pcb);
865 		return (NULL);
866 	}
867 	so->so_rcv.sb_lowat = head->so_rcv.sb_lowat;
868 	so->so_snd.sb_lowat = head->so_snd.sb_lowat;
869 	so->so_rcv.sb_timeo = head->so_rcv.sb_timeo;
870 	so->so_snd.sb_timeo = head->so_snd.sb_timeo;
871 	so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE;
872 	so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE;
873 
874 	soref(so);
875 
876 	return (so);
877 }
878 #endif	/* SCTP */
879 
880 int
881 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
882 {
883 	int error;
884 
885 	CURVNET_SET(so->so_vnet);
886 	error = (*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td);
887 	CURVNET_RESTORE();
888 	return (error);
889 }
890 
891 int
892 sobindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
893 {
894 	int error;
895 
896 	CURVNET_SET(so->so_vnet);
897 	error = (*so->so_proto->pr_usrreqs->pru_bindat)(fd, so, nam, td);
898 	CURVNET_RESTORE();
899 	return (error);
900 }
901 
902 /*
903  * solisten() transitions a socket from a non-listening state to a listening
904  * state, but can also be used to update the listen queue depth on an
905  * existing listen socket.  The protocol will call back into the sockets
906  * layer using solisten_proto_check() and solisten_proto() to check and set
907  * socket-layer listen state.  Call backs are used so that the protocol can
908  * acquire both protocol and socket layer locks in whatever order is required
909  * by the protocol.
910  *
911  * Protocol implementors are advised to hold the socket lock across the
912  * socket-layer test and set to avoid races at the socket layer.
913  */
914 int
915 solisten(struct socket *so, int backlog, struct thread *td)
916 {
917 	int error;
918 
919 	CURVNET_SET(so->so_vnet);
920 	error = (*so->so_proto->pr_usrreqs->pru_listen)(so, backlog, td);
921 	CURVNET_RESTORE();
922 	return (error);
923 }
924 
925 /*
926  * Prepare for a call to solisten_proto().  Acquire all socket buffer locks in
927  * order to interlock with socket I/O.
928  */
929 int
930 solisten_proto_check(struct socket *so)
931 {
932 	SOCK_LOCK_ASSERT(so);
933 
934 	if ((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
935 	    SS_ISDISCONNECTING)) != 0)
936 		return (EINVAL);
937 
938 	/*
939 	 * Sleeping is not permitted here, so simply fail if userspace is
940 	 * attempting to transmit or receive on the socket.  This kind of
941 	 * transient failure is not ideal, but it should occur only if userspace
942 	 * is misusing the socket interfaces.
943 	 */
944 	if (!sx_try_xlock(&so->so_snd_sx))
945 		return (EAGAIN);
946 	if (!sx_try_xlock(&so->so_rcv_sx)) {
947 		sx_xunlock(&so->so_snd_sx);
948 		return (EAGAIN);
949 	}
950 	mtx_lock(&so->so_snd_mtx);
951 	mtx_lock(&so->so_rcv_mtx);
952 
953 	/* Interlock with soo_aio_queue(). */
954 	if (!SOLISTENING(so) &&
955 	   ((so->so_snd.sb_flags & (SB_AIO | SB_AIO_RUNNING)) != 0 ||
956 	   (so->so_rcv.sb_flags & (SB_AIO | SB_AIO_RUNNING)) != 0)) {
957 		solisten_proto_abort(so);
958 		return (EINVAL);
959 	}
960 	return (0);
961 }
962 
963 /*
964  * Undo the setup done by solisten_proto_check().
965  */
966 void
967 solisten_proto_abort(struct socket *so)
968 {
969 	mtx_unlock(&so->so_snd_mtx);
970 	mtx_unlock(&so->so_rcv_mtx);
971 	sx_xunlock(&so->so_snd_sx);
972 	sx_xunlock(&so->so_rcv_sx);
973 }
974 
975 void
976 solisten_proto(struct socket *so, int backlog)
977 {
978 	int sbrcv_lowat, sbsnd_lowat;
979 	u_int sbrcv_hiwat, sbsnd_hiwat;
980 	short sbrcv_flags, sbsnd_flags;
981 	sbintime_t sbrcv_timeo, sbsnd_timeo;
982 
983 	SOCK_LOCK_ASSERT(so);
984 	KASSERT((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
985 	    SS_ISDISCONNECTING)) == 0,
986 	    ("%s: bad socket state %p", __func__, so));
987 
988 	if (SOLISTENING(so))
989 		goto listening;
990 
991 	/*
992 	 * Change this socket to listening state.
993 	 */
994 	sbrcv_lowat = so->so_rcv.sb_lowat;
995 	sbsnd_lowat = so->so_snd.sb_lowat;
996 	sbrcv_hiwat = so->so_rcv.sb_hiwat;
997 	sbsnd_hiwat = so->so_snd.sb_hiwat;
998 	sbrcv_flags = so->so_rcv.sb_flags;
999 	sbsnd_flags = so->so_snd.sb_flags;
1000 	sbrcv_timeo = so->so_rcv.sb_timeo;
1001 	sbsnd_timeo = so->so_snd.sb_timeo;
1002 
1003 	sbdestroy(so, SO_SND);
1004 	sbdestroy(so, SO_RCV);
1005 
1006 #ifdef INVARIANTS
1007 	bzero(&so->so_rcv,
1008 	    sizeof(struct socket) - offsetof(struct socket, so_rcv));
1009 #endif
1010 
1011 	so->sol_sbrcv_lowat = sbrcv_lowat;
1012 	so->sol_sbsnd_lowat = sbsnd_lowat;
1013 	so->sol_sbrcv_hiwat = sbrcv_hiwat;
1014 	so->sol_sbsnd_hiwat = sbsnd_hiwat;
1015 	so->sol_sbrcv_flags = sbrcv_flags;
1016 	so->sol_sbsnd_flags = sbsnd_flags;
1017 	so->sol_sbrcv_timeo = sbrcv_timeo;
1018 	so->sol_sbsnd_timeo = sbsnd_timeo;
1019 
1020 	so->sol_qlen = so->sol_incqlen = 0;
1021 	TAILQ_INIT(&so->sol_incomp);
1022 	TAILQ_INIT(&so->sol_comp);
1023 
1024 	so->sol_accept_filter = NULL;
1025 	so->sol_accept_filter_arg = NULL;
1026 	so->sol_accept_filter_str = NULL;
1027 
1028 	so->sol_upcall = NULL;
1029 	so->sol_upcallarg = NULL;
1030 
1031 	so->so_options |= SO_ACCEPTCONN;
1032 
1033 listening:
1034 	if (backlog < 0 || backlog > somaxconn)
1035 		backlog = somaxconn;
1036 	so->sol_qlimit = backlog;
1037 
1038 	mtx_unlock(&so->so_snd_mtx);
1039 	mtx_unlock(&so->so_rcv_mtx);
1040 	sx_xunlock(&so->so_snd_sx);
1041 	sx_xunlock(&so->so_rcv_sx);
1042 }
1043 
1044 /*
1045  * Wakeup listeners/subsystems once we have a complete connection.
1046  * Enters with lock, returns unlocked.
1047  */
1048 void
1049 solisten_wakeup(struct socket *sol)
1050 {
1051 
1052 	if (sol->sol_upcall != NULL)
1053 		(void )sol->sol_upcall(sol, sol->sol_upcallarg, M_NOWAIT);
1054 	else {
1055 		selwakeuppri(&sol->so_rdsel, PSOCK);
1056 		KNOTE_LOCKED(&sol->so_rdsel.si_note, 0);
1057 	}
1058 	SOLISTEN_UNLOCK(sol);
1059 	wakeup_one(&sol->sol_comp);
1060 	if ((sol->so_state & SS_ASYNC) && sol->so_sigio != NULL)
1061 		pgsigio(&sol->so_sigio, SIGIO, 0);
1062 }
1063 
1064 /*
1065  * Return single connection off a listening socket queue.  Main consumer of
1066  * the function is kern_accept4().  Some modules, that do their own accept
1067  * management also use the function.  The socket reference held by the
1068  * listen queue is handed to the caller.
1069  *
1070  * Listening socket must be locked on entry and is returned unlocked on
1071  * return.
1072  * The flags argument is set of accept4(2) flags and ACCEPT4_INHERIT.
1073  */
1074 int
1075 solisten_dequeue(struct socket *head, struct socket **ret, int flags)
1076 {
1077 	struct socket *so;
1078 	int error;
1079 
1080 	SOLISTEN_LOCK_ASSERT(head);
1081 
1082 	while (!(head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp) &&
1083 	    head->so_error == 0) {
1084 		error = msleep(&head->sol_comp, SOCK_MTX(head), PSOCK | PCATCH,
1085 		    "accept", 0);
1086 		if (error != 0) {
1087 			SOLISTEN_UNLOCK(head);
1088 			return (error);
1089 		}
1090 	}
1091 	if (head->so_error) {
1092 		error = head->so_error;
1093 		head->so_error = 0;
1094 	} else if ((head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp))
1095 		error = EWOULDBLOCK;
1096 	else
1097 		error = 0;
1098 	if (error) {
1099 		SOLISTEN_UNLOCK(head);
1100 		return (error);
1101 	}
1102 	so = TAILQ_FIRST(&head->sol_comp);
1103 	SOCK_LOCK(so);
1104 	KASSERT(so->so_qstate == SQ_COMP,
1105 	    ("%s: so %p not SQ_COMP", __func__, so));
1106 	head->sol_qlen--;
1107 	so->so_qstate = SQ_NONE;
1108 	so->so_listen = NULL;
1109 	TAILQ_REMOVE(&head->sol_comp, so, so_list);
1110 	if (flags & ACCEPT4_INHERIT)
1111 		so->so_state |= (head->so_state & SS_NBIO);
1112 	else
1113 		so->so_state |= (flags & SOCK_NONBLOCK) ? SS_NBIO : 0;
1114 	SOCK_UNLOCK(so);
1115 	sorele_locked(head);
1116 
1117 	*ret = so;
1118 	return (0);
1119 }
1120 
1121 /*
1122  * Free socket upon release of the very last reference.
1123  */
1124 static void
1125 sofree(struct socket *so)
1126 {
1127 	struct protosw *pr = so->so_proto;
1128 
1129 	SOCK_LOCK_ASSERT(so);
1130 	KASSERT(refcount_load(&so->so_count) == 0,
1131 	    ("%s: so %p has references", __func__, so));
1132 	KASSERT(SOLISTENING(so) || so->so_qstate == SQ_NONE,
1133 	    ("%s: so %p is on listen queue", __func__, so));
1134 
1135 	SOCK_UNLOCK(so);
1136 
1137 	if (so->so_dtor != NULL)
1138 		so->so_dtor(so);
1139 
1140 	VNET_SO_ASSERT(so);
1141 	if ((pr->pr_flags & PR_RIGHTS) && !SOLISTENING(so)) {
1142 		MPASS(pr->pr_domain->dom_dispose != NULL);
1143 		(*pr->pr_domain->dom_dispose)(so);
1144 	}
1145 	if (pr->pr_usrreqs->pru_detach != NULL)
1146 		(*pr->pr_usrreqs->pru_detach)(so);
1147 
1148 	/*
1149 	 * From this point on, we assume that no other references to this
1150 	 * socket exist anywhere else in the stack.  Therefore, no locks need
1151 	 * to be acquired or held.
1152 	 */
1153 	if (!(pr->pr_flags & PR_SOCKBUF) && !SOLISTENING(so)) {
1154 		sbdestroy(so, SO_SND);
1155 		sbdestroy(so, SO_RCV);
1156 	}
1157 	seldrain(&so->so_rdsel);
1158 	seldrain(&so->so_wrsel);
1159 	knlist_destroy(&so->so_rdsel.si_note);
1160 	knlist_destroy(&so->so_wrsel.si_note);
1161 	sodealloc(so);
1162 }
1163 
1164 /*
1165  * Release a reference on a socket while holding the socket lock.
1166  * Unlocks the socket lock before returning.
1167  */
1168 void
1169 sorele_locked(struct socket *so)
1170 {
1171 	SOCK_LOCK_ASSERT(so);
1172 	if (refcount_release(&so->so_count))
1173 		sofree(so);
1174 	else
1175 		SOCK_UNLOCK(so);
1176 }
1177 
1178 /*
1179  * Close a socket on last file table reference removal.  Initiate disconnect
1180  * if connected.  Free socket when disconnect complete.
1181  *
1182  * This function will sorele() the socket.  Note that soclose() may be called
1183  * prior to the ref count reaching zero.  The actual socket structure will
1184  * not be freed until the ref count reaches zero.
1185  */
1186 int
1187 soclose(struct socket *so)
1188 {
1189 	struct accept_queue lqueue;
1190 	int error = 0;
1191 	bool listening, last __diagused;
1192 
1193 	CURVNET_SET(so->so_vnet);
1194 	funsetown(&so->so_sigio);
1195 	if (so->so_state & SS_ISCONNECTED) {
1196 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
1197 			error = sodisconnect(so);
1198 			if (error) {
1199 				if (error == ENOTCONN)
1200 					error = 0;
1201 				goto drop;
1202 			}
1203 		}
1204 
1205 		if ((so->so_options & SO_LINGER) != 0 && so->so_linger != 0) {
1206 			if ((so->so_state & SS_ISDISCONNECTING) &&
1207 			    (so->so_state & SS_NBIO))
1208 				goto drop;
1209 			while (so->so_state & SS_ISCONNECTED) {
1210 				error = tsleep(&so->so_timeo,
1211 				    PSOCK | PCATCH, "soclos",
1212 				    so->so_linger * hz);
1213 				if (error)
1214 					break;
1215 			}
1216 		}
1217 	}
1218 
1219 drop:
1220 	if (so->so_proto->pr_usrreqs->pru_close != NULL)
1221 		(*so->so_proto->pr_usrreqs->pru_close)(so);
1222 
1223 	SOCK_LOCK(so);
1224 	if ((listening = SOLISTENING(so))) {
1225 		struct socket *sp;
1226 
1227 		TAILQ_INIT(&lqueue);
1228 		TAILQ_SWAP(&lqueue, &so->sol_incomp, socket, so_list);
1229 		TAILQ_CONCAT(&lqueue, &so->sol_comp, so_list);
1230 
1231 		so->sol_qlen = so->sol_incqlen = 0;
1232 
1233 		TAILQ_FOREACH(sp, &lqueue, so_list) {
1234 			SOCK_LOCK(sp);
1235 			sp->so_qstate = SQ_NONE;
1236 			sp->so_listen = NULL;
1237 			SOCK_UNLOCK(sp);
1238 			last = refcount_release(&so->so_count);
1239 			KASSERT(!last, ("%s: released last reference for %p",
1240 			    __func__, so));
1241 		}
1242 	}
1243 	sorele_locked(so);
1244 	if (listening) {
1245 		struct socket *sp, *tsp;
1246 
1247 		TAILQ_FOREACH_SAFE(sp, &lqueue, so_list, tsp)
1248 			soabort(sp);
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.  Likely this thread holds the last
1262  * reference on the socket and soabort() will proceed with sofree().  But
1263  * it might be not the last, as the sockets on the listen queues are seen
1264  * from the protocol side.
1265  *
1266  * This interface will call into the protocol code, so must not be called
1267  * with any socket locks held.  Protocols do call it while holding their own
1268  * recursible protocol mutexes, but this is something that should be subject
1269  * to review in the future.
1270  *
1271  * Usually socket should have a single reference left, but this is not a
1272  * requirement.  In the past, when we have had named references for file
1273  * descriptor and protocol, we asserted that none of them are being held.
1274  */
1275 void
1276 soabort(struct socket *so)
1277 {
1278 
1279 	VNET_SO_ASSERT(so);
1280 
1281 	if (so->so_proto->pr_usrreqs->pru_abort != NULL)
1282 		(*so->so_proto->pr_usrreqs->pru_abort)(so);
1283 	SOCK_LOCK(so);
1284 	sorele_locked(so);
1285 }
1286 
1287 int
1288 soaccept(struct socket *so, struct sockaddr **nam)
1289 {
1290 	int error;
1291 
1292 	CURVNET_SET(so->so_vnet);
1293 	error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam);
1294 	CURVNET_RESTORE();
1295 	return (error);
1296 }
1297 
1298 int
1299 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
1300 {
1301 
1302 	return (soconnectat(AT_FDCWD, so, nam, td));
1303 }
1304 
1305 int
1306 soconnectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
1307 {
1308 	int error;
1309 
1310 	CURVNET_SET(so->so_vnet);
1311 	/*
1312 	 * If protocol is connection-based, can only connect once.
