xref: /freebsd/sys/kern/uipc_socket.c (revision 643ac419fafba89f5adda0e0ea75b538727453fb)
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 
1900 	mp = mp0;
1901 	if (psa != NULL)
1902 		*psa = NULL;
1903 	if (controlp != NULL)
1904 		*controlp = NULL;
1905 	if (flagsp != NULL)
1906 		flags = *flagsp &~ MSG_EOR;
1907 	else
1908 		flags = 0;
1909 	if (flags & MSG_OOB)
1910 		return (soreceive_rcvoob(so, uio, flags));
1911 	if (mp != NULL)
1912 		*mp = NULL;
1913 	if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING)
1914 	    && uio->uio_resid) {
1915 		VNET_SO_ASSERT(so);
1916 		(*pr->pr_usrreqs->pru_rcvd)(so, 0);
1917 	}
1918 
1919 	error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags));
1920 	if (error)
1921 		return (error);
1922 
1923 restart:
1924 	SOCKBUF_LOCK(&so->so_rcv);
1925 	m = so->so_rcv.sb_mb;
1926 	/*
1927 	 * If we have less data than requested, block awaiting more (subject
1928 	 * to any timeout) if:
1929 	 *   1. the current count is less than the low water mark, or
1930 	 *   2. MSG_DONTWAIT is not set
1931 	 */
1932 	if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1933 	    sbavail(&so->so_rcv) < uio->uio_resid) &&
1934 	    sbavail(&so->so_rcv) < so->so_rcv.sb_lowat &&
1935 	    m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
1936 		KASSERT(m != NULL || !sbavail(&so->so_rcv),
1937 		    ("receive: m == %p sbavail == %u",
1938 		    m, sbavail(&so->so_rcv)));
1939 		if (so->so_error || so->so_rerror) {
1940 			if (m != NULL)
1941 				goto dontblock;
1942 			if (so->so_error)
1943 				error = so->so_error;
1944 			else
1945 				error = so->so_rerror;
1946 			if ((flags & MSG_PEEK) == 0) {
1947 				if (so->so_error)
1948 					so->so_error = 0;
1949 				else
1950 					so->so_rerror = 0;
1951 			}
1952 			SOCKBUF_UNLOCK(&so->so_rcv);
1953 			goto release;
1954 		}
1955 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1956 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1957 			if (m != NULL)
1958 				goto dontblock;
1959 #ifdef KERN_TLS
1960 			else if (so->so_rcv.sb_tlsdcc == 0 &&
1961 			    so->so_rcv.sb_tlscc == 0) {
1962 #else
1963 			else {
1964 #endif
1965 				SOCKBUF_UNLOCK(&so->so_rcv);
1966 				goto release;
1967 			}
1968 		}
1969 		for (; m != NULL; m = m->m_next)
1970 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
1971 				m = so->so_rcv.sb_mb;
1972 				goto dontblock;
1973 			}
1974 		if ((so->so_state & (SS_ISCONNECTING | SS_ISCONNECTED |
1975 		    SS_ISDISCONNECTING | SS_ISDISCONNECTED)) == 0 &&
1976 		    (so->so_proto->pr_flags & PR_CONNREQUIRED) != 0) {
1977 			SOCKBUF_UNLOCK(&so->so_rcv);
1978 			error = ENOTCONN;
1979 			goto release;
1980 		}
1981 		if (uio->uio_resid == 0) {
1982 			SOCKBUF_UNLOCK(&so->so_rcv);
1983 			goto release;
1984 		}
1985 		if ((so->so_state & SS_NBIO) ||
1986 		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
1987 			SOCKBUF_UNLOCK(&so->so_rcv);
1988 			error = EWOULDBLOCK;
1989 			goto release;
1990 		}
1991 		SBLASTRECORDCHK(&so->so_rcv);
1992 		SBLASTMBUFCHK(&so->so_rcv);
1993 		error = sbwait(so, SO_RCV);
1994 		SOCKBUF_UNLOCK(&so->so_rcv);
1995 		if (error)
1996 			goto release;
1997 		goto restart;
1998 	}
1999 dontblock:
2000 	/*
2001 	 * From this point onward, we maintain 'nextrecord' as a cache of the
2002 	 * pointer to the next record in the socket buffer.  We must keep the
2003 	 * various socket buffer pointers and local stack versions of the
2004 	 * pointers in sync, pushing out modifications before dropping the
2005 	 * socket buffer mutex, and re-reading them when picking it up.
2006 	 *
2007 	 * Otherwise, we will race with the network stack appending new data
2008 	 * or records onto the socket buffer by using inconsistent/stale
2009 	 * versions of the field, possibly resulting in socket buffer
2010 	 * corruption.
2011 	 *
2012 	 * By holding the high-level sblock(), we prevent simultaneous
2013 	 * readers from pulling off the front of the socket buffer.
2014 	 */
2015 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2016 	if (uio->uio_td)
2017 		uio->uio_td->td_ru.ru_msgrcv++;
2018 	KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb"));
2019 	SBLASTRECORDCHK(&so->so_rcv);
2020 	SBLASTMBUFCHK(&so->so_rcv);
2021 	nextrecord = m->m_nextpkt;
2022 	if (pr->pr_flags & PR_ADDR) {
2023 		KASSERT(m->m_type == MT_SONAME,
2024 		    ("m->m_type == %d", m->m_type));
2025 		orig_resid = 0;
2026 		if (psa != NULL)
2027 			*psa = sodupsockaddr(mtod(m, struct sockaddr *),
2028 			    M_NOWAIT);
2029 		if (flags & MSG_PEEK) {
2030 			m = m->m_next;
2031 		} else {
2032 			sbfree(&so->so_rcv, m);
2033 			so->so_rcv.sb_mb = m_free(m);
2034 			m = so->so_rcv.sb_mb;
2035 			sockbuf_pushsync(&so->so_rcv, nextrecord);
2036 		}
2037 	}
2038 
2039 	/*
2040 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
2041 	 * in the first mbuf chain on the socket buffer.  If MSG_PEEK, we
2042 	 * just copy the data; if !MSG_PEEK, we call into the protocol to
2043 	 * perform externalization (or freeing if controlp == NULL).
2044 	 */
2045 	if (m != NULL && m->m_type == MT_CONTROL) {
2046 		struct mbuf *cm = NULL, *cmn;
2047 		struct mbuf **cme = &cm;
2048 #ifdef KERN_TLS
2049 		struct cmsghdr *cmsg;
2050 		struct tls_get_record tgr;
2051 
2052 		/*
2053 		 * For MSG_TLSAPPDATA, check for an alert record.
2054 		 * If found, return ENXIO without removing
2055 		 * it from the receive queue.  This allows a subsequent
2056 		 * call without MSG_TLSAPPDATA to receive it.
2057 		 * Note that, for TLS, there should only be a single
2058 		 * control mbuf with the TLS_GET_RECORD message in it.
2059 		 */
2060 		if (flags & MSG_TLSAPPDATA) {
2061 			cmsg = mtod(m, struct cmsghdr *);
2062 			if (cmsg->cmsg_type == TLS_GET_RECORD &&
2063 			    cmsg->cmsg_len == CMSG_LEN(sizeof(tgr))) {
2064 				memcpy(&tgr, CMSG_DATA(cmsg), sizeof(tgr));
2065 				if (__predict_false(tgr.tls_type ==
2066 				    TLS_RLTYPE_ALERT)) {
2067 					SOCKBUF_UNLOCK(&so->so_rcv);
2068 					error = ENXIO;
2069 					goto release;
2070 				}
2071 			}
2072 		}
2073 #endif
2074 
2075 		do {
2076 			if (flags & MSG_PEEK) {
2077 				if (controlp != NULL) {
2078 					*controlp = m_copym(m, 0, m->m_len,
2079 					    M_NOWAIT);
2080 					controlp = &(*controlp)->m_next;
2081 				}
2082 				m = m->m_next;
2083 			} else {
2084 				sbfree(&so->so_rcv, m);
2085 				so->so_rcv.sb_mb = m->m_next;
2086 				m->m_next = NULL;
2087 				*cme = m;
2088 				cme = &(*cme)->m_next;
2089 				m = so->so_rcv.sb_mb;
2090 			}
2091 		} while (m != NULL && m->m_type == MT_CONTROL);
2092 		if ((flags & MSG_PEEK) == 0)
2093 			sockbuf_pushsync(&so->so_rcv, nextrecord);
2094 		while (cm != NULL) {
2095 			cmn = cm->m_next;
2096 			cm->m_next = NULL;
2097 			if (pr->pr_domain->dom_externalize != NULL) {
2098 				SOCKBUF_UNLOCK(&so->so_rcv);
2099 				VNET_SO_ASSERT(so);
2100 				error = (*pr->pr_domain->dom_externalize)
2101 				    (cm, controlp, flags);
2102 				SOCKBUF_LOCK(&so->so_rcv);
2103 			} else if (controlp != NULL)
2104 				*controlp = cm;
2105 			else
2106 				m_freem(cm);
2107 			if (controlp != NULL) {
2108 				while (*controlp != NULL)
2109 					controlp = &(*controlp)->m_next;
2110 			}
2111 			cm = cmn;
2112 		}
2113 		if (m != NULL)
2114 			nextrecord = so->so_rcv.sb_mb->m_nextpkt;
2115 		else
2116 			nextrecord = so->so_rcv.sb_mb;
2117 		orig_resid = 0;
2118 	}
2119 	if (m != NULL) {
2120 		if ((flags & MSG_PEEK) == 0) {
2121 			KASSERT(m->m_nextpkt == nextrecord,
2122 			    ("soreceive: post-control, nextrecord !sync"));
2123 			if (nextrecord == NULL) {
2124 				KASSERT(so->so_rcv.sb_mb == m,
2125 				    ("soreceive: post-control, sb_mb!=m"));
2126 				KASSERT(so->so_rcv.sb_lastrecord == m,
2127 				    ("soreceive: post-control, lastrecord!=m"));
2128 			}
2129 		}
2130 		type = m->m_type;
2131 		if (type == MT_OOBDATA)
2132 			flags |= MSG_OOB;
2133 	} else {
2134 		if ((flags & MSG_PEEK) == 0) {
2135 			KASSERT(so->so_rcv.sb_mb == nextrecord,
2136 			    ("soreceive: sb_mb != nextrecord"));
2137 			if (so->so_rcv.sb_mb == NULL) {
2138 				KASSERT(so->so_rcv.sb_lastrecord == NULL,
2139 				    ("soreceive: sb_lastercord != NULL"));
2140 			}
2141 		}
2142 	}
2143 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2144 	SBLASTRECORDCHK(&so->so_rcv);
2145 	SBLASTMBUFCHK(&so->so_rcv);
2146 
2147 	/*
2148 	 * Now continue to read any data mbufs off of the head of the socket
2149 	 * buffer until the read request is satisfied.  Note that 'type' is
2150 	 * used to store the type of any mbuf reads that have happened so far
2151 	 * such that soreceive() can stop reading if the type changes, which
2152 	 * causes soreceive() to return only one of regular data and inline
2153 	 * out-of-band data in a single socket receive operation.
2154 	 */
2155 	moff = 0;
2156 	offset = 0;
2157 	while (m != NULL && !(m->m_flags & M_NOTAVAIL) && uio->uio_resid > 0
2158 	    && error == 0) {
2159 		/*
2160 		 * If the type of mbuf has changed since the last mbuf
2161 		 * examined ('type'), end the receive operation.
