xref: /freebsd/sys/kern/uipc_socket.c (revision 2668e76d6e764c5c361156ffa3d39eb02ce8e5d9)
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 
35 /*
36  * Comments on the socket life cycle:
37  *
38  * soalloc() sets of socket layer state for a socket, called only by
39  * socreate() and sonewconn().  Socket layer private.
40  *
41  * sodealloc() tears down socket layer state for a socket, called only by
42  * sofree() and sonewconn().  Socket layer private.
43  *
44  * pru_attach() associates protocol layer state with an allocated socket;
45  * called only once, may fail, aborting socket allocation.  This is called
46  * from socreate() and sonewconn().  Socket layer private.
47  *
48  * pru_detach() disassociates protocol layer state from an attached socket,
49  * and will be called exactly once for sockets in which pru_attach() has
50  * been successfully called.  If pru_attach() returned an error,
51  * pru_detach() will not be called.  Socket layer private.
52  *
53  * pru_abort() and pru_close() notify the protocol layer that the last
54  * consumer of a socket is starting to tear down the socket, and that the
55  * protocol should terminate the connection.  Historically, pru_abort() also
56  * detached protocol state from the socket state, but this is no longer the
57  * case.
58  *
59  * socreate() creates a socket and attaches protocol state.  This is a public
60  * interface that may be used by socket layer consumers to create new
61  * sockets.
62  *
63  * sonewconn() creates a socket and attaches protocol state.  This is a
64  * public interface  that may be used by protocols to create new sockets when
65  * a new connection is received and will be available for accept() on a
66  * listen socket.
67  *
68  * soclose() destroys a socket after possibly waiting for it to disconnect.
69  * This is a public interface that socket consumers should use to close and
70  * release a socket when done with it.
71  *
72  * soabort() destroys a socket without waiting for it to disconnect (used
73  * only for incoming connections that are already partially or fully
74  * connected).  This is used internally by the socket layer when clearing
75  * listen socket queues (due to overflow or close on the listen socket), but
76  * is also a public interface protocols may use to abort connections in
77  * their incomplete listen queues should they no longer be required.  Sockets
78  * placed in completed connection listen queues should not be aborted for
79  * reasons described in the comment above the soclose() implementation.  This
80  * is not a general purpose close routine, and except in the specific
81  * circumstances described here, should not be used.
82  *
83  * sofree() will free a socket and its protocol state if all references on
84  * the socket have been released, and is the public interface to attempt to
85  * free a socket when a reference is removed.  This is a socket layer private
86  * interface.
87  *
88  * NOTE: In addition to socreate() and soclose(), which provide a single
89  * socket reference to the consumer to be managed as required, there are two
90  * calls to explicitly manage socket references, soref(), and sorele().
91  * Currently, these are generally required only when transitioning a socket
92  * from a listen queue to a file descriptor, in order to prevent garbage
93  * collection of the socket at an untimely moment.  For a number of reasons,
94  * these interfaces are not preferred, and should be avoided.
95  *
96  * NOTE: With regard to VNETs the general rule is that callers do not set
97  * curvnet. Exceptions to this rule include soabort(), sodisconnect(),
98  * sofree(), sorele(), sonewconn() and sorflush(), which are usually called
99  * from a pre-set VNET context.  sopoll() currently does not need a VNET
100  * context to be set.
101  */
102 
103 #include <sys/cdefs.h>
104 #include "opt_inet.h"
105 #include "opt_inet6.h"
106 #include "opt_kern_tls.h"
107 #include "opt_ktrace.h"
108 #include "opt_sctp.h"
109 
110 #include <sys/param.h>
111 #include <sys/systm.h>
112 #include <sys/capsicum.h>
113 #include <sys/fcntl.h>
114 #include <sys/limits.h>
115 #include <sys/lock.h>
116 #include <sys/mac.h>
117 #include <sys/malloc.h>
118 #include <sys/mbuf.h>
119 #include <sys/mutex.h>
120 #include <sys/domain.h>
121 #include <sys/file.h>			/* for struct knote */
122 #include <sys/hhook.h>
123 #include <sys/kernel.h>
124 #include <sys/khelp.h>
125 #include <sys/ktls.h>
126 #include <sys/event.h>
127 #include <sys/eventhandler.h>
128 #include <sys/poll.h>
129 #include <sys/proc.h>
130 #include <sys/protosw.h>
131 #include <sys/sbuf.h>
132 #include <sys/socket.h>
133 #include <sys/socketvar.h>
134 #include <sys/resourcevar.h>
135 #include <net/route.h>
136 #include <sys/signalvar.h>
137 #include <sys/stat.h>
138 #include <sys/sx.h>
139 #include <sys/sysctl.h>
140 #include <sys/taskqueue.h>
141 #include <sys/uio.h>
142 #include <sys/un.h>
143 #include <sys/unpcb.h>
144 #include <sys/jail.h>
145 #include <sys/syslog.h>
146 #include <netinet/in.h>
147 #include <netinet/in_pcb.h>
148 #include <netinet/tcp.h>
149 
150 #include <net/vnet.h>
151 
152 #include <security/mac/mac_framework.h>
153 
154 #include <vm/uma.h>
155 
156 #ifdef COMPAT_FREEBSD32
157 #include <sys/mount.h>
158 #include <sys/sysent.h>
159 #include <compat/freebsd32/freebsd32.h>
160 #endif
161 
162 static int	soreceive_rcvoob(struct socket *so, struct uio *uio,
163 		    int flags);
164 static void	so_rdknl_lock(void *);
165 static void	so_rdknl_unlock(void *);
166 static void	so_rdknl_assert_lock(void *, int);
167 static void	so_wrknl_lock(void *);
168 static void	so_wrknl_unlock(void *);
169 static void	so_wrknl_assert_lock(void *, int);
170 
171 static void	filt_sordetach(struct knote *kn);
172 static int	filt_soread(struct knote *kn, long hint);
173 static void	filt_sowdetach(struct knote *kn);
174 static int	filt_sowrite(struct knote *kn, long hint);
175 static int	filt_soempty(struct knote *kn, long hint);
176 fo_kqfilter_t	soo_kqfilter;
177 
178 static struct filterops soread_filtops = {
179 	.f_isfd = 1,
180 	.f_detach = filt_sordetach,
181 	.f_event = filt_soread,
182 };
183 static struct filterops sowrite_filtops = {
184 	.f_isfd = 1,
185 	.f_detach = filt_sowdetach,
186 	.f_event = filt_sowrite,
187 };
188 static struct filterops soempty_filtops = {
189 	.f_isfd = 1,
190 	.f_detach = filt_sowdetach,
191 	.f_event = filt_soempty,
192 };
193 
194 so_gen_t	so_gencnt;	/* generation count for sockets */
195 
196 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
197 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
198 
199 #define	VNET_SO_ASSERT(so)						\
200 	VNET_ASSERT(curvnet != NULL,					\
201 	    ("%s:%d curvnet is NULL, so=%p", __func__, __LINE__, (so)));
202 
203 #ifdef SOCKET_HHOOK
204 VNET_DEFINE(struct hhook_head *, socket_hhh[HHOOK_SOCKET_LAST + 1]);
205 #define	V_socket_hhh		VNET(socket_hhh)
206 static inline int hhook_run_socket(struct socket *, void *, int32_t);
207 #endif
208 
209 /*
210  * Limit on the number of connections in the listen queue waiting
211  * for accept(2).
212  * NB: The original sysctl somaxconn is still available but hidden
213  * to prevent confusion about the actual purpose of this number.
214  */
215 static u_int somaxconn = SOMAXCONN;
216 
217 static int
218 sysctl_somaxconn(SYSCTL_HANDLER_ARGS)
219 {
220 	int error;
221 	int val;
222 
223 	val = somaxconn;
224 	error = sysctl_handle_int(oidp, &val, 0, req);
225 	if (error || !req->newptr )
226 		return (error);
227 
228 	/*
229 	 * The purpose of the UINT_MAX / 3 limit, is so that the formula
230 	 *   3 * so_qlimit / 2
231 	 * below, will not overflow.
232          */
233 
234 	if (val < 1 || val > UINT_MAX / 3)
235 		return (EINVAL);
236 
237 	somaxconn = val;
238 	return (0);
239 }
240 SYSCTL_PROC(_kern_ipc, OID_AUTO, soacceptqueue,
241     CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_MPSAFE, 0, sizeof(int),
242     sysctl_somaxconn, "I",
243     "Maximum listen socket pending connection accept queue size");
244 SYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn,
245     CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_SKIP | CTLFLAG_MPSAFE, 0,
246     sizeof(int), sysctl_somaxconn, "I",
247     "Maximum listen socket pending connection accept queue size (compat)");
248 
249 static int numopensockets;
250 SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD,
251     &numopensockets, 0, "Number of open sockets");
252 
253 /*
254  * so_global_mtx protects so_gencnt, numopensockets, and the per-socket
255  * so_gencnt field.
256  */
257 static struct mtx so_global_mtx;
258 MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF);
259 
260 /*
261  * General IPC sysctl name space, used by sockets and a variety of other IPC
262  * types.
263  */
264 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
265     "IPC");
266 
267 /*
268  * Initialize the socket subsystem and set up the socket
269  * memory allocator.
270  */
271 static uma_zone_t socket_zone;
272 int	maxsockets;
273 
274 static void
275 socket_zone_change(void *tag)
276 {
277 
278 	maxsockets = uma_zone_set_max(socket_zone, maxsockets);
279 }
280 
281 static void
282 socket_init(void *tag)
283 {
284 
285 	socket_zone = uma_zcreate("socket", sizeof(struct socket), NULL, NULL,
286 	    NULL, NULL, UMA_ALIGN_PTR, 0);
287 	maxsockets = uma_zone_set_max(socket_zone, maxsockets);
288 	uma_zone_set_warning(socket_zone, "kern.ipc.maxsockets limit reached");
289 	EVENTHANDLER_REGISTER(maxsockets_change, socket_zone_change, NULL,
290 	    EVENTHANDLER_PRI_FIRST);
291 }
292 SYSINIT(socket, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, socket_init, NULL);
293 
294 #ifdef SOCKET_HHOOK
295 static void
296 socket_hhook_register(int subtype)
297 {
298 
299 	if (hhook_head_register(HHOOK_TYPE_SOCKET, subtype,
300 	    &V_socket_hhh[subtype],
301 	    HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
302 		printf("%s: WARNING: unable to register hook\n", __func__);
303 }
304 
305 static void
306 socket_hhook_deregister(int subtype)
307 {
308 
309 	if (hhook_head_deregister(V_socket_hhh[subtype]) != 0)
310 		printf("%s: WARNING: unable to deregister hook\n", __func__);
311 }
312 
313 static void
314 socket_vnet_init(const void *unused __unused)
315 {
316 	int i;
317 
318 	/* We expect a contiguous range */
319 	for (i = 0; i <= HHOOK_SOCKET_LAST; i++)
320 		socket_hhook_register(i);
321 }
322 VNET_SYSINIT(socket_vnet_init, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
323     socket_vnet_init, NULL);
324 
325 static void
326 socket_vnet_uninit(const void *unused __unused)
327 {
328 	int i;
329 
330 	for (i = 0; i <= HHOOK_SOCKET_LAST; i++)
331 		socket_hhook_deregister(i);
332 }
333 VNET_SYSUNINIT(socket_vnet_uninit, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
334     socket_vnet_uninit, NULL);
335 #endif	/* SOCKET_HHOOK */
336 
337 /*
338  * Initialise maxsockets.  This SYSINIT must be run after
339  * tunable_mbinit().
340  */
341 static void
342 init_maxsockets(void *ignored)
343 {
344 
345 	TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
346 	maxsockets = imax(maxsockets, maxfiles);
347 }
348 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL);
349 
350 /*
351  * Sysctl to get and set the maximum global sockets limit.  Notify protocols
352  * of the change so that they can update their dependent limits as required.
353  */
354 static int
355 sysctl_maxsockets(SYSCTL_HANDLER_ARGS)
356 {
357 	int error, newmaxsockets;
358 
359 	newmaxsockets = maxsockets;
360 	error = sysctl_handle_int(oidp, &newmaxsockets, 0, req);
361 	if (error == 0 && req->newptr && newmaxsockets != maxsockets) {
362 		if (newmaxsockets > maxsockets &&
363 		    newmaxsockets <= maxfiles) {
364 			maxsockets = newmaxsockets;
365 			EVENTHANDLER_INVOKE(maxsockets_change);
366 		} else
367 			error = EINVAL;
368 	}
369 	return (error);
370 }
371 SYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets,
372     CTLTYPE_INT | CTLFLAG_RWTUN | CTLFLAG_NOFETCH | CTLFLAG_MPSAFE,
373     &maxsockets, 0, sysctl_maxsockets, "IU",
374     "Maximum number of sockets available");
375 
376 /*
377  * Socket operation routines.  These routines are called by the routines in
378  * sys_socket.c or from a system process, and implement the semantics of
379  * socket operations by switching out to the protocol specific routines.
380  */
381 
382 /*
383  * Get a socket structure from our zone, and initialize it.  Note that it
384  * would probably be better to allocate socket and PCB at the same time, but
385  * I'm not convinced that all the protocols can be easily modified to do
386  * this.
387  *
388  * soalloc() returns a socket with a ref count of 0.
389  */
390 static struct socket *
391 soalloc(struct vnet *vnet)
392 {
393 	struct socket *so;
394 
395 	so = uma_zalloc(socket_zone, M_NOWAIT | M_ZERO);
396 	if (so == NULL)
397 		return (NULL);
398 #ifdef MAC
399 	if (mac_socket_init(so, M_NOWAIT) != 0) {
400 		uma_zfree(socket_zone, so);
401 		return (NULL);
402 	}
403 #endif
404 	if (khelp_init_osd(HELPER_CLASS_SOCKET, &so->osd)) {
405 		uma_zfree(socket_zone, so);
406 		return (NULL);
407 	}
408 
409 	/*
410 	 * The socket locking protocol allows to lock 2 sockets at a time,
411 	 * however, the first one must be a listening socket.  WITNESS lacks
412 	 * a feature to change class of an existing lock, so we use DUPOK.
413 	 */
414 	mtx_init(&so->so_lock, "socket", NULL, MTX_DEF | MTX_DUPOK);
415 	mtx_init(&so->so_snd_mtx, "so_snd", NULL, MTX_DEF);
416 	mtx_init(&so->so_rcv_mtx, "so_rcv", NULL, MTX_DEF);
417 	so->so_rcv.sb_sel = &so->so_rdsel;
418 	so->so_snd.sb_sel = &so->so_wrsel;
419 	sx_init(&so->so_snd_sx, "so_snd_sx");
420 	sx_init(&so->so_rcv_sx, "so_rcv_sx");
421 	TAILQ_INIT(&so->so_snd.sb_aiojobq);
422 	TAILQ_INIT(&so->so_rcv.sb_aiojobq);
423 	TASK_INIT(&so->so_snd.sb_aiotask, 0, soaio_snd, so);
424 	TASK_INIT(&so->so_rcv.sb_aiotask, 0, soaio_rcv, so);
425 #ifdef VIMAGE
426 	VNET_ASSERT(vnet != NULL, ("%s:%d vnet is NULL, so=%p",
427 	    __func__, __LINE__, so));
428 	so->so_vnet = vnet;
429 #endif
430 #ifdef SOCKET_HHOOK
431 	/* We shouldn't need the so_global_mtx */
432 	if (hhook_run_socket(so, NULL, HHOOK_SOCKET_CREATE)) {
433 		/* Do we need more comprehensive error returns? */
434 		uma_zfree(socket_zone, so);
435 		return (NULL);
436 	}
437 #endif
438 	mtx_lock(&so_global_mtx);
439 	so->so_gencnt = ++so_gencnt;
440 	++numopensockets;
441 #ifdef VIMAGE
442 	vnet->vnet_sockcnt++;
443 #endif
444 	mtx_unlock(&so_global_mtx);
445 
446 	return (so);
447 }
448 
449 /*
450  * Free the storage associated with a socket at the socket layer, tear down
451  * locks, labels, etc.  All protocol state is assumed already to have been
452  * torn down (and possibly never set up) by the caller.
453  */
454 void
455 sodealloc(struct socket *so)
456 {
457 
458 	KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count));
459 	KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL"));
460 
461 	mtx_lock(&so_global_mtx);
462 	so->so_gencnt = ++so_gencnt;
463 	--numopensockets;	/* Could be below, but faster here. */
464 #ifdef VIMAGE
465 	VNET_ASSERT(so->so_vnet != NULL, ("%s:%d so_vnet is NULL, so=%p",
466 	    __func__, __LINE__, so));
467 	so->so_vnet->vnet_sockcnt--;
468 #endif
469 	mtx_unlock(&so_global_mtx);
470 #ifdef MAC
471 	mac_socket_destroy(so);
472 #endif
473 #ifdef SOCKET_HHOOK
474 	hhook_run_socket(so, NULL, HHOOK_SOCKET_CLOSE);
475 #endif
476 
477 	khelp_destroy_osd(&so->osd);
478 	if (SOLISTENING(so)) {
479 		if (so->sol_accept_filter != NULL)
480 			accept_filt_setopt(so, NULL);
481 	} else {
482 		if (so->so_rcv.sb_hiwat)
483 			(void)chgsbsize(so->so_cred->cr_uidinfo,
484 			    &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY);
485 		if (so->so_snd.sb_hiwat)
486 			(void)chgsbsize(so->so_cred->cr_uidinfo,
487 			    &so->so_snd.sb_hiwat, 0, RLIM_INFINITY);
488 		sx_destroy(&so->so_snd_sx);
489 		sx_destroy(&so->so_rcv_sx);
490 		mtx_destroy(&so->so_snd_mtx);
491 		mtx_destroy(&so->so_rcv_mtx);
492 	}
493 	crfree(so->so_cred);
494 	mtx_destroy(&so->so_lock);
495 	uma_zfree(socket_zone, so);
496 }
497 
498 /*
499  * socreate returns a socket with a ref count of 1 and a file descriptor
500  * reference.  The socket should be closed with soclose().
501  */
502 int
503 socreate(int dom, struct socket **aso, int type, int proto,
504     struct ucred *cred, struct thread *td)
505 {
506 	struct protosw *prp;
507 	struct socket *so;
508 	int error;
509 
510 	/*
511 	 * XXX: divert(4) historically abused PF_INET.  Keep this compatibility
512 	 * shim until all applications have been updated.