1313 	 * Otherwise, if connected, try to disconnect first.  This allows
1314 	 * user to disconnect by connecting to, e.g., a null address.
1315 	 */
1316 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
1317 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
1318 	    (error = sodisconnect(so)))) {
1319 		error = EISCONN;
1320 	} else {
1321 		/*
1322 		 * Prevent accumulated error from previous connection from
1323 		 * biting us.
1324 		 */
1325 		so->so_error = 0;
1326 		if (fd == AT_FDCWD) {
1327 			error = (*so->so_proto->pr_usrreqs->pru_connect)(so,
1328 			    nam, td);
1329 		} else {
1330 			error = (*so->so_proto->pr_usrreqs->pru_connectat)(fd,
1331 			    so, nam, td);
1332 		}
1333 	}
1334 	CURVNET_RESTORE();
1335 
1336 	return (error);
1337 }
1338 
1339 int
1340 soconnect2(struct socket *so1, struct socket *so2)
1341 {
1342 	int error;
1343 
1344 	CURVNET_SET(so1->so_vnet);
1345 	error = (*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2);
1346 	CURVNET_RESTORE();
1347 	return (error);
1348 }
1349 
1350 int
1351 sodisconnect(struct socket *so)
1352 {
1353 	int error;
1354 
1355 	if ((so->so_state & SS_ISCONNECTED) == 0)
1356 		return (ENOTCONN);
1357 	if (so->so_state & SS_ISDISCONNECTING)
1358 		return (EALREADY);
1359 	VNET_SO_ASSERT(so);
1360 	error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so);
1361 	return (error);
1362 }
1363 
1364 int
1365 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio,
1366     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1367 {
1368 	long space;
1369 	ssize_t resid;
1370 	int clen = 0, error, dontroute;
1371 
1372 	KASSERT(so->so_type == SOCK_DGRAM, ("sosend_dgram: !SOCK_DGRAM"));
1373 	KASSERT(so->so_proto->pr_flags & PR_ATOMIC,
1374 	    ("sosend_dgram: !PR_ATOMIC"));
1375 
1376 	if (uio != NULL)
1377 		resid = uio->uio_resid;
1378 	else
1379 		resid = top->m_pkthdr.len;
1380 	/*
1381 	 * In theory resid should be unsigned.  However, space must be
1382 	 * signed, as it might be less than 0 if we over-committed, and we
1383 	 * must use a signed comparison of space and resid.  On the other
1384 	 * hand, a negative resid causes us to loop sending 0-length
1385 	 * segments to the protocol.
1386 	 */
1387 	if (resid < 0) {
1388 		error = EINVAL;
1389 		goto out;
1390 	}
1391 
1392 	dontroute =
1393 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0;
1394 	if (td != NULL)
1395 		td->td_ru.ru_msgsnd++;
1396 	if (control != NULL)
1397 		clen = control->m_len;
1398 
1399 	SOCKBUF_LOCK(&so->so_snd);
1400 	if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1401 		SOCKBUF_UNLOCK(&so->so_snd);
1402 		error = EPIPE;
1403 		goto out;
1404 	}
1405 	if (so->so_error) {
1406 		error = so->so_error;
1407 		so->so_error = 0;
1408 		SOCKBUF_UNLOCK(&so->so_snd);
1409 		goto out;
1410 	}
1411 	if ((so->so_state & SS_ISCONNECTED) == 0) {
1412 		/*
1413 		 * `sendto' and `sendmsg' is allowed on a connection-based
1414 		 * socket if it supports implied connect.  Return ENOTCONN if
1415 		 * not connected and no address is supplied.
1416 		 */
1417 		if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1418 		    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1419 			if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1420 			    !(resid == 0 && clen != 0)) {
1421 				SOCKBUF_UNLOCK(&so->so_snd);
1422 				error = ENOTCONN;
1423 				goto out;
1424 			}
1425 		} else if (addr == NULL) {
1426 			if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1427 				error = ENOTCONN;
1428 			else
1429 				error = EDESTADDRREQ;
1430 			SOCKBUF_UNLOCK(&so->so_snd);
1431 			goto out;
1432 		}
1433 	}
1434 
1435 	/*
1436 	 * Do we need MSG_OOB support in SOCK_DGRAM?  Signs here may be a
1437 	 * problem and need fixing.
1438 	 */
1439 	space = sbspace(&so->so_snd);
1440 	if (flags & MSG_OOB)
1441 		space += 1024;
1442 	space -= clen;
1443 	SOCKBUF_UNLOCK(&so->so_snd);
1444 	if (resid > space) {
1445 		error = EMSGSIZE;
1446 		goto out;
1447 	}
1448 	if (uio == NULL) {
1449 		resid = 0;
1450 		if (flags & MSG_EOR)
1451 			top->m_flags |= M_EOR;
1452 	} else {
1453 		/*
1454 		 * Copy the data from userland into a mbuf chain.
1455 		 * If no data is to be copied in, a single empty mbuf
1456 		 * is returned.
1457 		 */
1458 		top = m_uiotombuf(uio, M_WAITOK, space, max_hdr,
1459 		    (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0)));
1460 		if (top == NULL) {
1461 			error = EFAULT;	/* only possible error */
1462 			goto out;
1463 		}
1464 		space -= resid - uio->uio_resid;
1465 		resid = uio->uio_resid;
1466 	}
1467 	KASSERT(resid == 0, ("sosend_dgram: resid != 0"));
1468 	/*
1469 	 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock
1470 	 * than with.
1471 	 */
1472 	if (dontroute) {
1473 		SOCK_LOCK(so);
1474 		so->so_options |= SO_DONTROUTE;
1475 		SOCK_UNLOCK(so);
1476 	}
1477 	/*
1478 	 * XXX all the SBS_CANTSENDMORE checks previously done could be out
1479 	 * of date.  We could have received a reset packet in an interrupt or
1480 	 * maybe we slept while doing page faults in uiomove() etc.  We could
1481 	 * probably recheck again inside the locking protection here, but
1482 	 * there are probably other places that this also happens.  We must
1483 	 * rethink this.
1484 	 */
1485 	VNET_SO_ASSERT(so);
1486 	error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1487 	    (flags & MSG_OOB) ? PRUS_OOB :
1488 	/*
1489 	 * If the user set MSG_EOF, the protocol understands this flag and
1490 	 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND.
1491 	 */
1492 	    ((flags & MSG_EOF) &&
1493 	     (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1494 	     (resid <= 0)) ?
1495 		PRUS_EOF :
1496 		/* If there is more to send set PRUS_MORETOCOME */
1497 		(flags & MSG_MORETOCOME) ||
1498 		(resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1499 		top, addr, control, td);
1500 	if (dontroute) {
1501 		SOCK_LOCK(so);
1502 		so->so_options &= ~SO_DONTROUTE;
1503 		SOCK_UNLOCK(so);
1504 	}
1505 	clen = 0;
1506 	control = NULL;
1507 	top = NULL;
1508 out:
1509 	if (top != NULL)
1510 		m_freem(top);
1511 	if (control != NULL)
1512 		m_freem(control);
1513 	return (error);
1514 }
1515 
1516 /*
1517  * Send on a socket.  If send must go all at once and message is larger than
1518  * send buffering, then hard error.  Lock against other senders.  If must go
1519  * all at once and not enough room now, then inform user that this would
1520  * block and do nothing.  Otherwise, if nonblocking, send as much as
1521  * possible.  The data to be sent is described by "uio" if nonzero, otherwise
1522  * by the mbuf chain "top" (which must be null if uio is not).  Data provided
1523  * in mbuf chain must be small enough to send all at once.
1524  *
1525  * Returns nonzero on error, timeout or signal; callers must check for short
1526  * counts if EINTR/ERESTART are returned.  Data and control buffers are freed
1527  * on return.
1528  */
1529 int
1530 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio,
1531     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1532 {
1533 	long space;
1534 	ssize_t resid;
1535 	int clen = 0, error, dontroute;
1536 	int atomic = sosendallatonce(so) || top;
1537 	int pru_flag;
1538 #ifdef KERN_TLS
1539 	struct ktls_session *tls;
1540 	int tls_enq_cnt, tls_pruflag;
1541 	uint8_t tls_rtype;
1542 
1543 	tls = NULL;
1544 	tls_rtype = TLS_RLTYPE_APP;
1545 #endif
1546 	if (uio != NULL)
1547 		resid = uio->uio_resid;
1548 	else if ((top->m_flags & M_PKTHDR) != 0)
1549 		resid = top->m_pkthdr.len;
1550 	else
1551 		resid = m_length(top, NULL);
1552 	/*
1553 	 * In theory resid should be unsigned.  However, space must be
1554 	 * signed, as it might be less than 0 if we over-committed, and we
1555 	 * must use a signed comparison of space and resid.  On the other
1556 	 * hand, a negative resid causes us to loop sending 0-length
1557 	 * segments to the protocol.
1558 	 *
1559 	 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
1560 	 * type sockets since that's an error.
1561 	 */
1562 	if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
1563 		error = EINVAL;
1564 		goto out;
1565 	}
1566 
1567 	dontroute =
1568 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
1569 	    (so->so_proto->pr_flags & PR_ATOMIC);
1570 	if (td != NULL)
1571 		td->td_ru.ru_msgsnd++;
1572 	if (control != NULL)
1573 		clen = control->m_len;
1574 
1575 	error = SOCK_IO_SEND_LOCK(so, SBLOCKWAIT(flags));
1576 	if (error)
1577 		goto out;
1578 
1579 #ifdef KERN_TLS
1580 	tls_pruflag = 0;
1581 	tls = ktls_hold(so->so_snd.sb_tls_info);
1582 	if (tls != NULL) {
1583 		if (tls->mode == TCP_TLS_MODE_SW)
1584 			tls_pruflag = PRUS_NOTREADY;
1585 
1586 		if (control != NULL) {
1587 			struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1588 
1589 			if (clen >= sizeof(*cm) &&
1590 			    cm->cmsg_type == TLS_SET_RECORD_TYPE) {
1591 				tls_rtype = *((uint8_t *)CMSG_DATA(cm));
1592 				clen = 0;
1593 				m_freem(control);
1594 				control = NULL;
1595 				atomic = 1;
1596 			}
1597 		}
1598 
1599 		if (resid == 0 && !ktls_permit_empty_frames(tls)) {
1600 			error = EINVAL;
1601 			goto release;
1602 		}
1603 	}
1604 #endif
1605 
1606 restart:
1607 	do {
1608 		SOCKBUF_LOCK(&so->so_snd);
1609 		if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1610 			SOCKBUF_UNLOCK(&so->so_snd);
1611 			error = EPIPE;
1612 			goto release;
1613 		}
1614 		if (so->so_error) {
1615 			error = so->so_error;
1616 			so->so_error = 0;
1617 			SOCKBUF_UNLOCK(&so->so_snd);
1618 			goto release;
1619 		}
1620 		if ((so->so_state & SS_ISCONNECTED) == 0) {
1621 			/*
1622 			 * `sendto' and `sendmsg' is allowed on a connection-
1623 			 * based socket if it supports implied connect.
1624 			 * Return ENOTCONN if not connected and no address is
1625 			 * supplied.
1626 			 */
1627 			if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1628 			    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1629 				if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1630 				    !(resid == 0 && clen != 0)) {
1631 					SOCKBUF_UNLOCK(&so->so_snd);
1632 					error = ENOTCONN;
1633 					goto release;
1634 				}
1635 			} else if (addr == NULL) {
1636 				SOCKBUF_UNLOCK(&so->so_snd);
1637 				if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1638 					error = ENOTCONN;
1639 				else
1640 					error = EDESTADDRREQ;
1641 				goto release;
1642 			}
1643 		}
1644 		space = sbspace(&so->so_snd);
1645 		if (flags & MSG_OOB)
1646 			space += 1024;
1647 		if ((atomic && resid > so->so_snd.sb_hiwat) ||
1648 		    clen > so->so_snd.sb_hiwat) {
1649 			SOCKBUF_UNLOCK(&so->so_snd);
1650 			error = EMSGSIZE;
1651 			goto release;
1652 		}
1653 		if (space < resid + clen &&
1654 		    (atomic || space < so->so_snd.sb_lowat || space < clen)) {
1655 			if ((so->so_state & SS_NBIO) ||
1656 			    (flags & (MSG_NBIO | MSG_DONTWAIT)) != 0) {
1657 				SOCKBUF_UNLOCK(&so->so_snd);
1658 				error = EWOULDBLOCK;
1659 				goto release;
1660 			}
1661 			error = sbwait(so, SO_SND);
1662 			SOCKBUF_UNLOCK(&so->so_snd);
1663 			if (error)
1664 				goto release;
1665 			goto restart;
1666 		}
1667 		SOCKBUF_UNLOCK(&so->so_snd);
1668 		space -= clen;
1669 		do {
1670 			if (uio == NULL) {
1671 				resid = 0;
1672 				if (flags & MSG_EOR)
1673 					top->m_flags |= M_EOR;
1674 #ifdef KERN_TLS
1675 				if (tls != NULL) {
1676 					ktls_frame(top, tls, &tls_enq_cnt,
1677 					    tls_rtype);
1678 					tls_rtype = TLS_RLTYPE_APP;
1679 				}
1680 #endif
1681 			} else {
1682 				/*
1683 				 * Copy the data from userland into a mbuf
1684 				 * chain.  If resid is 0, which can happen
1685 				 * only if we have control to send, then
1686 				 * a single empty mbuf is returned.  This
1687 				 * is a workaround to prevent protocol send
1688 				 * methods to panic.
1689 				 */
1690 #ifdef KERN_TLS
1691 				if (tls != NULL) {
1692 					top = m_uiotombuf(uio, M_WAITOK, space,
1693 					    tls->params.max_frame_len,
1694 					    M_EXTPG |
1695 					    ((flags & MSG_EOR) ? M_EOR : 0));
1696 					if (top != NULL) {
1697 						ktls_frame(top, tls,
1698 						    &tls_enq_cnt, tls_rtype);
1699 					}
1700 					tls_rtype = TLS_RLTYPE_APP;
1701 				} else
1702 #endif
1703 					top = m_uiotombuf(uio, M_WAITOK, space,
1704 					    (atomic ? max_hdr : 0),
1705 					    (atomic ? M_PKTHDR : 0) |
1706 					    ((flags & MSG_EOR) ? M_EOR : 0));
1707 				if (top == NULL) {
1708 					error = EFAULT; /* only possible error */
1709 					goto release;
1710 				}
1711 				space -= resid - uio->uio_resid;
1712 				resid = uio->uio_resid;
1713 			}
1714 			if (dontroute) {
1715 				SOCK_LOCK(so);
1716 				so->so_options |= SO_DONTROUTE;
1717 				SOCK_UNLOCK(so);
1718 			}
1719 			/*
1720 			 * XXX all the SBS_CANTSENDMORE checks previously
1721 			 * done could be out of date.  We could have received
1722 			 * a reset packet in an interrupt or maybe we slept
1723 			 * while doing page faults in uiomove() etc.  We
1724 			 * could probably recheck again inside the locking
1725 			 * protection here, but there are probably other
1726 			 * places that this also happens.  We must rethink
1727 			 * this.
1728 			 */
1729 			VNET_SO_ASSERT(so);
1730 
1731 			pru_flag = (flags & MSG_OOB) ? PRUS_OOB :
1732 			/*
1733 			 * If the user set MSG_EOF, the protocol understands
1734 			 * this flag and nothing left to send then use
1735 			 * PRU_SEND_EOF instead of PRU_SEND.
1736 			 */
1737 			    ((flags & MSG_EOF) &&
1738 			     (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1739 			     (resid <= 0)) ?