2162 		 */
2163 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2164 		if (m->m_type == MT_OOBDATA || m->m_type == MT_CONTROL) {
2165 			if (type != m->m_type)
2166 				break;
2167 		} else if (type == MT_OOBDATA)
2168 			break;
2169 		else
2170 		    KASSERT(m->m_type == MT_DATA,
2171 			("m->m_type == %d", m->m_type));
2172 		so->so_rcv.sb_state &= ~SBS_RCVATMARK;
2173 		len = uio->uio_resid;
2174 		if (so->so_oobmark && len > so->so_oobmark - offset)
2175 			len = so->so_oobmark - offset;
2176 		if (len > m->m_len - moff)
2177 			len = m->m_len - moff;
2178 		/*
2179 		 * If mp is set, just pass back the mbufs.  Otherwise copy
2180 		 * them out via the uio, then free.  Sockbuf must be
2181 		 * consistent here (points to current mbuf, it points to next
2182 		 * record) when we drop priority; we must note any additions
2183 		 * to the sockbuf when we block interrupts again.
2184 		 */
2185 		if (mp == NULL) {
2186 			SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2187 			SBLASTRECORDCHK(&so->so_rcv);
2188 			SBLASTMBUFCHK(&so->so_rcv);
2189 			SOCKBUF_UNLOCK(&so->so_rcv);
2190 			if ((m->m_flags & M_EXTPG) != 0)
2191 				error = m_unmapped_uiomove(m, moff, uio,
2192 				    (int)len);
2193 			else
2194 				error = uiomove(mtod(m, char *) + moff,
2195 				    (int)len, uio);
2196 			SOCKBUF_LOCK(&so->so_rcv);
2197 			if (error) {
2198 				/*
2199 				 * The MT_SONAME mbuf has already been removed
2200 				 * from the record, so it is necessary to
2201 				 * remove the data mbufs, if any, to preserve
2202 				 * the invariant in the case of PR_ADDR that
2203 				 * requires MT_SONAME mbufs at the head of
2204 				 * each record.
2205 				 */
2206 				if (pr->pr_flags & PR_ATOMIC &&
2207 				    ((flags & MSG_PEEK) == 0))
2208 					(void)sbdroprecord_locked(&so->so_rcv);
2209 				SOCKBUF_UNLOCK(&so->so_rcv);
2210 				goto release;
2211 			}
2212 		} else
2213 			uio->uio_resid -= len;
2214 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2215 		if (len == m->m_len - moff) {
2216 			if (m->m_flags & M_EOR)
2217 				flags |= MSG_EOR;
2218 			if (flags & MSG_PEEK) {
2219 				m = m->m_next;
2220 				moff = 0;
2221 			} else {
2222 				nextrecord = m->m_nextpkt;
2223 				sbfree(&so->so_rcv, m);
2224 				if (mp != NULL) {
2225 					m->m_nextpkt = NULL;
2226 					*mp = m;
2227 					mp = &m->m_next;
2228 					so->so_rcv.sb_mb = m = m->m_next;
2229 					*mp = NULL;
2230 				} else {
2231 					so->so_rcv.sb_mb = m_free(m);
2232 					m = so->so_rcv.sb_mb;
2233 				}
2234 				sockbuf_pushsync(&so->so_rcv, nextrecord);
2235 				SBLASTRECORDCHK(&so->so_rcv);
2236 				SBLASTMBUFCHK(&so->so_rcv);
2237 			}
2238 		} else {
2239 			if (flags & MSG_PEEK)
2240 				moff += len;
2241 			else {
2242 				if (mp != NULL) {
2243 					if (flags & MSG_DONTWAIT) {
2244 						*mp = m_copym(m, 0, len,
2245 						    M_NOWAIT);
2246 						if (*mp == NULL) {
2247 							/*
2248 							 * m_copym() couldn't
2249 							 * allocate an mbuf.
2250 							 * Adjust uio_resid back
2251 							 * (it was adjusted
2252 							 * down by len bytes,
2253 							 * which we didn't end
2254 							 * up "copying" over).
2255 							 */
2256 							uio->uio_resid += len;
2257 							break;
2258 						}
2259 					} else {
2260 						SOCKBUF_UNLOCK(&so->so_rcv);
2261 						*mp = m_copym(m, 0, len,
2262 						    M_WAITOK);
2263 						SOCKBUF_LOCK(&so->so_rcv);
2264 					}
2265 				}
2266 				sbcut_locked(&so->so_rcv, len);
2267 			}
2268 		}
2269 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2270 		if (so->so_oobmark) {
2271 			if ((flags & MSG_PEEK) == 0) {
2272 				so->so_oobmark -= len;
2273 				if (so->so_oobmark == 0) {
2274 					so->so_rcv.sb_state |= SBS_RCVATMARK;
2275 					break;
2276 				}
2277 			} else {
2278 				offset += len;
2279 				if (offset == so->so_oobmark)
2280 					break;
2281 			}
2282 		}
2283 		if (flags & MSG_EOR)
2284 			break;
2285 		/*
2286 		 * If the MSG_WAITALL flag is set (for non-atomic socket), we
2287 		 * must not quit until "uio->uio_resid == 0" or an error
2288 		 * termination.  If a signal/timeout occurs, return with a
2289 		 * short count but without error.  Keep sockbuf locked
2290 		 * against other readers.
2291 		 */
2292 		while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
2293 		    !sosendallatonce(so) && nextrecord == NULL) {
2294 			SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2295 			if (so->so_error || so->so_rerror ||
2296 			    so->so_rcv.sb_state & SBS_CANTRCVMORE)
2297 				break;
2298 			/*
2299 			 * Notify the protocol that some data has been
2300 			 * drained before blocking.
2301 			 */
2302 			if (pr->pr_flags & PR_WANTRCVD) {
2303 				SOCKBUF_UNLOCK(&so->so_rcv);
2304 				VNET_SO_ASSERT(so);
2305 				(*pr->pr_usrreqs->pru_rcvd)(so, flags);
2306 				SOCKBUF_LOCK(&so->so_rcv);
2307 			}
2308 			SBLASTRECORDCHK(&so->so_rcv);
2309 			SBLASTMBUFCHK(&so->so_rcv);
2310 			/*
2311 			 * We could receive some data while was notifying
2312 			 * the protocol. Skip blocking in this case.
2313 			 */
2314 			if (so->so_rcv.sb_mb == NULL) {
2315 				error = sbwait(so, SO_RCV);
2316 				if (error) {
2317 					SOCKBUF_UNLOCK(&so->so_rcv);
2318 					goto release;
2319 				}
2320 			}
2321 			m = so->so_rcv.sb_mb;
2322 			if (m != NULL)
2323 				nextrecord = m->m_nextpkt;
2324 		}
2325 	}
2326 
2327 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2328 	if (m != NULL && pr->pr_flags & PR_ATOMIC) {
2329 		flags |= MSG_TRUNC;
2330 		if ((flags & MSG_PEEK) == 0)
2331 			(void) sbdroprecord_locked(&so->so_rcv);
2332 	}
2333 	if ((flags & MSG_PEEK) == 0) {
2334 		if (m == NULL) {
2335 			/*
2336 			 * First part is an inline SB_EMPTY_FIXUP().  Second
2337 			 * part makes sure sb_lastrecord is up-to-date if
2338 			 * there is still data in the socket buffer.
2339 			 */
2340 			so->so_rcv.sb_mb = nextrecord;
2341 			if (so->so_rcv.sb_mb == NULL) {
2342 				so->so_rcv.sb_mbtail = NULL;
2343 				so->so_rcv.sb_lastrecord = NULL;
2344 			} else if (nextrecord->m_nextpkt == NULL)
2345 				so->so_rcv.sb_lastrecord = nextrecord;
2346 		}
2347 		SBLASTRECORDCHK(&so->so_rcv);
2348 		SBLASTMBUFCHK(&so->so_rcv);
2349 		/*
2350 		 * If soreceive() is being done from the socket callback,
2351 		 * then don't need to generate ACK to peer to update window,
2352 		 * since ACK will be generated on return to TCP.
2353 		 */
2354 		if (!(flags & MSG_SOCALLBCK) &&
2355 		    (pr->pr_flags & PR_WANTRCVD)) {
2356 			SOCKBUF_UNLOCK(&so->so_rcv);
2357 			VNET_SO_ASSERT(so);
2358 			(*pr->pr_usrreqs->pru_rcvd)(so, flags);
2359 			SOCKBUF_LOCK(&so->so_rcv);
2360 		}
2361 	}
2362 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2363 	if (orig_resid == uio->uio_resid && orig_resid &&
2364 	    (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
2365 		SOCKBUF_UNLOCK(&so->so_rcv);
2366 		goto restart;
2367 	}
2368 	SOCKBUF_UNLOCK(&so->so_rcv);
2369 
2370 	if (flagsp != NULL)
2371 		*flagsp |= flags;
2372 release:
2373 	SOCK_IO_RECV_UNLOCK(so);
2374 	return (error);
2375 }
2376 
2377 /*
2378  * Optimized version of soreceive() for stream (TCP) sockets.
2379  */
2380 int
2381 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio,
2382     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2383 {
2384 	int len = 0, error = 0, flags, oresid;
2385 	struct sockbuf *sb;
2386 	struct mbuf *m, *n = NULL;
2387 
2388 	/* We only do stream sockets. */
2389 	if (so->so_type != SOCK_STREAM)
2390 		return (EINVAL);
2391 	if (psa != NULL)
2392 		*psa = NULL;
2393 	if (flagsp != NULL)
2394 		flags = *flagsp &~ MSG_EOR;
2395 	else
2396 		flags = 0;
2397 	if (controlp != NULL)
2398 		*controlp = NULL;
2399 	if (flags & MSG_OOB)
2400 		return (soreceive_rcvoob(so, uio, flags));
2401 	if (mp0 != NULL)
2402 		*mp0 = NULL;
2403 
2404 	sb = &so->so_rcv;
2405 
2406 #ifdef KERN_TLS
2407 	/*
2408 	 * KTLS store TLS records as records with a control message to
2409 	 * describe the framing.
2410 	 *
2411 	 * We check once here before acquiring locks to optimize the
2412 	 * common case.
2413 	 */
2414 	if (sb->sb_tls_info != NULL)
2415 		return (soreceive_generic(so, psa, uio, mp0, controlp,
2416 		    flagsp));
2417 #endif
2418 
2419 	/* Prevent other readers from entering the socket. */
2420 	error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags));
2421 	if (error)
2422 		return (error);
2423 	SOCKBUF_LOCK(sb);
2424 
2425 #ifdef KERN_TLS
2426 	if (sb->sb_tls_info != NULL) {
2427 		SOCKBUF_UNLOCK(sb);
2428 		SOCK_IO_RECV_UNLOCK(so);
2429 		return (soreceive_generic(so, psa, uio, mp0, controlp,
2430 		    flagsp));
2431 	}
2432 #endif
2433 
2434 	/* Easy one, no space to copyout anything. */
2435 	if (uio->uio_resid == 0) {
2436 		error = EINVAL;
2437 		goto out;
2438 	}
2439 	oresid = uio->uio_resid;
2440 
2441 	/* We will never ever get anything unless we are or were connected. */
2442 	if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) {
2443 		error = ENOTCONN;
2444 		goto out;
2445 	}
2446 
2447 restart:
2448 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2449 
2450 	/* Abort if socket has reported problems. */
2451 	if (so->so_error) {
2452 		if (sbavail(sb) > 0)
2453 			goto deliver;
2454 		if (oresid > uio->uio_resid)
2455 			goto out;
2456 		error = so->so_error;
2457 		if (!(flags & MSG_PEEK))
2458 			so->so_error = 0;
2459 		goto out;
2460 	}
2461 
2462 	/* Door is closed.  Deliver what is left, if any. */
2463 	if (sb->sb_state & SBS_CANTRCVMORE) {
2464 		if (sbavail(sb) > 0)
2465 			goto deliver;
2466 		else
2467 			goto out;
2468 	}
2469 
2470 	/* Socket buffer is empty and we shall not block. */
2471 	if (sbavail(sb) == 0 &&
2472 	    ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) {
2473 		error = EAGAIN;
2474 		goto out;
2475 	}
2476 
2477 	/* Socket buffer got some data that we shall deliver now. */
2478 	if (sbavail(sb) > 0 && !(flags & MSG_WAITALL) &&
2479 	    ((so->so_state & SS_NBIO) ||
2480 	     (flags & (MSG_DONTWAIT|MSG_NBIO)) ||
2481 	     sbavail(sb) >= sb->sb_lowat ||
2482 	     sbavail(sb) >= uio->uio_resid ||
2483 	     sbavail(sb) >= sb->sb_hiwat) ) {
2484 		goto deliver;
2485 	}
2486 
2487 	/* On MSG_WAITALL we must wait until all data or error arrives. */
2488 	if ((flags & MSG_WAITALL) &&
2489 	    (sbavail(sb) >= uio->uio_resid || sbavail(sb) >= sb->sb_hiwat))
2490 		goto deliver;
2491 
2492 	/*
2493 	 * Wait and block until (more) data comes in.