513 	 */
514 	if (__predict_false(dom == PF_INET && type == SOCK_RAW &&
515 	    proto == IPPROTO_DIVERT)) {
516 		dom = PF_DIVERT;
517 		printf("%s uses obsolete way to create divert(4) socket\n",
518 		    td->td_proc->p_comm);
519 	}
520 
521 	prp = pffindproto(dom, type, proto);
522 	if (prp == NULL) {
523 		/* No support for domain. */
524 		if (pffinddomain(dom) == NULL)
525 			return (EAFNOSUPPORT);
526 		/* No support for socket type. */
527 		if (proto == 0 && type != 0)
528 			return (EPROTOTYPE);
529 		return (EPROTONOSUPPORT);
530 	}
531 
532 	MPASS(prp->pr_attach);
533 
534 	if ((prp->pr_flags & PR_CAPATTACH) == 0) {
535 		if (CAP_TRACING(td))
536 			ktrcapfail(CAPFAIL_PROTO, &proto);
537 		if (IN_CAPABILITY_MODE(td))
538 			return (ECAPMODE);
539 	}
540 
541 	if (prison_check_af(cred, prp->pr_domain->dom_family) != 0)
542 		return (EPROTONOSUPPORT);
543 
544 	so = soalloc(CRED_TO_VNET(cred));
545 	if (so == NULL)
546 		return (ENOBUFS);
547 
548 	so->so_type = type;
549 	so->so_cred = crhold(cred);
550 	if ((prp->pr_domain->dom_family == PF_INET) ||
551 	    (prp->pr_domain->dom_family == PF_INET6) ||
552 	    (prp->pr_domain->dom_family == PF_ROUTE))
553 		so->so_fibnum = td->td_proc->p_fibnum;
554 	else
555 		so->so_fibnum = 0;
556 	so->so_proto = prp;
557 #ifdef MAC
558 	mac_socket_create(cred, so);
559 #endif
560 	knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
561 	    so_rdknl_assert_lock);
562 	knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
563 	    so_wrknl_assert_lock);
564 	if ((prp->pr_flags & PR_SOCKBUF) == 0) {
565 		so->so_snd.sb_mtx = &so->so_snd_mtx;
566 		so->so_rcv.sb_mtx = &so->so_rcv_mtx;
567 	}
568 	/*
569 	 * Auto-sizing of socket buffers is managed by the protocols and
570 	 * the appropriate flags must be set in the pru_attach function.
571 	 */
572 	CURVNET_SET(so->so_vnet);
573 	error = prp->pr_attach(so, proto, td);
574 	CURVNET_RESTORE();
575 	if (error) {
576 		sodealloc(so);
577 		return (error);
578 	}
579 	soref(so);
580 	*aso = so;
581 	return (0);
582 }
583 
584 #ifdef REGRESSION
585 static int regression_sonewconn_earlytest = 1;
586 SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW,
587     &regression_sonewconn_earlytest, 0, "Perform early sonewconn limit test");
588 #endif
589 
590 static int sooverprio = LOG_DEBUG;
591 SYSCTL_INT(_kern_ipc, OID_AUTO, sooverprio, CTLFLAG_RW,
592     &sooverprio, 0, "Log priority for listen socket overflows: 0..7 or -1 to disable");
593 
594 static struct timeval overinterval = { 60, 0 };
595 SYSCTL_TIMEVAL_SEC(_kern_ipc, OID_AUTO, sooverinterval, CTLFLAG_RW,
596     &overinterval,
597     "Delay in seconds between warnings for listen socket overflows");
598 
599 /*
600  * When an attempt at a new connection is noted on a socket which supports
601  * accept(2), the protocol has two options:
602  * 1) Call legacy sonewconn() function, which would call protocol attach
603  *    method, same as used for socket(2).
604  * 2) Call solisten_clone(), do attach that is specific to a cloned connection,
605  *    and then call solisten_enqueue().
606  *
607  * Note: the ref count on the socket is 0 on return.
608  */
609 struct socket *
610 solisten_clone(struct socket *head)
611 {
612 	struct sbuf descrsb;
613 	struct socket *so;
614 	int len, overcount;
615 	u_int qlen;
616 	const char localprefix[] = "local:";
617 	char descrbuf[SUNPATHLEN + sizeof(localprefix)];
618 #if defined(INET6)
619 	char addrbuf[INET6_ADDRSTRLEN];
620 #elif defined(INET)
621 	char addrbuf[INET_ADDRSTRLEN];
622 #endif
623 	bool dolog, over;
624 
625 	SOLISTEN_LOCK(head);
626 	over = (head->sol_qlen > 3 * head->sol_qlimit / 2);
627 #ifdef REGRESSION
628 	if (regression_sonewconn_earlytest && over) {
629 #else
630 	if (over) {
631 #endif
632 		head->sol_overcount++;
633 		dolog = (sooverprio >= 0) &&
634 			!!ratecheck(&head->sol_lastover, &overinterval);
635 
636 		/*
637 		 * If we're going to log, copy the overflow count and queue
638 		 * length from the listen socket before dropping the lock.
639 		 * Also, reset the overflow count.
640 		 */
641 		if (dolog) {
642 			overcount = head->sol_overcount;
643 			head->sol_overcount = 0;
644 			qlen = head->sol_qlen;
645 		}
646 		SOLISTEN_UNLOCK(head);
647 
648 		if (dolog) {
649 			/*
650 			 * Try to print something descriptive about the
651 			 * socket for the error message.
652 			 */
653 			sbuf_new(&descrsb, descrbuf, sizeof(descrbuf),
654 			    SBUF_FIXEDLEN);
655 			switch (head->so_proto->pr_domain->dom_family) {
656 #if defined(INET) || defined(INET6)
657 #ifdef INET
658 			case AF_INET:
659 #endif
660 #ifdef INET6
661 			case AF_INET6:
662 				if (head->so_proto->pr_domain->dom_family ==
663 				    AF_INET6 ||
664 				    (sotoinpcb(head)->inp_inc.inc_flags &
665 				    INC_ISIPV6)) {
666 					ip6_sprintf(addrbuf,
667 					    &sotoinpcb(head)->inp_inc.inc6_laddr);
668 					sbuf_printf(&descrsb, "[%s]", addrbuf);
669 				} else
670 #endif
671 				{
672 #ifdef INET
673 					inet_ntoa_r(
674 					    sotoinpcb(head)->inp_inc.inc_laddr,
675 					    addrbuf);
676 					sbuf_cat(&descrsb, addrbuf);
677 #endif
678 				}
679 				sbuf_printf(&descrsb, ":%hu (proto %u)",
680 				    ntohs(sotoinpcb(head)->inp_inc.inc_lport),
681 				    head->so_proto->pr_protocol);
682 				break;
683 #endif /* INET || INET6 */
684 			case AF_UNIX:
685 				sbuf_cat(&descrsb, localprefix);
686 				if (sotounpcb(head)->unp_addr != NULL)
687 					len =
688 					    sotounpcb(head)->unp_addr->sun_len -
689 					    offsetof(struct sockaddr_un,
690 					    sun_path);
691 				else
692 					len = 0;
693 				if (len > 0)
694 					sbuf_bcat(&descrsb,
695 					    sotounpcb(head)->unp_addr->sun_path,
696 					    len);
697 				else
698 					sbuf_cat(&descrsb, "(unknown)");
699 				break;
700 			}
701 
702 			/*
703 			 * If we can't print something more specific, at least
704 			 * print the domain name.
705 			 */
706 			if (sbuf_finish(&descrsb) != 0 ||
707 			    sbuf_len(&descrsb) <= 0) {
708 				sbuf_clear(&descrsb);
709 				sbuf_cat(&descrsb,
710 				    head->so_proto->pr_domain->dom_name ?:
711 				    "unknown");
712 				sbuf_finish(&descrsb);
713 			}
714 			KASSERT(sbuf_len(&descrsb) > 0,
715 			    ("%s: sbuf creation failed", __func__));
716 			/*
717 			 * Preserve the historic listen queue overflow log
718 			 * message, that starts with "sonewconn:".  It has
719 			 * been known to sysadmins for years and also test
720 			 * sys/kern/sonewconn_overflow checks for it.
721 			 */
722 			if (head->so_cred == 0) {
723 				log(LOG_PRI(sooverprio),
724 				    "sonewconn: pcb %p (%s): "
725 				    "Listen queue overflow: %i already in "
726 				    "queue awaiting acceptance (%d "
727 				    "occurrences)\n", head->so_pcb,
728 				    sbuf_data(&descrsb),
729 			    	qlen, overcount);
730 			} else {
731 				log(LOG_PRI(sooverprio),
732 				    "sonewconn: pcb %p (%s): "
733 				    "Listen queue overflow: "
734 				    "%i already in queue awaiting acceptance "
735 				    "(%d occurrences), euid %d, rgid %d, jail %s\n",
736 				    head->so_pcb, sbuf_data(&descrsb), qlen,
737 				    overcount, head->so_cred->cr_uid,
738 				    head->so_cred->cr_rgid,
739 				    head->so_cred->cr_prison ?
740 					head->so_cred->cr_prison->pr_name :
741 					"not_jailed");
742 			}
743 			sbuf_delete(&descrsb);
744 
745 			overcount = 0;
746 		}
747 
748 		return (NULL);
749 	}
750 	SOLISTEN_UNLOCK(head);
751 	VNET_ASSERT(head->so_vnet != NULL, ("%s: so %p vnet is NULL",
752 	    __func__, head));
753 	so = soalloc(head->so_vnet);
754 	if (so == NULL) {
755 		log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: "
756 		    "limit reached or out of memory\n",
757 		    __func__, head->so_pcb);
758 		return (NULL);
759 	}
760 	so->so_listen = head;
761 	so->so_type = head->so_type;
762 	/*
763 	 * POSIX is ambiguous on what options an accept(2)ed socket should
764 	 * inherit from the listener.  Words "create a new socket" may be
765 	 * interpreted as not inheriting anything.  Best programming practice
766 	 * for application developers is to not rely on such inheritance.
767 	 * FreeBSD had historically inherited all so_options excluding
768 	 * SO_ACCEPTCONN, which virtually means all SOL_SOCKET level options,
769 	 * including those completely irrelevant to a new born socket.  For
770 	 * compatibility with older versions we will inherit a list of
771 	 * meaningful options.
772 	 * The crucial bit to inherit is SO_ACCEPTFILTER.  We need it present
773 	 * in the child socket for soisconnected() promoting socket from the
774 	 * incomplete queue to complete.  It will be cleared before the child
775 	 * gets available to accept(2).
776 	 */
777 	so->so_options = head->so_options & (SO_ACCEPTFILTER | SO_KEEPALIVE |
778 	    SO_DONTROUTE | SO_LINGER | SO_OOBINLINE | SO_NOSIGPIPE);
779 	so->so_linger = head->so_linger;
780 	so->so_state = head->so_state;
781 	so->so_fibnum = head->so_fibnum;
782 	so->so_proto = head->so_proto;
783 	so->so_cred = crhold(head->so_cred);
784 #ifdef SOCKET_HHOOK
785 	if (V_socket_hhh[HHOOK_SOCKET_NEWCONN]->hhh_nhooks > 0) {
786 		if (hhook_run_socket(so, head, HHOOK_SOCKET_NEWCONN)) {
787 			sodealloc(so);
788 			log(LOG_DEBUG, "%s: hhook run failed\n", __func__);
789 			return (NULL);
790 		}
791 	}
792 #endif
793 #ifdef MAC
794 	mac_socket_newconn(head, so);
795 #endif
796 	knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
797 	    so_rdknl_assert_lock);
798 	knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
799 	    so_wrknl_assert_lock);
800 	VNET_SO_ASSERT(head);
801 	if (soreserve(so, head->sol_sbsnd_hiwat, head->sol_sbrcv_hiwat)) {
802 		sodealloc(so);
803 		log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n",
804 		    __func__, head->so_pcb);
805 		return (NULL);
806 	}
807 	so->so_rcv.sb_lowat = head->sol_sbrcv_lowat;
808 	so->so_snd.sb_lowat = head->sol_sbsnd_lowat;
809 	so->so_rcv.sb_timeo = head->sol_sbrcv_timeo;
810 	so->so_snd.sb_timeo = head->sol_sbsnd_timeo;
811 	so->so_rcv.sb_flags = head->sol_sbrcv_flags & SB_AUTOSIZE;
812 	so->so_snd.sb_flags = head->sol_sbsnd_flags & SB_AUTOSIZE;
813 	if ((so->so_proto->pr_flags & PR_SOCKBUF) == 0) {
814 		so->so_snd.sb_mtx = &so->so_snd_mtx;
815 		so->so_rcv.sb_mtx = &so->so_rcv_mtx;
816 	}
817 
818 	return (so);
819 }
820 
821 /* Connstatus may be 0 or SS_ISCONNECTED. */
822 struct socket *
823 sonewconn(struct socket *head, int connstatus)
824 {
825 	struct socket *so;
826 
827 	if ((so = solisten_clone(head)) == NULL)
828 		return (NULL);
829 
830 	if (so->so_proto->pr_attach(so, 0, NULL) != 0) {
831 		sodealloc(so);
832 		log(LOG_DEBUG, "%s: pcb %p: pr_attach() failed\n",
833 		    __func__, head->so_pcb);
834 		return (NULL);
835 	}
836 
837 	(void)solisten_enqueue(so, connstatus);
838 
839 	return (so);
840 }
841 
842 /*
843  * Enqueue socket cloned by solisten_clone() to the listen queue of the
844  * listener it has been cloned from.
845  *
846  * Return 'true' if socket landed on complete queue, otherwise 'false'.
847  */
848 bool
849 solisten_enqueue(struct socket *so, int connstatus)
850 {
851 	struct socket *head = so->so_listen;
852 
853 	MPASS(refcount_load(&so->so_count) == 0);
854 	refcount_init(&so->so_count, 1);
855 
856 	SOLISTEN_LOCK(head);
857 	if (head->sol_accept_filter != NULL)
858 		connstatus = 0;
859 	so->so_state |= connstatus;
860 	soref(head); /* A socket on (in)complete queue refs head. */
861 	if (connstatus) {
862 		TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list);
863 		so->so_qstate = SQ_COMP;
864 		head->sol_qlen++;
865 		solisten_wakeup(head);	/* unlocks */
866 		return (true);
867 	} else {
868 		/*
869 		 * Keep removing sockets from the head until there's room for
870 		 * us to insert on the tail.  In pre-locking revisions, this
871 		 * was a simple if(), but as we could be racing with other
872 		 * threads and soabort() requires dropping locks, we must
873 		 * loop waiting for the condition to be true.
874 		 */
875 		while (head->sol_incqlen > head->sol_qlimit) {
876 			struct socket *sp;
877 
878 			sp = TAILQ_FIRST(&head->sol_incomp);
879 			TAILQ_REMOVE(&head->sol_incomp, sp, so_list);
880 			head->sol_incqlen--;
881 			SOCK_LOCK(sp);
882 			sp->so_qstate = SQ_NONE;
883 			sp->so_listen = NULL;
884 			SOCK_UNLOCK(sp);
885 			sorele_locked(head);	/* does SOLISTEN_UNLOCK, head stays */
886 			soabort(sp);
887 			SOLISTEN_LOCK(head);
888 		}
889 		TAILQ_INSERT_TAIL(&head->sol_incomp, so, so_list);
890 		so->so_qstate = SQ_INCOMP;
891 		head->sol_incqlen++;
892 		SOLISTEN_UNLOCK(head);
893 		return (false);
894 	}
895 }
896 
897 #if defined(SCTP) || defined(SCTP_SUPPORT)
898 /*
899  * Socket part of sctp_peeloff().  Detach a new socket from an
900  * association.  The new socket is returned with a reference.
901  *
902  * XXXGL: reduce copy-paste with solisten_clone().
903  */
904 struct socket *
905 sopeeloff(struct socket *head)
906 {
907 	struct socket *so;
908 
909 	VNET_ASSERT(head->so_vnet != NULL, ("%s:%d so_vnet is NULL, head=%p",
910 	    __func__, __LINE__, head));
911 	so = soalloc(head->so_vnet);
912 	if (so == NULL) {
913 		log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: "
914 		    "limit reached or out of memory\n",
915 		    __func__, head->so_pcb);
916 		return (NULL);
917 	}
918 	so->so_type = head->so_type;
919 	so->so_options = head->so_options;
920 	so->so_linger = head->so_linger;
921 	so->so_state = (head->so_state & SS_NBIO) | SS_ISCONNECTED;
922 	so->so_fibnum = head->so_fibnum;
923 	so->so_proto = head->so_proto;
924 	so->so_cred = crhold(head->so_cred);
925 #ifdef MAC
926 	mac_socket_newconn(head, so);
927 #endif
928 	knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
929 	    so_rdknl_assert_lock);
930 	knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
931 	    so_wrknl_assert_lock);
932 	VNET_SO_ASSERT(head);
933 	if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat)) {
934 		sodealloc(so);
935 		log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n",
936 		    __func__, head->so_pcb);
937 		return (NULL);
938 	}
939 	if ((*so->so_proto->pr_attach)(so, 0, NULL)) {
940 		sodealloc(so);
941 		log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n",
942 		    __func__, head->so_pcb);
943 		return (NULL);
944 	}
945 	so->so_rcv.sb_lowat = head->so_rcv.sb_lowat;
946 	so->so_snd.sb_lowat = head->so_snd.sb_lowat;
947 	so->so_rcv.sb_timeo = head->so_rcv.sb_timeo;
948 	so->so_snd.sb_timeo = head->so_snd.sb_timeo;
949 	so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE;
950 	so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE;
951 	if ((so->so_proto->pr_flags & PR_SOCKBUF) == 0) {
952 		so->so_snd.sb_mtx = &so->so_snd_mtx;
953 		so->so_rcv.sb_mtx = &so->so_rcv_mtx;
954 	}
955 
956 	soref(so);
957 
958 	return (so);
959 }
960 #endif	/* SCTP */
961 
962 int
963 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
964 {
965 	int error;
966 
967 	CURVNET_SET(so->so_vnet);
968 	error = so->so_proto->pr_bind(so, nam, td);
969 	CURVNET_RESTORE();
970 	return (error);
971 }
972 
973 int
974 sobindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
975 {
976 	int error;
977 
978 	CURVNET_SET(so->so_vnet);
979 	error = so->so_proto->pr_bindat(fd, so, nam, td);
980 	CURVNET_RESTORE();
981 	return (error);
982 }
983 
984 /*
985  * solisten() transitions a socket from a non-listening state to a listening
986  * state, but can also be used to update the listen queue depth on an
987  * existing listen socket.  The protocol will call back into the sockets
988  * layer using solisten_proto_check() and solisten_proto() to check and set
989  * socket-layer listen state.  Call backs are used so that the protocol can
990  * acquire both protocol and socket layer locks in whatever order is required
991  * by the protocol.