1740 				PRUS_EOF :
1741 			/* If there is more to send set PRUS_MORETOCOME. */
1742 			    (flags & MSG_MORETOCOME) ||
1743 			    (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0;
1744 
1745 #ifdef KERN_TLS
1746 			pru_flag |= tls_pruflag;
1747 #endif
1748 
1749 			error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1750 			    pru_flag, top, addr, control, td);
1751 
1752 			if (dontroute) {
1753 				SOCK_LOCK(so);
1754 				so->so_options &= ~SO_DONTROUTE;
1755 				SOCK_UNLOCK(so);
1756 			}
1757 
1758 #ifdef KERN_TLS
1759 			if (tls != NULL && tls->mode == TCP_TLS_MODE_SW) {
1760 				if (error != 0) {
1761 					m_freem(top);
1762 					top = NULL;
1763 				} else {
1764 					soref(so);
1765 					ktls_enqueue(top, so, tls_enq_cnt);
1766 				}
1767 			}
1768 #endif
1769 			clen = 0;
1770 			control = NULL;
1771 			top = NULL;
1772 			if (error)
1773 				goto release;
1774 		} while (resid && space > 0);
1775 	} while (resid);
1776 
1777 release:
1778 	SOCK_IO_SEND_UNLOCK(so);
1779 out:
1780 #ifdef KERN_TLS
1781 	if (tls != NULL)
1782 		ktls_free(tls);
1783 #endif
1784 	if (top != NULL)
1785 		m_freem(top);
1786 	if (control != NULL)
1787 		m_freem(control);
1788 	return (error);
1789 }
1790 
1791 int
1792 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
1793     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1794 {
1795 	int error;
1796 
1797 	CURVNET_SET(so->so_vnet);
1798 	error = so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio,
1799 	    top, control, flags, td);
1800 	CURVNET_RESTORE();
1801 	return (error);
1802 }
1803 
1804 /*
1805  * The part of soreceive() that implements reading non-inline out-of-band
1806  * data from a socket.  For more complete comments, see soreceive(), from
1807  * which this code originated.
1808  *
1809  * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
1810  * unable to return an mbuf chain to the caller.
1811  */
1812 static int
1813 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags)
1814 {
1815 	struct protosw *pr = so->so_proto;
1816 	struct mbuf *m;
1817 	int error;
1818 
1819 	KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
1820 	VNET_SO_ASSERT(so);
1821 
1822 	m = m_get(M_WAITOK, MT_DATA);
1823 	error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK);
1824 	if (error)
1825 		goto bad;
1826 	do {
1827 		error = uiomove(mtod(m, void *),
1828 		    (int) min(uio->uio_resid, m->m_len), uio);
1829 		m = m_free(m);
1830 	} while (uio->uio_resid && error == 0 && m);
1831 bad:
1832 	if (m != NULL)
1833 		m_freem(m);
1834 	return (error);
1835 }
1836 
1837 /*
1838  * Following replacement or removal of the first mbuf on the first mbuf chain
1839  * of a socket buffer, push necessary state changes back into the socket
1840  * buffer so that other consumers see the values consistently.  'nextrecord'
1841  * is the callers locally stored value of the original value of
1842  * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes.
1843  * NOTE: 'nextrecord' may be NULL.
1844  */
1845 static __inline void
1846 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord)
1847 {
1848 
1849 	SOCKBUF_LOCK_ASSERT(sb);
1850 	/*
1851 	 * First, update for the new value of nextrecord.  If necessary, make
1852 	 * it the first record.
1853 	 */
1854 	if (sb->sb_mb != NULL)
1855 		sb->sb_mb->m_nextpkt = nextrecord;
1856 	else
1857 		sb->sb_mb = nextrecord;
1858 
1859 	/*
1860 	 * Now update any dependent socket buffer fields to reflect the new
1861 	 * state.  This is an expanded inline of SB_EMPTY_FIXUP(), with the
1862 	 * addition of a second clause that takes care of the case where
1863 	 * sb_mb has been updated, but remains the last record.
1864 	 */
1865 	if (sb->sb_mb == NULL) {
1866 		sb->sb_mbtail = NULL;
1867 		sb->sb_lastrecord = NULL;
1868 	} else if (sb->sb_mb->m_nextpkt == NULL)
1869 		sb->sb_lastrecord = sb->sb_mb;
1870 }
1871 
1872 /*
1873  * Implement receive operations on a socket.  We depend on the way that
1874  * records are added to the sockbuf by sbappend.  In particular, each record
1875  * (mbufs linked through m_next) must begin with an address if the protocol
1876  * so specifies, followed by an optional mbuf or mbufs containing ancillary
1877  * data, and then zero or more mbufs of data.  In order to allow parallelism
1878  * between network receive and copying to user space, as well as avoid
1879  * sleeping with a mutex held, we release the socket buffer mutex during the
1880  * user space copy.  Although the sockbuf is locked, new data may still be
1881  * appended, and thus we must maintain consistency of the sockbuf during that
1882  * time.
1883  *
1884  * The caller may receive the data as a single mbuf chain by supplying an
1885  * mbuf **mp0 for use in returning the chain.  The uio is then used only for
1886  * the count in uio_resid.
1887  */
1888 int
1889 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio,
1890     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1891 {
1892 	struct mbuf *m, **mp;
1893 	int flags, error, offset;
1894 	ssize_t len;
1895 	struct protosw *pr = so->so_proto;
1896 	struct mbuf *nextrecord;
1897 	int moff, type = 0;
1898 	ssize_t orig_resid = uio->uio_resid;
1899 	bool report_real_len = false;
1900 
1901 	mp = mp0;
1902 	if (psa != NULL)
1903 		*psa = NULL;
1904 	if (controlp != NULL)
1905 		*controlp = NULL;
1906 	if (flagsp != NULL) {
1907 		report_real_len = *flagsp & MSG_TRUNC;
1908 		*flagsp &= ~MSG_TRUNC;
1909 		flags = *flagsp &~ MSG_EOR;
1910 	} else
1911 		flags = 0;
1912 	if (flags & MSG_OOB)
1913 		return (soreceive_rcvoob(so, uio, flags));
1914 	if (mp != NULL)
1915 		*mp = NULL;
1916 	if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING)
1917 	    && uio->uio_resid) {
1918 		VNET_SO_ASSERT(so);
1919 		(*pr->pr_usrreqs->pru_rcvd)(so, 0);
1920 	}
1921 
1922 	error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags));
1923 	if (error)
1924 		return (error);
1925 
1926 restart:
1927 	SOCKBUF_LOCK(&so->so_rcv);
1928 	m = so->so_rcv.sb_mb;
1929 	/*
1930 	 * If we have less data than requested, block awaiting more (subject
1931 	 * to any timeout) if:
1932 	 *   1. the current count is less than the low water mark, or
1933 	 *   2. MSG_DONTWAIT is not set
1934 	 */
1935 	if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1936 	    sbavail(&so->so_rcv) < uio->uio_resid) &&
1937 	    sbavail(&so->so_rcv) < so->so_rcv.sb_lowat &&
1938 	    m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
1939 		KASSERT(m != NULL || !sbavail(&so->so_rcv),
1940 		    ("receive: m == %p sbavail == %u",
1941 		    m, sbavail(&so->so_rcv)));
1942 		if (so->so_error || so->so_rerror) {
1943 			if (m != NULL)
1944 				goto dontblock;
1945 			if (so->so_error)
1946 				error = so->so_error;
1947 			else
1948 				error = so->so_rerror;
1949 			if ((flags & MSG_PEEK) == 0) {
1950 				if (so->so_error)
1951 					so->so_error = 0;
1952 				else
1953 					so->so_rerror = 0;
1954 			}
1955 			SOCKBUF_UNLOCK(&so->so_rcv);
1956 			goto release;
1957 		}
1958 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1959 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1960 			if (m != NULL)
1961 				goto dontblock;
1962 #ifdef KERN_TLS
1963 			else if (so->so_rcv.sb_tlsdcc == 0 &&
1964 			    so->so_rcv.sb_tlscc == 0) {
1965 #else
1966 			else {
1967 #endif
1968 				SOCKBUF_UNLOCK(&so->so_rcv);
1969 				goto release;
1970 			}
1971 		}
1972 		for (; m != NULL; m = m->m_next)
1973 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
1974 				m = so->so_rcv.sb_mb;
1975 				goto dontblock;
1976 			}
1977 		if ((so->so_state & (SS_ISCONNECTING | SS_ISCONNECTED |
1978 		    SS_ISDISCONNECTING | SS_ISDISCONNECTED)) == 0 &&
1979 		    (so->so_proto->pr_flags & PR_CONNREQUIRED) != 0) {
1980 			SOCKBUF_UNLOCK(&so->so_rcv);
1981 			error = ENOTCONN;
1982 			goto release;
1983 		}
1984 		if (uio->uio_resid == 0 && !report_real_len) {
1985 			SOCKBUF_UNLOCK(&so->so_rcv);
1986 			goto release;
1987 		}
1988 		if ((so->so_state & SS_NBIO) ||
1989 		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
1990 			SOCKBUF_UNLOCK(&so->so_rcv);
1991 			error = EWOULDBLOCK;
1992 			goto release;
1993 		}
1994 		SBLASTRECORDCHK(&so->so_rcv);
1995 		SBLASTMBUFCHK(&so->so_rcv);
1996 		error = sbwait(so, SO_RCV);
1997 		SOCKBUF_UNLOCK(&so->so_rcv);
1998 		if (error)
1999 			goto release;
2000 		goto restart;
2001 	}
2002 dontblock:
2003 	/*
2004 	 * From this point onward, we maintain 'nextrecord' as a cache of the
2005 	 * pointer to the next record in the socket buffer.  We must keep the
2006 	 * various socket buffer pointers and local stack versions of the
2007 	 * pointers in sync, pushing out modifications before dropping the
2008 	 * socket buffer mutex, and re-reading them when picking it up.
2009 	 *
2010 	 * Otherwise, we will race with the network stack appending new data
2011 	 * or records onto the socket buffer by using inconsistent/stale
2012 	 * versions of the field, possibly resulting in socket buffer
2013 	 * corruption.
2014 	 *
2015 	 * By holding the high-level sblock(), we prevent simultaneous
2016 	 * readers from pulling off the front of the socket buffer.
2017 	 */
2018 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2019 	if (uio->uio_td)
2020 		uio->uio_td->td_ru.ru_msgrcv++;
2021 	KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb"));
2022 	SBLASTRECORDCHK(&so->so_rcv);
2023 	SBLASTMBUFCHK(&so->so_rcv);
2024 	nextrecord = m->m_nextpkt;
2025 	if (pr->pr_flags & PR_ADDR) {
2026 		KASSERT(m->m_type == MT_SONAME,
2027 		    ("m->m_type == %d", m->m_type));
2028 		orig_resid = 0;
2029 		if (psa != NULL)
2030 			*psa = sodupsockaddr(mtod(m, struct sockaddr *),
2031 			    M_NOWAIT);
2032 		if (flags & MSG_PEEK) {
2033 			m = m->m_next;
2034 		} else {
2035 			sbfree(&so->so_rcv, m);
2036 			so->so_rcv.sb_mb = m_free(m);
2037 			m = so->so_rcv.sb_mb;
2038 			sockbuf_pushsync(&so->so_rcv, nextrecord);
2039 		}
2040 	}
2041 
2042 	/*
2043 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
2044 	 * in the first mbuf chain on the socket buffer.  If MSG_PEEK, we
2045 	 * just copy the data; if !MSG_PEEK, we call into the protocol to
2046 	 * perform externalization (or freeing if controlp == NULL).
2047 	 */
2048 	if (m != NULL && m->m_type == MT_CONTROL) {
2049 		struct mbuf *cm = NULL, *cmn;
2050 		struct mbuf **cme = &cm;
2051 #ifdef KERN_TLS
2052 		struct cmsghdr *cmsg;
2053 		struct tls_get_record tgr;
2054 
2055 		/*
2056 		 * For MSG_TLSAPPDATA, check for an alert record.
2057 		 * If found, return ENXIO without removing
2058 		 * it from the receive queue.  This allows a subsequent
2059 		 * call without MSG_TLSAPPDATA to receive it.
2060 		 * Note that, for TLS, there should only be a single
2061 		 * control mbuf with the TLS_GET_RECORD message in it.
2062 		 */
2063 		if (flags & MSG_TLSAPPDATA) {
2064 			cmsg = mtod(m, struct cmsghdr *);
2065 			if (cmsg->cmsg_type == TLS_GET_RECORD &&
2066 			    cmsg->cmsg_len == CMSG_LEN(sizeof(tgr))) {
2067 				memcpy(&tgr, CMSG_DATA(cmsg), sizeof(tgr));
2068 				if (__predict_false(tgr.tls_type ==
2069 				    TLS_RLTYPE_ALERT)) {
2070 					SOCKBUF_UNLOCK(&so->so_rcv);
2071 					error = ENXIO;
2072 					goto release;
2073 				}
2074 			}
2075 		}
2076 #endif
2077 
2078 		do {
2079 			if (flags & MSG_PEEK) {
2080 				if (controlp != NULL) {
2081 					*controlp = m_copym(m, 0, m->m_len,
2082 					    M_NOWAIT);
2083 					controlp = &(*controlp)->m_next;
2084 				}
2085 				m = m->m_next;
2086 			} else {
2087 				sbfree(&so->so_rcv, m);
2088 				so->so_rcv.sb_mb = m->m_next;
2089 				m->m_next = NULL;
2090 				*cme = m;
2091 				cme = &(*cme)->m_next;
2092 				m = so->so_rcv.sb_mb;
2093 			}
2094 		} while (m != NULL && m->m_type == MT_CONTROL);
2095 		if ((flags & MSG_PEEK) == 0)
2096 			sockbuf_pushsync(&so->so_rcv, nextrecord);
2097 		while (cm != NULL) {
2098 			cmn = cm->m_next;
2099 			cm->m_next = NULL;
2100 			if (pr->pr_domain->dom_externalize != NULL) {
2101 				SOCKBUF_UNLOCK(&so->so_rcv);
2102 				VNET_SO_ASSERT(so);
2103 				error = (*pr->pr_domain->dom_externalize)
2104 				    (cm, controlp, flags);
2105 				SOCKBUF_LOCK(&so->so_rcv);
2106 			} else if (controlp != NULL)
2107 				*controlp = cm;
2108 			else
2109 				m_freem(cm);
2110 			if (controlp != NULL) {
2111 				while (*controlp != NULL)
2112 					controlp = &(*controlp)->m_next;
2113 			}
2114 			cm = cmn;
2115 		}
2116 		if (m != NULL)
2117 			nextrecord = so->so_rcv.sb_mb->m_nextpkt;
2118 		else
2119 			nextrecord = so->so_rcv.sb_mb;
2120 		orig_resid = 0;
2121 	}
2122 	if (m != NULL) {
2123 		if ((flags & MSG_PEEK) == 0) {
2124 			KASSERT(m->m_nextpkt == nextrecord,
2125 			    ("soreceive: post-control, nextrecord !sync"));
2126 			if (nextrecord == NULL) {
2127 				KASSERT(so->so_rcv.sb_mb == m,
2128 				    ("soreceive: post-control, sb_mb!=m"));
2129 				KASSERT(so->so_rcv.sb_lastrecord == m,
2130 				    ("soreceive: post-control, lastrecord!=m"));
2131 			}
2132 		}
2133 		type = m->m_type;
2134 		if (type == MT_OOBDATA)
2135 			flags |= MSG_OOB;
2136 	} else {
2137 		if ((flags & MSG_PEEK) == 0) {
2138 			KASSERT(so->so_rcv.sb_mb == nextrecord,
2139 			    ("soreceive: sb_mb != nextrecord"));
2140 			if (so->so_rcv.sb_mb == NULL) {
2141 				KASSERT(so->so_rcv.sb_lastrecord == NULL,
2142 				    ("soreceive: sb_lastercord != NULL"));
2143 			}
2144 		}
2145 	}
2146 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2147 	SBLASTRECORDCHK(&so->so_rcv);
2148 	SBLASTMBUFCHK(&so->so_rcv);
2149 
2150 	/*
2151 	 * Now continue to read any data mbufs off of the head of the socket
2152 	 * buffer until the read request is satisfied.  Note that 'type' is
2153 	 * used to store the type of any mbuf reads that have happened so far
2154 	 * such that soreceive() can stop reading if the type changes, which
2155 	 * causes soreceive() to return only one of regular data and inline
2156 	 * out-of-band data in a single socket receive operation.