2494 	 * NB: Drops the sockbuf lock during wait.
2495 	 */
2496 	error = sbwait(so, SO_RCV);
2497 	if (error)
2498 		goto out;
2499 	goto restart;
2500 
2501 deliver:
2502 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2503 	KASSERT(sbavail(sb) > 0, ("%s: sockbuf empty", __func__));
2504 	KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__));
2505 
2506 	/* Statistics. */
2507 	if (uio->uio_td)
2508 		uio->uio_td->td_ru.ru_msgrcv++;
2509 
2510 	/* Fill uio until full or current end of socket buffer is reached. */
2511 	len = min(uio->uio_resid, sbavail(sb));
2512 	if (mp0 != NULL) {
2513 		/* Dequeue as many mbufs as possible. */
2514 		if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) {
2515 			if (*mp0 == NULL)
2516 				*mp0 = sb->sb_mb;
2517 			else
2518 				m_cat(*mp0, sb->sb_mb);
2519 			for (m = sb->sb_mb;
2520 			     m != NULL && m->m_len <= len;
2521 			     m = m->m_next) {
2522 				KASSERT(!(m->m_flags & M_NOTAVAIL),
2523 				    ("%s: m %p not available", __func__, m));
2524 				len -= m->m_len;
2525 				uio->uio_resid -= m->m_len;
2526 				sbfree(sb, m);
2527 				n = m;
2528 			}
2529 			n->m_next = NULL;
2530 			sb->sb_mb = m;
2531 			sb->sb_lastrecord = sb->sb_mb;
2532 			if (sb->sb_mb == NULL)
2533 				SB_EMPTY_FIXUP(sb);
2534 		}
2535 		/* Copy the remainder. */
2536 		if (len > 0) {
2537 			KASSERT(sb->sb_mb != NULL,
2538 			    ("%s: len > 0 && sb->sb_mb empty", __func__));
2539 
2540 			m = m_copym(sb->sb_mb, 0, len, M_NOWAIT);
2541 			if (m == NULL)
2542 				len = 0;	/* Don't flush data from sockbuf. */
2543 			else
2544 				uio->uio_resid -= len;
2545 			if (*mp0 != NULL)
2546 				m_cat(*mp0, m);
2547 			else
2548 				*mp0 = m;
2549 			if (*mp0 == NULL) {
2550 				error = ENOBUFS;
2551 				goto out;
2552 			}
2553 		}
2554 	} else {
2555 		/* NB: Must unlock socket buffer as uiomove may sleep. */
2556 		SOCKBUF_UNLOCK(sb);
2557 		error = m_mbuftouio(uio, sb->sb_mb, len);
2558 		SOCKBUF_LOCK(sb);
2559 		if (error)
2560 			goto out;
2561 	}
2562 	SBLASTRECORDCHK(sb);
2563 	SBLASTMBUFCHK(sb);
2564 
2565 	/*
2566 	 * Remove the delivered data from the socket buffer unless we
2567 	 * were only peeking.
2568 	 */
2569 	if (!(flags & MSG_PEEK)) {
2570 		if (len > 0)
2571 			sbdrop_locked(sb, len);
2572 
2573 		/* Notify protocol that we drained some data. */
2574 		if ((so->so_proto->pr_flags & PR_WANTRCVD) &&
2575 		    (((flags & MSG_WAITALL) && uio->uio_resid > 0) ||
2576 		     !(flags & MSG_SOCALLBCK))) {
2577 			SOCKBUF_UNLOCK(sb);
2578 			VNET_SO_ASSERT(so);
2579 			(*so->so_proto->pr_usrreqs->pru_rcvd)(so, flags);
2580 			SOCKBUF_LOCK(sb);
2581 		}
2582 	}
2583 
2584 	/*
2585 	 * For MSG_WAITALL we may have to loop again and wait for
2586 	 * more data to come in.
2587 	 */
2588 	if ((flags & MSG_WAITALL) && uio->uio_resid > 0)
2589 		goto restart;
2590 out:
2591 	SBLASTRECORDCHK(sb);
2592 	SBLASTMBUFCHK(sb);
2593 	SOCKBUF_UNLOCK(sb);
2594 	SOCK_IO_RECV_UNLOCK(so);
2595 	return (error);
2596 }
2597 
2598 /*
2599  * Optimized version of soreceive() for simple datagram cases from userspace.
2600  * Unlike in the stream case, we're able to drop a datagram if copyout()
2601  * fails, and because we handle datagrams atomically, we don't need to use a
2602  * sleep lock to prevent I/O interlacing.
2603  */
2604 int
2605 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio,
2606     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2607 {
2608 	struct mbuf *m, *m2;
2609 	int flags, error;
2610 	ssize_t len;
2611 	struct protosw *pr = so->so_proto;
2612 	struct mbuf *nextrecord;
2613 
2614 	if (psa != NULL)
2615 		*psa = NULL;
2616 	if (controlp != NULL)
2617 		*controlp = NULL;
2618 	if (flagsp != NULL)
2619 		flags = *flagsp &~ MSG_EOR;
2620 	else
2621 		flags = 0;
2622 
2623 	/*
2624 	 * For any complicated cases, fall back to the full
2625 	 * soreceive_generic().
2626 	 */
2627 	if (mp0 != NULL || (flags & MSG_PEEK) || (flags & MSG_OOB))
2628 		return (soreceive_generic(so, psa, uio, mp0, controlp,
2629 		    flagsp));
2630 
2631 	/*
2632 	 * Enforce restrictions on use.
2633 	 */
2634 	KASSERT((pr->pr_flags & PR_WANTRCVD) == 0,
2635 	    ("soreceive_dgram: wantrcvd"));
2636 	KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic"));
2637 	KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0,
2638 	    ("soreceive_dgram: SBS_RCVATMARK"));
2639 	KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0,
2640 	    ("soreceive_dgram: P_CONNREQUIRED"));
2641 
2642 	/*
2643 	 * Loop blocking while waiting for a datagram.
2644 	 */
2645 	SOCKBUF_LOCK(&so->so_rcv);
2646 	while ((m = so->so_rcv.sb_mb) == NULL) {
2647 		KASSERT(sbavail(&so->so_rcv) == 0,
2648 		    ("soreceive_dgram: sb_mb NULL but sbavail %u",
2649 		    sbavail(&so->so_rcv)));
2650 		if (so->so_error) {
2651 			error = so->so_error;
2652 			so->so_error = 0;
2653 			SOCKBUF_UNLOCK(&so->so_rcv);
2654 			return (error);
2655 		}
2656 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE ||
2657 		    uio->uio_resid == 0) {
2658 			SOCKBUF_UNLOCK(&so->so_rcv);
2659 			return (0);
2660 		}
2661 		if ((so->so_state & SS_NBIO) ||
2662 		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2663 			SOCKBUF_UNLOCK(&so->so_rcv);
2664 			return (EWOULDBLOCK);
2665 		}
2666 		SBLASTRECORDCHK(&so->so_rcv);
2667 		SBLASTMBUFCHK(&so->so_rcv);
2668 		error = sbwait(so, SO_RCV);
2669 		if (error) {
2670 			SOCKBUF_UNLOCK(&so->so_rcv);
2671 			return (error);
2672 		}
2673 	}
2674 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2675 
2676 	if (uio->uio_td)
2677 		uio->uio_td->td_ru.ru_msgrcv++;
2678 	SBLASTRECORDCHK(&so->so_rcv);
2679 	SBLASTMBUFCHK(&so->so_rcv);
2680 	nextrecord = m->m_nextpkt;
2681 	if (nextrecord == NULL) {
2682 		KASSERT(so->so_rcv.sb_lastrecord == m,
2683 		    ("soreceive_dgram: lastrecord != m"));
2684 	}
2685 
2686 	KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord,
2687 	    ("soreceive_dgram: m_nextpkt != nextrecord"));
2688 
2689 	/*
2690 	 * Pull 'm' and its chain off the front of the packet queue.
2691 	 */
2692 	so->so_rcv.sb_mb = NULL;
2693 	sockbuf_pushsync(&so->so_rcv, nextrecord);
2694 
2695 	/*
2696 	 * Walk 'm's chain and free that many bytes from the socket buffer.
2697 	 */
2698 	for (m2 = m; m2 != NULL; m2 = m2->m_next)
2699 		sbfree(&so->so_rcv, m2);
2700 
2701 	/*
2702 	 * Do a few last checks before we let go of the lock.
2703 	 */
2704 	SBLASTRECORDCHK(&so->so_rcv);
2705 	SBLASTMBUFCHK(&so->so_rcv);
2706 	SOCKBUF_UNLOCK(&so->so_rcv);
2707 
2708 	if (pr->pr_flags & PR_ADDR) {
2709 		KASSERT(m->m_type == MT_SONAME,
2710 		    ("m->m_type == %d", m->m_type));
2711 		if (psa != NULL)
2712 			*psa = sodupsockaddr(mtod(m, struct sockaddr *),
2713 			    M_NOWAIT);
2714 		m = m_free(m);
2715 	}
2716 	if (m == NULL) {
2717 		/* XXXRW: Can this happen? */
2718 		return (0);
2719 	}
2720 
2721 	/*
2722 	 * Packet to copyout() is now in 'm' and it is disconnected from the
2723 	 * queue.
2724 	 *
2725 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
2726 	 * in the first mbuf chain on the socket buffer.  We call into the
2727 	 * protocol to perform externalization (or freeing if controlp ==
2728 	 * NULL). In some cases there can be only MT_CONTROL mbufs without
2729 	 * MT_DATA mbufs.