992  *
993  * Protocol implementors are advised to hold the socket lock across the
994  * socket-layer test and set to avoid races at the socket layer.
995  */
996 int
997 solisten(struct socket *so, int backlog, struct thread *td)
998 {
999 	int error;
1000 
1001 	CURVNET_SET(so->so_vnet);
1002 	error = so->so_proto->pr_listen(so, backlog, td);
1003 	CURVNET_RESTORE();
1004 	return (error);
1005 }
1006 
1007 /*
1008  * Prepare for a call to solisten_proto().  Acquire all socket buffer locks in
1009  * order to interlock with socket I/O.
1010  */
1011 int
1012 solisten_proto_check(struct socket *so)
1013 {
1014 	SOCK_LOCK_ASSERT(so);
1015 
1016 	if ((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
1017 	    SS_ISDISCONNECTING)) != 0)
1018 		return (EINVAL);
1019 
1020 	/*
1021 	 * Sleeping is not permitted here, so simply fail if userspace is
1022 	 * attempting to transmit or receive on the socket.  This kind of
1023 	 * transient failure is not ideal, but it should occur only if userspace
1024 	 * is misusing the socket interfaces.
1025 	 */
1026 	if (!sx_try_xlock(&so->so_snd_sx))
1027 		return (EAGAIN);
1028 	if (!sx_try_xlock(&so->so_rcv_sx)) {
1029 		sx_xunlock(&so->so_snd_sx);
1030 		return (EAGAIN);
1031 	}
1032 	mtx_lock(&so->so_snd_mtx);
1033 	mtx_lock(&so->so_rcv_mtx);
1034 
1035 	/* Interlock with soo_aio_queue() and KTLS. */
1036 	if (!SOLISTENING(so)) {
1037 		bool ktls;
1038 
1039 #ifdef KERN_TLS
1040 		ktls = so->so_snd.sb_tls_info != NULL ||
1041 		    so->so_rcv.sb_tls_info != NULL;
1042 #else
1043 		ktls = false;
1044 #endif
1045 		if (ktls ||
1046 		    (so->so_snd.sb_flags & (SB_AIO | SB_AIO_RUNNING)) != 0 ||
1047 		    (so->so_rcv.sb_flags & (SB_AIO | SB_AIO_RUNNING)) != 0) {
1048 			solisten_proto_abort(so);
1049 			return (EINVAL);
1050 		}
1051 	}
1052 
1053 	return (0);
1054 }
1055 
1056 /*
1057  * Undo the setup done by solisten_proto_check().
1058  */
1059 void
1060 solisten_proto_abort(struct socket *so)
1061 {
1062 	mtx_unlock(&so->so_snd_mtx);
1063 	mtx_unlock(&so->so_rcv_mtx);
1064 	sx_xunlock(&so->so_snd_sx);
1065 	sx_xunlock(&so->so_rcv_sx);
1066 }
1067 
1068 void
1069 solisten_proto(struct socket *so, int backlog)
1070 {
1071 	int sbrcv_lowat, sbsnd_lowat;
1072 	u_int sbrcv_hiwat, sbsnd_hiwat;
1073 	short sbrcv_flags, sbsnd_flags;
1074 	sbintime_t sbrcv_timeo, sbsnd_timeo;
1075 
1076 	SOCK_LOCK_ASSERT(so);
1077 	KASSERT((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
1078 	    SS_ISDISCONNECTING)) == 0,
1079 	    ("%s: bad socket state %p", __func__, so));
1080 
1081 	if (SOLISTENING(so))
1082 		goto listening;
1083 
1084 	/*
1085 	 * Change this socket to listening state.
1086 	 */
1087 	sbrcv_lowat = so->so_rcv.sb_lowat;
1088 	sbsnd_lowat = so->so_snd.sb_lowat;
1089 	sbrcv_hiwat = so->so_rcv.sb_hiwat;
1090 	sbsnd_hiwat = so->so_snd.sb_hiwat;
1091 	sbrcv_flags = so->so_rcv.sb_flags;
1092 	sbsnd_flags = so->so_snd.sb_flags;
1093 	sbrcv_timeo = so->so_rcv.sb_timeo;
1094 	sbsnd_timeo = so->so_snd.sb_timeo;
1095 
1096 	if (!(so->so_proto->pr_flags & PR_SOCKBUF)) {
1097 		sbdestroy(so, SO_SND);
1098 		sbdestroy(so, SO_RCV);
1099 	}
1100 
1101 #ifdef INVARIANTS
1102 	bzero(&so->so_rcv,
1103 	    sizeof(struct socket) - offsetof(struct socket, so_rcv));
1104 #endif
1105 
1106 	so->sol_sbrcv_lowat = sbrcv_lowat;
1107 	so->sol_sbsnd_lowat = sbsnd_lowat;
1108 	so->sol_sbrcv_hiwat = sbrcv_hiwat;
1109 	so->sol_sbsnd_hiwat = sbsnd_hiwat;
1110 	so->sol_sbrcv_flags = sbrcv_flags;
1111 	so->sol_sbsnd_flags = sbsnd_flags;
1112 	so->sol_sbrcv_timeo = sbrcv_timeo;
1113 	so->sol_sbsnd_timeo = sbsnd_timeo;
1114 
1115 	so->sol_qlen = so->sol_incqlen = 0;
1116 	TAILQ_INIT(&so->sol_incomp);
1117 	TAILQ_INIT(&so->sol_comp);
1118 
1119 	so->sol_accept_filter = NULL;
1120 	so->sol_accept_filter_arg = NULL;
1121 	so->sol_accept_filter_str = NULL;
1122 
1123 	so->sol_upcall = NULL;
1124 	so->sol_upcallarg = NULL;
1125 
1126 	so->so_options |= SO_ACCEPTCONN;
1127 
1128 listening:
1129 	if (backlog < 0 || backlog > somaxconn)
1130 		backlog = somaxconn;
1131 	so->sol_qlimit = backlog;
1132 
1133 	mtx_unlock(&so->so_snd_mtx);
1134 	mtx_unlock(&so->so_rcv_mtx);
1135 	sx_xunlock(&so->so_snd_sx);
1136 	sx_xunlock(&so->so_rcv_sx);
1137 }
1138 
1139 /*
1140  * Wakeup listeners/subsystems once we have a complete connection.
1141  * Enters with lock, returns unlocked.
1142  */
1143 void
1144 solisten_wakeup(struct socket *sol)
1145 {
1146 
1147 	if (sol->sol_upcall != NULL)
1148 		(void )sol->sol_upcall(sol, sol->sol_upcallarg, M_NOWAIT);
1149 	else {
1150 		selwakeuppri(&sol->so_rdsel, PSOCK);
1151 		KNOTE_LOCKED(&sol->so_rdsel.si_note, 0);
1152 	}
1153 	SOLISTEN_UNLOCK(sol);
1154 	wakeup_one(&sol->sol_comp);
1155 	if ((sol->so_state & SS_ASYNC) && sol->so_sigio != NULL)
1156 		pgsigio(&sol->so_sigio, SIGIO, 0);
1157 }
1158 
1159 /*
1160  * Return single connection off a listening socket queue.  Main consumer of
1161  * the function is kern_accept4().  Some modules, that do their own accept
1162  * management also use the function.  The socket reference held by the
1163  * listen queue is handed to the caller.
1164  *
1165  * Listening socket must be locked on entry and is returned unlocked on
1166  * return.
1167  * The flags argument is set of accept4(2) flags and ACCEPT4_INHERIT.
1168  */
1169 int
1170 solisten_dequeue(struct socket *head, struct socket **ret, int flags)
1171 {
1172 	struct socket *so;
1173 	int error;
1174 
1175 	SOLISTEN_LOCK_ASSERT(head);
1176 
1177 	while (!(head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp) &&
1178 	    head->so_error == 0) {
1179 		error = msleep(&head->sol_comp, SOCK_MTX(head), PSOCK | PCATCH,
1180 		    "accept", 0);
1181 		if (error != 0) {
1182 			SOLISTEN_UNLOCK(head);
1183 			return (error);
1184 		}
1185 	}
1186 	if (head->so_error) {
1187 		error = head->so_error;
1188 		head->so_error = 0;
1189 	} else if ((head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp))
1190 		error = EWOULDBLOCK;
1191 	else
1192 		error = 0;
1193 	if (error) {
1194 		SOLISTEN_UNLOCK(head);
1195 		return (error);
1196 	}
1197 	so = TAILQ_FIRST(&head->sol_comp);
1198 	SOCK_LOCK(so);
1199 	KASSERT(so->so_qstate == SQ_COMP,
1200 	    ("%s: so %p not SQ_COMP", __func__, so));
1201 	head->sol_qlen--;
1202 	so->so_qstate = SQ_NONE;
1203 	so->so_listen = NULL;
1204 	TAILQ_REMOVE(&head->sol_comp, so, so_list);
1205 	if (flags & ACCEPT4_INHERIT)
1206 		so->so_state |= (head->so_state & SS_NBIO);
1207 	else
1208 		so->so_state |= (flags & SOCK_NONBLOCK) ? SS_NBIO : 0;
1209 	SOCK_UNLOCK(so);
1210 	sorele_locked(head);
1211 
1212 	*ret = so;
1213 	return (0);
1214 }
1215 
1216 /*
1217  * Free socket upon release of the very last reference.
1218  */
1219 static void
1220 sofree(struct socket *so)
1221 {
1222 	struct protosw *pr = so->so_proto;
1223 
1224 	SOCK_LOCK_ASSERT(so);
1225 	KASSERT(refcount_load(&so->so_count) == 0,
1226 	    ("%s: so %p has references", __func__, so));
1227 	KASSERT(SOLISTENING(so) || so->so_qstate == SQ_NONE,
1228 	    ("%s: so %p is on listen queue", __func__, so));
1229 
1230 	SOCK_UNLOCK(so);
1231 
1232 	if (so->so_dtor != NULL)
1233 		so->so_dtor(so);
1234 
1235 	VNET_SO_ASSERT(so);
1236 	if (pr->pr_detach != NULL)
1237 		pr->pr_detach(so);
1238 
1239 	/*
1240 	 * From this point on, we assume that no other references to this
1241 	 * socket exist anywhere else in the stack.  Therefore, no locks need
1242 	 * to be acquired or held.
1243 	 */
1244 	if (!(pr->pr_flags & PR_SOCKBUF) && !SOLISTENING(so)) {
1245 		sbdestroy(so, SO_SND);
1246 		sbdestroy(so, SO_RCV);
1247 	}
1248 	seldrain(&so->so_rdsel);
1249 	seldrain(&so->so_wrsel);
1250 	knlist_destroy(&so->so_rdsel.si_note);
1251 	knlist_destroy(&so->so_wrsel.si_note);
1252 	sodealloc(so);
1253 }
1254 
1255 /*
1256  * Release a reference on a socket while holding the socket lock.
1257  * Unlocks the socket lock before returning.
1258  */
1259 void
1260 sorele_locked(struct socket *so)
1261 {
1262 	SOCK_LOCK_ASSERT(so);
1263 	if (refcount_release(&so->so_count))
1264 		sofree(so);
1265 	else
1266 		SOCK_UNLOCK(so);
1267 }
1268 
1269 /*
1270  * Close a socket on last file table reference removal.  Initiate disconnect
1271  * if connected.  Free socket when disconnect complete.
1272  *
1273  * This function will sorele() the socket.  Note that soclose() may be called
1274  * prior to the ref count reaching zero.  The actual socket structure will
1275  * not be freed until the ref count reaches zero.
1276  */
1277 int
1278 soclose(struct socket *so)
1279 {
1280 	struct accept_queue lqueue;
1281 	int error = 0;
1282 	bool listening, last __diagused;
1283 
1284 	CURVNET_SET(so->so_vnet);
1285 	funsetown(&so->so_sigio);
1286 	if (so->so_state & SS_ISCONNECTED) {
1287 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
1288 			error = sodisconnect(so);
1289 			if (error) {
1290 				if (error == ENOTCONN)
1291 					error = 0;
1292 				goto drop;
1293 			}
1294 		}
1295 
1296 		if ((so->so_options & SO_LINGER) != 0 && so->so_linger != 0) {
1297 			if ((so->so_state & SS_ISDISCONNECTING) &&
1298 			    (so->so_state & SS_NBIO))
1299 				goto drop;
1300 			while (so->so_state & SS_ISCONNECTED) {
1301 				error = tsleep(&so->so_timeo,
1302 				    PSOCK | PCATCH, "soclos",
1303 				    so->so_linger * hz);
1304 				if (error)
1305 					break;
1306 			}
1307 		}
1308 	}
1309 
1310 drop:
1311 	if (so->so_proto->pr_close != NULL)
1312 		so->so_proto->pr_close(so);
1313 
1314 	SOCK_LOCK(so);
1315 	if ((listening = SOLISTENING(so))) {
1316 		struct socket *sp;
1317 
1318 		TAILQ_INIT(&lqueue);
1319 		TAILQ_SWAP(&lqueue, &so->sol_incomp, socket, so_list);
1320 		TAILQ_CONCAT(&lqueue, &so->sol_comp, so_list);
1321 
1322 		so->sol_qlen = so->sol_incqlen = 0;
1323 
1324 		TAILQ_FOREACH(sp, &lqueue, so_list) {
1325 			SOCK_LOCK(sp);
1326 			sp->so_qstate = SQ_NONE;
1327 			sp->so_listen = NULL;
1328 			SOCK_UNLOCK(sp);
1329 			last = refcount_release(&so->so_count);
1330 			KASSERT(!last, ("%s: released last reference for %p",
1331 			    __func__, so));
1332 		}
1333 	}
1334 	sorele_locked(so);
1335 	if (listening) {
1336 		struct socket *sp, *tsp;
1337 
1338 		TAILQ_FOREACH_SAFE(sp, &lqueue, so_list, tsp)
1339 			soabort(sp);
1340 	}
1341 	CURVNET_RESTORE();
1342 	return (error);
1343 }
1344 
1345 /*
1346  * soabort() is used to abruptly tear down a connection, such as when a
1347  * resource limit is reached (listen queue depth exceeded), or if a listen
1348  * socket is closed while there are sockets waiting to be accepted.
1349  *
1350  * This interface is tricky, because it is called on an unreferenced socket,
1351  * and must be called only by a thread that has actually removed the socket
1352  * from the listen queue it was on.  Likely this thread holds the last
1353  * reference on the socket and soabort() will proceed with sofree().  But
1354  * it might be not the last, as the sockets on the listen queues are seen
1355  * from the protocol side.
1356  *
1357  * This interface will call into the protocol code, so must not be called
1358  * with any socket locks held.  Protocols do call it while holding their own
1359  * recursible protocol mutexes, but this is something that should be subject
1360  * to review in the future.
1361  *
1362  * Usually socket should have a single reference left, but this is not a
1363  * requirement.  In the past, when we have had named references for file
1364  * descriptor and protocol, we asserted that none of them are being held.
1365  */
1366 void
1367 soabort(struct socket *so)
1368 {
1369 
1370 	VNET_SO_ASSERT(so);
1371 
1372 	if (so->so_proto->pr_abort != NULL)
1373 		so->so_proto->pr_abort(so);
1374 	SOCK_LOCK(so);
1375 	sorele_locked(so);
1376 }
1377 
1378 int
1379 soaccept(struct socket *so, struct sockaddr *sa)
1380 {
1381 #ifdef INVARIANTS
1382 	u_char len = sa->sa_len;
1383 #endif
1384 	int error;
1385 
1386 	CURVNET_SET(so->so_vnet);
1387 	error = so->so_proto->pr_accept(so, sa);
1388 	KASSERT(sa->sa_len <= len,
1389 	    ("%s: protocol %p sockaddr overflow", __func__, so->so_proto));
1390 	CURVNET_RESTORE();
1391 	return (error);
1392 }
1393 
1394 int
1395 sopeeraddr(struct socket *so, struct sockaddr *sa)
1396 {
1397 #ifdef INVARIANTS
1398 	u_char len = sa->sa_len;
1399 #endif
1400 	int error;
1401 
1402 	CURVNET_SET(so->so_vnet);
1403 	error = so->so_proto->pr_peeraddr(so, sa);
1404 	KASSERT(sa->sa_len <= len,
1405 	    ("%s: protocol %p sockaddr overflow", __func__, so->so_proto));
1406 	CURVNET_RESTORE();
1407 
1408 	return (error);
1409 }
1410 
1411 int
1412 sosockaddr(struct socket *so, struct sockaddr *sa)
1413 {
1414 #ifdef INVARIANTS
1415 	u_char len = sa->sa_len;
1416 #endif
1417 	int error;
1418 
1419 	CURVNET_SET(so->so_vnet);
1420 	error = so->so_proto->pr_sockaddr(so, sa);
1421 	KASSERT(sa->sa_len <= len,
1422 	    ("%s: protocol %p sockaddr overflow", __func__, so->so_proto));
1423 	CURVNET_RESTORE();
1424 
1425 	return (error);
1426 }
1427 
1428 int
1429 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
1430 {
1431 
1432 	return (soconnectat(AT_FDCWD, so, nam, td));
1433 }
1434 
1435 int
1436 soconnectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
1437 {
1438 	int error;
1439 
1440 	CURVNET_SET(so->so_vnet);
1441 
1442 	/*
1443 	 * If protocol is connection-based, can only connect once.
1444 	 * Otherwise, if connected, try to disconnect first.  This allows
1445 	 * user to disconnect by connecting to, e.g., a null address.
1446 	 *
1447 	 * Note, this check is racy and may need to be re-evaluated at the
1448 	 * protocol layer.
1449 	 */
1450 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
1451 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
1452 	    (error = sodisconnect(so)))) {
1453 		error = EISCONN;
1454 	} else {
1455 		/*
1456 		 * Prevent accumulated error from previous connection from
1457 		 * biting us.