2157 	 */
2158 	moff = 0;
2159 	offset = 0;
2160 	while (m != NULL && !(m->m_flags & M_NOTAVAIL) && uio->uio_resid > 0
2161 	    && error == 0) {
2162 		/*
2163 		 * If the type of mbuf has changed since the last mbuf
2164 		 * examined ('type'), end the receive operation.
2165 		 */
2166 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2167 		if (m->m_type == MT_OOBDATA || m->m_type == MT_CONTROL) {
2168 			if (type != m->m_type)
2169 				break;
2170 		} else if (type == MT_OOBDATA)
2171 			break;
2172 		else
2173 		    KASSERT(m->m_type == MT_DATA,
2174 			("m->m_type == %d", m->m_type));
2175 		so->so_rcv.sb_state &= ~SBS_RCVATMARK;
2176 		len = uio->uio_resid;
2177 		if (so->so_oobmark && len > so->so_oobmark - offset)
2178 			len = so->so_oobmark - offset;
2179 		if (len > m->m_len - moff)
2180 			len = m->m_len - moff;
2181 		/*
2182 		 * If mp is set, just pass back the mbufs.  Otherwise copy
2183 		 * them out via the uio, then free.  Sockbuf must be
2184 		 * consistent here (points to current mbuf, it points to next
2185 		 * record) when we drop priority; we must note any additions
2186 		 * to the sockbuf when we block interrupts again.
2187 		 */
2188 		if (mp == NULL) {
2189 			SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2190 			SBLASTRECORDCHK(&so->so_rcv);
2191 			SBLASTMBUFCHK(&so->so_rcv);
2192 			SOCKBUF_UNLOCK(&so->so_rcv);
2193 			if ((m->m_flags & M_EXTPG) != 0)
2194 				error = m_unmapped_uiomove(m, moff, uio,
2195 				    (int)len);
2196 			else
2197 				error = uiomove(mtod(m, char *) + moff,
2198 				    (int)len, uio);
2199 			SOCKBUF_LOCK(&so->so_rcv);
2200 			if (error) {
2201 				/*
2202 				 * The MT_SONAME mbuf has already been removed
2203 				 * from the record, so it is necessary to
2204 				 * remove the data mbufs, if any, to preserve
2205 				 * the invariant in the case of PR_ADDR that
2206 				 * requires MT_SONAME mbufs at the head of
2207 				 * each record.
2208 				 */
2209 				if (pr->pr_flags & PR_ATOMIC &&
2210 				    ((flags & MSG_PEEK) == 0))
2211 					(void)sbdroprecord_locked(&so->so_rcv);
2212 				SOCKBUF_UNLOCK(&so->so_rcv);
2213 				goto release;
2214 			}
2215 		} else
2216 			uio->uio_resid -= len;
2217 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2218 		if (len == m->m_len - moff) {
2219 			if (m->m_flags & M_EOR)
2220 				flags |= MSG_EOR;
2221 			if (flags & MSG_PEEK) {
2222 				m = m->m_next;
2223 				moff = 0;
2224 			} else {
2225 				nextrecord = m->m_nextpkt;
2226 				sbfree(&so->so_rcv, m);
2227 				if (mp != NULL) {
2228 					m->m_nextpkt = NULL;
2229 					*mp = m;
2230 					mp = &m->m_next;
2231 					so->so_rcv.sb_mb = m = m->m_next;
2232 					*mp = NULL;
2233 				} else {
2234 					so->so_rcv.sb_mb = m_free(m);
2235 					m = so->so_rcv.sb_mb;
2236 				}
2237 				sockbuf_pushsync(&so->so_rcv, nextrecord);
2238 				SBLASTRECORDCHK(&so->so_rcv);
2239 				SBLASTMBUFCHK(&so->so_rcv);
2240 			}
2241 		} else {
2242 			if (flags & MSG_PEEK)
2243 				moff += len;
2244 			else {
2245 				if (mp != NULL) {
2246 					if (flags & MSG_DONTWAIT) {
2247 						*mp = m_copym(m, 0, len,
2248 						    M_NOWAIT);
2249 						if (*mp == NULL) {
2250 							/*
2251 							 * m_copym() couldn't
2252 							 * allocate an mbuf.
2253 							 * Adjust uio_resid back
2254 							 * (it was adjusted
2255 							 * down by len bytes,
2256 							 * which we didn't end
2257 							 * up "copying" over).
2258 							 */
2259 							uio->uio_resid += len;
2260 							break;
2261 						}
2262 					} else {
2263 						SOCKBUF_UNLOCK(&so->so_rcv);
2264 						*mp = m_copym(m, 0, len,
2265 						    M_WAITOK);
2266 						SOCKBUF_LOCK(&so->so_rcv);
2267 					}
2268 				}
2269 				sbcut_locked(&so->so_rcv, len);
2270 			}
2271 		}
2272 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2273 		if (so->so_oobmark) {
2274 			if ((flags & MSG_PEEK) == 0) {
2275 				so->so_oobmark -= len;
2276 				if (so->so_oobmark == 0) {
2277 					so->so_rcv.sb_state |= SBS_RCVATMARK;
2278 					break;
2279 				}
2280 			} else {
2281 				offset += len;
2282 				if (offset == so->so_oobmark)
2283 					break;
2284 			}
2285 		}
2286 		if (flags & MSG_EOR)
2287 			break;
2288 		/*
2289 		 * If the MSG_WAITALL flag is set (for non-atomic socket), we
2290 		 * must not quit until "uio->uio_resid == 0" or an error
2291 		 * termination.  If a signal/timeout occurs, return with a
2292 		 * short count but without error.  Keep sockbuf locked
2293 		 * against other readers.
2294 		 */
2295 		while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
2296 		    !sosendallatonce(so) && nextrecord == NULL) {
2297 			SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2298 			if (so->so_error || so->so_rerror ||
2299 			    so->so_rcv.sb_state & SBS_CANTRCVMORE)
2300 				break;
2301 			/*
2302 			 * Notify the protocol that some data has been
2303 			 * drained before blocking.
2304 			 */
2305 			if (pr->pr_flags & PR_WANTRCVD) {
2306 				SOCKBUF_UNLOCK(&so->so_rcv);
2307 				VNET_SO_ASSERT(so);
2308 				(*pr->pr_usrreqs->pru_rcvd)(so, flags);
2309 				SOCKBUF_LOCK(&so->so_rcv);
2310 			}
2311 			SBLASTRECORDCHK(&so->so_rcv);
2312 			SBLASTMBUFCHK(&so->so_rcv);
2313 			/*
2314 			 * We could receive some data while was notifying
2315 			 * the protocol. Skip blocking in this case.
2316 			 */
2317 			if (so->so_rcv.sb_mb == NULL) {
2318 				error = sbwait(so, SO_RCV);
2319 				if (error) {
2320 					SOCKBUF_UNLOCK(&so->so_rcv);
2321 					goto release;
2322 				}
2323 			}
2324 			m = so->so_rcv.sb_mb;
2325 			if (m != NULL)
2326 				nextrecord = m->m_nextpkt;
2327 		}
2328 	}
2329 
2330 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2331 	if (m != NULL && pr->pr_flags & PR_ATOMIC) {
2332 		if (report_real_len)
2333 			uio->uio_resid -= m_length(m, NULL) - moff;
2334 		flags |= MSG_TRUNC;
2335 		if ((flags & MSG_PEEK) == 0)
2336 			(void) sbdroprecord_locked(&so->so_rcv);
2337 	}
2338 	if ((flags & MSG_PEEK) == 0) {
2339 		if (m == NULL) {
2340 			/*
2341 			 * First part is an inline SB_EMPTY_FIXUP().  Second
2342 			 * part makes sure sb_lastrecord is up-to-date if
2343 			 * there is still data in the socket buffer.
2344 			 */
2345 			so->so_rcv.sb_mb = nextrecord;
2346 			if (so->so_rcv.sb_mb == NULL) {
2347 				so->so_rcv.sb_mbtail = NULL;
2348 				so->so_rcv.sb_lastrecord = NULL;
2349 			} else if (nextrecord->m_nextpkt == NULL)
2350 				so->so_rcv.sb_lastrecord = nextrecord;
2351 		}
2352 		SBLASTRECORDCHK(&so->so_rcv);
2353 		SBLASTMBUFCHK(&so->so_rcv);
2354 		/*
2355 		 * If soreceive() is being done from the socket callback,
2356 		 * then don't need to generate ACK to peer to update window,
2357 		 * since ACK will be generated on return to TCP.
2358 		 */
2359 		if (!(flags & MSG_SOCALLBCK) &&
2360 		    (pr->pr_flags & PR_WANTRCVD)) {
2361 			SOCKBUF_UNLOCK(&so->so_rcv);
2362 			VNET_SO_ASSERT(so);
2363 			(*pr->pr_usrreqs->pru_rcvd)(so, flags);
2364 			SOCKBUF_LOCK(&so->so_rcv);
2365 		}
2366 	}
2367 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2368 	if (orig_resid == uio->uio_resid && orig_resid &&
2369 	    (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
2370 		SOCKBUF_UNLOCK(&so->so_rcv);
2371 		goto restart;
2372 	}
2373 	SOCKBUF_UNLOCK(&so->so_rcv);
2374 
2375 	if (flagsp != NULL)
2376 		*flagsp |= flags;
2377 release:
2378 	SOCK_IO_RECV_UNLOCK(so);
2379 	return (error);
2380 }
2381 
2382 /*
2383  * Optimized version of soreceive() for stream (TCP) sockets.
2384  */
2385 int
2386 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio,
2387     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2388 {
2389 	int len = 0, error = 0, flags, oresid;
2390 	struct sockbuf *sb;
2391 	struct mbuf *m, *n = NULL;
2392 
2393 	/* We only do stream sockets. */
2394 	if (so->so_type != SOCK_STREAM)
2395 		return (EINVAL);
2396 	if (psa != NULL)
2397 		*psa = NULL;
2398 	if (flagsp != NULL)
2399 		flags = *flagsp &~ MSG_EOR;
2400 	else
2401 		flags = 0;
2402 	if (controlp != NULL)
2403 		*controlp = NULL;
2404 	if (flags & MSG_OOB)
2405 		return (soreceive_rcvoob(so, uio, flags));
2406 	if (mp0 != NULL)
2407 		*mp0 = NULL;
2408 
2409 	sb = &so->so_rcv;
2410 
2411 #ifdef KERN_TLS
2412 	/*
2413 	 * KTLS store TLS records as records with a control message to
2414 	 * describe the framing.
2415 	 *
2416 	 * We check once here before acquiring locks to optimize the
2417 	 * common case.
2418 	 */
2419 	if (sb->sb_tls_info != NULL)
2420 		return (soreceive_generic(so, psa, uio, mp0, controlp,
2421 		    flagsp));
2422 #endif
2423 
2424 	/* Prevent other readers from entering the socket. */
2425 	error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags));
2426 	if (error)
2427 		return (error);
2428 	SOCKBUF_LOCK(sb);
2429 
2430 #ifdef KERN_TLS
2431 	if (sb->sb_tls_info != NULL) {
2432 		SOCKBUF_UNLOCK(sb);
2433 		SOCK_IO_RECV_UNLOCK(so);
2434 		return (soreceive_generic(so, psa, uio, mp0, controlp,
2435 		    flagsp));
2436 	}
2437 #endif
2438 
2439 	/* Easy one, no space to copyout anything. */
2440 	if (uio->uio_resid == 0) {
2441 		error = EINVAL;
2442 		goto out;
2443 	}
2444 	oresid = uio->uio_resid;
2445 
2446 	/* We will never ever get anything unless we are or were connected. */
2447 	if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) {
2448 		error = ENOTCONN;
2449 		goto out;
2450 	}
2451 
2452 restart:
2453 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2454 
2455 	/* Abort if socket has reported problems. */
2456 	if (so->so_error) {
2457 		if (sbavail(sb) > 0)
2458 			goto deliver;
2459 		if (oresid > uio->uio_resid)
2460 			goto out;
2461 		error = so->so_error;
2462 		if (!(flags & MSG_PEEK))
2463 			so->so_error = 0;
2464 		goto out;
2465 	}
2466 
2467 	/* Door is closed.  Deliver what is left, if any. */
2468 	if (sb->sb_state & SBS_CANTRCVMORE) {
2469 		if (sbavail(sb) > 0)
2470 			goto deliver;
2471 		else
2472 			goto out;
2473 	}
2474 
2475 	/* Socket buffer is empty and we shall not block. */
2476 	if (sbavail(sb) == 0 &&
2477 	    ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) {
2478 		error = EAGAIN;
2479 		goto out;
2480 	}
2481 
2482 	/* Socket buffer got some data that we shall deliver now. */
2483 	if (sbavail(sb) > 0 && !(flags & MSG_WAITALL) &&
2484 	    ((so->so_state & SS_NBIO) ||
2485 	     (flags & (MSG_DONTWAIT|MSG_NBIO)) ||
2486 	     sbavail(sb) >= sb->sb_lowat ||
2487 	     sbavail(sb) >= uio->uio_resid ||
2488 	     sbavail(sb) >= sb->sb_hiwat) ) {
2489 		goto deliver;
2490 	}
2491 
2492 	/* On MSG_WAITALL we must wait until all data or error arrives. */
2493 	if ((flags & MSG_WAITALL) &&
2494 	    (sbavail(sb) >= uio->uio_resid || sbavail(sb) >= sb->sb_hiwat))
2495 		goto deliver;
2496 
2497 	/*
2498 	 * Wait and block until (more) data comes in.
2499 	 * NB: Drops the sockbuf lock during wait.
2500 	 */
2501 	error = sbwait(so, SO_RCV);
2502 	if (error)
2503 		goto out;
2504 	goto restart;
2505 
2506 deliver:
2507 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2508 	KASSERT(sbavail(sb) > 0, ("%s: sockbuf empty", __func__));
2509 	KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__));
2510 
2511 	/* Statistics. */
2512 	if (uio->uio_td)
2513 		uio->uio_td->td_ru.ru_msgrcv++;
2514 
2515 	/* Fill uio until full or current end of socket buffer is reached. */
2516 	len = min(uio->uio_resid, sbavail(sb));
2517 	if (mp0 != NULL) {
2518 		/* Dequeue as many mbufs as possible. */
2519 		if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) {
2520 			if (*mp0 == NULL)
2521 				*mp0 = sb->sb_mb;
2522 			else
2523 				m_cat(*mp0, sb->sb_mb);
2524 			for (m = sb->sb_mb;
2525 			     m != NULL && m->m_len <= len;
2526 			     m = m->m_next) {
2527 				KASSERT(!(m->m_flags & M_NOTAVAIL),
2528 				    ("%s: m %p not available", __func__, m));
2529 				len -= m->m_len;
2530 				uio->uio_resid -= m->m_len;
2531 				sbfree(sb, m);
2532 				n = m;
2533 			}
2534 			n->m_next = NULL;
2535 			sb->sb_mb = m;
2536 			sb->sb_lastrecord = sb->sb_mb;
2537 			if (sb->sb_mb == NULL)
2538 				SB_EMPTY_FIXUP(sb);
2539 		}
2540 		/* Copy the remainder. */
2541 		if (len > 0) {
2542 			KASSERT(sb->sb_mb != NULL,
2543 			    ("%s: len > 0 && sb->sb_mb empty", __func__));
2544 
2545 			m = m_copym(sb->sb_mb, 0, len, M_NOWAIT);
2546 			if (m == NULL)
2547 				len = 0;	/* Don't flush data from sockbuf. */
2548 			else
2549 				uio->uio_resid -= len;
2550 			if (*mp0 != NULL)
2551 				m_cat(*mp0, m);
2552 			else
2553 				*mp0 = m;
2554 			if (*mp0 == NULL) {
2555 				error = ENOBUFS;
2556 				goto out;
2557 			}
2558 		}
2559 	} else {
2560 		/* NB: Must unlock socket buffer as uiomove may sleep. */
2561 		SOCKBUF_UNLOCK(sb);
2562 		error = m_mbuftouio(uio, sb->sb_mb, len);
2563 		SOCKBUF_LOCK(sb);
2564 		if (error)
2565 			goto out;
2566 	}
2567 	SBLASTRECORDCHK(sb);
2568 	SBLASTMBUFCHK(sb);
2569 
2570 	/*
2571 	 * Remove the delivered data from the socket buffer unless we
2572 	 * were only peeking.