2730 	 */
2731 	if (m->m_type == MT_CONTROL) {
2732 		struct mbuf *cm = NULL, *cmn;
2733 		struct mbuf **cme = &cm;
2734 
2735 		do {
2736 			m2 = m->m_next;
2737 			m->m_next = NULL;
2738 			*cme = m;
2739 			cme = &(*cme)->m_next;
2740 			m = m2;
2741 		} while (m != NULL && m->m_type == MT_CONTROL);
2742 		while (cm != NULL) {
2743 			cmn = cm->m_next;
2744 			cm->m_next = NULL;
2745 			if (pr->pr_domain->dom_externalize != NULL) {
2746 				error = (*pr->pr_domain->dom_externalize)
2747 				    (cm, controlp, flags);
2748 			} else if (controlp != NULL)
2749 				*controlp = cm;
2750 			else
2751 				m_freem(cm);
2752 			if (controlp != NULL) {
2753 				while (*controlp != NULL)
2754 					controlp = &(*controlp)->m_next;
2755 			}
2756 			cm = cmn;
2757 		}
2758 	}
2759 	KASSERT(m == NULL || m->m_type == MT_DATA,
2760 	    ("soreceive_dgram: !data"));
2761 	while (m != NULL && uio->uio_resid > 0) {
2762 		len = uio->uio_resid;
2763 		if (len > m->m_len)
2764 			len = m->m_len;
2765 		error = uiomove(mtod(m, char *), (int)len, uio);
2766 		if (error) {
2767 			m_freem(m);
2768 			return (error);
2769 		}
2770 		if (len == m->m_len)
2771 			m = m_free(m);
2772 		else {
2773 			m->m_data += len;
2774 			m->m_len -= len;
2775 		}
2776 	}
2777 	if (m != NULL) {
2778 		flags |= MSG_TRUNC;
2779 		m_freem(m);
2780 	}
2781 	if (flagsp != NULL)
2782 		*flagsp |= flags;
2783 	return (0);
2784 }
2785 
2786 int
2787 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
2788     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2789 {
2790 	int error;
2791 
2792 	CURVNET_SET(so->so_vnet);
2793 	error = (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio,
2794 	    mp0, controlp, flagsp));
2795 	CURVNET_RESTORE();
2796 	return (error);
2797 }
2798 
2799 int
2800 soshutdown(struct socket *so, int how)
2801 {
2802 	struct protosw *pr;
2803 	int error, soerror_enotconn;
2804 
2805 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
2806 		return (EINVAL);
2807 
2808 	soerror_enotconn = 0;
2809 	SOCK_LOCK(so);
2810 	if ((so->so_state &
2811 	    (SS_ISCONNECTED | SS_ISCONNECTING | SS_ISDISCONNECTING)) == 0) {
2812 		/*
2813 		 * POSIX mandates us to return ENOTCONN when shutdown(2) is
2814 		 * invoked on a datagram sockets, however historically we would
2815 		 * actually tear socket down. This is known to be leveraged by
2816 		 * some applications to unblock process waiting in recvXXX(2)
2817 		 * by other process that it shares that socket with. Try to meet
2818 		 * both backward-compatibility and POSIX requirements by forcing
2819 		 * ENOTCONN but still asking protocol to perform pru_shutdown().
2820 		 */
2821 		if (so->so_type != SOCK_DGRAM && !SOLISTENING(so)) {
2822 			SOCK_UNLOCK(so);
2823 			return (ENOTCONN);
2824 		}
2825 		soerror_enotconn = 1;
2826 	}
2827 
2828 	if (SOLISTENING(so)) {
2829 		if (how != SHUT_WR) {
2830 			so->so_error = ECONNABORTED;
2831 			solisten_wakeup(so);	/* unlocks so */
2832 		} else {
2833 			SOCK_UNLOCK(so);
2834 		}
2835 		goto done;
2836 	}
2837 	SOCK_UNLOCK(so);
2838 
2839 	CURVNET_SET(so->so_vnet);
2840 	pr = so->so_proto;
2841 	if (pr->pr_usrreqs->pru_flush != NULL)
2842 		(*pr->pr_usrreqs->pru_flush)(so, how);
2843 	if (how != SHUT_WR)
2844 		sorflush(so);
2845 	if (how != SHUT_RD) {
2846 		error = (*pr->pr_usrreqs->pru_shutdown)(so);
2847 		wakeup(&so->so_timeo);
2848 		CURVNET_RESTORE();
2849 		return ((error == 0 && soerror_enotconn) ? ENOTCONN : error);
2850 	}
2851 	wakeup(&so->so_timeo);
2852 	CURVNET_RESTORE();
2853 
2854 done:
2855 	return (soerror_enotconn ? ENOTCONN : 0);
2856 }
2857 
2858 void
2859 sorflush(struct socket *so)
2860 {
2861 	struct protosw *pr;
2862 	int error;
2863 
2864 	VNET_SO_ASSERT(so);
2865 
2866 	/*
2867 	 * Dislodge threads currently blocked in receive and wait to acquire
2868 	 * a lock against other simultaneous readers before clearing the
2869 	 * socket buffer.  Don't let our acquire be interrupted by a signal
2870 	 * despite any existing socket disposition on interruptable waiting.
2871 	 */
2872 	socantrcvmore(so);
2873 
2874 	error = SOCK_IO_RECV_LOCK(so, SBL_WAIT | SBL_NOINTR);
2875 	if (error != 0) {
2876 		KASSERT(SOLISTENING(so),
2877 		    ("%s: soiolock(%p) failed", __func__, so));
2878 		return;
2879 	}
2880 
2881 	pr = so->so_proto;
2882 	if (pr->pr_flags & PR_RIGHTS) {
2883 		MPASS(pr->pr_domain->dom_dispose != NULL);
2884 		(*pr->pr_domain->dom_dispose)(so);
2885 	} else {
2886 		sbrelease(so, SO_RCV);
2887 		SOCK_IO_RECV_UNLOCK(so);
2888 	}
2889 
2890 }
2891 
2892 /*
2893  * Wrapper for Socket established helper hook.
2894  * Parameters: socket, context of the hook point, hook id.
2895  */
2896 static int inline
2897 hhook_run_socket(struct socket *so, void *hctx, int32_t h_id)
2898 {
2899 	struct socket_hhook_data hhook_data = {
2900 		.so = so,
2901 		.hctx = hctx,
2902 		.m = NULL,
2903 		.status = 0
2904 	};
2905 
2906 	CURVNET_SET(so->so_vnet);
2907 	HHOOKS_RUN_IF(V_socket_hhh[h_id], &hhook_data, &so->osd);
2908 	CURVNET_RESTORE();
2909 
2910 	/* Ugly but needed, since hhooks return void for now */
2911 	return (hhook_data.status);
2912 }
2913 
2914 /*
2915  * Perhaps this routine, and sooptcopyout(), below, ought to come in an
2916  * additional variant to handle the case where the option value needs to be
2917  * some kind of integer, but not a specific size.  In addition to their use
2918  * here, these functions are also called by the protocol-level pr_ctloutput()
2919  * routines.
2920  */
2921 int
2922 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2923 {
2924 	size_t	valsize;
2925 
2926 	/*
2927 	 * If the user gives us more than we wanted, we ignore it, but if we
2928 	 * don't get the minimum length the caller wants, we return EINVAL.
2929 	 * On success, sopt->sopt_valsize is set to however much we actually
2930 	 * retrieved.
2931 	 */
2932 	if ((valsize = sopt->sopt_valsize) < minlen)
2933 		return EINVAL;
2934 	if (valsize > len)
2935 		sopt->sopt_valsize = valsize = len;
2936 
2937 	if (sopt->sopt_td != NULL)
2938 		return (copyin(sopt->sopt_val, buf, valsize));
2939 
2940 	bcopy(sopt->sopt_val, buf, valsize);
2941 	return (0);
2942 }
2943 
2944 /*
2945  * Kernel version of setsockopt(2).
2946  *
2947  * XXX: optlen is size_t, not socklen_t
2948  */
2949 int
2950 so_setsockopt(struct socket *so, int level, int optname, void *optval,
2951     size_t optlen)
2952 {
2953 	struct sockopt sopt;
2954 
2955 	sopt.sopt_level = level;
2956 	sopt.sopt_name = optname;
2957 	sopt.sopt_dir = SOPT_SET;
2958 	sopt.sopt_val = optval;
2959 	sopt.sopt_valsize = optlen;
2960 	sopt.sopt_td = NULL;
2961 	return (sosetopt(so, &sopt));
2962 }
2963 
2964 int
2965 sosetopt(struct socket *so, struct sockopt *sopt)
2966 {
2967 	int	error, optval;
2968 	struct	linger l;
2969 	struct	timeval tv;
2970 	sbintime_t val, *valp;
2971 	uint32_t val32;
2972 #ifdef MAC
2973 	struct mac extmac;
2974 #endif
2975 
2976 	CURVNET_SET(so->so_vnet);
2977 	error = 0;
2978 	if (sopt->sopt_level != SOL_SOCKET) {
2979 		if (so->so_proto->pr_ctloutput != NULL)
2980 			error = (*so->so_proto->pr_ctloutput)(so, sopt);
2981 		else
2982 			error = ENOPROTOOPT;
2983 	} else {
2984 		switch (sopt->sopt_name) {
2985 		case SO_ACCEPTFILTER:
2986 			error = accept_filt_setopt(so, sopt);
2987 			if (error)
2988 				goto bad;
2989 			break;
2990 
2991 		case SO_LINGER:
2992 			error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
2993 			if (error)
2994 				goto bad;
2995 			if (l.l_linger < 0 ||
2996 			    l.l_linger > USHRT_MAX ||
2997 			    l.l_linger > (INT_MAX / hz)) {
2998 				error = EDOM;
2999 				goto bad;
3000 			}
3001 			SOCK_LOCK(so);
3002 			so->so_linger = l.l_linger;
3003 			if (l.l_onoff)
3004 				so->so_options |= SO_LINGER;
3005 			else
3006 				so->so_options &= ~SO_LINGER;
3007 			SOCK_UNLOCK(so);
3008 			break;
3009 
3010 		case SO_DEBUG:
3011 		case SO_KEEPALIVE:
3012 		case SO_DONTROUTE:
3013 		case SO_USELOOPBACK:
3014 		case SO_BROADCAST:
3015 		case SO_REUSEADDR:
3016 		case SO_REUSEPORT:
3017 		case SO_REUSEPORT_LB:
3018 		case SO_OOBINLINE:
3019 		case SO_TIMESTAMP:
3020 		case SO_BINTIME:
3021 		case SO_NOSIGPIPE:
3022 		case SO_NO_DDP:
3023 		case SO_NO_OFFLOAD:
3024 		case SO_RERROR:
3025 			error = sooptcopyin(sopt, &optval, sizeof optval,
3026 			    sizeof optval);
3027 			if (error)
3028 				goto bad;
3029 			SOCK_LOCK(so);
3030 			if (optval)
3031 				so->so_options |= sopt->sopt_name;
3032 			else
3033 				so->so_options &= ~sopt->sopt_name;
3034 			SOCK_UNLOCK(so);
3035 			break;
3036 
3037 		case SO_SETFIB:
3038 			error = sooptcopyin(sopt, &optval, sizeof optval,
3039 			    sizeof optval);
3040 			if (error)
3041 				goto bad;
3042 
3043 			if (optval < 0 || optval >= rt_numfibs) {
3044 				error = EINVAL;
3045 				goto bad;
3046 			}
3047 			if (((so->so_proto->pr_domain->dom_family == PF_INET) ||
3048 			   (so->so_proto->pr_domain->dom_family == PF_INET6) ||
3049 			   (so->so_proto->pr_domain->dom_family == PF_ROUTE)))
3050 				so->so_fibnum = optval;
3051 			else
3052 				so->so_fibnum = 0;
3053 			break;
3054 
3055 		case SO_USER_COOKIE:
3056 			error = sooptcopyin(sopt, &val32, sizeof val32,
3057 			    sizeof val32);
3058 			if (error)
3059 				goto bad;
3060 			so->so_user_cookie = val32;
3061 			break;
3062 
3063 		case SO_SNDBUF:
3064 		case SO_RCVBUF:
3065 		case SO_SNDLOWAT:
3066 		case SO_RCVLOWAT:
3067 			error = sooptcopyin(sopt, &optval, sizeof optval,
3068 			    sizeof optval);
3069 			if (error)
3070 				goto bad;
3071 
3072 			/*
3073 			 * Values < 1 make no sense for any of these options,
3074 			 * so disallow them.