1458 		 */
1459 		so->so_error = 0;
1460 		if (fd == AT_FDCWD) {
1461 			error = so->so_proto->pr_connect(so, nam, td);
1462 		} else {
1463 			error = so->so_proto->pr_connectat(fd, so, nam, td);
1464 		}
1465 	}
1466 	CURVNET_RESTORE();
1467 
1468 	return (error);
1469 }
1470 
1471 int
1472 soconnect2(struct socket *so1, struct socket *so2)
1473 {
1474 	int error;
1475 
1476 	CURVNET_SET(so1->so_vnet);
1477 	error = so1->so_proto->pr_connect2(so1, so2);
1478 	CURVNET_RESTORE();
1479 	return (error);
1480 }
1481 
1482 int
1483 sodisconnect(struct socket *so)
1484 {
1485 	int error;
1486 
1487 	if ((so->so_state & SS_ISCONNECTED) == 0)
1488 		return (ENOTCONN);
1489 	if (so->so_state & SS_ISDISCONNECTING)
1490 		return (EALREADY);
1491 	VNET_SO_ASSERT(so);
1492 	error = so->so_proto->pr_disconnect(so);
1493 	return (error);
1494 }
1495 
1496 int
1497 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio,
1498     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1499 {
1500 	long space;
1501 	ssize_t resid;
1502 	int clen = 0, error, dontroute;
1503 
1504 	KASSERT(so->so_type == SOCK_DGRAM, ("sosend_dgram: !SOCK_DGRAM"));
1505 	KASSERT(so->so_proto->pr_flags & PR_ATOMIC,
1506 	    ("sosend_dgram: !PR_ATOMIC"));
1507 
1508 	if (uio != NULL)
1509 		resid = uio->uio_resid;
1510 	else
1511 		resid = top->m_pkthdr.len;
1512 	/*
1513 	 * In theory resid should be unsigned.  However, space must be
1514 	 * signed, as it might be less than 0 if we over-committed, and we
1515 	 * must use a signed comparison of space and resid.  On the other
1516 	 * hand, a negative resid causes us to loop sending 0-length
1517 	 * segments to the protocol.
1518 	 */
1519 	if (resid < 0) {
1520 		error = EINVAL;
1521 		goto out;
1522 	}
1523 
1524 	dontroute =
1525 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0;
1526 	if (td != NULL)
1527 		td->td_ru.ru_msgsnd++;
1528 	if (control != NULL)
1529 		clen = control->m_len;
1530 
1531 	SOCKBUF_LOCK(&so->so_snd);
1532 	if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1533 		SOCKBUF_UNLOCK(&so->so_snd);
1534 		error = EPIPE;
1535 		goto out;
1536 	}
1537 	if (so->so_error) {
1538 		error = so->so_error;
1539 		so->so_error = 0;
1540 		SOCKBUF_UNLOCK(&so->so_snd);
1541 		goto out;
1542 	}
1543 	if ((so->so_state & SS_ISCONNECTED) == 0) {
1544 		/*
1545 		 * `sendto' and `sendmsg' is allowed on a connection-based
1546 		 * socket if it supports implied connect.  Return ENOTCONN if
1547 		 * not connected and no address is supplied.
1548 		 */
1549 		if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1550 		    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1551 			if (!(resid == 0 && clen != 0)) {
1552 				SOCKBUF_UNLOCK(&so->so_snd);
1553 				error = ENOTCONN;
1554 				goto out;
1555 			}
1556 		} else if (addr == NULL) {
1557 			if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1558 				error = ENOTCONN;
1559 			else
1560 				error = EDESTADDRREQ;
1561 			SOCKBUF_UNLOCK(&so->so_snd);
1562 			goto out;
1563 		}
1564 	}
1565 
1566 	/*
1567 	 * Do we need MSG_OOB support in SOCK_DGRAM?  Signs here may be a
1568 	 * problem and need fixing.
1569 	 */
1570 	space = sbspace(&so->so_snd);
1571 	if (flags & MSG_OOB)
1572 		space += 1024;
1573 	space -= clen;
1574 	SOCKBUF_UNLOCK(&so->so_snd);
1575 	if (resid > space) {
1576 		error = EMSGSIZE;
1577 		goto out;
1578 	}
1579 	if (uio == NULL) {
1580 		resid = 0;
1581 		if (flags & MSG_EOR)
1582 			top->m_flags |= M_EOR;
1583 	} else {
1584 		/*
1585 		 * Copy the data from userland into a mbuf chain.
1586 		 * If no data is to be copied in, a single empty mbuf
1587 		 * is returned.
1588 		 */
1589 		top = m_uiotombuf(uio, M_WAITOK, space, max_hdr,
1590 		    (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0)));
1591 		if (top == NULL) {
1592 			error = EFAULT;	/* only possible error */
1593 			goto out;
1594 		}
1595 		space -= resid - uio->uio_resid;
1596 		resid = uio->uio_resid;
1597 	}
1598 	KASSERT(resid == 0, ("sosend_dgram: resid != 0"));
1599 	/*
1600 	 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock
1601 	 * than with.
1602 	 */
1603 	if (dontroute) {
1604 		SOCK_LOCK(so);
1605 		so->so_options |= SO_DONTROUTE;
1606 		SOCK_UNLOCK(so);
1607 	}
1608 	/*
1609 	 * XXX all the SBS_CANTSENDMORE checks previously done could be out
1610 	 * of date.  We could have received a reset packet in an interrupt or
1611 	 * maybe we slept while doing page faults in uiomove() etc.  We could
1612 	 * probably recheck again inside the locking protection here, but
1613 	 * there are probably other places that this also happens.  We must
1614 	 * rethink this.
1615 	 */
1616 	VNET_SO_ASSERT(so);
1617 	error = so->so_proto->pr_send(so, (flags & MSG_OOB) ? PRUS_OOB :
1618 	/*
1619 	 * If the user set MSG_EOF, the protocol understands this flag and
1620 	 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND.
1621 	 */
1622 	    ((flags & MSG_EOF) &&
1623 	     (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1624 	     (resid <= 0)) ?
1625 		PRUS_EOF :
1626 		/* If there is more to send set PRUS_MORETOCOME */
1627 		(flags & MSG_MORETOCOME) ||
1628 		(resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1629 		top, addr, control, td);
1630 	if (dontroute) {
1631 		SOCK_LOCK(so);
1632 		so->so_options &= ~SO_DONTROUTE;
1633 		SOCK_UNLOCK(so);
1634 	}
1635 	clen = 0;
1636 	control = NULL;
1637 	top = NULL;
1638 out:
1639 	if (top != NULL)
1640 		m_freem(top);
1641 	if (control != NULL)
1642 		m_freem(control);
1643 	return (error);
1644 }
1645 
1646 /*
1647  * Send on a socket.  If send must go all at once and message is larger than
1648  * send buffering, then hard error.  Lock against other senders.  If must go
1649  * all at once and not enough room now, then inform user that this would
1650  * block and do nothing.  Otherwise, if nonblocking, send as much as
1651  * possible.  The data to be sent is described by "uio" if nonzero, otherwise
1652  * by the mbuf chain "top" (which must be null if uio is not).  Data provided
1653  * in mbuf chain must be small enough to send all at once.
1654  *
1655  * Returns nonzero on error, timeout or signal; callers must check for short
1656  * counts if EINTR/ERESTART are returned.  Data and control buffers are freed
1657  * on return.
1658  */
1659 static int
1660 sosend_generic_locked(struct socket *so, struct sockaddr *addr, struct uio *uio,
1661     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1662 {
1663 	long space;
1664 	ssize_t resid;
1665 	int clen = 0, error, dontroute;
1666 	int atomic = sosendallatonce(so) || top;
1667 	int pr_send_flag;
1668 #ifdef KERN_TLS
1669 	struct ktls_session *tls;
1670 	int tls_enq_cnt, tls_send_flag;
1671 	uint8_t tls_rtype;
1672 
1673 	tls = NULL;
1674 	tls_rtype = TLS_RLTYPE_APP;
1675 #endif
1676 
1677 	SOCK_IO_SEND_ASSERT_LOCKED(so);
1678 
1679 	if (uio != NULL)
1680 		resid = uio->uio_resid;
1681 	else if ((top->m_flags & M_PKTHDR) != 0)
1682 		resid = top->m_pkthdr.len;
1683 	else
1684 		resid = m_length(top, NULL);
1685 	/*
1686 	 * In theory resid should be unsigned.  However, space must be
1687 	 * signed, as it might be less than 0 if we over-committed, and we
1688 	 * must use a signed comparison of space and resid.  On the other
1689 	 * hand, a negative resid causes us to loop sending 0-length
1690 	 * segments to the protocol.
1691 	 *
1692 	 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
1693 	 * type sockets since that's an error.
1694 	 */
1695 	if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
1696 		error = EINVAL;
1697 		goto out;
1698 	}
1699 
1700 	dontroute =
1701 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
1702 	    (so->so_proto->pr_flags & PR_ATOMIC);
1703 	if (td != NULL)
1704 		td->td_ru.ru_msgsnd++;
1705 	if (control != NULL)
1706 		clen = control->m_len;
1707 
1708 #ifdef KERN_TLS
1709 	tls_send_flag = 0;
1710 	tls = ktls_hold(so->so_snd.sb_tls_info);
1711 	if (tls != NULL) {
1712 		if (tls->mode == TCP_TLS_MODE_SW)
1713 			tls_send_flag = PRUS_NOTREADY;
1714 
1715 		if (control != NULL) {
1716 			struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1717 
1718 			if (clen >= sizeof(*cm) &&
1719 			    cm->cmsg_type == TLS_SET_RECORD_TYPE) {
1720 				tls_rtype = *((uint8_t *)CMSG_DATA(cm));
1721 				clen = 0;
1722 				m_freem(control);
1723 				control = NULL;
1724 				atomic = 1;
1725 			}
1726 		}
1727 
1728 		if (resid == 0 && !ktls_permit_empty_frames(tls)) {
1729 			error = EINVAL;
1730 			goto out;
1731 		}
1732 	}
1733 #endif
1734 
1735 restart:
1736 	do {
1737 		SOCKBUF_LOCK(&so->so_snd);
1738 		if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1739 			SOCKBUF_UNLOCK(&so->so_snd);
1740 			error = EPIPE;
1741 			goto out;
1742 		}
1743 		if (so->so_error) {
1744 			error = so->so_error;
1745 			so->so_error = 0;
1746 			SOCKBUF_UNLOCK(&so->so_snd);
1747 			goto out;
1748 		}
1749 		if ((so->so_state & SS_ISCONNECTED) == 0) {
1750 			/*
1751 			 * `sendto' and `sendmsg' is allowed on a connection-
1752 			 * based socket if it supports implied connect.
1753 			 * Return ENOTCONN if not connected and no address is
1754 			 * supplied.
1755 			 */
1756 			if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1757 			    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1758 				if (!(resid == 0 && clen != 0)) {
1759 					SOCKBUF_UNLOCK(&so->so_snd);
1760 					error = ENOTCONN;
1761 					goto out;
1762 				}
1763 			} else if (addr == NULL) {
1764 				SOCKBUF_UNLOCK(&so->so_snd);
1765 				if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1766 					error = ENOTCONN;
1767 				else
1768 					error = EDESTADDRREQ;
1769 				goto out;
1770 			}
1771 		}
1772 		space = sbspace(&so->so_snd);
1773 		if (flags & MSG_OOB)
1774 			space += 1024;
1775 		if ((atomic && resid > so->so_snd.sb_hiwat) ||
1776 		    clen > so->so_snd.sb_hiwat) {
1777 			SOCKBUF_UNLOCK(&so->so_snd);
1778 			error = EMSGSIZE;
1779 			goto out;
1780 		}
1781 		if (space < resid + clen &&
1782 		    (atomic || space < so->so_snd.sb_lowat || space < clen)) {
1783 			if ((so->so_state & SS_NBIO) ||
1784 			    (flags & (MSG_NBIO | MSG_DONTWAIT)) != 0) {
1785 				SOCKBUF_UNLOCK(&so->so_snd);
1786 				error = EWOULDBLOCK;
1787 				goto out;
1788 			}
1789 			error = sbwait(so, SO_SND);
1790 			SOCKBUF_UNLOCK(&so->so_snd);
1791 			if (error)
1792 				goto out;
1793 			goto restart;
1794 		}
1795 		SOCKBUF_UNLOCK(&so->so_snd);
1796 		space -= clen;
1797 		do {
1798 			if (uio == NULL) {
1799 				resid = 0;
1800 				if (flags & MSG_EOR)
1801 					top->m_flags |= M_EOR;
1802 #ifdef KERN_TLS
1803 				if (tls != NULL) {
1804 					ktls_frame(top, tls, &tls_enq_cnt,
1805 					    tls_rtype);
1806 					tls_rtype = TLS_RLTYPE_APP;
1807 				}
1808 #endif
1809 			} else {
1810 				/*
1811 				 * Copy the data from userland into a mbuf
1812 				 * chain.  If resid is 0, which can happen
1813 				 * only if we have control to send, then
1814 				 * a single empty mbuf is returned.  This
1815 				 * is a workaround to prevent protocol send
1816 				 * methods to panic.
1817 				 */
1818 #ifdef KERN_TLS
1819 				if (tls != NULL) {
1820 					top = m_uiotombuf(uio, M_WAITOK, space,
1821 					    tls->params.max_frame_len,
1822 					    M_EXTPG |
1823 					    ((flags & MSG_EOR) ? M_EOR : 0));
1824 					if (top != NULL) {
1825 						ktls_frame(top, tls,
1826 						    &tls_enq_cnt, tls_rtype);
1827 					}
1828 					tls_rtype = TLS_RLTYPE_APP;
1829 				} else
1830 #endif
1831 					top = m_uiotombuf(uio, M_WAITOK, space,
1832 					    (atomic ? max_hdr : 0),
1833 					    (atomic ? M_PKTHDR : 0) |
1834 					    ((flags & MSG_EOR) ? M_EOR : 0));
1835 				if (top == NULL) {
1836 					error = EFAULT; /* only possible error */
1837 					goto out;
1838 				}
1839 				space -= resid - uio->uio_resid;
1840 				resid = uio->uio_resid;
1841 			}
1842 			if (dontroute) {
1843 				SOCK_LOCK(so);
1844 				so->so_options |= SO_DONTROUTE;
1845 				SOCK_UNLOCK(so);
1846 			}
1847 			/*
1848 			 * XXX all the SBS_CANTSENDMORE checks previously
1849 			 * done could be out of date.  We could have received
1850 			 * a reset packet in an interrupt or maybe we slept
1851 			 * while doing page faults in uiomove() etc.  We
1852 			 * could probably recheck again inside the locking
1853 			 * protection here, but there are probably other
1854 			 * places that this also happens.  We must rethink
1855 			 * this.
1856 			 */
1857 			VNET_SO_ASSERT(so);
1858 
1859 			pr_send_flag = (flags & MSG_OOB) ? PRUS_OOB :
1860 			/*
1861 			 * If the user set MSG_EOF, the protocol understands
1862 			 * this flag and nothing left to send then use
1863 			 * PRU_SEND_EOF instead of PRU_SEND.
1864 			 */
1865 			    ((flags & MSG_EOF) &&
1866 			     (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1867 			     (resid <= 0)) ?
1868 				PRUS_EOF :
1869 			/* If there is more to send set PRUS_MORETOCOME. */
1870 			    (flags & MSG_MORETOCOME) ||
1871 			    (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0;
1872 
1873 #ifdef KERN_TLS
1874 			pr_send_flag |= tls_send_flag;
1875 #endif
1876 
1877 			error = so->so_proto->pr_send(so, pr_send_flag, top,
1878 			    addr, control, td);
1879 
1880 			if (dontroute) {
1881 				SOCK_LOCK(so);
1882 				so->so_options &= ~SO_DONTROUTE;
1883 				SOCK_UNLOCK(so);
1884 			}
1885 
1886 #ifdef KERN_TLS
1887 			if (tls != NULL && tls->mode == TCP_TLS_MODE_SW) {
1888 				if (error != 0) {
1889 					m_freem(top);
1890 					top = NULL;
1891 				} else {
1892 					soref(so);
1893 					ktls_enqueue(top, so, tls_enq_cnt);
1894 				}
1895 			}
1896 #endif
1897 			clen = 0;
1898 			control = NULL;
1899 			top = NULL;
1900 			if (error)
1901 				goto out;
1902 		} while (resid && space > 0);
1903 	} while (resid);
1904 
1905 out:
1906 #ifdef KERN_TLS
1907 	if (tls != NULL)
1908 		ktls_free(tls);
1909 #endif
1910 	if (top != NULL)
1911 		m_freem(top);
1912 	if (control != NULL)
1913 		m_freem(control);
1914 	return (error);
1915 }
1916 
1917 int
1918 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio,
1919     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1920 {
1921 	int error;
1922 
1923 	error = SOCK_IO_SEND_LOCK(so, SBLOCKWAIT(flags));
1924 	if (error)
1925 		return (error);
1926 	error = sosend_generic_locked(so, addr, uio, top, control, flags, td);
1927 	SOCK_IO_SEND_UNLOCK(so);
1928 	return (error);
1929 }
1930 
1931 /*
1932  * Send to a socket from a kernel thread.
1933  *
1934  * XXXGL: in almost all cases uio is NULL and the mbuf is supplied.
1935  * Exception is nfs/bootp_subr.c.  It is arguable that the VNET context needs
1936  * to be set at all.  This function should just boil down to a static inline
1937  * calling the protocol method.
1938  */
1939 int
1940 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
1941     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1942 {
1943 	int error;
1944 
1945 	CURVNET_SET(so->so_vnet);
1946 	error = so->so_proto->pr_sosend(so, addr, uio,
1947 	    top, control, flags, td);
1948 	CURVNET_RESTORE();
1949 	return (error);
1950 }
1951 
1952 /*
1953  * send(2), write(2) or aio_write(2) on a socket.
1954  */
1955 int
1956 sousrsend(struct socket *so, struct sockaddr *addr, struct uio *uio,
1957     struct mbuf *control, int flags, struct proc *userproc)
1958 {
1959 	struct thread *td;
1960 	ssize_t len;
1961 	int error;
1962 
1963 	td = uio->uio_td;
1964 	len = uio->uio_resid;
1965 	CURVNET_SET(so->so_vnet);
1966 	error = so->so_proto->pr_sosend(so, addr, uio, NULL, control, flags,
1967 	    td);
1968 	CURVNET_RESTORE();
1969 	if (error != 0) {
1970 		/*
1971 		 * Clear transient errors for stream protocols if they made
1972 		 * some progress.  Make exclusion for aio(4) that would
1973 		 * schedule a new write in case of EWOULDBLOCK and clear
1974 		 * error itself.  See soaio_process_job().