2573 	 */
2574 	if (!(flags & MSG_PEEK)) {
2575 		if (len > 0)
2576 			sbdrop_locked(sb, len);
2577 
2578 		/* Notify protocol that we drained some data. */
2579 		if ((so->so_proto->pr_flags & PR_WANTRCVD) &&
2580 		    (((flags & MSG_WAITALL) && uio->uio_resid > 0) ||
2581 		     !(flags & MSG_SOCALLBCK))) {
2582 			SOCKBUF_UNLOCK(sb);
2583 			VNET_SO_ASSERT(so);
2584 			(*so->so_proto->pr_usrreqs->pru_rcvd)(so, flags);
2585 			SOCKBUF_LOCK(sb);
2586 		}
2587 	}
2588 
2589 	/*
2590 	 * For MSG_WAITALL we may have to loop again and wait for
2591 	 * more data to come in.
2592 	 */
2593 	if ((flags & MSG_WAITALL) && uio->uio_resid > 0)
2594 		goto restart;
2595 out:
2596 	SBLASTRECORDCHK(sb);
2597 	SBLASTMBUFCHK(sb);
2598 	SOCKBUF_UNLOCK(sb);
2599 	SOCK_IO_RECV_UNLOCK(so);
2600 	return (error);
2601 }
2602 
2603 /*
2604  * Optimized version of soreceive() for simple datagram cases from userspace.
2605  * Unlike in the stream case, we're able to drop a datagram if copyout()
2606  * fails, and because we handle datagrams atomically, we don't need to use a
2607  * sleep lock to prevent I/O interlacing.
2608  */
2609 int
2610 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio,
2611     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2612 {
2613 	struct mbuf *m, *m2;
2614 	int flags, error;
2615 	ssize_t len;
2616 	struct protosw *pr = so->so_proto;
2617 	struct mbuf *nextrecord;
2618 
2619 	if (psa != NULL)
2620 		*psa = NULL;
2621 	if (controlp != NULL)
2622 		*controlp = NULL;
2623 	if (flagsp != NULL)
2624 		flags = *flagsp &~ MSG_EOR;
2625 	else
2626 		flags = 0;
2627 
2628 	/*
2629 	 * For any complicated cases, fall back to the full
2630 	 * soreceive_generic().
2631 	 */
2632 	if (mp0 != NULL || (flags & (MSG_PEEK | MSG_OOB | MSG_TRUNC)))
2633 		return (soreceive_generic(so, psa, uio, mp0, controlp,
2634 		    flagsp));
2635 
2636 	/*
2637 	 * Enforce restrictions on use.
2638 	 */
2639 	KASSERT((pr->pr_flags & PR_WANTRCVD) == 0,
2640 	    ("soreceive_dgram: wantrcvd"));
2641 	KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic"));
2642 	KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0,
2643 	    ("soreceive_dgram: SBS_RCVATMARK"));
2644 	KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0,
2645 	    ("soreceive_dgram: P_CONNREQUIRED"));
2646 
2647 	/*
2648 	 * Loop blocking while waiting for a datagram.
2649 	 */
2650 	SOCKBUF_LOCK(&so->so_rcv);
2651 	while ((m = so->so_rcv.sb_mb) == NULL) {
2652 		KASSERT(sbavail(&so->so_rcv) == 0,
2653 		    ("soreceive_dgram: sb_mb NULL but sbavail %u",
2654 		    sbavail(&so->so_rcv)));
2655 		if (so->so_error) {
2656 			error = so->so_error;
2657 			so->so_error = 0;
2658 			SOCKBUF_UNLOCK(&so->so_rcv);
2659 			return (error);
2660 		}
2661 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE ||
2662 		    uio->uio_resid == 0) {
2663 			SOCKBUF_UNLOCK(&so->so_rcv);
2664 			return (0);
2665 		}
2666 		if ((so->so_state & SS_NBIO) ||
2667 		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2668 			SOCKBUF_UNLOCK(&so->so_rcv);
2669 			return (EWOULDBLOCK);
2670 		}
2671 		SBLASTRECORDCHK(&so->so_rcv);
2672 		SBLASTMBUFCHK(&so->so_rcv);
2673 		error = sbwait(so, SO_RCV);
2674 		if (error) {
2675 			SOCKBUF_UNLOCK(&so->so_rcv);
2676 			return (error);
2677 		}
2678 	}
2679 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2680 
2681 	if (uio->uio_td)
2682 		uio->uio_td->td_ru.ru_msgrcv++;
2683 	SBLASTRECORDCHK(&so->so_rcv);
2684 	SBLASTMBUFCHK(&so->so_rcv);
2685 	nextrecord = m->m_nextpkt;
2686 	if (nextrecord == NULL) {
2687 		KASSERT(so->so_rcv.sb_lastrecord == m,
2688 		    ("soreceive_dgram: lastrecord != m"));
2689 	}
2690 
2691 	KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord,
2692 	    ("soreceive_dgram: m_nextpkt != nextrecord"));
2693 
2694 	/*
2695 	 * Pull 'm' and its chain off the front of the packet queue.
2696 	 */
2697 	so->so_rcv.sb_mb = NULL;
2698 	sockbuf_pushsync(&so->so_rcv, nextrecord);
2699 
2700 	/*
2701 	 * Walk 'm's chain and free that many bytes from the socket buffer.
2702 	 */
2703 	for (m2 = m; m2 != NULL; m2 = m2->m_next)
2704 		sbfree(&so->so_rcv, m2);
2705 
2706 	/*
2707 	 * Do a few last checks before we let go of the lock.
2708 	 */
2709 	SBLASTRECORDCHK(&so->so_rcv);
2710 	SBLASTMBUFCHK(&so->so_rcv);
2711 	SOCKBUF_UNLOCK(&so->so_rcv);
2712 
2713 	if (pr->pr_flags & PR_ADDR) {
2714 		KASSERT(m->m_type == MT_SONAME,
2715 		    ("m->m_type == %d", m->m_type));
2716 		if (psa != NULL)
2717 			*psa = sodupsockaddr(mtod(m, struct sockaddr *),
2718 			    M_NOWAIT);
2719 		m = m_free(m);
2720 	}
2721 	if (m == NULL) {
2722 		/* XXXRW: Can this happen? */
2723 		return (0);
2724 	}
2725 
2726 	/*
2727 	 * Packet to copyout() is now in 'm' and it is disconnected from the
2728 	 * queue.
2729 	 *
2730 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
2731 	 * in the first mbuf chain on the socket buffer.  We call into the
2732 	 * protocol to perform externalization (or freeing if controlp ==
2733 	 * NULL). In some cases there can be only MT_CONTROL mbufs without
2734 	 * MT_DATA mbufs.
2735 	 */
2736 	if (m->m_type == MT_CONTROL) {
2737 		struct mbuf *cm = NULL, *cmn;
2738 		struct mbuf **cme = &cm;
2739 
2740 		do {
2741 			m2 = m->m_next;
2742 			m->m_next = NULL;
2743 			*cme = m;
2744 			cme = &(*cme)->m_next;
2745 			m = m2;
2746 		} while (m != NULL && m->m_type == MT_CONTROL);
2747 		while (cm != NULL) {
2748 			cmn = cm->m_next;
2749 			cm->m_next = NULL;
2750 			if (pr->pr_domain->dom_externalize != NULL) {
2751 				error = (*pr->pr_domain->dom_externalize)
2752 				    (cm, controlp, flags);
2753 			} else if (controlp != NULL)
2754 				*controlp = cm;
2755 			else
2756 				m_freem(cm);
2757 			if (controlp != NULL) {
2758 				while (*controlp != NULL)
2759 					controlp = &(*controlp)->m_next;
2760 			}
2761 			cm = cmn;
2762 		}
2763 	}
2764 	KASSERT(m == NULL || m->m_type == MT_DATA,
2765 	    ("soreceive_dgram: !data"));
2766 	while (m != NULL && uio->uio_resid > 0) {
2767 		len = uio->uio_resid;
2768 		if (len > m->m_len)
2769 			len = m->m_len;
2770 		error = uiomove(mtod(m, char *), (int)len, uio);
2771 		if (error) {
2772 			m_freem(m);
2773 			return (error);
2774 		}
2775 		if (len == m->m_len)
2776 			m = m_free(m);
2777 		else {
2778 			m->m_data += len;
2779 			m->m_len -= len;
2780 		}
2781 	}
2782 	if (m != NULL) {
2783 		flags |= MSG_TRUNC;
2784 		m_freem(m);
2785 	}
2786 	if (flagsp != NULL)
2787 		*flagsp |= flags;
2788 	return (0);
2789 }
2790 
2791 int
2792 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
2793     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2794 {
2795 	int error;
2796 
2797 	CURVNET_SET(so->so_vnet);
2798 	error = (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio,
2799 	    mp0, controlp, flagsp));
2800 	CURVNET_RESTORE();
2801 	return (error);
2802 }
2803 
2804 int
2805 soshutdown(struct socket *so, int how)
2806 {
2807 	struct protosw *pr;
2808 	int error, soerror_enotconn;
2809 
2810 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
2811 		return (EINVAL);
2812 
2813 	soerror_enotconn = 0;
2814 	SOCK_LOCK(so);
2815 	if ((so->so_state &
2816 	    (SS_ISCONNECTED | SS_ISCONNECTING | SS_ISDISCONNECTING)) == 0) {
2817 		/*
2818 		 * POSIX mandates us to return ENOTCONN when shutdown(2) is
2819 		 * invoked on a datagram sockets, however historically we would
2820 		 * actually tear socket down. This is known to be leveraged by
2821 		 * some applications to unblock process waiting in recvXXX(2)
2822 		 * by other process that it shares that socket with. Try to meet
2823 		 * both backward-compatibility and POSIX requirements by forcing
2824 		 * ENOTCONN but still asking protocol to perform pru_shutdown().
2825 		 */
2826 		if (so->so_type != SOCK_DGRAM && !SOLISTENING(so)) {
2827 			SOCK_UNLOCK(so);
2828 			return (ENOTCONN);
2829 		}
2830 		soerror_enotconn = 1;
2831 	}
2832 
2833 	if (SOLISTENING(so)) {
2834 		if (how != SHUT_WR) {
2835 			so->so_error = ECONNABORTED;
2836 			solisten_wakeup(so);	/* unlocks so */
2837 		} else {
2838 			SOCK_UNLOCK(so);
2839 		}
2840 		goto done;
2841 	}
2842 	SOCK_UNLOCK(so);
2843 
2844 	CURVNET_SET(so->so_vnet);
2845 	pr = so->so_proto;
2846 	if (pr->pr_usrreqs->pru_flush != NULL)
2847 		(*pr->pr_usrreqs->pru_flush)(so, how);
2848 	if (how != SHUT_WR)
2849 		sorflush(so);
2850 	if (how != SHUT_RD) {
2851 		error = (*pr->pr_usrreqs->pru_shutdown)(so);
2852 		wakeup(&so->so_timeo);
2853 		CURVNET_RESTORE();
2854 		return ((error == 0 && soerror_enotconn) ? ENOTCONN : error);
2855 	}
2856 	wakeup(&so->so_timeo);
2857 	CURVNET_RESTORE();
2858 
2859 done:
2860 	return (soerror_enotconn ? ENOTCONN : 0);
2861 }
2862 
2863 void
2864 sorflush(struct socket *so)
2865 {
2866 	struct protosw *pr;
2867 	int error;
2868 
2869 	VNET_SO_ASSERT(so);
2870 
2871 	/*
2872 	 * Dislodge threads currently blocked in receive and wait to acquire
2873 	 * a lock against other simultaneous readers before clearing the
2874 	 * socket buffer.  Don't let our acquire be interrupted by a signal
2875 	 * despite any existing socket disposition on interruptable waiting.
2876 	 */
2877 	socantrcvmore(so);
2878 
2879 	error = SOCK_IO_RECV_LOCK(so, SBL_WAIT | SBL_NOINTR);
2880 	if (error != 0) {
2881 		KASSERT(SOLISTENING(so),
2882 		    ("%s: soiolock(%p) failed", __func__, so));
2883 		return;
2884 	}
2885 
2886 	pr = so->so_proto;
2887 	if (pr->pr_flags & PR_RIGHTS) {
2888 		MPASS(pr->pr_domain->dom_dispose != NULL);
2889 		(*pr->pr_domain->dom_dispose)(so);
2890 	} else {
2891 		sbrelease(so, SO_RCV);
2892 		SOCK_IO_RECV_UNLOCK(so);
2893 	}
2894 
2895 }
2896 
2897 /*
2898  * Wrapper for Socket established helper hook.
2899  * Parameters: socket, context of the hook point, hook id.
2900  */
2901 static int inline
2902 hhook_run_socket(struct socket *so, void *hctx, int32_t h_id)
2903 {
2904 	struct socket_hhook_data hhook_data = {
2905 		.so = so,
2906 		.hctx = hctx,
2907 		.m = NULL,
2908 		.status = 0
2909 	};
2910 
2911 	CURVNET_SET(so->so_vnet);
2912 	HHOOKS_RUN_IF(V_socket_hhh[h_id], &hhook_data, &so->osd);
2913 	CURVNET_RESTORE();
2914 
2915 	/* Ugly but needed, since hhooks return void for now */
2916 	return (hhook_data.status);
2917 }
2918 
2919 /*
2920  * Perhaps this routine, and sooptcopyout(), below, ought to come in an
2921  * additional variant to handle the case where the option value needs to be
2922  * some kind of integer, but not a specific size.  In addition to their use
2923  * here, these functions are also called by the protocol-level pr_ctloutput()
2924  * routines.
2925  */
2926 int
2927 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2928 {
2929 	size_t	valsize;
2930 
2931 	/*
2932 	 * If the user gives us more than we wanted, we ignore it, but if we
2933 	 * don't get the minimum length the caller wants, we return EINVAL.
2934 	 * On success, sopt->sopt_valsize is set to however much we actually
2935 	 * retrieved.
2936 	 */
2937 	if ((valsize = sopt->sopt_valsize) < minlen)
2938 		return EINVAL;
2939 	if (valsize > len)
2940 		sopt->sopt_valsize = valsize = len;
2941 
2942 	if (sopt->sopt_td != NULL)
2943 		return (copyin(sopt->sopt_val, buf, valsize));
2944 
2945 	bcopy(sopt->sopt_val, buf, valsize);
2946 	return (0);
2947 }
2948 
2949 /*
2950  * Kernel version of setsockopt(2).