3075 			 */
3076 			if (optval < 1) {
3077 				error = EINVAL;
3078 				goto bad;
3079 			}
3080 
3081 			error = sbsetopt(so, sopt->sopt_name, optval);
3082 			break;
3083 
3084 		case SO_SNDTIMEO:
3085 		case SO_RCVTIMEO:
3086 #ifdef COMPAT_FREEBSD32
3087 			if (SV_CURPROC_FLAG(SV_ILP32)) {
3088 				struct timeval32 tv32;
3089 
3090 				error = sooptcopyin(sopt, &tv32, sizeof tv32,
3091 				    sizeof tv32);
3092 				CP(tv32, tv, tv_sec);
3093 				CP(tv32, tv, tv_usec);
3094 			} else
3095 #endif
3096 				error = sooptcopyin(sopt, &tv, sizeof tv,
3097 				    sizeof tv);
3098 			if (error)
3099 				goto bad;
3100 			if (tv.tv_sec < 0 || tv.tv_usec < 0 ||
3101 			    tv.tv_usec >= 1000000) {
3102 				error = EDOM;
3103 				goto bad;
3104 			}
3105 			if (tv.tv_sec > INT32_MAX)
3106 				val = SBT_MAX;
3107 			else
3108 				val = tvtosbt(tv);
3109 			SOCK_LOCK(so);
3110 			valp = sopt->sopt_name == SO_SNDTIMEO ?
3111 			    (SOLISTENING(so) ? &so->sol_sbsnd_timeo :
3112 			    &so->so_snd.sb_timeo) :
3113 			    (SOLISTENING(so) ? &so->sol_sbrcv_timeo :
3114 			    &so->so_rcv.sb_timeo);
3115 			*valp = val;
3116 			SOCK_UNLOCK(so);
3117 			break;
3118 
3119 		case SO_LABEL:
3120 #ifdef MAC
3121 			error = sooptcopyin(sopt, &extmac, sizeof extmac,
3122 			    sizeof extmac);
3123 			if (error)
3124 				goto bad;
3125 			error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
3126 			    so, &extmac);
3127 #else
3128 			error = EOPNOTSUPP;
3129 #endif
3130 			break;
3131 
3132 		case SO_TS_CLOCK:
3133 			error = sooptcopyin(sopt, &optval, sizeof optval,
3134 			    sizeof optval);
3135 			if (error)
3136 				goto bad;
3137 			if (optval < 0 || optval > SO_TS_CLOCK_MAX) {
3138 				error = EINVAL;
3139 				goto bad;
3140 			}
3141 			so->so_ts_clock = optval;
3142 			break;
3143 
3144 		case SO_MAX_PACING_RATE:
3145 			error = sooptcopyin(sopt, &val32, sizeof(val32),
3146 			    sizeof(val32));
3147 			if (error)
3148 				goto bad;
3149 			so->so_max_pacing_rate = val32;
3150 			break;
3151 
3152 		default:
3153 			if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
3154 				error = hhook_run_socket(so, sopt,
3155 				    HHOOK_SOCKET_OPT);
3156 			else
3157 				error = ENOPROTOOPT;
3158 			break;
3159 		}
3160 		if (error == 0 && so->so_proto->pr_ctloutput != NULL)
3161 			(void)(*so->so_proto->pr_ctloutput)(so, sopt);
3162 	}
3163 bad:
3164 	CURVNET_RESTORE();
3165 	return (error);
3166 }
3167 
3168 /*
3169  * Helper routine for getsockopt.
3170  */
3171 int
3172 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
3173 {
3174 	int	error;
3175 	size_t	valsize;
3176 
3177 	error = 0;
3178 
3179 	/*
3180 	 * Documented get behavior is that we always return a value, possibly
3181 	 * truncated to fit in the user's buffer.  Traditional behavior is
3182 	 * that we always tell the user precisely how much we copied, rather
3183 	 * than something useful like the total amount we had available for
3184 	 * her.  Note that this interface is not idempotent; the entire
3185 	 * answer must be generated ahead of time.
3186 	 */
3187 	valsize = min(len, sopt->sopt_valsize);
3188 	sopt->sopt_valsize = valsize;
3189 	if (sopt->sopt_val != NULL) {
3190 		if (sopt->sopt_td != NULL)
3191 			error = copyout(buf, sopt->sopt_val, valsize);
3192 		else
3193 			bcopy(buf, sopt->sopt_val, valsize);
3194 	}
3195 	return (error);
3196 }
3197 
3198 int
3199 sogetopt(struct socket *so, struct sockopt *sopt)
3200 {
3201 	int	error, optval;
3202 	struct	linger l;
3203 	struct	timeval tv;
3204 #ifdef MAC
3205 	struct mac extmac;
3206 #endif
3207 
3208 	CURVNET_SET(so->so_vnet);
3209 	error = 0;
3210 	if (sopt->sopt_level != SOL_SOCKET) {
3211 		if (so->so_proto->pr_ctloutput != NULL)
3212 			error = (*so->so_proto->pr_ctloutput)(so, sopt);
3213 		else
3214 			error = ENOPROTOOPT;
3215 		CURVNET_RESTORE();
3216 		return (error);
3217 	} else {
3218 		switch (sopt->sopt_name) {
3219 		case SO_ACCEPTFILTER:
3220 			error = accept_filt_getopt(so, sopt);
3221 			break;
3222 
3223 		case SO_LINGER:
3224 			SOCK_LOCK(so);
3225 			l.l_onoff = so->so_options & SO_LINGER;
3226 			l.l_linger = so->so_linger;
3227 			SOCK_UNLOCK(so);
3228 			error = sooptcopyout(sopt, &l, sizeof l);
3229 			break;
3230 
3231 		case SO_USELOOPBACK:
3232 		case SO_DONTROUTE:
3233 		case SO_DEBUG:
3234 		case SO_KEEPALIVE:
3235 		case SO_REUSEADDR:
3236 		case SO_REUSEPORT:
3237 		case SO_REUSEPORT_LB:
3238 		case SO_BROADCAST:
3239 		case SO_OOBINLINE:
3240 		case SO_ACCEPTCONN:
3241 		case SO_TIMESTAMP:
3242 		case SO_BINTIME:
3243 		case SO_NOSIGPIPE:
3244 		case SO_NO_DDP:
3245 		case SO_NO_OFFLOAD:
3246 		case SO_RERROR:
3247 			optval = so->so_options & sopt->sopt_name;
3248 integer:
3249 			error = sooptcopyout(sopt, &optval, sizeof optval);
3250 			break;
3251 
3252 		case SO_DOMAIN:
3253 			optval = so->so_proto->pr_domain->dom_family;
3254 			goto integer;
3255 
3256 		case SO_TYPE:
3257 			optval = so->so_type;
3258 			goto integer;
3259 
3260 		case SO_PROTOCOL:
3261 			optval = so->so_proto->pr_protocol;
3262 			goto integer;
3263 
3264 		case SO_ERROR:
3265 			SOCK_LOCK(so);
3266 			if (so->so_error) {
3267 				optval = so->so_error;
3268 				so->so_error = 0;
3269 			} else {
3270 				optval = so->so_rerror;
3271 				so->so_rerror = 0;
3272 			}
3273 			SOCK_UNLOCK(so);
3274 			goto integer;
3275 
3276 		case SO_SNDBUF:
3277 			optval = SOLISTENING(so) ? so->sol_sbsnd_hiwat :
3278 			    so->so_snd.sb_hiwat;
3279 			goto integer;
3280 
3281 		case SO_RCVBUF:
3282 			optval = SOLISTENING(so) ? so->sol_sbrcv_hiwat :
3283 			    so->so_rcv.sb_hiwat;
3284 			goto integer;
3285 
3286 		case SO_SNDLOWAT:
3287 			optval = SOLISTENING(so) ? so->sol_sbsnd_lowat :
3288 			    so->so_snd.sb_lowat;
3289 			goto integer;
3290 
3291 		case SO_RCVLOWAT:
3292 			optval = SOLISTENING(so) ? so->sol_sbrcv_lowat :
3293 			    so->so_rcv.sb_lowat;
3294 			goto integer;
3295 
3296 		case SO_SNDTIMEO:
3297 		case SO_RCVTIMEO:
3298 			SOCK_LOCK(so);
3299 			tv = sbttotv(sopt->sopt_name == SO_SNDTIMEO ?