1975 		 */
1976 		if (uio->uio_resid != len &&
1977 		    (so->so_proto->pr_flags & PR_ATOMIC) == 0 &&
1978 		    userproc == NULL &&
1979 		    (error == ERESTART || error == EINTR ||
1980 		    error == EWOULDBLOCK))
1981 			error = 0;
1982 		/* Generation of SIGPIPE can be controlled per socket. */
1983 		if (error == EPIPE && (so->so_options & SO_NOSIGPIPE) == 0 &&
1984 		    (flags & MSG_NOSIGNAL) == 0) {
1985 			if (userproc != NULL) {
1986 				/* aio(4) job */
1987 				PROC_LOCK(userproc);
1988 				kern_psignal(userproc, SIGPIPE);
1989 				PROC_UNLOCK(userproc);
1990 			} else {
1991 				PROC_LOCK(td->td_proc);
1992 				tdsignal(td, SIGPIPE);
1993 				PROC_UNLOCK(td->td_proc);
1994 			}
1995 		}
1996 	}
1997 	return (error);
1998 }
1999 
2000 /*
2001  * The part of soreceive() that implements reading non-inline out-of-band
2002  * data from a socket.  For more complete comments, see soreceive(), from
2003  * which this code originated.
2004  *
2005  * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
2006  * unable to return an mbuf chain to the caller.
2007  */
2008 static int
2009 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags)
2010 {
2011 	struct protosw *pr = so->so_proto;
2012 	struct mbuf *m;
2013 	int error;
2014 
2015 	KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
2016 	VNET_SO_ASSERT(so);
2017 
2018 	m = m_get(M_WAITOK, MT_DATA);
2019 	error = pr->pr_rcvoob(so, m, flags & MSG_PEEK);
2020 	if (error)
2021 		goto bad;
2022 	do {
2023 		error = uiomove(mtod(m, void *),
2024 		    (int) min(uio->uio_resid, m->m_len), uio);
2025 		m = m_free(m);
2026 	} while (uio->uio_resid && error == 0 && m);
2027 bad:
2028 	if (m != NULL)
2029 		m_freem(m);
2030 	return (error);
2031 }
2032 
2033 /*
2034  * Following replacement or removal of the first mbuf on the first mbuf chain
2035  * of a socket buffer, push necessary state changes back into the socket
2036  * buffer so that other consumers see the values consistently.  'nextrecord'
2037  * is the callers locally stored value of the original value of
2038  * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes.
2039  * NOTE: 'nextrecord' may be NULL.
2040  */
2041 static __inline void
2042 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord)
2043 {
2044 
2045 	SOCKBUF_LOCK_ASSERT(sb);
2046 	/*
2047 	 * First, update for the new value of nextrecord.  If necessary, make
2048 	 * it the first record.
2049 	 */
2050 	if (sb->sb_mb != NULL)
2051 		sb->sb_mb->m_nextpkt = nextrecord;
2052 	else
2053 		sb->sb_mb = nextrecord;
2054 
2055 	/*
2056 	 * Now update any dependent socket buffer fields to reflect the new
2057 	 * state.  This is an expanded inline of SB_EMPTY_FIXUP(), with the
2058 	 * addition of a second clause that takes care of the case where
2059 	 * sb_mb has been updated, but remains the last record.
2060 	 */
2061 	if (sb->sb_mb == NULL) {
2062 		sb->sb_mbtail = NULL;
2063 		sb->sb_lastrecord = NULL;
2064 	} else if (sb->sb_mb->m_nextpkt == NULL)
2065 		sb->sb_lastrecord = sb->sb_mb;
2066 }
2067 
2068 /*
2069  * Implement receive operations on a socket.  We depend on the way that
2070  * records are added to the sockbuf by sbappend.  In particular, each record
2071  * (mbufs linked through m_next) must begin with an address if the protocol
2072  * so specifies, followed by an optional mbuf or mbufs containing ancillary
2073  * data, and then zero or more mbufs of data.  In order to allow parallelism
2074  * between network receive and copying to user space, as well as avoid
2075  * sleeping with a mutex held, we release the socket buffer mutex during the
2076  * user space copy.  Although the sockbuf is locked, new data may still be
2077  * appended, and thus we must maintain consistency of the sockbuf during that
2078  * time.
2079  *
2080  * The caller may receive the data as a single mbuf chain by supplying an
2081  * mbuf **mp0 for use in returning the chain.  The uio is then used only for
2082  * the count in uio_resid.
2083  */
2084 static int
2085 soreceive_generic_locked(struct socket *so, struct sockaddr **psa,
2086     struct uio *uio, struct mbuf **mp, struct mbuf **controlp, int *flagsp)
2087 {
2088 	struct mbuf *m;
2089 	int flags, error, offset;
2090 	ssize_t len;
2091 	struct protosw *pr = so->so_proto;
2092 	struct mbuf *nextrecord;
2093 	int moff, type = 0;
2094 	ssize_t orig_resid = uio->uio_resid;
2095 	bool report_real_len = false;
2096 
2097 	SOCK_IO_RECV_ASSERT_LOCKED(so);
2098 
2099 	error = 0;
2100 	if (flagsp != NULL) {
2101 		report_real_len = *flagsp & MSG_TRUNC;
2102 		*flagsp &= ~MSG_TRUNC;
2103 		flags = *flagsp &~ MSG_EOR;
2104 	} else
2105 		flags = 0;
2106 
2107 restart:
2108 	SOCKBUF_LOCK(&so->so_rcv);
2109 	m = so->so_rcv.sb_mb;
2110 	/*
2111 	 * If we have less data than requested, block awaiting more (subject
2112 	 * to any timeout) if:
2113 	 *   1. the current count is less than the low water mark, or
2114 	 *   2. MSG_DONTWAIT is not set
2115 	 */
2116 	if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
2117 	    sbavail(&so->so_rcv) < uio->uio_resid) &&
2118 	    sbavail(&so->so_rcv) < so->so_rcv.sb_lowat &&
2119 	    m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
2120 		KASSERT(m != NULL || !sbavail(&so->so_rcv),
2121 		    ("receive: m == %p sbavail == %u",
2122 		    m, sbavail(&so->so_rcv)));
2123 		if (so->so_error || so->so_rerror) {
2124 			if (m != NULL)
2125 				goto dontblock;
2126 			if (so->so_error)
2127 				error = so->so_error;
2128 			else
2129 				error = so->so_rerror;
2130 			if ((flags & MSG_PEEK) == 0) {
2131 				if (so->so_error)
2132 					so->so_error = 0;
2133 				else
2134 					so->so_rerror = 0;
2135 			}
2136 			SOCKBUF_UNLOCK(&so->so_rcv);
2137 			goto release;
2138 		}
2139 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2140 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2141 			if (m != NULL)
2142 				goto dontblock;
2143 #ifdef KERN_TLS
2144 			else if (so->so_rcv.sb_tlsdcc == 0 &&
2145 			    so->so_rcv.sb_tlscc == 0) {
2146 #else
2147 			else {
2148 #endif
2149 				SOCKBUF_UNLOCK(&so->so_rcv);
2150 				goto release;
2151 			}
2152 		}
2153 		for (; m != NULL; m = m->m_next)
2154 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
2155 				m = so->so_rcv.sb_mb;
2156 				goto dontblock;
2157 			}
2158 		if ((so->so_state & (SS_ISCONNECTING | SS_ISCONNECTED |
2159 		    SS_ISDISCONNECTING | SS_ISDISCONNECTED)) == 0 &&
2160 		    (so->so_proto->pr_flags & PR_CONNREQUIRED) != 0) {
2161 			SOCKBUF_UNLOCK(&so->so_rcv);
2162 			error = ENOTCONN;
2163 			goto release;
2164 		}
2165 		if (uio->uio_resid == 0 && !report_real_len) {
2166 			SOCKBUF_UNLOCK(&so->so_rcv);
2167 			goto release;
2168 		}
2169 		if ((so->so_state & SS_NBIO) ||
2170 		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2171 			SOCKBUF_UNLOCK(&so->so_rcv);
2172 			error = EWOULDBLOCK;
2173 			goto release;
2174 		}
2175 		SBLASTRECORDCHK(&so->so_rcv);
2176 		SBLASTMBUFCHK(&so->so_rcv);
2177 		error = sbwait(so, SO_RCV);
2178 		SOCKBUF_UNLOCK(&so->so_rcv);
2179 		if (error)
2180 			goto release;
2181 		goto restart;
2182 	}
2183 dontblock:
2184 	/*
2185 	 * From this point onward, we maintain 'nextrecord' as a cache of the
2186 	 * pointer to the next record in the socket buffer.  We must keep the
2187 	 * various socket buffer pointers and local stack versions of the
2188 	 * pointers in sync, pushing out modifications before dropping the
2189 	 * socket buffer mutex, and re-reading them when picking it up.
2190 	 *
2191 	 * Otherwise, we will race with the network stack appending new data
2192 	 * or records onto the socket buffer by using inconsistent/stale
2193 	 * versions of the field, possibly resulting in socket buffer
2194 	 * corruption.
2195 	 *
2196 	 * By holding the high-level sblock(), we prevent simultaneous
2197 	 * readers from pulling off the front of the socket buffer.
2198 	 */
2199 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2200 	if (uio->uio_td)
2201 		uio->uio_td->td_ru.ru_msgrcv++;
2202 	KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb"));
2203 	SBLASTRECORDCHK(&so->so_rcv);
2204 	SBLASTMBUFCHK(&so->so_rcv);
2205 	nextrecord = m->m_nextpkt;
2206 	if (pr->pr_flags & PR_ADDR) {
2207 		KASSERT(m->m_type == MT_SONAME,
2208 		    ("m->m_type == %d", m->m_type));
2209 		orig_resid = 0;
2210 		if (psa != NULL)
2211 			*psa = sodupsockaddr(mtod(m, struct sockaddr *),
2212 			    M_NOWAIT);
2213 		if (flags & MSG_PEEK) {
2214 			m = m->m_next;
2215 		} else {
2216 			sbfree(&so->so_rcv, m);
2217 			so->so_rcv.sb_mb = m_free(m);
2218 			m = so->so_rcv.sb_mb;
2219 			sockbuf_pushsync(&so->so_rcv, nextrecord);
2220 		}
2221 	}
2222 
2223 	/*
2224 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
2225 	 * in the first mbuf chain on the socket buffer.  If MSG_PEEK, we
2226 	 * just copy the data; if !MSG_PEEK, we call into the protocol to
2227 	 * perform externalization (or freeing if controlp == NULL).
2228 	 */
2229 	if (m != NULL && m->m_type == MT_CONTROL) {
2230 		struct mbuf *cm = NULL, *cmn;
2231 		struct mbuf **cme = &cm;
2232 #ifdef KERN_TLS
2233 		struct cmsghdr *cmsg;
2234 		struct tls_get_record tgr;
2235 
2236 		/*
2237 		 * For MSG_TLSAPPDATA, check for an alert record.
2238 		 * If found, return ENXIO without removing
2239 		 * it from the receive queue.  This allows a subsequent
2240 		 * call without MSG_TLSAPPDATA to receive it.
2241 		 * Note that, for TLS, there should only be a single
2242 		 * control mbuf with the TLS_GET_RECORD message in it.
2243 		 */
2244 		if (flags & MSG_TLSAPPDATA) {
2245 			cmsg = mtod(m, struct cmsghdr *);
2246 			if (cmsg->cmsg_type == TLS_GET_RECORD &&
2247 			    cmsg->cmsg_len == CMSG_LEN(sizeof(tgr))) {
2248 				memcpy(&tgr, CMSG_DATA(cmsg), sizeof(tgr));
2249 				if (__predict_false(tgr.tls_type ==
2250 				    TLS_RLTYPE_ALERT)) {
2251 					SOCKBUF_UNLOCK(&so->so_rcv);
2252 					error = ENXIO;
2253 					goto release;
2254 				}
2255 			}
2256 		}
2257 #endif
2258 
2259 		do {
2260 			if (flags & MSG_PEEK) {
2261 				if (controlp != NULL) {
2262 					*controlp = m_copym(m, 0, m->m_len,
2263 					    M_NOWAIT);
2264 					controlp = &(*controlp)->m_next;
2265 				}
2266 				m = m->m_next;
2267 			} else {
2268 				sbfree(&so->so_rcv, m);
2269 				so->so_rcv.sb_mb = m->m_next;
2270 				m->m_next = NULL;
2271 				*cme = m;
2272 				cme = &(*cme)->m_next;
2273 				m = so->so_rcv.sb_mb;
2274 			}
2275 		} while (m != NULL && m->m_type == MT_CONTROL);
2276 		if ((flags & MSG_PEEK) == 0)
2277 			sockbuf_pushsync(&so->so_rcv, nextrecord);
2278 		while (cm != NULL) {
2279 			cmn = cm->m_next;
2280 			cm->m_next = NULL;
2281 			if (pr->pr_domain->dom_externalize != NULL) {
2282 				SOCKBUF_UNLOCK(&so->so_rcv);
2283 				VNET_SO_ASSERT(so);
2284 				error = (*pr->pr_domain->dom_externalize)
2285 				    (cm, controlp, flags);
2286 				SOCKBUF_LOCK(&so->so_rcv);
2287 			} else if (controlp != NULL)
2288 				*controlp = cm;
2289 			else
2290 				m_freem(cm);
2291 			if (controlp != NULL) {
2292 				while (*controlp != NULL)
2293 					controlp = &(*controlp)->m_next;
2294 			}
2295 			cm = cmn;
2296 		}
2297 		if (m != NULL)
2298 			nextrecord = so->so_rcv.sb_mb->m_nextpkt;
2299 		else
2300 			nextrecord = so->so_rcv.sb_mb;
2301 		orig_resid = 0;
2302 	}
2303 	if (m != NULL) {
2304 		if ((flags & MSG_PEEK) == 0) {
2305 			KASSERT(m->m_nextpkt == nextrecord,
2306 			    ("soreceive: post-control, nextrecord !sync"));
2307 			if (nextrecord == NULL) {
2308 				KASSERT(so->so_rcv.sb_mb == m,
2309 				    ("soreceive: post-control, sb_mb!=m"));
2310 				KASSERT(so->so_rcv.sb_lastrecord == m,
2311 				    ("soreceive: post-control, lastrecord!=m"));
2312 			}
2313 		}
2314 		type = m->m_type;
2315 		if (type == MT_OOBDATA)
2316 			flags |= MSG_OOB;
2317 	} else {
2318 		if ((flags & MSG_PEEK) == 0) {
2319 			KASSERT(so->so_rcv.sb_mb == nextrecord,
2320 			    ("soreceive: sb_mb != nextrecord"));
2321 			if (so->so_rcv.sb_mb == NULL) {
2322 				KASSERT(so->so_rcv.sb_lastrecord == NULL,
2323 				    ("soreceive: sb_lastercord != NULL"));
2324 			}
2325 		}
2326 	}
2327 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2328 	SBLASTRECORDCHK(&so->so_rcv);
2329 	SBLASTMBUFCHK(&so->so_rcv);
2330 
2331 	/*
2332 	 * Now continue to read any data mbufs off of the head of the socket
2333 	 * buffer until the read request is satisfied.  Note that 'type' is
2334 	 * used to store the type of any mbuf reads that have happened so far
2335 	 * such that soreceive() can stop reading if the type changes, which
2336 	 * causes soreceive() to return only one of regular data and inline
2337 	 * out-of-band data in a single socket receive operation.
2338 	 */
2339 	moff = 0;
2340 	offset = 0;
2341 	while (m != NULL && !(m->m_flags & M_NOTAVAIL) && uio->uio_resid > 0
2342 	    && error == 0) {
2343 		/*
2344 		 * If the type of mbuf has changed since the last mbuf
2345 		 * examined ('type'), end the receive operation.
2346 		 */
2347 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2348 		if (m->m_type == MT_OOBDATA || m->m_type == MT_CONTROL) {
2349 			if (type != m->m_type)
2350 				break;
2351 		} else if (type == MT_OOBDATA)
2352 			break;
2353 		else
2354 		    KASSERT(m->m_type == MT_DATA,
2355 			("m->m_type == %d", m->m_type));
2356 		so->so_rcv.sb_state &= ~SBS_RCVATMARK;
2357 		len = uio->uio_resid;
2358 		if (so->so_oobmark && len > so->so_oobmark - offset)
2359 			len = so->so_oobmark - offset;
2360 		if (len > m->m_len - moff)
2361 			len = m->m_len - moff;
2362 		/*
2363 		 * If mp is set, just pass back the mbufs.  Otherwise copy
2364 		 * them out via the uio, then free.  Sockbuf must be
2365 		 * consistent here (points to current mbuf, it points to next
2366 		 * record) when we drop priority; we must note any additions
2367 		 * to the sockbuf when we block interrupts again.
2368 		 */
2369 		if (mp == NULL) {
2370 			SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2371 			SBLASTRECORDCHK(&so->so_rcv);
2372 			SBLASTMBUFCHK(&so->so_rcv);
2373 			SOCKBUF_UNLOCK(&so->so_rcv);
2374 			if ((m->m_flags & M_EXTPG) != 0)
2375 				error = m_unmapped_uiomove(m, moff, uio,
2376 				    (int)len);
2377 			else
2378 				error = uiomove(mtod(m, char *) + moff,
2379 				    (int)len, uio);
2380 			SOCKBUF_LOCK(&so->so_rcv);
2381 			if (error) {
2382 				/*
2383 				 * The MT_SONAME mbuf has already been removed
2384 				 * from the record, so it is necessary to
2385 				 * remove the data mbufs, if any, to preserve
2386 				 * the invariant in the case of PR_ADDR that
2387 				 * requires MT_SONAME mbufs at the head of
2388 				 * each record.
2389 				 */
2390 				if (pr->pr_flags & PR_ATOMIC &&
2391 				    ((flags & MSG_PEEK) == 0))
2392 					(void)sbdroprecord_locked(&so->so_rcv);
2393 				SOCKBUF_UNLOCK(&so->so_rcv);
2394 				goto release;
2395 			}
2396 		} else
2397 			uio->uio_resid -= len;
2398 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2399 		if (len == m->m_len - moff) {
2400 			if (m->m_flags & M_EOR)
2401 				flags |= MSG_EOR;
2402 			if (flags & MSG_PEEK) {
2403 				m = m->m_next;
2404 				moff = 0;
2405 			} else {
2406 				nextrecord = m->m_nextpkt;
2407 				sbfree(&so->so_rcv, m);
2408 				if (mp != NULL) {
2409 					m->m_nextpkt = NULL;
2410 					*mp = m;
2411 					mp = &m->m_next;
2412 					so->so_rcv.sb_mb = m = m->m_next;
2413 					*mp = NULL;
2414 				} else {
2415 					so->so_rcv.sb_mb = m_free(m);
2416 					m = so->so_rcv.sb_mb;
2417 				}
2418 				sockbuf_pushsync(&so->so_rcv, nextrecord);
2419 				SBLASTRECORDCHK(&so->so_rcv);
2420 				SBLASTMBUFCHK(&so->so_rcv);
2421 			}
2422 		} else {
2423 			if (flags & MSG_PEEK)
2424 				moff += len;
2425 			else {
2426 				if (mp != NULL) {
2427 					if (flags & MSG_DONTWAIT) {
2428 						*mp = m_copym(m, 0, len,
2429 						    M_NOWAIT);
2430 						if (*mp == NULL) {
2431 							/*
2432 							 * m_copym() couldn't
2433 							 * allocate an mbuf.