2951  *
2952  * XXX: optlen is size_t, not socklen_t
2953  */
2954 int
2955 so_setsockopt(struct socket *so, int level, int optname, void *optval,
2956     size_t optlen)
2957 {
2958 	struct sockopt sopt;
2959 
2960 	sopt.sopt_level = level;
2961 	sopt.sopt_name = optname;
2962 	sopt.sopt_dir = SOPT_SET;
2963 	sopt.sopt_val = optval;
2964 	sopt.sopt_valsize = optlen;
2965 	sopt.sopt_td = NULL;
2966 	return (sosetopt(so, &sopt));
2967 }
2968 
2969 int
2970 sosetopt(struct socket *so, struct sockopt *sopt)
2971 {
2972 	int	error, optval;
2973 	struct	linger l;
2974 	struct	timeval tv;
2975 	sbintime_t val, *valp;
2976 	uint32_t val32;
2977 #ifdef MAC
2978 	struct mac extmac;
2979 #endif
2980 
2981 	CURVNET_SET(so->so_vnet);
2982 	error = 0;
2983 	if (sopt->sopt_level != SOL_SOCKET) {
2984 		if (so->so_proto->pr_ctloutput != NULL)
2985 			error = (*so->so_proto->pr_ctloutput)(so, sopt);
2986 		else
2987 			error = ENOPROTOOPT;
2988 	} else {
2989 		switch (sopt->sopt_name) {
2990 		case SO_ACCEPTFILTER:
2991 			error = accept_filt_setopt(so, sopt);
2992 			if (error)
2993 				goto bad;
2994 			break;
2995 
2996 		case SO_LINGER:
2997 			error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
2998 			if (error)
2999 				goto bad;
3000 			if (l.l_linger < 0 ||
3001 			    l.l_linger > USHRT_MAX ||
3002 			    l.l_linger > (INT_MAX / hz)) {
3003 				error = EDOM;
3004 				goto bad;
3005 			}
3006 			SOCK_LOCK(so);
3007 			so->so_linger = l.l_linger;
3008 			if (l.l_onoff)
3009 				so->so_options |= SO_LINGER;
3010 			else
3011 				so->so_options &= ~SO_LINGER;
3012 			SOCK_UNLOCK(so);
3013 			break;
3014 
3015 		case SO_DEBUG:
3016 		case SO_KEEPALIVE:
3017 		case SO_DONTROUTE:
3018 		case SO_USELOOPBACK:
3019 		case SO_BROADCAST:
3020 		case SO_REUSEADDR:
3021 		case SO_REUSEPORT:
3022 		case SO_REUSEPORT_LB:
3023 		case SO_OOBINLINE:
3024 		case SO_TIMESTAMP:
3025 		case SO_BINTIME:
3026 		case SO_NOSIGPIPE:
3027 		case SO_NO_DDP:
3028 		case SO_NO_OFFLOAD:
3029 		case SO_RERROR:
3030 			error = sooptcopyin(sopt, &optval, sizeof optval,
3031 			    sizeof optval);
3032 			if (error)
3033 				goto bad;
3034 			SOCK_LOCK(so);
3035 			if (optval)
3036 				so->so_options |= sopt->sopt_name;
3037 			else
3038 				so->so_options &= ~sopt->sopt_name;
3039 			SOCK_UNLOCK(so);
3040 			break;
3041 
3042 		case SO_SETFIB:
3043 			error = sooptcopyin(sopt, &optval, sizeof optval,
3044 			    sizeof optval);
3045 			if (error)
3046 				goto bad;
3047 
3048 			if (optval < 0 || optval >= rt_numfibs) {
3049 				error = EINVAL;
3050 				goto bad;
3051 			}
3052 			if (((so->so_proto->pr_domain->dom_family == PF_INET) ||
3053 			   (so->so_proto->pr_domain->dom_family == PF_INET6) ||
3054 			   (so->so_proto->pr_domain->dom_family == PF_ROUTE)))
3055 				so->so_fibnum = optval;
3056 			else
3057 				so->so_fibnum = 0;
3058 			break;
3059 
3060 		case SO_USER_COOKIE:
3061 			error = sooptcopyin(sopt, &val32, sizeof val32,
3062 			    sizeof val32);
3063 			if (error)
3064 				goto bad;
3065 			so->so_user_cookie = val32;
3066 			break;
3067 
3068 		case SO_SNDBUF:
3069 		case SO_RCVBUF:
3070 		case SO_SNDLOWAT:
3071 		case SO_RCVLOWAT:
3072 			error = sooptcopyin(sopt, &optval, sizeof optval,
3073 			    sizeof optval);
3074 			if (error)
3075 				goto bad;
3076 
3077 			/*
3078 			 * Values < 1 make no sense for any of these options,
3079 			 * so disallow them.
3080 			 */
3081 			if (optval < 1) {
3082 				error = EINVAL;
3083 				goto bad;
3084 			}
3085 
3086 			error = sbsetopt(so, sopt->sopt_name, optval);
3087 			break;
3088 
3089 		case SO_SNDTIMEO:
3090 		case SO_RCVTIMEO:
3091 #ifdef COMPAT_FREEBSD32
3092 			if (SV_CURPROC_FLAG(SV_ILP32)) {
3093 				struct timeval32 tv32;
3094 
3095 				error = sooptcopyin(sopt, &tv32, sizeof tv32,
3096 				    sizeof tv32);
3097 				CP(tv32, tv, tv_sec);
3098 				CP(tv32, tv, tv_usec);
3099 			} else
3100 #endif
3101 				error = sooptcopyin(sopt, &tv, sizeof tv,
3102 				    sizeof tv);
3103 			if (error)
3104 				goto bad;
3105 			if (tv.tv_sec < 0 || tv.tv_usec < 0 ||
3106 			    tv.tv_usec >= 1000000) {
3107 				error = EDOM;
3108 				goto bad;
3109 			}
3110 			if (tv.tv_sec > INT32_MAX)
3111 				val = SBT_MAX;
3112 			else
3113 				val = tvtosbt(tv);
3114 			SOCK_LOCK(so);
3115 			valp = sopt->sopt_name == SO_SNDTIMEO ?
3116 			    (SOLISTENING(so) ? &so->sol_sbsnd_timeo :
3117 			    &so->so_snd.sb_timeo) :
3118 			    (SOLISTENING(so) ? &so->sol_sbrcv_timeo :
3119 			    &so->so_rcv.sb_timeo);
3120 			*valp = val;
3121 			SOCK_UNLOCK(so);
3122 			break;
3123 
3124 		case SO_LABEL:
3125 #ifdef MAC
3126 			error = sooptcopyin(sopt, &extmac, sizeof extmac,
3127 			    sizeof extmac);
3128 			if (error)
3129 				goto bad;
3130 			error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
3131 			    so, &extmac);
3132 #else
3133 			error = EOPNOTSUPP;
3134 #endif
3135 			break;
3136 
3137 		case SO_TS_CLOCK:
3138 			error = sooptcopyin(sopt, &optval, sizeof optval,
3139 			    sizeof optval);
3140 			if (error)
3141 				goto bad;
3142 			if (optval < 0 || optval > SO_TS_CLOCK_MAX) {
3143 				error = EINVAL;
3144 				goto bad;
3145 			}
3146 			so->so_ts_clock = optval;
3147 			break;
3148 
3149 		case SO_MAX_PACING_RATE:
3150 			error = sooptcopyin(sopt, &val32, sizeof(val32),
3151 			    sizeof(val32));
3152 			if (error)
3153 				goto bad;
3154 			so->so_max_pacing_rate = val32;
3155 			break;
3156 
3157 		default:
3158 			if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
3159 				error = hhook_run_socket(so, sopt,
3160 				    HHOOK_SOCKET_OPT);
3161 			else
3162 				error = ENOPROTOOPT;
3163 			break;
3164 		}
3165 		if (error == 0 && so->so_proto->pr_ctloutput != NULL)
3166 			(void)(*so->so_proto->pr_ctloutput)(so, sopt);
3167 	}
3168 bad:
3169 	CURVNET_RESTORE();
3170 	return (error);
3171 }
3172 
3173 /*
3174  * Helper routine for getsockopt.
3175  */
3176 int
3177 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
3178 {
3179 	int	error;
3180 	size_t	valsize;
3181 
3182 	error = 0;
3183 
3184 	/*
3185 	 * Documented get behavior is that we always return a value, possibly
3186 	 * truncated to fit in the user's buffer.  Traditional behavior is
3187 	 * that we always tell the user precisely how much we copied, rather
3188 	 * than something useful like the total amount we had available for
3189 	 * her.  Note that this interface is not idempotent; the entire
3190 	 * answer must be generated ahead of time.
3191 	 */
3192 	valsize = min(len, sopt->sopt_valsize);
3193 	sopt->sopt_valsize = valsize;
3194 	if (sopt->sopt_val != NULL) {
3195 		if (sopt->sopt_td != NULL)
3196 			error = copyout(buf, sopt->sopt_val, valsize);
3197 		else
3198 			bcopy(buf, sopt->sopt_val, valsize);
3199 	}
3200 	return (error);
3201 }
3202 
3203 int
3204 sogetopt(struct socket *so, struct sockopt *sopt)
3205 {
3206 	int	error, optval;
3207 	struct	linger l;
3208 	struct	timeval tv;
3209 #ifdef MAC
3210 	struct mac extmac;
3211 #endif
3212 
3213 	CURVNET_SET(so->so_vnet);
3214 	error = 0;
3215 	if (sopt->sopt_level != SOL_SOCKET) {
3216 		if (so->so_proto->pr_ctloutput != NULL)
3217 			error = (*so->so_proto->pr_ctloutput)(so, sopt);
3218 		else
3219 			error = ENOPROTOOPT;
3220 		CURVNET_RESTORE();
3221 		return (error);
3222 	} else {
3223 		switch (sopt->sopt_name) {
3224 		case SO_ACCEPTFILTER:
3225 			error = accept_filt_getopt(so, sopt);
3226 			break;
3227 
3228 		case SO_LINGER:
3229 			SOCK_LOCK(so);
3230 			l.l_onoff = so->so_options & SO_LINGER;
3231 			l.l_linger = so->so_linger;
3232 			SOCK_UNLOCK(so);
3233 			error = sooptcopyout(sopt, &l, sizeof l);
3234 			break;
3235 
3236 		case SO_USELOOPBACK:
3237 		case SO_DONTROUTE:
3238 		case SO_DEBUG:
3239 		case SO_KEEPALIVE:
3240 		case SO_REUSEADDR:
3241 		case SO_REUSEPORT:
3242 		case SO_REUSEPORT_LB:
3243 		case SO_BROADCAST:
3244 		case SO_OOBINLINE:
3245 		case SO_ACCEPTCONN:
3246 		case SO_TIMESTAMP:
3247 		case SO_BINTIME:
3248 		case SO_NOSIGPIPE:
3249 		case SO_NO_DDP:
3250 		case SO_NO_OFFLOAD:
3251 		case SO_RERROR:
3252 			optval = so->so_options & sopt->sopt_name;
3253 integer:
3254 			error = sooptcopyout(sopt, &optval, sizeof optval);
3255 			break;
3256 
3257 		case SO_DOMAIN:
3258 			optval = so->so_proto->pr_domain->dom_family;
3259 			goto integer;
3260 
3261 		case SO_TYPE:
3262 			optval = so->so_type;
3263 			goto integer;
3264 
3265 		case SO_PROTOCOL:
3266 			optval = so->so_proto->pr_protocol;
3267 			goto integer;
3268 
3269 		case SO_ERROR:
3270 			SOCK_LOCK(so);
3271 			if (so->so_error) {
3272 				optval = so->so_error;
3273 				so->so_error = 0;
3274 			} else {
3275 				optval = so->so_rerror;
3276 				so->so_rerror = 0;
3277 			}
3278 			SOCK_UNLOCK(so);
3279 			goto integer;
3280 
3281 		case SO_SNDBUF:
3282 			optval = SOLISTENING(so) ? so->sol_sbsnd_hiwat :
3283 			    so->so_snd.sb_hiwat;
3284 			goto integer;
3285 
3286 		case SO_RCVBUF:
3287 			optval = SOLISTENING(so) ? so->sol_sbrcv_hiwat :
3288 			    so->so_rcv.sb_hiwat;
3289 			goto integer;
3290 
3291 		case SO_SNDLOWAT:
3292 			optval = SOLISTENING(so) ? so->sol_sbsnd_lowat :
3293 			    so->so_snd.sb_lowat;
3294 			goto integer;
3295 
3296 		case SO_RCVLOWAT:
3297 			optval = SOLISTENING(so) ? so->sol_sbrcv_lowat :
3298 			    so->so_rcv.sb_lowat;
3299 			goto integer;
3300 
3301 		case SO_SNDTIMEO:
3302 		case SO_RCVTIMEO:
3303 			SOCK_LOCK(so);
3304 			tv = sbttotv(sopt->sopt_name == SO_SNDTIMEO ?
3305 			    (SOLISTENING(so) ? so->sol_sbsnd_timeo :
3306 			    so->so_snd.sb_timeo) :
3307 			    (SOLISTENING(so) ? so->sol_sbrcv_timeo :
3308 			    so->so_rcv.sb_timeo));
3309 			SOCK_UNLOCK(so);
3310 #ifdef COMPAT_FREEBSD32
3311 			if (SV_CURPROC_FLAG(SV_ILP32)) {
3312 				struct timeval32 tv32;
3313 
3314 				CP(tv, tv32, tv_sec);
3315 				CP(tv, tv32, tv_usec);
3316 				error = sooptcopyout(sopt, &tv32, sizeof tv32);
3317 			} else
3318 #endif
3319 				error = sooptcopyout(sopt, &tv, sizeof tv);
3320 			break;
3321 
3322 		case SO_LABEL:
3323 #ifdef MAC
3324 			error = sooptcopyin(sopt, &extmac, sizeof(extmac),
3325 			    sizeof(extmac));
3326 			if (error)
3327 				goto bad;
3328 			error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
3329 			    so, &extmac);
3330 			if (error)
3331 				goto bad;
3332 			error = sooptcopyout(sopt, &extmac, sizeof extmac);
3333 #else
3334 			error = EOPNOTSUPP;
3335 #endif
3336 			break;
3337 
3338 		case SO_PEERLABEL:
3339 #ifdef MAC
3340 			error = sooptcopyin(sopt, &extmac, sizeof(extmac),
3341 			    sizeof(extmac));
3342 			if (error)
3343 				goto bad;
3344 			error = mac_getsockopt_peerlabel(
3345 			    sopt->sopt_td->td_ucred, so, &extmac);
3346 			if (error)
3347 				goto bad;
3348 			error = sooptcopyout(sopt, &extmac, sizeof extmac);
3349 #else
3350 			error = EOPNOTSUPP;
3351 #endif
3352 			break;
3353 
3354 		case SO_LISTENQLIMIT:
3355 			optval = SOLISTENING(so) ? so->sol_qlimit : 0;
3356 			goto integer;
3357 
3358 		case SO_LISTENQLEN:
3359 			optval = SOLISTENING(so) ? so->sol_qlen : 0;
3360 			goto integer;
3361 
3362 		case SO_LISTENINCQLEN:
3363 			optval = SOLISTENING(so) ? so->sol_incqlen : 0;
3364 			goto integer;
3365 
3366 		case SO_TS_CLOCK:
3367 			optval = so->so_ts_clock;
3368 			goto integer;
3369 
3370 		case SO_MAX_PACING_RATE:
3371 			optval = so->so_max_pacing_rate;
3372 			goto integer;
3373 
3374 		default:
3375 			if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
3376 				error = hhook_run_socket(so, sopt,
3377 				    HHOOK_SOCKET_OPT);
3378 			else
3379 				error = ENOPROTOOPT;
3380 			break;
3381 		}
3382 	}
3383 #ifdef MAC
3384 bad:
3385 #endif
3386 	CURVNET_RESTORE();
3387 	return (error);
3388 }
3389 
3390 int
3391 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
3392 {
3393 	struct mbuf *m, *m_prev;
3394 	int sopt_size = sopt->sopt_valsize;
3395 
3396 	MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
3397 	if (m == NULL)
3398 		return ENOBUFS;
3399 	if (sopt_size > MLEN) {
3400 		MCLGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT);
3401 		if ((m->m_flags & M_EXT) == 0) {
3402 			m_free(m);
3403 			return ENOBUFS;
3404 		}
3405 		m->m_len = min(MCLBYTES, sopt_size);
3406 	} else {
3407 		m->m_len = min(MLEN, sopt_size);
3408 	}
3409 	sopt_size -= m->m_len;
3410 	*mp = m;
3411 	m_prev = m;
3412 
3413 	while (sopt_size) {
3414 		MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
3415 		if (m == NULL) {
3416 			m_freem(*mp);
3417 			return ENOBUFS;
3418 		}
3419 		if (sopt_size > MLEN) {
3420 			MCLGET(m, sopt->sopt_td != NULL ? M_WAITOK :
3421 			    M_NOWAIT);
3422 			if ((m->m_flags & M_EXT) == 0) {
3423 				m_freem(m);
3424 				m_freem(*mp);
3425 				return ENOBUFS;
3426 			}
3427 			m->m_len = min(MCLBYTES, sopt_size);
3428 		} else {
3429 			m->m_len = min(MLEN, sopt_size);
3430 		}
3431 		sopt_size -= m->m_len;
3432 		m_prev->m_next = m;
3433 		m_prev = m;
3434 	}
3435 	return (0);
3436 }
3437 
3438 int
3439 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
3440 {
3441 	struct mbuf *m0 = m;
3442 
3443 	if (sopt->sopt_val == NULL)
3444 		return (0);
3445 	while (m != NULL && sopt->sopt_valsize >= m->m_len) {
3446 		if (sopt->sopt_td != NULL) {
3447 			int error;
3448 
3449 			error = copyin(sopt->sopt_val, mtod(m, char *),
3450 			    m->m_len);
3451 			if (error != 0) {
3452 				m_freem(m0);
3453 				return(error);
3454 			}
3455 		} else
3456 			bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
3457 		sopt->sopt_valsize -= m->m_len;
3458 		sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
3459 		m = m->m_next;
3460 	}
3461 	if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
3462 		panic("ip6_sooptmcopyin");
3463 	return (0);
3464 }
3465 
3466 int
3467 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
3468 {
3469 	struct mbuf *m0 = m;
3470 	size_t valsize = 0;
3471 
3472 	if (sopt->sopt_val == NULL)
3473 		return (0);
3474 	while (m != NULL && sopt->sopt_valsize >= m->m_len) {
3475 		if (sopt->sopt_td != NULL) {
3476 			int error;
3477 
3478 			error = copyout(mtod(m, char *), sopt->sopt_val,
3479 			    m->m_len);
3480 			if (error != 0) {
3481 				m_freem(m0);
3482 				return(error);
3483 			}
3484 		} else
3485 			bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
3486 		sopt->sopt_valsize -= m->m_len;
3487 		sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
3488 		valsize += m->m_len;
3489 		m = m->m_next;
3490 	}
3491 	if (m != NULL) {
3492 		/* enough soopt buffer should be given from user-land */
3493 		m_freem(m0);
3494 		return(EINVAL);
3495 	}
3496 	sopt->sopt_valsize = valsize;
3497 	return (0);
3498 }
3499 
3500 /*
3501  * sohasoutofband(): protocol notifies socket layer of the arrival of new
3502  * out-of-band data, which will then notify socket consumers.