3300 			    (SOLISTENING(so) ? so->sol_sbsnd_timeo :
3301 			    so->so_snd.sb_timeo) :
3302 			    (SOLISTENING(so) ? so->sol_sbrcv_timeo :
3303 			    so->so_rcv.sb_timeo));
3304 			SOCK_UNLOCK(so);
3305 #ifdef COMPAT_FREEBSD32
3306 			if (SV_CURPROC_FLAG(SV_ILP32)) {
3307 				struct timeval32 tv32;
3308 
3309 				CP(tv, tv32, tv_sec);
3310 				CP(tv, tv32, tv_usec);
3311 				error = sooptcopyout(sopt, &tv32, sizeof tv32);
3312 			} else
3313 #endif
3314 				error = sooptcopyout(sopt, &tv, sizeof tv);
3315 			break;
3316 
3317 		case SO_LABEL:
3318 #ifdef MAC
3319 			error = sooptcopyin(sopt, &extmac, sizeof(extmac),
3320 			    sizeof(extmac));
3321 			if (error)
3322 				goto bad;
3323 			error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
3324 			    so, &extmac);
3325 			if (error)
3326 				goto bad;
3327 			error = sooptcopyout(sopt, &extmac, sizeof extmac);
3328 #else
3329 			error = EOPNOTSUPP;
3330 #endif
3331 			break;
3332 
3333 		case SO_PEERLABEL:
3334 #ifdef MAC
3335 			error = sooptcopyin(sopt, &extmac, sizeof(extmac),
3336 			    sizeof(extmac));
3337 			if (error)
3338 				goto bad;
3339 			error = mac_getsockopt_peerlabel(
3340 			    sopt->sopt_td->td_ucred, so, &extmac);
3341 			if (error)
3342 				goto bad;
3343 			error = sooptcopyout(sopt, &extmac, sizeof extmac);
3344 #else
3345 			error = EOPNOTSUPP;
3346 #endif
3347 			break;
3348 
3349 		case SO_LISTENQLIMIT:
3350 			optval = SOLISTENING(so) ? so->sol_qlimit : 0;
3351 			goto integer;
3352 
3353 		case SO_LISTENQLEN:
3354 			optval = SOLISTENING(so) ? so->sol_qlen : 0;
3355 			goto integer;
3356 
3357 		case SO_LISTENINCQLEN:
3358 			optval = SOLISTENING(so) ? so->sol_incqlen : 0;
3359 			goto integer;
3360 
3361 		case SO_TS_CLOCK:
3362 			optval = so->so_ts_clock;
3363 			goto integer;
3364 
3365 		case SO_MAX_PACING_RATE:
3366 			optval = so->so_max_pacing_rate;
3367 			goto integer;
3368 
3369 		default:
3370 			if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
3371 				error = hhook_run_socket(so, sopt,
3372 				    HHOOK_SOCKET_OPT);
3373 			else
3374 				error = ENOPROTOOPT;
3375 			break;
3376 		}
3377 	}
3378 #ifdef MAC
3379 bad:
3380 #endif
3381 	CURVNET_RESTORE();
3382 	return (error);
3383 }
3384 
3385 int
3386 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
3387 {
3388 	struct mbuf *m, *m_prev;
3389 	int sopt_size = sopt->sopt_valsize;
3390 
3391 	MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
3392 	if (m == NULL)
3393 		return ENOBUFS;
3394 	if (sopt_size > MLEN) {
3395 		MCLGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT);
3396 		if ((m->m_flags & M_EXT) == 0) {
3397 			m_free(m);
3398 			return ENOBUFS;
3399 		}
3400 		m->m_len = min(MCLBYTES, sopt_size);
3401 	} else {
3402 		m->m_len = min(MLEN, sopt_size);
3403 	}
3404 	sopt_size -= m->m_len;
3405 	*mp = m;
3406 	m_prev = m;
3407 
3408 	while (sopt_size) {
3409 		MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
3410 		if (m == NULL) {
3411 			m_freem(*mp);
3412 			return ENOBUFS;
3413 		}
3414 		if (sopt_size > MLEN) {
3415 			MCLGET(m, sopt->sopt_td != NULL ? M_WAITOK :
3416 			    M_NOWAIT);
3417 			if ((m->m_flags & M_EXT) == 0) {
3418 				m_freem(m);
3419 				m_freem(*mp);
3420 				return ENOBUFS;
3421 			}
3422 			m->m_len = min(MCLBYTES, sopt_size);
3423 		} else {
3424 			m->m_len = min(MLEN, sopt_size);
3425 		}
3426 		sopt_size -= m->m_len;
3427 		m_prev->m_next = m;
3428 		m_prev = m;
3429 	}
3430 	return (0);
3431 }
3432 
3433 int
3434 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
3435 {
3436 	struct mbuf *m0 = m;
3437 
3438 	if (sopt->sopt_val == NULL)
3439 		return (0);
3440 	while (m != NULL && sopt->sopt_valsize >= m->m_len) {
3441 		if (sopt->sopt_td != NULL) {
3442 			int error;
3443 
3444 			error = copyin(sopt->sopt_val, mtod(m, char *),
3445 			    m->m_len);
3446 			if (error != 0) {
3447 				m_freem(m0);
3448 				return(error);
3449 			}
3450 		} else
3451 			bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
3452 		sopt->sopt_valsize -= m->m_len;
3453 		sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
3454 		m = m->m_next;
3455 	}
3456 	if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
3457 		panic("ip6_sooptmcopyin");
3458 	return (0);
3459 }
3460 
3461 int
3462 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
3463 {
3464 	struct mbuf *m0 = m;
3465 	size_t valsize = 0;
3466 
3467 	if (sopt->sopt_val == NULL)
3468 		return (0);
3469 	while (m != NULL && sopt->sopt_valsize >= m->m_len) {
3470 		if (sopt->sopt_td != NULL) {
3471 			int error;
3472 
3473 			error = copyout(mtod(m, char *), sopt->sopt_val,
3474 			    m->m_len);
3475 			if (error != 0) {
3476 				m_freem(m0);
3477 				return(error);
3478 			}
3479 		} else
3480 			bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
3481 		sopt->sopt_valsize -= m->m_len;
3482 		sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
3483 		valsize += m->m_len;
3484 		m = m->m_next;
3485 	}
3486 	if (m != NULL) {
3487 		/* enough soopt buffer should be given from user-land */
3488 		m_freem(m0);
3489 		return(EINVAL);
3490 	}
3491 	sopt->sopt_valsize = valsize;
3492 	return (0);
3493 }
3494 
3495 /*
3496  * sohasoutofband(): protocol notifies socket layer of the arrival of new
3497  * out-of-band data, which will then notify socket consumers.
3498  */
3499 void
3500 sohasoutofband(struct socket *so)
3501 {
3502 
3503 	if (so->so_sigio != NULL)
3504 		pgsigio(&so->so_sigio, SIGURG, 0);
3505 	selwakeuppri(&so->so_rdsel, PSOCK);
3506 }
3507 
3508 int
3509 sopoll(struct socket *so, int events, struct ucred *active_cred,
3510     struct thread *td)
3511 {
3512 
3513 	/*
3514 	 * We do not need to set or assert curvnet as long as everyone uses
3515 	 * sopoll_generic().
3516 	 */
3517 	return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred,
3518 	    td));
3519 }
3520 
3521 int
3522 sopoll_generic(struct socket *so, int events, struct ucred *active_cred,
3523     struct thread *td)
3524 {
3525 	int revents;
3526 
3527 	SOCK_LOCK(so);
3528 	if (SOLISTENING(so)) {
3529 		if (!(events & (POLLIN | POLLRDNORM)))
3530 			revents = 0;
3531 		else if (!TAILQ_EMPTY(&so->sol_comp))
3532 			revents = events & (POLLIN | POLLRDNORM);
3533 		else if ((events & POLLINIGNEOF) == 0 && so->so_error)
3534 			revents = (events & (POLLIN | POLLRDNORM)) | POLLHUP;
3535 		else {
3536 			selrecord(td, &so->so_rdsel);
3537 			revents = 0;
3538 		}
3539 	} else {
3540 		revents = 0;
3541 		SOCK_SENDBUF_LOCK(so);
3542 		SOCK_RECVBUF_LOCK(so);
3543 		if (events & (POLLIN | POLLRDNORM))
3544 			if (soreadabledata(so))
3545 				revents |= events & (POLLIN | POLLRDNORM);
3546 		if (events & (POLLOUT | POLLWRNORM))
3547 			if (sowriteable(so))
3548 				revents |= events & (POLLOUT | POLLWRNORM);
3549 		if (events & (POLLPRI | POLLRDBAND))
3550 			if (so->so_oobmark ||
3551 			    (so->so_rcv.sb_state & SBS_RCVATMARK))
3552 				revents |= events & (POLLPRI | POLLRDBAND);
3553 		if ((events & POLLINIGNEOF) == 0) {
3554 			if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3555 				revents |= events & (POLLIN | POLLRDNORM);
3556 				if (so->so_snd.sb_state & SBS_CANTSENDMORE)
3557 					revents |= POLLHUP;
3558 			}
3559 		}
3560 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
3561 			revents |= events & POLLRDHUP;
3562 		if (revents == 0) {
3563 			if (events &
3564 			    (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND | POLLRDHUP)) {
3565 				selrecord(td, &so->so_rdsel);
3566 				so->so_rcv.sb_flags |= SB_SEL;
3567 			}
3568 			if (events & (POLLOUT | POLLWRNORM)) {
3569 				selrecord(td, &so->so_wrsel);
3570 				so->so_snd.sb_flags |= SB_SEL;
3571 			}
3572 		}
3573 		SOCK_RECVBUF_UNLOCK(so);
3574 		SOCK_SENDBUF_UNLOCK(so);
3575 	}
3576 	SOCK_UNLOCK(so);
3577 	return (revents);
3578 }
3579 
3580 int
3581 soo_kqfilter(struct file *fp, struct knote *kn)
3582 {
3583 	struct socket *so = kn->kn_fp->f_data;
3584 	struct sockbuf *sb;
3585 	sb_which which;
3586 	struct knlist *knl;
3587 
3588 	switch (kn->kn_filter) {
3589 	case EVFILT_READ:
3590 		kn->kn_fop = &soread_filtops;
3591 		knl = &so->so_rdsel.si_note;
3592 		sb = &so->so_rcv;
3593 		which = SO_RCV;
3594 		break;
3595 	case EVFILT_WRITE:
3596 		kn->kn_fop = &sowrite_filtops;
3597 		knl = &so->so_wrsel.si_note;
3598 		sb = &so->so_snd;
3599 		which = SO_SND;
3600 		break;
3601 	case EVFILT_EMPTY:
3602 		kn->kn_fop = &soempty_filtops;
3603 		knl = &so->so_wrsel.si_note;
3604 		sb = &so->so_snd;
3605 		which = SO_SND;
3606 		break;
3607 	default:
3608 		return (EINVAL);
3609 	}
3610 
3611 	SOCK_LOCK(so);
3612 	if (SOLISTENING(so)) {
3613 		knlist_add(knl, kn, 1);
3614 	} else {
3615 		SOCK_BUF_LOCK(so, which);
3616 		knlist_add(knl, kn, 1);
3617 		sb->sb_flags |= SB_KNOTE;
3618 		SOCK_BUF_UNLOCK(so, which);
3619 	}
3620 	SOCK_UNLOCK(so);
3621 	return (0);
3622 }
3623 
3624 /*
3625  * Some routines that return EOPNOTSUPP for entry points that are not
3626  * supported by a protocol.  Fill in as needed.
3627  */
3628 int
3629 pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
3630 {
3631 
3632 	return EOPNOTSUPP;
3633 }
3634 
3635 int
3636 pru_aio_queue_notsupp(struct socket *so, struct kaiocb *job)
3637 {
3638 
3639 	return EOPNOTSUPP;
3640 }
3641 
3642 int
3643 pru_attach_notsupp(struct socket *so, int proto, struct thread *td)
3644 {
3645 
3646 	return EOPNOTSUPP;
3647 }
3648 
3649 int
3650 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3651 {
3652 
3653 	return EOPNOTSUPP;
3654 }
3655 
3656 int
3657 pru_bindat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3658     struct thread *td)
3659 {
3660 
3661 	return EOPNOTSUPP;
3662 }
3663 
3664 int
3665 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3666 {
3667 
3668 	return EOPNOTSUPP;
3669 }
3670 
3671 int
3672 pru_connectat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3673     struct thread *td)
3674 {
3675 
3676 	return EOPNOTSUPP;
3677 }
3678 
3679 int
3680 pru_connect2_notsupp(struct socket *so1, struct socket *so2)
3681 {
3682 
3683 	return EOPNOTSUPP;
3684 }
3685 
3686 int
3687 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
3688     struct ifnet *ifp, struct thread *td)
3689 {
3690 
3691 	return EOPNOTSUPP;
3692 }
3693 
3694 int
3695 pru_disconnect_notsupp(struct socket *so)
3696 {
3697 
3698 	return EOPNOTSUPP;
3699 }
3700 
3701 int
3702 pru_listen_notsupp(struct socket *so, int backlog, struct thread *td)
3703 {
3704 
3705 	return EOPNOTSUPP;
3706 }
3707 
3708 int
3709 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam)
3710 {
3711 
3712 	return EOPNOTSUPP;
3713 }
3714 
3715 int
3716 pru_rcvd_notsupp(struct socket *so, int flags)
3717 {
3718 
3719 	return EOPNOTSUPP;
3720 }
3721 
3722 int
3723 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
3724 {
3725 
3726 	return EOPNOTSUPP;
3727 }
3728 
3729 int
3730 pru_send_notsupp(struct socket *so, int flags, struct mbuf *m,
3731     struct sockaddr *addr, struct mbuf *control, struct thread *td)
3732 {
3733 
3734 	if (control != NULL)
3735 		m_freem(control);
3736 	if ((flags & PRUS_NOTREADY) == 0)
3737 		m_freem(m);
3738 	return (EOPNOTSUPP);
3739 }
3740 
3741 int
3742 pru_ready_notsupp(struct socket *so, struct mbuf *m, int count)
3743 {
3744 
3745 	return (EOPNOTSUPP);
3746 }
3747 
3748 /*
3749  * This isn't really a ``null'' operation, but it's the default one and
3750  * doesn't do anything destructive.