2434 							 * Adjust uio_resid back
2435 							 * (it was adjusted
2436 							 * down by len bytes,
2437 							 * which we didn't end
2438 							 * up "copying" over).
2439 							 */
2440 							uio->uio_resid += len;
2441 							break;
2442 						}
2443 					} else {
2444 						SOCKBUF_UNLOCK(&so->so_rcv);
2445 						*mp = m_copym(m, 0, len,
2446 						    M_WAITOK);
2447 						SOCKBUF_LOCK(&so->so_rcv);
2448 					}
2449 				}
2450 				sbcut_locked(&so->so_rcv, len);
2451 			}
2452 		}
2453 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2454 		if (so->so_oobmark) {
2455 			if ((flags & MSG_PEEK) == 0) {
2456 				so->so_oobmark -= len;
2457 				if (so->so_oobmark == 0) {
2458 					so->so_rcv.sb_state |= SBS_RCVATMARK;
2459 					break;
2460 				}
2461 			} else {
2462 				offset += len;
2463 				if (offset == so->so_oobmark)
2464 					break;
2465 			}
2466 		}
2467 		if (flags & MSG_EOR)
2468 			break;
2469 		/*
2470 		 * If the MSG_WAITALL flag is set (for non-atomic socket), we
2471 		 * must not quit until "uio->uio_resid == 0" or an error
2472 		 * termination.  If a signal/timeout occurs, return with a
2473 		 * short count but without error.  Keep sockbuf locked
2474 		 * against other readers.
2475 		 */
2476 		while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
2477 		    !sosendallatonce(so) && nextrecord == NULL) {
2478 			SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2479 			if (so->so_error || so->so_rerror ||
2480 			    so->so_rcv.sb_state & SBS_CANTRCVMORE)
2481 				break;
2482 			/*
2483 			 * Notify the protocol that some data has been
2484 			 * drained before blocking.
2485 			 */
2486 			if (pr->pr_flags & PR_WANTRCVD) {
2487 				SOCKBUF_UNLOCK(&so->so_rcv);
2488 				VNET_SO_ASSERT(so);
2489 				pr->pr_rcvd(so, flags);
2490 				SOCKBUF_LOCK(&so->so_rcv);
2491 				if (__predict_false(so->so_rcv.sb_mb == NULL &&
2492 				    (so->so_error || so->so_rerror ||
2493 				    so->so_rcv.sb_state & SBS_CANTRCVMORE)))
2494 					break;
2495 			}
2496 			SBLASTRECORDCHK(&so->so_rcv);
2497 			SBLASTMBUFCHK(&so->so_rcv);
2498 			/*
2499 			 * We could receive some data while was notifying
2500 			 * the protocol. Skip blocking in this case.
2501 			 */
2502 			if (so->so_rcv.sb_mb == NULL) {
2503 				error = sbwait(so, SO_RCV);
2504 				if (error) {
2505 					SOCKBUF_UNLOCK(&so->so_rcv);
2506 					goto release;
2507 				}
2508 			}
2509 			m = so->so_rcv.sb_mb;
2510 			if (m != NULL)
2511 				nextrecord = m->m_nextpkt;
2512 		}
2513 	}
2514 
2515 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2516 	if (m != NULL && pr->pr_flags & PR_ATOMIC) {
2517 		if (report_real_len)
2518 			uio->uio_resid -= m_length(m, NULL) - moff;
2519 		flags |= MSG_TRUNC;
2520 		if ((flags & MSG_PEEK) == 0)
2521 			(void) sbdroprecord_locked(&so->so_rcv);
2522 	}
2523 	if ((flags & MSG_PEEK) == 0) {
2524 		if (m == NULL) {
2525 			/*
2526 			 * First part is an inline SB_EMPTY_FIXUP().  Second
2527 			 * part makes sure sb_lastrecord is up-to-date if
2528 			 * there is still data in the socket buffer.
2529 			 */
2530 			so->so_rcv.sb_mb = nextrecord;
2531 			if (so->so_rcv.sb_mb == NULL) {
2532 				so->so_rcv.sb_mbtail = NULL;
2533 				so->so_rcv.sb_lastrecord = NULL;
2534 			} else if (nextrecord->m_nextpkt == NULL)
2535 				so->so_rcv.sb_lastrecord = nextrecord;
2536 		}
2537 		SBLASTRECORDCHK(&so->so_rcv);
2538 		SBLASTMBUFCHK(&so->so_rcv);
2539 		/*
2540 		 * If soreceive() is being done from the socket callback,
2541 		 * then don't need to generate ACK to peer to update window,
2542 		 * since ACK will be generated on return to TCP.
2543 		 */
2544 		if (!(flags & MSG_SOCALLBCK) &&
2545 		    (pr->pr_flags & PR_WANTRCVD)) {
2546 			SOCKBUF_UNLOCK(&so->so_rcv);
2547 			VNET_SO_ASSERT(so);
2548 			pr->pr_rcvd(so, flags);
2549 			SOCKBUF_LOCK(&so->so_rcv);
2550 		}
2551 	}
2552 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2553 	if (orig_resid == uio->uio_resid && orig_resid &&
2554 	    (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
2555 		SOCKBUF_UNLOCK(&so->so_rcv);
2556 		goto restart;
2557 	}
2558 	SOCKBUF_UNLOCK(&so->so_rcv);
2559 
2560 	if (flagsp != NULL)
2561 		*flagsp |= flags;
2562 release:
2563 	return (error);
2564 }
2565 
2566 int
2567 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio,
2568     struct mbuf **mp, struct mbuf **controlp, int *flagsp)
2569 {
2570 	int error, flags;
2571 
2572 	if (psa != NULL)
2573 		*psa = NULL;
2574 	if (controlp != NULL)
2575 		*controlp = NULL;
2576 	if (flagsp != NULL) {
2577 		flags = *flagsp;
2578 		if ((flags & MSG_OOB) != 0)
2579 			return (soreceive_rcvoob(so, uio, flags));
2580 	} else {
2581 		flags = 0;
2582 	}
2583 	if (mp != NULL)
2584 		*mp = NULL;
2585 
2586 	error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags));
2587 	if (error)
2588 		return (error);
2589 	error = soreceive_generic_locked(so, psa, uio, mp, controlp, flagsp);
2590 	SOCK_IO_RECV_UNLOCK(so);
2591 	return (error);
2592 }
2593 
2594 /*
2595  * Optimized version of soreceive() for stream (TCP) sockets.
2596  */
2597 static int
2598 soreceive_stream_locked(struct socket *so, struct sockbuf *sb,
2599     struct sockaddr **psa, struct uio *uio, struct mbuf **mp0,
2600     struct mbuf **controlp, int flags)
2601 {
2602 	int len = 0, error = 0, oresid;
2603 	struct mbuf *m, *n = NULL;
2604 
2605 	SOCK_IO_RECV_ASSERT_LOCKED(so);
2606 
2607 	/* Easy one, no space to copyout anything. */
2608 	if (uio->uio_resid == 0)
2609 		return (EINVAL);
2610 	oresid = uio->uio_resid;
2611 
2612 	SOCKBUF_LOCK(sb);
2613 	/* We will never ever get anything unless we are or were connected. */
2614 	if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) {
2615 		error = ENOTCONN;
2616 		goto out;
2617 	}
2618 
2619 restart:
2620 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2621 
2622 	/* Abort if socket has reported problems. */
2623 	if (so->so_error) {
2624 		if (sbavail(sb) > 0)
2625 			goto deliver;
2626 		if (oresid > uio->uio_resid)
2627 			goto out;
2628 		error = so->so_error;
2629 		if (!(flags & MSG_PEEK))
2630 			so->so_error = 0;
2631 		goto out;
2632 	}
2633 
2634 	/* Door is closed.  Deliver what is left, if any. */
2635 	if (sb->sb_state & SBS_CANTRCVMORE) {
2636 		if (sbavail(sb) > 0)
2637 			goto deliver;
2638 		else
2639 			goto out;
2640 	}
2641 
2642 	/* Socket buffer is empty and we shall not block. */
2643 	if (sbavail(sb) == 0 &&
2644 	    ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) {
2645 		error = EAGAIN;
2646 		goto out;
2647 	}
2648 
2649 	/* Socket buffer got some data that we shall deliver now. */
2650 	if (sbavail(sb) > 0 && !(flags & MSG_WAITALL) &&
2651 	    ((so->so_state & SS_NBIO) ||
2652 	     (flags & (MSG_DONTWAIT|MSG_NBIO)) ||
2653 	     sbavail(sb) >= sb->sb_lowat ||
2654 	     sbavail(sb) >= uio->uio_resid ||
2655 	     sbavail(sb) >= sb->sb_hiwat) ) {
2656 		goto deliver;
2657 	}
2658 
2659 	/* On MSG_WAITALL we must wait until all data or error arrives. */
2660 	if ((flags & MSG_WAITALL) &&
2661 	    (sbavail(sb) >= uio->uio_resid || sbavail(sb) >= sb->sb_hiwat))
2662 		goto deliver;
2663 
2664 	/*
2665 	 * Wait and block until (more) data comes in.
2666 	 * NB: Drops the sockbuf lock during wait.
2667 	 */
2668 	error = sbwait(so, SO_RCV);
2669 	if (error)
2670 		goto out;
2671 	goto restart;
2672 
2673 deliver:
2674 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2675 	KASSERT(sbavail(sb) > 0, ("%s: sockbuf empty", __func__));
2676 	KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__));
2677 
2678 	/* Statistics. */
2679 	if (uio->uio_td)
2680 		uio->uio_td->td_ru.ru_msgrcv++;
2681 
2682 	/* Fill uio until full or current end of socket buffer is reached. */
2683 	len = min(uio->uio_resid, sbavail(sb));
2684 	if (mp0 != NULL) {
2685 		/* Dequeue as many mbufs as possible. */
2686 		if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) {
2687 			if (*mp0 == NULL)
2688 				*mp0 = sb->sb_mb;
2689 			else
2690 				m_cat(*mp0, sb->sb_mb);
2691 			for (m = sb->sb_mb;
2692 			     m != NULL && m->m_len <= len;
2693 			     m = m->m_next) {
2694 				KASSERT(!(m->m_flags & M_NOTAVAIL),
2695 				    ("%s: m %p not available", __func__, m));
2696 				len -= m->m_len;
2697 				uio->uio_resid -= m->m_len;
2698 				sbfree(sb, m);
2699 				n = m;
2700 			}
2701 			n->m_next = NULL;
2702 			sb->sb_mb = m;
2703 			sb->sb_lastrecord = sb->sb_mb;
2704 			if (sb->sb_mb == NULL)
2705 				SB_EMPTY_FIXUP(sb);
2706 		}
2707 		/* Copy the remainder. */
2708 		if (len > 0) {
2709 			KASSERT(sb->sb_mb != NULL,
2710 			    ("%s: len > 0 && sb->sb_mb empty", __func__));
2711 
2712 			m = m_copym(sb->sb_mb, 0, len, M_NOWAIT);
2713 			if (m == NULL)
2714 				len = 0;	/* Don't flush data from sockbuf. */
2715 			else
2716 				uio->uio_resid -= len;
2717 			if (*mp0 != NULL)
2718 				m_cat(*mp0, m);
2719 			else
2720 				*mp0 = m;
2721 			if (*mp0 == NULL) {
2722 				error = ENOBUFS;
2723 				goto out;
2724 			}
2725 		}
2726 	} else {
2727 		/* NB: Must unlock socket buffer as uiomove may sleep. */
2728 		SOCKBUF_UNLOCK(sb);
2729 		error = m_mbuftouio(uio, sb->sb_mb, len);
2730 		SOCKBUF_LOCK(sb);
2731 		if (error)
2732 			goto out;
2733 	}
2734 	SBLASTRECORDCHK(sb);
2735 	SBLASTMBUFCHK(sb);
2736 
2737 	/*
2738 	 * Remove the delivered data from the socket buffer unless we
2739 	 * were only peeking.
2740 	 */
2741 	if (!(flags & MSG_PEEK)) {
2742 		if (len > 0)
2743 			sbdrop_locked(sb, len);
2744 
2745 		/* Notify protocol that we drained some data. */
2746 		if ((so->so_proto->pr_flags & PR_WANTRCVD) &&
2747 		    (((flags & MSG_WAITALL) && uio->uio_resid > 0) ||
2748 		     !(flags & MSG_SOCALLBCK))) {
2749 			SOCKBUF_UNLOCK(sb);
2750 			VNET_SO_ASSERT(so);
2751 			so->so_proto->pr_rcvd(so, flags);
2752 			SOCKBUF_LOCK(sb);
2753 		}
2754 	}
2755 
2756 	/*
2757 	 * For MSG_WAITALL we may have to loop again and wait for
2758 	 * more data to come in.
2759 	 */
2760 	if ((flags & MSG_WAITALL) && uio->uio_resid > 0)
2761 		goto restart;
2762 out:
2763 	SBLASTRECORDCHK(sb);
2764 	SBLASTMBUFCHK(sb);
2765 	SOCKBUF_UNLOCK(sb);
2766 	return (error);
2767 }
2768 
2769 int
2770 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio,
2771     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2772 {
2773 	struct sockbuf *sb;
2774 	int error, flags;
2775 
2776 	sb = &so->so_rcv;
2777 
2778 	/* We only do stream sockets. */
2779 	if (so->so_type != SOCK_STREAM)
2780 		return (EINVAL);
2781 	if (psa != NULL)
2782 		*psa = NULL;
2783 	if (flagsp != NULL)
2784 		flags = *flagsp & ~MSG_EOR;
2785 	else
2786 		flags = 0;
2787 	if (controlp != NULL)
2788 		*controlp = NULL;
2789 	if (flags & MSG_OOB)
2790 		return (soreceive_rcvoob(so, uio, flags));
2791 	if (mp0 != NULL)
2792 		*mp0 = NULL;
2793 
2794 #ifdef KERN_TLS
2795 	/*
2796 	 * KTLS store TLS records as records with a control message to
2797 	 * describe the framing.
2798 	 *
2799 	 * We check once here before acquiring locks to optimize the
2800 	 * common case.
2801 	 */
2802 	if (sb->sb_tls_info != NULL)
2803 		return (soreceive_generic(so, psa, uio, mp0, controlp,
2804 		    flagsp));
2805 #endif
2806 
2807 	/*
2808 	 * Prevent other threads from reading from the socket.  This lock may be
2809 	 * dropped in order to sleep waiting for data to arrive.
2810 	 */
2811 	error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags));
2812 	if (error)
2813 		return (error);
2814 #ifdef KERN_TLS
2815 	if (__predict_false(sb->sb_tls_info != NULL)) {
2816 		SOCK_IO_RECV_UNLOCK(so);
2817 		return (soreceive_generic(so, psa, uio, mp0, controlp,
2818 		    flagsp));
2819 	}
2820 #endif
2821 	error = soreceive_stream_locked(so, sb, psa, uio, mp0, controlp, flags);
2822 	SOCK_IO_RECV_UNLOCK(so);
2823 	return (error);
2824 }
2825 
2826 /*
2827  * Optimized version of soreceive() for simple datagram cases from userspace.
2828  * Unlike in the stream case, we're able to drop a datagram if copyout()
2829  * fails, and because we handle datagrams atomically, we don't need to use a
2830  * sleep lock to prevent I/O interlacing.
2831  */
2832 int
2833 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio,
2834     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2835 {
2836 	struct mbuf *m, *m2;
2837 	int flags, error;
2838 	ssize_t len;
2839 	struct protosw *pr = so->so_proto;
2840 	struct mbuf *nextrecord;
2841 
2842 	if (psa != NULL)
2843 		*psa = NULL;
2844 	if (controlp != NULL)
2845 		*controlp = NULL;
2846 	if (flagsp != NULL)
2847 		flags = *flagsp &~ MSG_EOR;
2848 	else
2849 		flags = 0;
2850 
2851 	/*
2852 	 * For any complicated cases, fall back to the full
2853 	 * soreceive_generic().
2854 	 */
2855 	if (mp0 != NULL || (flags & (MSG_PEEK | MSG_OOB | MSG_TRUNC)))
2856 		return (soreceive_generic(so, psa, uio, mp0, controlp,
2857 		    flagsp));
2858 
2859 	/*
2860 	 * Enforce restrictions on use.
2861 	 */
2862 	KASSERT((pr->pr_flags & PR_WANTRCVD) == 0,
2863 	    ("soreceive_dgram: wantrcvd"));
2864 	KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic"));
2865 	KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0,
2866 	    ("soreceive_dgram: SBS_RCVATMARK"));
2867 	KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0,
2868 	    ("soreceive_dgram: P_CONNREQUIRED"));
2869 
2870 	/*
2871 	 * Loop blocking while waiting for a datagram.
2872 	 */
2873 	SOCKBUF_LOCK(&so->so_rcv);
2874 	while ((m = so->so_rcv.sb_mb) == NULL) {
2875 		KASSERT(sbavail(&so->so_rcv) == 0,
2876 		    ("soreceive_dgram: sb_mb NULL but sbavail %u",
2877 		    sbavail(&so->so_rcv)));
2878 		if (so->so_error) {
2879 			error = so->so_error;
2880 			so->so_error = 0;
2881 			SOCKBUF_UNLOCK(&so->so_rcv);
2882 			return (error);
2883 		}
2884 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE ||
2885 		    uio->uio_resid == 0) {
2886 			SOCKBUF_UNLOCK(&so->so_rcv);
2887 			return (0);
2888 		}
2889 		if ((so->so_state & SS_NBIO) ||
2890 		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2891 			SOCKBUF_UNLOCK(&so->so_rcv);
2892 			return (EWOULDBLOCK);
2893 		}
2894 		SBLASTRECORDCHK(&so->so_rcv);
2895 		SBLASTMBUFCHK(&so->so_rcv);
2896 		error = sbwait(so, SO_RCV);
2897 		if (error) {
2898 			SOCKBUF_UNLOCK(&so->so_rcv);
2899 			return (error);
2900 		}
2901 	}
2902 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2903 
2904 	if (uio->uio_td)
2905 		uio->uio_td->td_ru.ru_msgrcv++;
2906 	SBLASTRECORDCHK(&so->so_rcv);
2907 	SBLASTMBUFCHK(&so->so_rcv);
2908 	nextrecord = m->m_nextpkt;
2909 	if (nextrecord == NULL) {
2910 		KASSERT(so->so_rcv.sb_lastrecord == m,
2911 		    ("soreceive_dgram: lastrecord != m"));
2912 	}
2913 
2914 	KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord,
2915 	    ("soreceive_dgram: m_nextpkt != nextrecord"));
2916 
2917 	/*
2918 	 * Pull 'm' and its chain off the front of the packet queue.