3503  */
3504 void
3505 sohasoutofband(struct socket *so)
3506 {
3507 
3508 	if (so->so_sigio != NULL)
3509 		pgsigio(&so->so_sigio, SIGURG, 0);
3510 	selwakeuppri(&so->so_rdsel, PSOCK);
3511 }
3512 
3513 int
3514 sopoll(struct socket *so, int events, struct ucred *active_cred,
3515     struct thread *td)
3516 {
3517 
3518 	/*
3519 	 * We do not need to set or assert curvnet as long as everyone uses
3520 	 * sopoll_generic().
3521 	 */
3522 	return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred,
3523 	    td));
3524 }
3525 
3526 int
3527 sopoll_generic(struct socket *so, int events, struct ucred *active_cred,
3528     struct thread *td)
3529 {
3530 	int revents;
3531 
3532 	SOCK_LOCK(so);
3533 	if (SOLISTENING(so)) {
3534 		if (!(events & (POLLIN | POLLRDNORM)))
3535 			revents = 0;
3536 		else if (!TAILQ_EMPTY(&so->sol_comp))
3537 			revents = events & (POLLIN | POLLRDNORM);
3538 		else if ((events & POLLINIGNEOF) == 0 && so->so_error)
3539 			revents = (events & (POLLIN | POLLRDNORM)) | POLLHUP;
3540 		else {
3541 			selrecord(td, &so->so_rdsel);
3542 			revents = 0;
3543 		}
3544 	} else {
3545 		revents = 0;
3546 		SOCK_SENDBUF_LOCK(so);
3547 		SOCK_RECVBUF_LOCK(so);
3548 		if (events & (POLLIN | POLLRDNORM))
3549 			if (soreadabledata(so))
3550 				revents |= events & (POLLIN | POLLRDNORM);
3551 		if (events & (POLLOUT | POLLWRNORM))
3552 			if (sowriteable(so))
3553 				revents |= events & (POLLOUT | POLLWRNORM);
3554 		if (events & (POLLPRI | POLLRDBAND))
3555 			if (so->so_oobmark ||
3556 			    (so->so_rcv.sb_state & SBS_RCVATMARK))
3557 				revents |= events & (POLLPRI | POLLRDBAND);
3558 		if ((events & POLLINIGNEOF) == 0) {
3559 			if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3560 				revents |= events & (POLLIN | POLLRDNORM);
3561 				if (so->so_snd.sb_state & SBS_CANTSENDMORE)
3562 					revents |= POLLHUP;
3563 			}
3564 		}
3565 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
3566 			revents |= events & POLLRDHUP;
3567 		if (revents == 0) {
3568 			if (events &
3569 			    (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND | POLLRDHUP)) {
3570 				selrecord(td, &so->so_rdsel);
3571 				so->so_rcv.sb_flags |= SB_SEL;
3572 			}
3573 			if (events & (POLLOUT | POLLWRNORM)) {
3574 				selrecord(td, &so->so_wrsel);
3575 				so->so_snd.sb_flags |= SB_SEL;
3576 			}
3577 		}
3578 		SOCK_RECVBUF_UNLOCK(so);
3579 		SOCK_SENDBUF_UNLOCK(so);
3580 	}
3581 	SOCK_UNLOCK(so);
3582 	return (revents);
3583 }
3584 
3585 int
3586 soo_kqfilter(struct file *fp, struct knote *kn)
3587 {
3588 	struct socket *so = kn->kn_fp->f_data;
3589 	struct sockbuf *sb;
3590 	sb_which which;
3591 	struct knlist *knl;
3592 
3593 	switch (kn->kn_filter) {
3594 	case EVFILT_READ:
3595 		kn->kn_fop = &soread_filtops;
3596 		knl = &so->so_rdsel.si_note;
3597 		sb = &so->so_rcv;
3598 		which = SO_RCV;
3599 		break;
3600 	case EVFILT_WRITE:
3601 		kn->kn_fop = &sowrite_filtops;
3602 		knl = &so->so_wrsel.si_note;
3603 		sb = &so->so_snd;
3604 		which = SO_SND;
3605 		break;
3606 	case EVFILT_EMPTY:
3607 		kn->kn_fop = &soempty_filtops;
3608 		knl = &so->so_wrsel.si_note;
3609 		sb = &so->so_snd;
3610 		which = SO_SND;
3611 		break;
3612 	default:
3613 		return (EINVAL);
3614 	}
3615 
3616 	SOCK_LOCK(so);
3617 	if (SOLISTENING(so)) {
3618 		knlist_add(knl, kn, 1);
3619 	} else {
3620 		SOCK_BUF_LOCK(so, which);
3621 		knlist_add(knl, kn, 1);
3622 		sb->sb_flags |= SB_KNOTE;
3623 		SOCK_BUF_UNLOCK(so, which);
3624 	}
3625 	SOCK_UNLOCK(so);
3626 	return (0);
3627 }
3628 
3629 /*
3630  * Some routines that return EOPNOTSUPP for entry points that are not
3631  * supported by a protocol.  Fill in as needed.
3632  */
3633 int
3634 pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
3635 {
3636 
3637 	return EOPNOTSUPP;
3638 }
3639 
3640 int
3641 pru_aio_queue_notsupp(struct socket *so, struct kaiocb *job)
3642 {
3643 
3644 	return EOPNOTSUPP;
3645 }
3646 
3647 int
3648 pru_attach_notsupp(struct socket *so, int proto, struct thread *td)
3649 {
3650 
3651 	return EOPNOTSUPP;
3652 }
3653 
3654 int
3655 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3656 {
3657 
3658 	return EOPNOTSUPP;
3659 }
3660 
3661 int
3662 pru_bindat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3663     struct thread *td)
3664 {
3665 
3666 	return EOPNOTSUPP;
3667 }
3668 
3669 int
3670 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3671 {
3672 
3673 	return EOPNOTSUPP;
3674 }
3675 
3676 int
3677 pru_connectat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3678     struct thread *td)
3679 {
3680 
3681 	return EOPNOTSUPP;
3682 }
3683 
3684 int
3685 pru_connect2_notsupp(struct socket *so1, struct socket *so2)
3686 {
3687 
3688 	return EOPNOTSUPP;
3689 }
3690 
3691 int
3692 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
3693     struct ifnet *ifp, struct thread *td)
3694 {
3695 
3696 	return EOPNOTSUPP;
3697 }
3698 
3699 int
3700 pru_disconnect_notsupp(struct socket *so)
3701 {
3702 
3703 	return EOPNOTSUPP;
3704 }
3705 
3706 int
3707 pru_listen_notsupp(struct socket *so, int backlog, struct thread *td)
3708 {
3709 
3710 	return EOPNOTSUPP;
3711 }
3712 
3713 int
3714 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam)
3715 {
3716 
3717 	return EOPNOTSUPP;
3718 }
3719 
3720 int
3721 pru_rcvd_notsupp(struct socket *so, int flags)
3722 {
3723 
3724 	return EOPNOTSUPP;
3725 }
3726 
3727 int
3728 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
3729 {
3730 
3731 	return EOPNOTSUPP;
3732 }
3733 
3734 int
3735 pru_send_notsupp(struct socket *so, int flags, struct mbuf *m,
3736     struct sockaddr *addr, struct mbuf *control, struct thread *td)
3737 {
3738 
3739 	if (control != NULL)
3740 		m_freem(control);
3741 	if ((flags & PRUS_NOTREADY) == 0)
3742 		m_freem(m);
3743 	return (EOPNOTSUPP);
3744 }
3745 
3746 int
3747 pru_ready_notsupp(struct socket *so, struct mbuf *m, int count)
3748 {
3749 
3750 	return (EOPNOTSUPP);
3751 }
3752 
3753 /*
3754  * This isn't really a ``null'' operation, but it's the default one and
3755  * doesn't do anything destructive.
3756  */
3757 int
3758 pru_sense_null(struct socket *so, struct stat *sb)
3759 {
3760 
3761 	sb->st_blksize = so->so_snd.sb_hiwat;
3762 	return 0;
3763 }
3764 
3765 int
3766 pru_shutdown_notsupp(struct socket *so)
3767 {
3768 
3769 	return EOPNOTSUPP;
3770 }
3771 
3772 int
3773 pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam)
3774 {
3775 
3776 	return EOPNOTSUPP;
3777 }
3778 
3779 int
3780 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio,
3781     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
3782 {
3783 
3784 	return EOPNOTSUPP;
3785 }
3786 
3787 int
3788 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr,
3789     struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
3790 {
3791 
3792 	return EOPNOTSUPP;
3793 }
3794 
3795 int
3796 pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred,
3797     struct thread *td)
3798 {
3799 
3800 	return EOPNOTSUPP;
3801 }
3802 
3803 static void
3804 filt_sordetach(struct knote *kn)
3805 {
3806 	struct socket *so = kn->kn_fp->f_data;
3807 
3808 	so_rdknl_lock(so);
3809 	knlist_remove(&so->so_rdsel.si_note, kn, 1);
3810 	if (!SOLISTENING(so) && knlist_empty(&so->so_rdsel.si_note))
3811 		so->so_rcv.sb_flags &= ~SB_KNOTE;
3812 	so_rdknl_unlock(so);
3813 }
3814 
3815 /*ARGSUSED*/
3816 static int
3817 filt_soread(struct knote *kn, long hint)
3818 {
3819 	struct socket *so;
3820 
3821 	so = kn->kn_fp->f_data;
3822 
3823 	if (SOLISTENING(so)) {
3824 		SOCK_LOCK_ASSERT(so);
3825 		kn->kn_data = so->sol_qlen;
3826 		if (so->so_error) {
3827 			kn->kn_flags |= EV_EOF;
3828 			kn->kn_fflags = so->so_error;
3829 			return (1);
3830 		}
3831 		return (!TAILQ_EMPTY(&so->sol_comp));
3832 	}
3833 
3834 	SOCK_RECVBUF_LOCK_ASSERT(so);
3835 
3836 	kn->kn_data = sbavail(&so->so_rcv) - so->so_rcv.sb_ctl;
3837 	if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3838 		kn->kn_flags |= EV_EOF;
3839 		kn->kn_fflags = so->so_error;
3840 		return (1);
3841 	} else if (so->so_error || so->so_rerror)
3842 		return (1);
3843 
3844 	if (kn->kn_sfflags & NOTE_LOWAT) {
3845 		if (kn->kn_data >= kn->kn_sdata)
3846 			return (1);
3847 	} else if (sbavail(&so->so_rcv) >= so->so_rcv.sb_lowat)
3848 		return (1);
3849 
3850 	/* This hook returning non-zero indicates an event, not error */
3851 	return (hhook_run_socket(so, NULL, HHOOK_FILT_SOREAD));
3852 }
3853 
3854 static void
3855 filt_sowdetach(struct knote *kn)
3856 {
3857 	struct socket *so = kn->kn_fp->f_data;
3858 
3859 	so_wrknl_lock(so);
3860 	knlist_remove(&so->so_wrsel.si_note, kn, 1);
3861 	if (!SOLISTENING(so) && knlist_empty(&so->so_wrsel.si_note))
3862 		so->so_snd.sb_flags &= ~SB_KNOTE;
3863 	so_wrknl_unlock(so);
3864 }
3865 
3866 /*ARGSUSED*/
3867 static int
3868 filt_sowrite(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 (0);
3876 
3877 	SOCK_SENDBUF_LOCK_ASSERT(so);
3878 	kn->kn_data = sbspace(&so->so_snd);
3879 
3880 	hhook_run_socket(so, kn, HHOOK_FILT_SOWRITE);
3881 
3882 	if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
3883 		kn->kn_flags |= EV_EOF;
3884 		kn->kn_fflags = so->so_error;
3885 		return (1);
3886 	} else if (so->so_error)	/* temporary udp error */
3887 		return (1);
3888 	else if (((so->so_state & SS_ISCONNECTED) == 0) &&
3889 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
3890 		return (0);
3891 	else if (kn->kn_sfflags & NOTE_LOWAT)
3892 		return (kn->kn_data >= kn->kn_sdata);
3893 	else
3894 		return (kn->kn_data >= so->so_snd.sb_lowat);
3895 }
3896 
3897 static int
3898 filt_soempty(struct knote *kn, long hint)
3899 {
3900 	struct socket *so;
3901 
3902 	so = kn->kn_fp->f_data;
3903 
3904 	if (SOLISTENING(so))
3905 		return (1);
3906 
3907 	SOCK_SENDBUF_LOCK_ASSERT(so);
3908 	kn->kn_data = sbused(&so->so_snd);
3909 
3910 	if (kn->kn_data == 0)
3911 		return (1);
3912 	else
3913 		return (0);
3914 }
3915 
3916 int
3917 socheckuid(struct socket *so, uid_t uid)
3918 {
3919 
3920 	if (so == NULL)
3921 		return (EPERM);
3922 	if (so->so_cred->cr_uid != uid)
3923 		return (EPERM);
3924 	return (0);
3925 }
3926 
3927 /*
3928  * These functions are used by protocols to notify the socket layer (and its
3929  * consumers) of state changes in the sockets driven by protocol-side events.
3930  */
3931 
3932 /*
3933  * Procedures to manipulate state flags of socket and do appropriate wakeups.
3934  *
3935  * Normal sequence from the active (originating) side is that
3936  * soisconnecting() is called during processing of connect() call, resulting
3937  * in an eventual call to soisconnected() if/when the connection is
3938  * established.  When the connection is torn down soisdisconnecting() is
3939  * called during processing of disconnect() call, and soisdisconnected() is
3940  * called when the connection to the peer is totally severed.  The semantics
3941  * of these routines are such that connectionless protocols can call
3942  * soisconnected() and soisdisconnected() only, bypassing the in-progress
3943  * calls when setting up a ``connection'' takes no time.
3944  *
3945  * From the passive side, a socket is created with two queues of sockets:
3946  * so_incomp for connections in progress and so_comp for connections already
3947  * made and awaiting user acceptance.  As a protocol is preparing incoming
3948  * connections, it creates a socket structure queued on so_incomp by calling
3949  * sonewconn().  When the connection is established, soisconnected() is
3950  * called, and transfers the socket structure to so_comp, making it available
3951  * to accept().
3952  *
3953  * If a socket is closed with sockets on either so_incomp or so_comp, these
3954  * sockets are dropped.