3751  */
3752 int
3753 pru_sense_null(struct socket *so, struct stat *sb)
3754 {
3755 
3756 	sb->st_blksize = so->so_snd.sb_hiwat;
3757 	return 0;
3758 }
3759 
3760 int
3761 pru_shutdown_notsupp(struct socket *so)
3762 {
3763 
3764 	return EOPNOTSUPP;
3765 }
3766 
3767 int
3768 pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam)
3769 {
3770 
3771 	return EOPNOTSUPP;
3772 }
3773 
3774 int
3775 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio,
3776     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
3777 {
3778 
3779 	return EOPNOTSUPP;
3780 }
3781 
3782 int
3783 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr,
3784     struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
3785 {
3786 
3787 	return EOPNOTSUPP;
3788 }
3789 
3790 int
3791 pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred,
3792     struct thread *td)
3793 {
3794 
3795 	return EOPNOTSUPP;
3796 }
3797 
3798 static void
3799 filt_sordetach(struct knote *kn)
3800 {
3801 	struct socket *so = kn->kn_fp->f_data;
3802 
3803 	so_rdknl_lock(so);
3804 	knlist_remove(&so->so_rdsel.si_note, kn, 1);
3805 	if (!SOLISTENING(so) && knlist_empty(&so->so_rdsel.si_note))
3806 		so->so_rcv.sb_flags &= ~SB_KNOTE;
3807 	so_rdknl_unlock(so);
3808 }
3809 
3810 /*ARGSUSED*/
3811 static int
3812 filt_soread(struct knote *kn, long hint)
3813 {
3814 	struct socket *so;
3815 
3816 	so = kn->kn_fp->f_data;
3817 
3818 	if (SOLISTENING(so)) {
3819 		SOCK_LOCK_ASSERT(so);
3820 		kn->kn_data = so->sol_qlen;
3821 		if (so->so_error) {
3822 			kn->kn_flags |= EV_EOF;
3823 			kn->kn_fflags = so->so_error;
3824 			return (1);
3825 		}
3826 		return (!TAILQ_EMPTY(&so->sol_comp));
3827 	}
3828 
3829 	SOCK_RECVBUF_LOCK_ASSERT(so);
3830 
3831 	kn->kn_data = sbavail(&so->so_rcv) - so->so_rcv.sb_ctl;
3832 	if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3833 		kn->kn_flags |= EV_EOF;
3834 		kn->kn_fflags = so->so_error;
3835 		return (1);
3836 	} else if (so->so_error || so->so_rerror)
3837 		return (1);
3838 
3839 	if (kn->kn_sfflags & NOTE_LOWAT) {
3840 		if (kn->kn_data >= kn->kn_sdata)
3841 			return (1);
3842 	} else if (sbavail(&so->so_rcv) >= so->so_rcv.sb_lowat)
3843 		return (1);
3844 
3845 	/* This hook returning non-zero indicates an event, not error */
3846 	return (hhook_run_socket(so, NULL, HHOOK_FILT_SOREAD));
3847 }
3848 
3849 static void
3850 filt_sowdetach(struct knote *kn)
3851 {
3852 	struct socket *so = kn->kn_fp->f_data;
3853 
3854 	so_wrknl_lock(so);
3855 	knlist_remove(&so->so_wrsel.si_note, kn, 1);
3856 	if (!SOLISTENING(so) && knlist_empty(&so->so_wrsel.si_note))
3857 		so->so_snd.sb_flags &= ~SB_KNOTE;
3858 	so_wrknl_unlock(so);
3859 }
3860 
3861 /*ARGSUSED*/
3862 static int
3863 filt_sowrite(struct knote *kn, long hint)
3864 {
3865 	struct socket *so;
3866 
3867 	so = kn->kn_fp->f_data;
3868 
3869 	if (SOLISTENING(so))
3870 		return (0);
3871 
3872 	SOCK_SENDBUF_LOCK_ASSERT(so);
3873 	kn->kn_data = sbspace(&so->so_snd);
3874 
3875 	hhook_run_socket(so, kn, HHOOK_FILT_SOWRITE);
3876 
3877 	if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
3878 		kn->kn_flags |= EV_EOF;
3879 		kn->kn_fflags = so->so_error;
3880 		return (1);
3881 	} else if (so->so_error)	/* temporary udp error */
3882 		return (1);
3883 	else if (((so->so_state & SS_ISCONNECTED) == 0) &&
3884 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
3885 		return (0);
3886 	else if (kn->kn_sfflags & NOTE_LOWAT)
3887 		return (kn->kn_data >= kn->kn_sdata);
3888 	else
3889 		return (kn->kn_data >= so->so_snd.sb_lowat);
3890 }
3891 
3892 static int
3893 filt_soempty(struct knote *kn, long hint)
3894 {
3895 	struct socket *so;
3896 
3897 	so = kn->kn_fp->f_data;
3898 
3899 	if (SOLISTENING(so))
3900 		return (1);
3901 
3902 	SOCK_SENDBUF_LOCK_ASSERT(so);
3903 	kn->kn_data = sbused(&so->so_snd);
3904 
3905 	if (kn->kn_data == 0)
3906 		return (1);
3907 	else
3908 		return (0);
3909 }
3910 
3911 int
3912 socheckuid(struct socket *so, uid_t uid)
3913 {
3914 
3915 	if (so == NULL)
3916 		return (EPERM);
3917 	if (so->so_cred->cr_uid != uid)
3918 		return (EPERM);
3919 	return (0);
3920 }
3921 
3922 /*
3923  * These functions are used by protocols to notify the socket layer (and its
3924  * consumers) of state changes in the sockets driven by protocol-side events.
3925  */
3926 
3927 /*
3928  * Procedures to manipulate state flags of socket and do appropriate wakeups.
3929  *
3930  * Normal sequence from the active (originating) side is that
3931  * soisconnecting() is called during processing of connect() call, resulting
3932  * in an eventual call to soisconnected() if/when the connection is
3933  * established.  When the connection is torn down soisdisconnecting() is
3934  * called during processing of disconnect() call, and soisdisconnected() is
3935  * called when the connection to the peer is totally severed.  The semantics
3936  * of these routines are such that connectionless protocols can call
3937  * soisconnected() and soisdisconnected() only, bypassing the in-progress
3938  * calls when setting up a ``connection'' takes no time.
3939  *
3940  * From the passive side, a socket is created with two queues of sockets:
3941  * so_incomp for connections in progress and so_comp for connections already
3942  * made and awaiting user acceptance.  As a protocol is preparing incoming
3943  * connections, it creates a socket structure queued on so_incomp by calling
3944  * sonewconn().  When the connection is established, soisconnected() is
3945  * called, and transfers the socket structure to so_comp, making it available
3946  * to accept().
3947  *
3948  * If a socket is closed with sockets on either so_incomp or so_comp, these
3949  * sockets are dropped.
3950  *
3951  * If higher-level protocols are implemented in the kernel, the wakeups done
3952  * here will sometimes cause software-interrupt process scheduling.
3953  */
3954 void
3955 soisconnecting(struct socket *so)
3956 {
3957 
3958 	SOCK_LOCK(so);
3959 	so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
3960 	so->so_state |= SS_ISCONNECTING;
3961 	SOCK_UNLOCK(so);
3962 }
3963 
3964 void
3965 soisconnected(struct socket *so)
3966 {
3967 	bool last __diagused;
3968 
3969 	SOCK_LOCK(so);
3970 	so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
3971 	so->so_state |= SS_ISCONNECTED;
3972 
3973 	if (so->so_qstate == SQ_INCOMP) {
3974 		struct socket *head = so->so_listen;
3975 		int ret;
3976 
3977 		KASSERT(head, ("%s: so %p on incomp of NULL", __func__, so));
3978 		/*
3979 		 * Promoting a socket from incomplete queue to complete, we
3980 		 * need to go through reverse order of locking.  We first do
3981 		 * trylock, and if that doesn't succeed, we go the hard way
3982 		 * leaving a reference and rechecking consistency after proper
3983 		 * locking.
3984 		 */
3985 		if (__predict_false(SOLISTEN_TRYLOCK(head) == 0)) {
3986 			soref(head);
3987 			SOCK_UNLOCK(so);
3988 			SOLISTEN_LOCK(head);
3989 			SOCK_LOCK(so);
3990 			if (__predict_false(head != so->so_listen)) {
3991 				/*
3992 				 * The socket went off the listen queue,
3993 				 * should be lost race to close(2) of sol.
3994 				 * The socket is about to soabort().
3995 				 */
3996 				SOCK_UNLOCK(so);
3997 				sorele_locked(head);
3998 				return;
3999 			}
4000 			last = refcount_release(&head->so_count);
4001 			KASSERT(!last, ("%s: released last reference for %p",
4002 			    __func__, head));
4003 		}
4004 again:
4005 		if ((so->so_options & SO_ACCEPTFILTER) == 0) {
4006 			TAILQ_REMOVE(&head->sol_incomp, so, so_list);
4007 			head->sol_incqlen--;
4008 			TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list);
4009 			head->sol_qlen++;
4010 			so->so_qstate = SQ_COMP;
4011 			SOCK_UNLOCK(so);
4012 			solisten_wakeup(head);	/* unlocks */
4013 		} else {
4014 			SOCK_RECVBUF_LOCK(so);
4015 			soupcall_set(so, SO_RCV,
4016 			    head->sol_accept_filter->accf_callback,
4017 			    head->sol_accept_filter_arg);
4018 			so->so_options &= ~SO_ACCEPTFILTER;
4019 			ret = head->sol_accept_filter->accf_callback(so,
4020 			    head->sol_accept_filter_arg, M_NOWAIT);
4021 			if (ret == SU_ISCONNECTED) {
4022 				soupcall_clear(so, SO_RCV);
4023 				SOCK_RECVBUF_UNLOCK(so);
4024 				goto again;
4025 			}
4026 			SOCK_RECVBUF_UNLOCK(so);
4027 			SOCK_UNLOCK(so);
4028 			SOLISTEN_UNLOCK(head);
4029 		}
4030 		return;
4031 	}
4032 	SOCK_UNLOCK(so);
4033 	wakeup(&so->so_timeo);
4034 	sorwakeup(so);
4035 	sowwakeup(so);
4036 }
4037 
4038 void
4039 soisdisconnecting(struct socket *so)
4040 {
4041 
4042 	SOCK_LOCK(so);
4043 	so->so_state &= ~SS_ISCONNECTING;
4044 	so->so_state |= SS_ISDISCONNECTING;
4045 
4046 	if (!SOLISTENING(so)) {
4047 		SOCK_RECVBUF_LOCK(so);
4048 		socantrcvmore_locked(so);
4049 		SOCK_SENDBUF_LOCK(so);
4050 		socantsendmore_locked(so);
4051 	}
4052 	SOCK_UNLOCK(so);
4053 	wakeup(&so->so_timeo);
4054 }
4055 
4056 void
4057 soisdisconnected(struct socket *so)
4058 {
4059 
4060 	SOCK_LOCK(so);
4061 
4062 	/*
4063 	 * There is at least one reader of so_state that does not
4064 	 * acquire socket lock, namely soreceive_generic().  Ensure
4065 	 * that it never sees all flags that track connection status
4066 	 * cleared, by ordering the update with a barrier semantic of
4067 	 * our release thread fence.