2919 	 */
2920 	so->so_rcv.sb_mb = NULL;
2921 	sockbuf_pushsync(&so->so_rcv, nextrecord);
2922 
2923 	/*
2924 	 * Walk 'm's chain and free that many bytes from the socket buffer.
2925 	 */
2926 	for (m2 = m; m2 != NULL; m2 = m2->m_next)
2927 		sbfree(&so->so_rcv, m2);
2928 
2929 	/*
2930 	 * Do a few last checks before we let go of the lock.
2931 	 */
2932 	SBLASTRECORDCHK(&so->so_rcv);
2933 	SBLASTMBUFCHK(&so->so_rcv);
2934 	SOCKBUF_UNLOCK(&so->so_rcv);
2935 
2936 	if (pr->pr_flags & PR_ADDR) {
2937 		KASSERT(m->m_type == MT_SONAME,
2938 		    ("m->m_type == %d", m->m_type));
2939 		if (psa != NULL)
2940 			*psa = sodupsockaddr(mtod(m, struct sockaddr *),
2941 			    M_WAITOK);
2942 		m = m_free(m);
2943 	}
2944 	KASSERT(m, ("%s: no data or control after soname", __func__));
2945 
2946 	/*
2947 	 * Packet to copyout() is now in 'm' and it is disconnected from the
2948 	 * queue.
2949 	 *
2950 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
2951 	 * in the first mbuf chain on the socket buffer.  We call into the
2952 	 * protocol to perform externalization (or freeing if controlp ==
2953 	 * NULL). In some cases there can be only MT_CONTROL mbufs without
2954 	 * MT_DATA mbufs.
2955 	 */
2956 	if (m->m_type == MT_CONTROL) {
2957 		struct mbuf *cm = NULL, *cmn;
2958 		struct mbuf **cme = &cm;
2959 
2960 		do {
2961 			m2 = m->m_next;
2962 			m->m_next = NULL;
2963 			*cme = m;
2964 			cme = &(*cme)->m_next;
2965 			m = m2;
2966 		} while (m != NULL && m->m_type == MT_CONTROL);
2967 		while (cm != NULL) {
2968 			cmn = cm->m_next;
2969 			cm->m_next = NULL;
2970 			if (pr->pr_domain->dom_externalize != NULL) {
2971 				error = (*pr->pr_domain->dom_externalize)
2972 				    (cm, controlp, flags);
2973 			} else if (controlp != NULL)
2974 				*controlp = cm;
2975 			else
2976 				m_freem(cm);
2977 			if (controlp != NULL) {
2978 				while (*controlp != NULL)
2979 					controlp = &(*controlp)->m_next;
2980 			}
2981 			cm = cmn;
2982 		}
2983 	}
2984 	KASSERT(m == NULL || m->m_type == MT_DATA,
2985 	    ("soreceive_dgram: !data"));
2986 	while (m != NULL && uio->uio_resid > 0) {
2987 		len = uio->uio_resid;
2988 		if (len > m->m_len)
2989 			len = m->m_len;
2990 		error = uiomove(mtod(m, char *), (int)len, uio);
2991 		if (error) {
2992 			m_freem(m);
2993 			return (error);
2994 		}
2995 		if (len == m->m_len)
2996 			m = m_free(m);
2997 		else {
2998 			m->m_data += len;
2999 			m->m_len -= len;
3000 		}
3001 	}
3002 	if (m != NULL) {
3003 		flags |= MSG_TRUNC;
3004 		m_freem(m);
3005 	}
3006 	if (flagsp != NULL)
3007 		*flagsp |= flags;
3008 	return (0);
3009 }
3010 
3011 int
3012 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
3013     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
3014 {
3015 	int error;
3016 
3017 	CURVNET_SET(so->so_vnet);
3018 	error = so->so_proto->pr_soreceive(so, psa, uio, mp0, controlp, flagsp);
3019 	CURVNET_RESTORE();
3020 	return (error);
3021 }
3022 
3023 int
3024 soshutdown(struct socket *so, enum shutdown_how how)
3025 {
3026 	int error;
3027 
3028 	CURVNET_SET(so->so_vnet);
3029 	error = so->so_proto->pr_shutdown(so, how);
3030 	CURVNET_RESTORE();
3031 
3032 	return (error);
3033 }
3034 
3035 /*
3036  * Used by several pr_shutdown implementations that use generic socket buffers.
3037  */
3038 void
3039 sorflush(struct socket *so)
3040 {
3041 	int error;
3042 
3043 	VNET_SO_ASSERT(so);
3044 
3045 	/*
3046 	 * Dislodge threads currently blocked in receive and wait to acquire
3047 	 * a lock against other simultaneous readers before clearing the
3048 	 * socket buffer.  Don't let our acquire be interrupted by a signal
3049 	 * despite any existing socket disposition on interruptable waiting.
3050 	 *
3051 	 * The SOCK_IO_RECV_LOCK() is important here as there some pr_soreceive
3052 	 * methods that read the top of the socket buffer without acquisition
3053 	 * of the socket buffer mutex, assuming that top of the buffer
3054 	 * exclusively belongs to the read(2) syscall.  This is handy when
3055 	 * performing MSG_PEEK.
3056 	 */
3057 	socantrcvmore(so);
3058 
3059 	error = SOCK_IO_RECV_LOCK(so, SBL_WAIT | SBL_NOINTR);
3060 	if (error != 0) {
3061 		KASSERT(SOLISTENING(so),
3062 		    ("%s: soiolock(%p) failed", __func__, so));
3063 		return;
3064 	}
3065 
3066 	sbrelease(so, SO_RCV);
3067 	SOCK_IO_RECV_UNLOCK(so);
3068 
3069 }
3070 
3071 #ifdef SOCKET_HHOOK
3072 /*
3073  * Wrapper for Socket established helper hook.
3074  * Parameters: socket, context of the hook point, hook id.
3075  */
3076 static inline int
3077 hhook_run_socket(struct socket *so, void *hctx, int32_t h_id)
3078 {
3079 	struct socket_hhook_data hhook_data = {
3080 		.so = so,
3081 		.hctx = hctx,
3082 		.m = NULL,
3083 		.status = 0
3084 	};
3085 
3086 	CURVNET_SET(so->so_vnet);
3087 	HHOOKS_RUN_IF(V_socket_hhh[h_id], &hhook_data, &so->osd);
3088 	CURVNET_RESTORE();
3089 
3090 	/* Ugly but needed, since hhooks return void for now */
3091 	return (hhook_data.status);
3092 }
3093 #endif
3094 
3095 /*
3096  * Perhaps this routine, and sooptcopyout(), below, ought to come in an
3097  * additional variant to handle the case where the option value needs to be
3098  * some kind of integer, but not a specific size.  In addition to their use
3099  * here, these functions are also called by the protocol-level pr_ctloutput()
3100  * routines.
3101  */
3102 int
3103 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
3104 {
3105 	size_t	valsize;
3106 
3107 	/*
3108 	 * If the user gives us more than we wanted, we ignore it, but if we
3109 	 * don't get the minimum length the caller wants, we return EINVAL.
3110 	 * On success, sopt->sopt_valsize is set to however much we actually
3111 	 * retrieved.
3112 	 */
3113 	if ((valsize = sopt->sopt_valsize) < minlen)
3114 		return EINVAL;
3115 	if (valsize > len)
3116 		sopt->sopt_valsize = valsize = len;
3117 
3118 	if (sopt->sopt_td != NULL)
3119 		return (copyin(sopt->sopt_val, buf, valsize));
3120 
3121 	bcopy(sopt->sopt_val, buf, valsize);
3122 	return (0);
3123 }
3124 
3125 /*
3126  * Kernel version of setsockopt(2).
3127  *
3128  * XXX: optlen is size_t, not socklen_t
3129  */
3130 int
3131 so_setsockopt(struct socket *so, int level, int optname, void *optval,
3132     size_t optlen)
3133 {
3134 	struct sockopt sopt;
3135 
3136 	sopt.sopt_level = level;
3137 	sopt.sopt_name = optname;
3138 	sopt.sopt_dir = SOPT_SET;
3139 	sopt.sopt_val = optval;
3140 	sopt.sopt_valsize = optlen;
3141 	sopt.sopt_td = NULL;
3142 	return (sosetopt(so, &sopt));
3143 }
3144 
3145 int
3146 sosetopt(struct socket *so, struct sockopt *sopt)
3147 {
3148 	int	error, optval;
3149 	struct	linger l;
3150 	struct	timeval tv;
3151 	sbintime_t val, *valp;
3152 	uint32_t val32;
3153 #ifdef MAC
3154 	struct mac extmac;
3155 #endif
3156 
3157 	CURVNET_SET(so->so_vnet);
3158 	error = 0;
3159 	if (sopt->sopt_level != SOL_SOCKET) {
3160 		if (so->so_proto->pr_ctloutput != NULL)
3161 			error = (*so->so_proto->pr_ctloutput)(so, sopt);
3162 		else
3163 			error = ENOPROTOOPT;
3164 	} else {
3165 		switch (sopt->sopt_name) {
3166 		case SO_ACCEPTFILTER:
3167 			error = accept_filt_setopt(so, sopt);
3168 			if (error)
3169 				goto bad;
3170 			break;
3171 
3172 		case SO_LINGER:
3173 			error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
3174 			if (error)
3175 				goto bad;
3176 			if (l.l_linger < 0 ||
3177 			    l.l_linger > USHRT_MAX ||
3178 			    l.l_linger > (INT_MAX / hz)) {
3179 				error = EDOM;
3180 				goto bad;
3181 			}
3182 			SOCK_LOCK(so);
3183 			so->so_linger = l.l_linger;
3184 			if (l.l_onoff)
3185 				so->so_options |= SO_LINGER;
3186 			else
3187 				so->so_options &= ~SO_LINGER;
3188 			SOCK_UNLOCK(so);
3189 			break;
3190 
3191 		case SO_DEBUG:
3192 		case SO_KEEPALIVE:
3193 		case SO_DONTROUTE:
3194 		case SO_USELOOPBACK:
3195 		case SO_BROADCAST:
3196 		case SO_REUSEADDR:
3197 		case SO_REUSEPORT:
3198 		case SO_REUSEPORT_LB:
3199 		case SO_OOBINLINE:
3200 		case SO_TIMESTAMP:
3201 		case SO_BINTIME:
3202 		case SO_NOSIGPIPE:
3203 		case SO_NO_DDP:
3204 		case SO_NO_OFFLOAD:
3205 		case SO_RERROR:
3206 			error = sooptcopyin(sopt, &optval, sizeof optval,
3207 			    sizeof optval);
3208 			if (error)
3209 				goto bad;
3210 			SOCK_LOCK(so);
3211 			if (optval)
3212 				so->so_options |= sopt->sopt_name;
3213 			else
3214 				so->so_options &= ~sopt->sopt_name;
3215 			SOCK_UNLOCK(so);
3216 			break;
3217 
3218 		case SO_SETFIB:
3219 			error = sooptcopyin(sopt, &optval, sizeof optval,
3220 			    sizeof optval);
3221 			if (error)
3222 				goto bad;
3223 
3224 			if (optval < 0 || optval >= rt_numfibs) {
3225 				error = EINVAL;
3226 				goto bad;
3227 			}
3228 			if (((so->so_proto->pr_domain->dom_family == PF_INET) ||
3229 			   (so->so_proto->pr_domain->dom_family == PF_INET6) ||
3230 			   (so->so_proto->pr_domain->dom_family == PF_ROUTE)))
3231 				so->so_fibnum = optval;
3232 			else
3233 				so->so_fibnum = 0;
3234 			break;
3235 
3236 		case SO_USER_COOKIE:
3237 			error = sooptcopyin(sopt, &val32, sizeof val32,
3238 			    sizeof val32);
3239 			if (error)
3240 				goto bad;
3241 			so->so_user_cookie = val32;
3242 			break;
3243 
3244 		case SO_SNDBUF:
3245 		case SO_RCVBUF:
3246 		case SO_SNDLOWAT:
3247 		case SO_RCVLOWAT:
3248 			error = so->so_proto->pr_setsbopt(so, sopt);
3249 			if (error)
3250 				goto bad;
3251 			break;
3252 
3253 		case SO_SNDTIMEO:
3254 		case SO_RCVTIMEO:
3255 #ifdef COMPAT_FREEBSD32
3256 			if (SV_CURPROC_FLAG(SV_ILP32)) {
3257 				struct timeval32 tv32;
3258 
3259 				error = sooptcopyin(sopt, &tv32, sizeof tv32,
3260 				    sizeof tv32);
3261 				CP(tv32, tv, tv_sec);
3262 				CP(tv32, tv, tv_usec);
3263 			} else
3264 #endif
3265 				error = sooptcopyin(sopt, &tv, sizeof tv,
3266 				    sizeof tv);
3267 			if (error)
3268 				goto bad;
3269 			if (tv.tv_sec < 0 || tv.tv_usec < 0 ||
3270 			    tv.tv_usec >= 1000000) {
3271 				error = EDOM;
3272 				goto bad;
3273 			}
3274 			if (tv.tv_sec > INT32_MAX)
3275 				val = SBT_MAX;
3276 			else
3277 				val = tvtosbt(tv);
3278 			SOCK_LOCK(so);
3279 			valp = sopt->sopt_name == SO_SNDTIMEO ?
3280 			    (SOLISTENING(so) ? &so->sol_sbsnd_timeo :
3281 			    &so->so_snd.sb_timeo) :
3282 			    (SOLISTENING(so) ? &so->sol_sbrcv_timeo :
3283 			    &so->so_rcv.sb_timeo);
3284 			*valp = val;
3285 			SOCK_UNLOCK(so);
3286 			break;
3287 
3288 		case SO_LABEL:
3289 #ifdef MAC
3290 			error = sooptcopyin(sopt, &extmac, sizeof extmac,
3291 			    sizeof extmac);
3292 			if (error)
3293 				goto bad;
3294 			error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
3295 			    so, &extmac);
3296 #else
3297 			error = EOPNOTSUPP;
3298 #endif
3299 			break;
3300 
3301 		case SO_TS_CLOCK:
3302 			error = sooptcopyin(sopt, &optval, sizeof optval,
3303 			    sizeof optval);
3304 			if (error)
3305 				goto bad;
3306 			if (optval < 0 || optval > SO_TS_CLOCK_MAX) {
3307 				error = EINVAL;
3308 				goto bad;
3309 			}
3310 			so->so_ts_clock = optval;
3311 			break;
3312 
3313 		case SO_MAX_PACING_RATE:
3314 			error = sooptcopyin(sopt, &val32, sizeof(val32),
3315 			    sizeof(val32));
3316 			if (error)
3317 				goto bad;
3318 			so->so_max_pacing_rate = val32;
3319 			break;
3320 
3321 		default:
3322 #ifdef SOCKET_HHOOK
3323 			if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
3324 				error = hhook_run_socket(so, sopt,
3325 				    HHOOK_SOCKET_OPT);
3326 			else
3327 #endif
3328 				error = ENOPROTOOPT;
3329 			break;
3330 		}
3331 		if (error == 0 && so->so_proto->pr_ctloutput != NULL)
3332 			(void)(*so->so_proto->pr_ctloutput)(so, sopt);
3333 	}
3334 bad:
3335 	CURVNET_RESTORE();
3336 	return (error);
3337 }
3338 
3339 /*
3340  * Helper routine for getsockopt.
3341  */
3342 int
3343 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
3344 {
3345 	int	error;
3346 	size_t	valsize;
3347 
3348 	error = 0;
3349 
3350 	/*
3351 	 * Documented get behavior is that we always return a value, possibly
3352 	 * truncated to fit in the user's buffer.  Traditional behavior is
3353 	 * that we always tell the user precisely how much we copied, rather
3354 	 * than something useful like the total amount we had available for
3355 	 * her.  Note that this interface is not idempotent; the entire
3356 	 * answer must be generated ahead of time.
3357 	 */
3358 	valsize = min(len, sopt->sopt_valsize);
3359 	sopt->sopt_valsize = valsize;
3360 	if (sopt->sopt_val != NULL) {
3361 		if (sopt->sopt_td != NULL)
3362 			error = copyout(buf, sopt->sopt_val, valsize);
3363 		else
3364 			bcopy(buf, sopt->sopt_val, valsize);
3365 	}
3366 	return (error);
3367 }
3368 
3369 int
3370 sogetopt(struct socket *so, struct sockopt *sopt)
3371 {
3372 	int	error, optval;
3373 	struct	linger l;
3374 	struct	timeval tv;
3375 #ifdef MAC
3376 	struct mac extmac;
3377 #endif
3378 
3379 	CURVNET_SET(so->so_vnet);
3380 	error = 0;
3381 	if (sopt->sopt_level != SOL_SOCKET) {
3382 		if (so->so_proto->pr_ctloutput != NULL)
3383 			error = (*so->so_proto->pr_ctloutput)(so, sopt);
3384 		else
3385 			error = ENOPROTOOPT;
3386 		CURVNET_RESTORE();
3387 		return (error);
3388 	} else {
3389 		switch (sopt->sopt_name) {
3390 		case SO_ACCEPTFILTER:
3391 			error = accept_filt_getopt(so, sopt);
3392 			break;
3393 
3394 		case SO_LINGER:
3395 			SOCK_LOCK(so);
3396 			l.l_onoff = so->so_options & SO_LINGER;
3397 			l.l_linger = so->so_linger;
3398 			SOCK_UNLOCK(so);
3399 			error = sooptcopyout(sopt, &l, sizeof l);
3400 			break;
3401 
3402 		case SO_USELOOPBACK:
3403 		case SO_DONTROUTE:
3404 		case SO_DEBUG:
3405 		case SO_KEEPALIVE:
3406 		case SO_REUSEADDR:
3407 		case SO_REUSEPORT:
3408 		case SO_REUSEPORT_LB:
3409 		case SO_BROADCAST:
3410 		case SO_OOBINLINE:
3411 		case SO_ACCEPTCONN:
3412 		case SO_TIMESTAMP:
3413 		case SO_BINTIME:
3414 		case SO_NOSIGPIPE:
3415 		case SO_NO_DDP:
3416 		case SO_NO_OFFLOAD:
3417 		case SO_RERROR:
3418 			optval = so->so_options & sopt->sopt_name;
3419 integer:
3420 			error = sooptcopyout(sopt, &optval, sizeof optval);
3421 			break;
3422 
3423 		case SO_DOMAIN:
3424 			optval = so->so_proto->pr_domain->dom_family;
3425 			goto integer;
3426 
3427 		case SO_TYPE:
3428 			optval = so->so_type;
3429 			goto integer;
3430 
3431 		case SO_PROTOCOL:
3432 			optval = so->so_proto->pr_protocol;
3433 			goto integer;
3434 
3435 		case SO_ERROR:
3436 			SOCK_LOCK(so);
3437 			if (so->so_error) {
3438 				optval = so->so_error;
3439 				so->so_error = 0;
3440 			} else {
3441 				optval = so->so_rerror;
3442 				so->so_rerror = 0;
3443 			}
3444 			SOCK_UNLOCK(so);
3445 			goto integer;
3446 
3447 		case SO_SNDBUF:
3448 			optval = SOLISTENING(so) ? so->sol_sbsnd_hiwat :
3449 			    so->so_snd.sb_hiwat;
3450 			goto integer;
3451 
3452 		case SO_RCVBUF:
3453 			optval = SOLISTENING(so) ? so->sol_sbrcv_hiwat :
3454 			    so->so_rcv.sb_hiwat;
3455 			goto integer;
3456 
3457 		case SO_SNDLOWAT:
3458 			optval = SOLISTENING(so) ? so->sol_sbsnd_lowat :
3459 			    so->so_snd.sb_lowat;
3460 			goto integer;
3461 
3462 		case SO_RCVLOWAT:
3463 			optval = SOLISTENING(so) ? so->sol_sbrcv_lowat :
3464 			    so->so_rcv.sb_lowat;
3465 			goto integer;
3466 
3467 		case SO_SNDTIMEO:
3468 		case SO_RCVTIMEO:
3469 			SOCK_LOCK(so);
3470 			tv = sbttotv(sopt->sopt_name == SO_SNDTIMEO ?