3955  *
3956  * If higher-level protocols are implemented in the kernel, the wakeups done
3957  * here will sometimes cause software-interrupt process scheduling.
3958  */
3959 void
3960 soisconnecting(struct socket *so)
3961 {
3962 
3963 	SOCK_LOCK(so);
3964 	so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
3965 	so->so_state |= SS_ISCONNECTING;
3966 	SOCK_UNLOCK(so);
3967 }
3968 
3969 void
3970 soisconnected(struct socket *so)
3971 {
3972 	bool last __diagused;
3973 
3974 	SOCK_LOCK(so);
3975 	so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
3976 	so->so_state |= SS_ISCONNECTED;
3977 
3978 	if (so->so_qstate == SQ_INCOMP) {
3979 		struct socket *head = so->so_listen;
3980 		int ret;
3981 
3982 		KASSERT(head, ("%s: so %p on incomp of NULL", __func__, so));
3983 		/*
3984 		 * Promoting a socket from incomplete queue to complete, we
3985 		 * need to go through reverse order of locking.  We first do
3986 		 * trylock, and if that doesn't succeed, we go the hard way
3987 		 * leaving a reference and rechecking consistency after proper
3988 		 * locking.
3989 		 */
3990 		if (__predict_false(SOLISTEN_TRYLOCK(head) == 0)) {
3991 			soref(head);
3992 			SOCK_UNLOCK(so);
3993 			SOLISTEN_LOCK(head);
3994 			SOCK_LOCK(so);
3995 			if (__predict_false(head != so->so_listen)) {
3996 				/*
3997 				 * The socket went off the listen queue,
3998 				 * should be lost race to close(2) of sol.
3999 				 * The socket is about to soabort().
4000 				 */
4001 				SOCK_UNLOCK(so);
4002 				sorele_locked(head);
4003 				return;
4004 			}
4005 			last = refcount_release(&head->so_count);
4006 			KASSERT(!last, ("%s: released last reference for %p",
4007 			    __func__, head));
4008 		}
4009 again:
4010 		if ((so->so_options & SO_ACCEPTFILTER) == 0) {
4011 			TAILQ_REMOVE(&head->sol_incomp, so, so_list);
4012 			head->sol_incqlen--;
4013 			TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list);
4014 			head->sol_qlen++;
4015 			so->so_qstate = SQ_COMP;
4016 			SOCK_UNLOCK(so);
4017 			solisten_wakeup(head);	/* unlocks */
4018 		} else {
4019 			SOCK_RECVBUF_LOCK(so);
4020 			soupcall_set(so, SO_RCV,
4021 			    head->sol_accept_filter->accf_callback,
4022 			    head->sol_accept_filter_arg);
4023 			so->so_options &= ~SO_ACCEPTFILTER;
4024 			ret = head->sol_accept_filter->accf_callback(so,
4025 			    head->sol_accept_filter_arg, M_NOWAIT);
4026 			if (ret == SU_ISCONNECTED) {
4027 				soupcall_clear(so, SO_RCV);
4028 				SOCK_RECVBUF_UNLOCK(so);
4029 				goto again;
4030 			}
4031 			SOCK_RECVBUF_UNLOCK(so);
4032 			SOCK_UNLOCK(so);
4033 			SOLISTEN_UNLOCK(head);
4034 		}
4035 		return;
4036 	}
4037 	SOCK_UNLOCK(so);
4038 	wakeup(&so->so_timeo);
4039 	sorwakeup(so);
4040 	sowwakeup(so);
4041 }
4042 
4043 void
4044 soisdisconnecting(struct socket *so)
4045 {
4046 
4047 	SOCK_LOCK(so);
4048 	so->so_state &= ~SS_ISCONNECTING;
4049 	so->so_state |= SS_ISDISCONNECTING;
4050 
4051 	if (!SOLISTENING(so)) {
4052 		SOCK_RECVBUF_LOCK(so);
4053 		socantrcvmore_locked(so);
4054 		SOCK_SENDBUF_LOCK(so);
4055 		socantsendmore_locked(so);
4056 	}
4057 	SOCK_UNLOCK(so);
4058 	wakeup(&so->so_timeo);
4059 }
4060 
4061 void
4062 soisdisconnected(struct socket *so)
4063 {
4064 
4065 	SOCK_LOCK(so);
4066 
4067 	/*
4068 	 * There is at least one reader of so_state that does not
4069 	 * acquire socket lock, namely soreceive_generic().  Ensure
4070 	 * that it never sees all flags that track connection status
4071 	 * cleared, by ordering the update with a barrier semantic of
4072 	 * our release thread fence.
4073 	 */
4074 	so->so_state |= SS_ISDISCONNECTED;
4075 	atomic_thread_fence_rel();
4076 	so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
4077 
4078 	if (!SOLISTENING(so)) {
4079 		SOCK_UNLOCK(so);
4080 		SOCK_RECVBUF_LOCK(so);
4081 		socantrcvmore_locked(so);
4082 		SOCK_SENDBUF_LOCK(so);
4083 		sbdrop_locked(&so->so_snd, sbused(&so->so_snd));
4084 		socantsendmore_locked(so);
4085 	} else
4086 		SOCK_UNLOCK(so);
4087 	wakeup(&so->so_timeo);
4088 }
4089 
4090 int
4091 soiolock(struct socket *so, struct sx *sx, int flags)
4092 {
4093 	int error;
4094 
4095 	KASSERT((flags & SBL_VALID) == flags,
4096 	    ("soiolock: invalid flags %#x", flags));
4097 
4098 	if ((flags & SBL_WAIT) != 0) {
4099 		if ((flags & SBL_NOINTR) != 0) {
4100 			sx_xlock(sx);
4101 		} else {
4102 			error = sx_xlock_sig(sx);
4103 			if (error != 0)
4104 				return (error);
4105 		}
4106 	} else if (!sx_try_xlock(sx)) {
4107 		return (EWOULDBLOCK);
4108 	}
4109 
4110 	if (__predict_false(SOLISTENING(so))) {
4111 		sx_xunlock(sx);
4112 		return (ENOTCONN);
4113 	}
4114 	return (0);
4115 }
4116 
4117 void
4118 soiounlock(struct sx *sx)
4119 {
4120 	sx_xunlock(sx);
4121 }
4122 
4123 /*
4124  * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
4125  */
4126 struct sockaddr *
4127 sodupsockaddr(const struct sockaddr *sa, int mflags)
4128 {
4129 	struct sockaddr *sa2;
4130 
4131 	sa2 = malloc(sa->sa_len, M_SONAME, mflags);
4132 	if (sa2)
4133 		bcopy(sa, sa2, sa->sa_len);
4134 	return sa2;
4135 }
4136 
4137 /*
4138  * Register per-socket destructor.
4139  */
4140 void
4141 sodtor_set(struct socket *so, so_dtor_t *func)
4142 {
4143 
4144 	SOCK_LOCK_ASSERT(so);
4145 	so->so_dtor = func;
4146 }
4147 
4148 /*
4149  * Register per-socket buffer upcalls.
4150  */
4151 void
4152 soupcall_set(struct socket *so, sb_which which, so_upcall_t func, void *arg)
4153 {
4154 	struct sockbuf *sb;
4155 
4156 	KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so));
4157 
4158 	switch (which) {
4159 	case SO_RCV:
4160 		sb = &so->so_rcv;
4161 		break;
4162 	case SO_SND:
4163 		sb = &so->so_snd;
4164 		break;
4165 	}
4166 	SOCK_BUF_LOCK_ASSERT(so, which);
4167 	sb->sb_upcall = func;
4168 	sb->sb_upcallarg = arg;
4169 	sb->sb_flags |= SB_UPCALL;
4170 }
4171 
4172 void
4173 soupcall_clear(struct socket *so, sb_which which)
4174 {
4175 	struct sockbuf *sb;
4176 
4177 	KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so));
4178 
4179 	switch (which) {
4180 	case SO_RCV:
4181 		sb = &so->so_rcv;
4182 		break;
4183 	case SO_SND:
4184 		sb = &so->so_snd;
4185 		break;
4186 	}
4187 	SOCK_BUF_LOCK_ASSERT(so, which);
4188 	KASSERT(sb->sb_upcall != NULL,
4189 	    ("%s: so %p no upcall to clear", __func__, so));
4190 	sb->sb_upcall = NULL;
4191 	sb->sb_upcallarg = NULL;
4192 	sb->sb_flags &= ~SB_UPCALL;
4193 }
4194 
4195 void
4196 solisten_upcall_set(struct socket *so, so_upcall_t func, void *arg)
4197 {
4198 
4199 	SOLISTEN_LOCK_ASSERT(so);
4200 	so->sol_upcall = func;
4201 	so->sol_upcallarg = arg;
4202 }
4203 
4204 static void
4205 so_rdknl_lock(void *arg)
4206 {
4207 	struct socket *so = arg;
4208 
4209 retry:
4210 	if (SOLISTENING(so)) {
4211 		SOLISTEN_LOCK(so);
4212 	} else {
4213 		SOCK_RECVBUF_LOCK(so);
4214 		if (__predict_false(SOLISTENING(so))) {
4215 			SOCK_RECVBUF_UNLOCK(so);
4216 			goto retry;
4217 		}
4218 	}
4219 }
4220 
4221 static void
4222 so_rdknl_unlock(void *arg)
4223 {
4224 	struct socket *so = arg;
4225 
4226 	if (SOLISTENING(so))
4227 		SOLISTEN_UNLOCK(so);
4228 	else
4229 		SOCK_RECVBUF_UNLOCK(so);
4230 }
4231 
4232 static void
4233 so_rdknl_assert_lock(void *arg, int what)
4234 {
4235 	struct socket *so = arg;
4236 
4237 	if (what == LA_LOCKED) {
4238 		if (SOLISTENING(so))
4239 			SOLISTEN_LOCK_ASSERT(so);
4240 		else
4241 			SOCK_RECVBUF_LOCK_ASSERT(so);
4242 	} else {
4243 		if (SOLISTENING(so))
4244 			SOLISTEN_UNLOCK_ASSERT(so);
4245 		else
4246 			SOCK_RECVBUF_UNLOCK_ASSERT(so);
4247 	}
4248 }
4249 
4250 static void
4251 so_wrknl_lock(void *arg)
4252 {
4253 	struct socket *so = arg;
4254 
4255 retry:
4256 	if (SOLISTENING(so)) {
4257 		SOLISTEN_LOCK(so);
4258 	} else {
4259 		SOCK_SENDBUF_LOCK(so);
4260 		if (__predict_false(SOLISTENING(so))) {
4261 			SOCK_SENDBUF_UNLOCK(so);
4262 			goto retry;
4263 		}
4264 	}
4265 }
4266 
4267 static void
4268 so_wrknl_unlock(void *arg)
4269 {
4270 	struct socket *so = arg;
4271 
4272 	if (SOLISTENING(so))
4273 		SOLISTEN_UNLOCK(so);
4274 	else
4275 		SOCK_SENDBUF_UNLOCK(so);
4276 }
4277 
4278 static void
4279 so_wrknl_assert_lock(void *arg, int what)
4280 {
4281 	struct socket *so = arg;
4282 
4283 	if (what == LA_LOCKED) {
4284 		if (SOLISTENING(so))
4285 			SOLISTEN_LOCK_ASSERT(so);
4286 		else
4287 			SOCK_SENDBUF_LOCK_ASSERT(so);
4288 	} else {
4289 		if (SOLISTENING(so))
4290 			SOLISTEN_UNLOCK_ASSERT(so);
4291 		else
4292 			SOCK_SENDBUF_UNLOCK_ASSERT(so);
4293 	}
4294 }
4295 
4296 /*
4297  * Create an external-format (``xsocket'') structure using the information in
4298  * the kernel-format socket structure pointed to by so.  This is done to
4299  * reduce the spew of irrelevant information over this interface, to isolate
4300  * user code from changes in the kernel structure, and potentially to provide
4301  * information-hiding if we decide that some of this information should be
4302  * hidden from users.
4303  */
4304 void
4305 sotoxsocket(struct socket *so, struct xsocket *xso)
4306 {
4307 
4308 	bzero(xso, sizeof(*xso));
4309 	xso->xso_len = sizeof *xso;
4310 	xso->xso_so = (uintptr_t)so;
4311 	xso->so_type = so->so_type;
4312 	xso->so_options = so->so_options;
4313 	xso->so_linger = so->so_linger;
4314 	xso->so_state = so->so_state;
4315 	xso->so_pcb = (uintptr_t)so->so_pcb;
4316 	xso->xso_protocol = so->so_proto->pr_protocol;
4317 	xso->xso_family = so->so_proto->pr_domain->dom_family;
4318 	xso->so_timeo = so->so_timeo;
4319 	xso->so_error = so->so_error;
4320 	xso->so_uid = so->so_cred->cr_uid;
4321 	xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
4322 	if (SOLISTENING(so)) {
4323 		xso->so_qlen = so->sol_qlen;
4324 		xso->so_incqlen = so->sol_incqlen;
4325 		xso->so_qlimit = so->sol_qlimit;
4326 		xso->so_oobmark = 0;
4327 	} else {
4328 		xso->so_state |= so->so_qstate;
4329 		xso->so_qlen = xso->so_incqlen = xso->so_qlimit = 0;
4330 		xso->so_oobmark = so->so_oobmark;
4331 		sbtoxsockbuf(&so->so_snd, &xso->so_snd);
4332 		sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
4333 	}
4334 }
4335 
4336 struct sockbuf *
4337 so_sockbuf_rcv(struct socket *so)
4338 {
4339 
4340 	return (&so->so_rcv);
4341 }
4342 
4343 struct sockbuf *
4344 so_sockbuf_snd(struct socket *so)
4345 {
4346 
4347 	return (&so->so_snd);
4348 }
4349 
4350 int
4351 so_state_get(const struct socket *so)
4352 {
4353 
4354 	return (so->so_state);
4355 }
4356 
4357 void
4358 so_state_set(struct socket *so, int val)
4359 {
4360 
4361 	so->so_state = val;
4362 }
4363 
4364 int
4365 so_options_get(const struct socket *so)
4366 {
4367 
4368 	return (so->so_options);
4369 }
4370 
4371 void
4372 so_options_set(struct socket *so, int val)
4373 {
4374 
4375 	so->so_options = val;
4376 }
4377 
4378 int
4379 so_error_get(const struct socket *so)
4380 {
4381 
4382 	return (so->so_error);
4383 }
4384 
4385 void
4386 so_error_set(struct socket *so, int val)
4387 {
4388 
4389 	so->so_error = val;
4390 }
4391 
4392 int
4393 so_linger_get(const struct socket *so)
4394 {
4395 
4396 	return (so->so_linger);
4397 }
4398 
4399 void
4400 so_linger_set(struct socket *so, int val)
4401 {
4402 
4403 	KASSERT(val >= 0 && val <= USHRT_MAX && val <= (INT_MAX / hz),
4404 	    ("%s: val %d out of range", __func__, val));
4405 
4406 	so->so_linger = val;
4407 }
4408 
4409 struct protosw *
4410 so_protosw_get(const struct socket *so)
4411 {
4412 
4413 	return (so->so_proto);
4414 }
4415 
4416 void
4417 so_protosw_set(struct socket *so, struct protosw *val)
4418 {
4419 
4420 	so->so_proto = val;
4421 }
4422 
4423 void
4424 so_sorwakeup(struct socket *so)
4425 {
4426 
4427 	sorwakeup(so);
4428 }
4429 
4430 void
4431 so_sowwakeup(struct socket *so)
4432 {
4433 
4434 	sowwakeup(so);
4435 }
4436 
4437 void
4438 so_sorwakeup_locked(struct socket *so)
4439 {
4440 
4441 	sorwakeup_locked(so);
4442 }
4443 
4444 void
4445 so_sowwakeup_locked(struct socket *so)
4446 {
4447 
4448 	sowwakeup_locked(so);
4449 }
4450 
4451 void
4452 so_lock(struct socket *so)
4453 {
4454 
4455 	SOCK_LOCK(so);
4456 }
4457 
4458 void
4459 so_unlock(struct socket *so)
4460 {
4461 
4462 	SOCK_UNLOCK(so);
4463 }
4464