4068 	 */
4069 	so->so_state |= SS_ISDISCONNECTED;
4070 	atomic_thread_fence_rel();
4071 	so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
4072 
4073 	if (!SOLISTENING(so)) {
4074 		SOCK_UNLOCK(so);
4075 		SOCK_RECVBUF_LOCK(so);
4076 		socantrcvmore_locked(so);
4077 		SOCK_SENDBUF_LOCK(so);
4078 		sbdrop_locked(&so->so_snd, sbused(&so->so_snd));
4079 		socantsendmore_locked(so);
4080 	} else
4081 		SOCK_UNLOCK(so);
4082 	wakeup(&so->so_timeo);
4083 }
4084 
4085 int
4086 soiolock(struct socket *so, struct sx *sx, int flags)
4087 {
4088 	int error;
4089 
4090 	KASSERT((flags & SBL_VALID) == flags,
4091 	    ("soiolock: invalid flags %#x", flags));
4092 
4093 	if ((flags & SBL_WAIT) != 0) {
4094 		if ((flags & SBL_NOINTR) != 0) {
4095 			sx_xlock(sx);
4096 		} else {
4097 			error = sx_xlock_sig(sx);
4098 			if (error != 0)
4099 				return (error);
4100 		}
4101 	} else if (!sx_try_xlock(sx)) {
4102 		return (EWOULDBLOCK);
4103 	}
4104 
4105 	if (__predict_false(SOLISTENING(so))) {
4106 		sx_xunlock(sx);
4107 		return (ENOTCONN);
4108 	}
4109 	return (0);
4110 }
4111 
4112 void
4113 soiounlock(struct sx *sx)
4114 {
4115 	sx_xunlock(sx);
4116 }
4117 
4118 /*
4119  * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
4120  */
4121 struct sockaddr *
4122 sodupsockaddr(const struct sockaddr *sa, int mflags)
4123 {
4124 	struct sockaddr *sa2;
4125 
4126 	sa2 = malloc(sa->sa_len, M_SONAME, mflags);
4127 	if (sa2)
4128 		bcopy(sa, sa2, sa->sa_len);
4129 	return sa2;
4130 }
4131 
4132 /*
4133  * Register per-socket destructor.
4134  */
4135 void
4136 sodtor_set(struct socket *so, so_dtor_t *func)
4137 {
4138 
4139 	SOCK_LOCK_ASSERT(so);
4140 	so->so_dtor = func;
4141 }
4142 
4143 /*
4144  * Register per-socket buffer upcalls.
4145  */
4146 void
4147 soupcall_set(struct socket *so, sb_which which, so_upcall_t func, void *arg)
4148 {
4149 	struct sockbuf *sb;
4150 
4151 	KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so));
4152 
4153 	switch (which) {
4154 	case SO_RCV:
4155 		sb = &so->so_rcv;
4156 		break;
4157 	case SO_SND:
4158 		sb = &so->so_snd;
4159 		break;
4160 	}
4161 	SOCK_BUF_LOCK_ASSERT(so, which);
4162 	sb->sb_upcall = func;
4163 	sb->sb_upcallarg = arg;
4164 	sb->sb_flags |= SB_UPCALL;
4165 }
4166 
4167 void
4168 soupcall_clear(struct socket *so, sb_which which)
4169 {
4170 	struct sockbuf *sb;
4171 
4172 	KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so));
4173 
4174 	switch (which) {
4175 	case SO_RCV:
4176 		sb = &so->so_rcv;
4177 		break;
4178 	case SO_SND:
4179 		sb = &so->so_snd;
4180 		break;
4181 	}
4182 	SOCK_BUF_LOCK_ASSERT(so, which);
4183 	KASSERT(sb->sb_upcall != NULL,
4184 	    ("%s: so %p no upcall to clear", __func__, so));
4185 	sb->sb_upcall = NULL;
4186 	sb->sb_upcallarg = NULL;
4187 	sb->sb_flags &= ~SB_UPCALL;
4188 }
4189 
4190 void
4191 solisten_upcall_set(struct socket *so, so_upcall_t func, void *arg)
4192 {
4193 
4194 	SOLISTEN_LOCK_ASSERT(so);
4195 	so->sol_upcall = func;
4196 	so->sol_upcallarg = arg;
4197 }
4198 
4199 static void
4200 so_rdknl_lock(void *arg)
4201 {
4202 	struct socket *so = arg;
4203 
4204 retry:
4205 	if (SOLISTENING(so)) {
4206 		SOLISTEN_LOCK(so);
4207 	} else {
4208 		SOCK_RECVBUF_LOCK(so);
4209 		if (__predict_false(SOLISTENING(so))) {
4210 			SOCK_RECVBUF_UNLOCK(so);
4211 			goto retry;
4212 		}
4213 	}
4214 }
4215 
4216 static void
4217 so_rdknl_unlock(void *arg)
4218 {
4219 	struct socket *so = arg;
4220 
4221 	if (SOLISTENING(so))
4222 		SOLISTEN_UNLOCK(so);
4223 	else
4224 		SOCK_RECVBUF_UNLOCK(so);
4225 }
4226 
4227 static void
4228 so_rdknl_assert_lock(void *arg, int what)
4229 {
4230 	struct socket *so = arg;
4231 
4232 	if (what == LA_LOCKED) {
4233 		if (SOLISTENING(so))
4234 			SOLISTEN_LOCK_ASSERT(so);
4235 		else
4236 			SOCK_RECVBUF_LOCK_ASSERT(so);
4237 	} else {
4238 		if (SOLISTENING(so))
4239 			SOLISTEN_UNLOCK_ASSERT(so);
4240 		else
4241 			SOCK_RECVBUF_UNLOCK_ASSERT(so);
4242 	}
4243 }
4244 
4245 static void
4246 so_wrknl_lock(void *arg)
4247 {
4248 	struct socket *so = arg;
4249 
4250 retry:
4251 	if (SOLISTENING(so)) {
4252 		SOLISTEN_LOCK(so);
4253 	} else {
4254 		SOCK_SENDBUF_LOCK(so);
4255 		if (__predict_false(SOLISTENING(so))) {
4256 			SOCK_SENDBUF_UNLOCK(so);
4257 			goto retry;
4258 		}
4259 	}
4260 }
4261 
4262 static void
4263 so_wrknl_unlock(void *arg)
4264 {
4265 	struct socket *so = arg;
4266 
4267 	if (SOLISTENING(so))
4268 		SOLISTEN_UNLOCK(so);
4269 	else
4270 		SOCK_SENDBUF_UNLOCK(so);
4271 }
4272 
4273 static void
4274 so_wrknl_assert_lock(void *arg, int what)
4275 {
4276 	struct socket *so = arg;
4277 
4278 	if (what == LA_LOCKED) {
4279 		if (SOLISTENING(so))
4280 			SOLISTEN_LOCK_ASSERT(so);
4281 		else
4282 			SOCK_SENDBUF_LOCK_ASSERT(so);
4283 	} else {
4284 		if (SOLISTENING(so))
4285 			SOLISTEN_UNLOCK_ASSERT(so);
4286 		else
4287 			SOCK_SENDBUF_UNLOCK_ASSERT(so);
4288 	}
4289 }
4290 
4291 /*
4292  * Create an external-format (``xsocket'') structure using the information in
4293  * the kernel-format socket structure pointed to by so.  This is done to
4294  * reduce the spew of irrelevant information over this interface, to isolate
4295  * user code from changes in the kernel structure, and potentially to provide
4296  * information-hiding if we decide that some of this information should be
4297  * hidden from users.
4298  */
4299 void
4300 sotoxsocket(struct socket *so, struct xsocket *xso)
4301 {
4302 
4303 	bzero(xso, sizeof(*xso));
4304 	xso->xso_len = sizeof *xso;
4305 	xso->xso_so = (uintptr_t)so;
4306 	xso->so_type = so->so_type;
4307 	xso->so_options = so->so_options;
4308 	xso->so_linger = so->so_linger;
4309 	xso->so_state = so->so_state;
4310 	xso->so_pcb = (uintptr_t)so->so_pcb;
4311 	xso->xso_protocol = so->so_proto->pr_protocol;
4312 	xso->xso_family = so->so_proto->pr_domain->dom_family;
4313 	xso->so_timeo = so->so_timeo;
4314 	xso->so_error = so->so_error;
4315 	xso->so_uid = so->so_cred->cr_uid;
4316 	xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
4317 	if (SOLISTENING(so)) {
4318 		xso->so_qlen = so->sol_qlen;
4319 		xso->so_incqlen = so->sol_incqlen;
4320 		xso->so_qlimit = so->sol_qlimit;
4321 		xso->so_oobmark = 0;
4322 	} else {
4323 		xso->so_state |= so->so_qstate;
4324 		xso->so_qlen = xso->so_incqlen = xso->so_qlimit = 0;
4325 		xso->so_oobmark = so->so_oobmark;
4326 		sbtoxsockbuf(&so->so_snd, &xso->so_snd);
4327 		sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
4328 	}
4329 }
4330 
4331 struct sockbuf *
4332 so_sockbuf_rcv(struct socket *so)
4333 {
4334 
4335 	return (&so->so_rcv);
4336 }
4337 
4338 struct sockbuf *
4339 so_sockbuf_snd(struct socket *so)
4340 {
4341 
4342 	return (&so->so_snd);
4343 }
4344 
4345 int
4346 so_state_get(const struct socket *so)
4347 {
4348 
4349 	return (so->so_state);
4350 }
4351 
4352 void
4353 so_state_set(struct socket *so, int val)
4354 {
4355 
4356 	so->so_state = val;
4357 }
4358 
4359 int
4360 so_options_get(const struct socket *so)
4361 {
4362 
4363 	return (so->so_options);
4364 }
4365 
4366 void
4367 so_options_set(struct socket *so, int val)
4368 {
4369 
4370 	so->so_options = val;
4371 }
4372 
4373 int
4374 so_error_get(const struct socket *so)
4375 {
4376 
4377 	return (so->so_error);
4378 }
4379 
4380 void
4381 so_error_set(struct socket *so, int val)
4382 {
4383 
4384 	so->so_error = val;
4385 }
4386 
4387 int
4388 so_linger_get(const struct socket *so)
4389 {
4390 
4391 	return (so->so_linger);
4392 }
4393 
4394 void
4395 so_linger_set(struct socket *so, int val)
4396 {
4397 
4398 	KASSERT(val >= 0 && val <= USHRT_MAX && val <= (INT_MAX / hz),
4399 	    ("%s: val %d out of range", __func__, val));
4400 
4401 	so->so_linger = val;
4402 }
4403 
4404 struct protosw *
4405 so_protosw_get(const struct socket *so)
4406 {
4407 
4408 	return (so->so_proto);
4409 }
4410 
4411 void
4412 so_protosw_set(struct socket *so, struct protosw *val)
4413 {
4414 
4415 	so->so_proto = val;
4416 }
4417 
4418 void
4419 so_sorwakeup(struct socket *so)
4420 {
4421 
4422 	sorwakeup(so);
4423 }
4424 
4425 void
4426 so_sowwakeup(struct socket *so)
4427 {
4428 
4429 	sowwakeup(so);
4430 }
4431 
4432 void
4433 so_sorwakeup_locked(struct socket *so)
4434 {
4435 
4436 	sorwakeup_locked(so);
4437 }
4438 
4439 void
4440 so_sowwakeup_locked(struct socket *so)
4441 {
4442 
4443 	sowwakeup_locked(so);
4444 }
4445 
4446 void
4447 so_lock(struct socket *so)
4448 {
4449 
4450 	SOCK_LOCK(so);
4451 }
4452 
4453 void
4454 so_unlock(struct socket *so)
4455 {
4456 
4457 	SOCK_UNLOCK(so);
4458 }
4459