3471 			    (SOLISTENING(so) ? so->sol_sbsnd_timeo :
3472 			    so->so_snd.sb_timeo) :
3473 			    (SOLISTENING(so) ? so->sol_sbrcv_timeo :
3474 			    so->so_rcv.sb_timeo));
3475 			SOCK_UNLOCK(so);
3476 #ifdef COMPAT_FREEBSD32
3477 			if (SV_CURPROC_FLAG(SV_ILP32)) {
3478 				struct timeval32 tv32;
3479 
3480 				CP(tv, tv32, tv_sec);
3481 				CP(tv, tv32, tv_usec);
3482 				error = sooptcopyout(sopt, &tv32, sizeof tv32);
3483 			} else
3484 #endif
3485 				error = sooptcopyout(sopt, &tv, sizeof tv);
3486 			break;
3487 
3488 		case SO_LABEL:
3489 #ifdef MAC
3490 			error = sooptcopyin(sopt, &extmac, sizeof(extmac),
3491 			    sizeof(extmac));
3492 			if (error)
3493 				goto bad;
3494 			error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
3495 			    so, &extmac);
3496 			if (error)
3497 				goto bad;
3498 			/* Don't copy out extmac, it is unchanged. */
3499 #else
3500 			error = EOPNOTSUPP;
3501 #endif
3502 			break;
3503 
3504 		case SO_PEERLABEL:
3505 #ifdef MAC
3506 			error = sooptcopyin(sopt, &extmac, sizeof(extmac),
3507 			    sizeof(extmac));
3508 			if (error)
3509 				goto bad;
3510 			error = mac_getsockopt_peerlabel(
3511 			    sopt->sopt_td->td_ucred, so, &extmac);
3512 			if (error)
3513 				goto bad;
3514 			/* Don't copy out extmac, it is unchanged. */
3515 #else
3516 			error = EOPNOTSUPP;
3517 #endif
3518 			break;
3519 
3520 		case SO_LISTENQLIMIT:
3521 			optval = SOLISTENING(so) ? so->sol_qlimit : 0;
3522 			goto integer;
3523 
3524 		case SO_LISTENQLEN:
3525 			optval = SOLISTENING(so) ? so->sol_qlen : 0;
3526 			goto integer;
3527 
3528 		case SO_LISTENINCQLEN:
3529 			optval = SOLISTENING(so) ? so->sol_incqlen : 0;
3530 			goto integer;
3531 
3532 		case SO_TS_CLOCK:
3533 			optval = so->so_ts_clock;
3534 			goto integer;
3535 
3536 		case SO_MAX_PACING_RATE:
3537 			optval = so->so_max_pacing_rate;
3538 			goto integer;
3539 
3540 		default:
3541 #ifdef SOCKET_HHOOK
3542 			if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
3543 				error = hhook_run_socket(so, sopt,
3544 				    HHOOK_SOCKET_OPT);
3545 			else
3546 #endif
3547 				error = ENOPROTOOPT;
3548 			break;
3549 		}
3550 	}
3551 #ifdef MAC
3552 bad:
3553 #endif
3554 	CURVNET_RESTORE();
3555 	return (error);
3556 }
3557 
3558 int
3559 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
3560 {
3561 	struct mbuf *m, *m_prev;
3562 	int sopt_size = sopt->sopt_valsize;
3563 
3564 	MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
3565 	if (m == NULL)
3566 		return ENOBUFS;
3567 	if (sopt_size > MLEN) {
3568 		MCLGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT);
3569 		if ((m->m_flags & M_EXT) == 0) {
3570 			m_free(m);
3571 			return ENOBUFS;
3572 		}
3573 		m->m_len = min(MCLBYTES, sopt_size);
3574 	} else {
3575 		m->m_len = min(MLEN, sopt_size);
3576 	}
3577 	sopt_size -= m->m_len;
3578 	*mp = m;
3579 	m_prev = m;
3580 
3581 	while (sopt_size) {
3582 		MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
3583 		if (m == NULL) {
3584 			m_freem(*mp);
3585 			return ENOBUFS;
3586 		}
3587 		if (sopt_size > MLEN) {
3588 			MCLGET(m, sopt->sopt_td != NULL ? M_WAITOK :
3589 			    M_NOWAIT);
3590 			if ((m->m_flags & M_EXT) == 0) {
3591 				m_freem(m);
3592 				m_freem(*mp);
3593 				return ENOBUFS;
3594 			}
3595 			m->m_len = min(MCLBYTES, sopt_size);
3596 		} else {
3597 			m->m_len = min(MLEN, sopt_size);
3598 		}
3599 		sopt_size -= m->m_len;
3600 		m_prev->m_next = m;
3601 		m_prev = m;
3602 	}
3603 	return (0);
3604 }
3605 
3606 int
3607 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
3608 {
3609 	struct mbuf *m0 = m;
3610 
3611 	if (sopt->sopt_val == NULL)
3612 		return (0);
3613 	while (m != NULL && sopt->sopt_valsize >= m->m_len) {
3614 		if (sopt->sopt_td != NULL) {
3615 			int error;
3616 
3617 			error = copyin(sopt->sopt_val, mtod(m, char *),
3618 			    m->m_len);
3619 			if (error != 0) {
3620 				m_freem(m0);
3621 				return(error);
3622 			}
3623 		} else
3624 			bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
3625 		sopt->sopt_valsize -= m->m_len;
3626 		sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
3627 		m = m->m_next;
3628 	}
3629 	if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
3630 		panic("ip6_sooptmcopyin");
3631 	return (0);
3632 }
3633 
3634 int
3635 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
3636 {
3637 	struct mbuf *m0 = m;
3638 	size_t valsize = 0;
3639 
3640 	if (sopt->sopt_val == NULL)
3641 		return (0);
3642 	while (m != NULL && sopt->sopt_valsize >= m->m_len) {
3643 		if (sopt->sopt_td != NULL) {
3644 			int error;
3645 
3646 			error = copyout(mtod(m, char *), sopt->sopt_val,
3647 			    m->m_len);
3648 			if (error != 0) {
3649 				m_freem(m0);
3650 				return(error);
3651 			}
3652 		} else
3653 			bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
3654 		sopt->sopt_valsize -= m->m_len;
3655 		sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
3656 		valsize += m->m_len;
3657 		m = m->m_next;
3658 	}
3659 	if (m != NULL) {
3660 		/* enough soopt buffer should be given from user-land */
3661 		m_freem(m0);
3662 		return(EINVAL);
3663 	}
3664 	sopt->sopt_valsize = valsize;
3665 	return (0);
3666 }
3667 
3668 /*
3669  * sohasoutofband(): protocol notifies socket layer of the arrival of new
3670  * out-of-band data, which will then notify socket consumers.
3671  */
3672 void
3673 sohasoutofband(struct socket *so)
3674 {
3675 
3676 	if (so->so_sigio != NULL)
3677 		pgsigio(&so->so_sigio, SIGURG, 0);
3678 	selwakeuppri(&so->so_rdsel, PSOCK);
3679 }
3680 
3681 int
3682 sopoll(struct socket *so, int events, struct ucred *active_cred,
3683     struct thread *td)
3684 {
3685 
3686 	/*
3687 	 * We do not need to set or assert curvnet as long as everyone uses
3688 	 * sopoll_generic().
3689 	 */
3690 	return (so->so_proto->pr_sopoll(so, events, active_cred, td));
3691 }
3692 
3693 int
3694 sopoll_generic(struct socket *so, int events, struct ucred *active_cred,
3695     struct thread *td)
3696 {
3697 	int revents;
3698 
3699 	SOCK_LOCK(so);
3700 	if (SOLISTENING(so)) {
3701 		if (!(events & (POLLIN | POLLRDNORM)))
3702 			revents = 0;
3703 		else if (!TAILQ_EMPTY(&so->sol_comp))
3704 			revents = events & (POLLIN | POLLRDNORM);
3705 		else if ((events & POLLINIGNEOF) == 0 && so->so_error)
3706 			revents = (events & (POLLIN | POLLRDNORM)) | POLLHUP;
3707 		else {
3708 			selrecord(td, &so->so_rdsel);
3709 			revents = 0;
3710 		}
3711 	} else {
3712 		revents = 0;
3713 		SOCK_SENDBUF_LOCK(so);
3714 		SOCK_RECVBUF_LOCK(so);
3715 		if (events & (POLLIN | POLLRDNORM))
3716 			if (soreadabledata(so))
3717 				revents |= events & (POLLIN | POLLRDNORM);
3718 		if (events & (POLLOUT | POLLWRNORM))
3719 			if (sowriteable(so))
3720 				revents |= events & (POLLOUT | POLLWRNORM);
3721 		if (events & (POLLPRI | POLLRDBAND))
3722 			if (so->so_oobmark ||
3723 			    (so->so_rcv.sb_state & SBS_RCVATMARK))
3724 				revents |= events & (POLLPRI | POLLRDBAND);
3725 		if ((events & POLLINIGNEOF) == 0) {
3726 			if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3727 				revents |= events & (POLLIN | POLLRDNORM);
3728 				if (so->so_snd.sb_state & SBS_CANTSENDMORE)
3729 					revents |= POLLHUP;
3730 			}
3731 		}
3732 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
3733 			revents |= events & POLLRDHUP;
3734 		if (revents == 0) {
3735 			if (events &
3736 			    (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND | POLLRDHUP)) {
3737 				selrecord(td, &so->so_rdsel);
3738 				so->so_rcv.sb_flags |= SB_SEL;
3739 			}
3740 			if (events & (POLLOUT | POLLWRNORM)) {
3741 				selrecord(td, &so->so_wrsel);
3742 				so->so_snd.sb_flags |= SB_SEL;
3743 			}
3744 		}
3745 		SOCK_RECVBUF_UNLOCK(so);
3746 		SOCK_SENDBUF_UNLOCK(so);
3747 	}
3748 	SOCK_UNLOCK(so);
3749 	return (revents);
3750 }
3751 
3752 int
3753 soo_kqfilter(struct file *fp, struct knote *kn)
3754 {
3755 	struct socket *so = kn->kn_fp->f_data;
3756 	struct sockbuf *sb;
3757 	sb_which which;
3758 	struct knlist *knl;
3759 
3760 	switch (kn->kn_filter) {
3761 	case EVFILT_READ:
3762 		kn->kn_fop = &soread_filtops;
3763 		knl = &so->so_rdsel.si_note;
3764 		sb = &so->so_rcv;
3765 		which = SO_RCV;
3766 		break;
3767 	case EVFILT_WRITE:
3768 		kn->kn_fop = &sowrite_filtops;
3769 		knl = &so->so_wrsel.si_note;
3770 		sb = &so->so_snd;
3771 		which = SO_SND;
3772 		break;
3773 	case EVFILT_EMPTY:
3774 		kn->kn_fop = &soempty_filtops;
3775 		knl = &so->so_wrsel.si_note;
3776 		sb = &so->so_snd;
3777 		which = SO_SND;
3778 		break;
3779 	default:
3780 		return (EINVAL);
3781 	}
3782 
3783 	SOCK_LOCK(so);
3784 	if (SOLISTENING(so)) {
3785 		knlist_add(knl, kn, 1);
3786 	} else {
3787 		SOCK_BUF_LOCK(so, which);
3788 		knlist_add(knl, kn, 1);
3789 		sb->sb_flags |= SB_KNOTE;
3790 		SOCK_BUF_UNLOCK(so, which);
3791 	}
3792 	SOCK_UNLOCK(so);
3793 	return (0);
3794 }
3795 
3796 static void
3797 filt_sordetach(struct knote *kn)
3798 {
3799 	struct socket *so = kn->kn_fp->f_data;
3800 
3801 	so_rdknl_lock(so);
3802 	knlist_remove(&so->so_rdsel.si_note, kn, 1);
3803 	if (!SOLISTENING(so) && knlist_empty(&so->so_rdsel.si_note))
3804 		so->so_rcv.sb_flags &= ~SB_KNOTE;
3805 	so_rdknl_unlock(so);
3806 }
3807 
3808 /*ARGSUSED*/
3809 static int
3810 filt_soread(struct knote *kn, long hint)
3811 {
3812 	struct socket *so;
3813 
3814 	so = kn->kn_fp->f_data;
3815 
3816 	if (SOLISTENING(so)) {
3817 		SOCK_LOCK_ASSERT(so);
3818 		kn->kn_data = so->sol_qlen;
3819 		if (so->so_error) {
3820 			kn->kn_flags |= EV_EOF;
3821 			kn->kn_fflags = so->so_error;
3822 			return (1);
3823 		}
3824 		return (!TAILQ_EMPTY(&so->sol_comp));
3825 	}
3826 
3827 	SOCK_RECVBUF_LOCK_ASSERT(so);
3828 
3829 	kn->kn_data = sbavail(&so->so_rcv) - so->so_rcv.sb_ctl;
3830 	if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3831 		kn->kn_flags |= EV_EOF;
3832 		kn->kn_fflags = so->so_error;
3833 		return (1);
3834 	} else if (so->so_error || so->so_rerror)
3835 		return (1);
3836 
3837 	if (kn->kn_sfflags & NOTE_LOWAT) {
3838 		if (kn->kn_data >= kn->kn_sdata)
3839 			return (1);
3840 	} else if (sbavail(&so->so_rcv) >= so->so_rcv.sb_lowat)
3841 		return (1);
3842 
3843 #ifdef SOCKET_HHOOK
3844 	/* This hook returning non-zero indicates an event, not error */
3845 	return (hhook_run_socket(so, NULL, HHOOK_FILT_SOREAD));
3846 #else
3847 	return (0);
3848 #endif
3849 }
3850 
3851 static void
3852 filt_sowdetach(struct knote *kn)
3853 {
3854 	struct socket *so = kn->kn_fp->f_data;
3855 
3856 	so_wrknl_lock(so);
3857 	knlist_remove(&so->so_wrsel.si_note, kn, 1);
3858 	if (!SOLISTENING(so) && knlist_empty(&so->so_wrsel.si_note))
3859 		so->so_snd.sb_flags &= ~SB_KNOTE;
3860 	so_wrknl_unlock(so);
3861 }
3862 
3863 /*ARGSUSED*/
3864 static int
3865 filt_sowrite(struct knote *kn, long hint)
3866 {
3867 	struct socket *so;
3868 
3869 	so = kn->kn_fp->f_data;
3870 
3871 	if (SOLISTENING(so))
3872 		return (0);
3873 
3874 	SOCK_SENDBUF_LOCK_ASSERT(so);
3875 	kn->kn_data = sbspace(&so->so_snd);
3876 
3877 #ifdef SOCKET_HHOOK
3878 	hhook_run_socket(so, kn, HHOOK_FILT_SOWRITE);
3879 #endif
3880 
3881 	if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
3882 		kn->kn_flags |= EV_EOF;
3883 		kn->kn_fflags = so->so_error;
3884 		return (1);
3885 	} else if (so->so_error)	/* temporary udp error */
3886 		return (1);
3887 	else if (((so->so_state & SS_ISCONNECTED) == 0) &&
3888 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
3889 		return (0);
3890 	else if (kn->kn_sfflags & NOTE_LOWAT)
3891 		return (kn->kn_data >= kn->kn_sdata);
3892 	else
3893 		return (kn->kn_data >= so->so_snd.sb_lowat);
3894 }
3895 
3896 static int
3897 filt_soempty(struct knote *kn, long hint)
3898 {
3899 	struct socket *so;
3900 
3901 	so = kn->kn_fp->f_data;
3902 
3903 	if (SOLISTENING(so))
3904 		return (1);
3905 
3906 	SOCK_SENDBUF_LOCK_ASSERT(so);
3907 	kn->kn_data = sbused(&so->so_snd);
3908 
3909 	if (kn->kn_data == 0)
3910 		return (1);
3911 	else
3912 		return (0);
3913 }
3914 
3915 int
3916 socheckuid(struct socket *so, uid_t uid)
3917 {
3918 
3919 	if (so == NULL)
3920 		return (EPERM);
3921 	if (so->so_cred->cr_uid != uid)
3922 		return (EPERM);
3923 	return (0);
3924 }
3925 
3926 /*
3927  * These functions are used by protocols to notify the socket layer (and its
3928  * consumers) of state changes in the sockets driven by protocol-side events.
3929  */
3930 
3931 /*
3932  * Procedures to manipulate state flags of socket and do appropriate wakeups.
3933  *
3934  * Normal sequence from the active (originating) side is that
3935  * soisconnecting() is called during processing of connect() call, resulting
3936  * in an eventual call to soisconnected() if/when the connection is
3937  * established.  When the connection is torn down soisdisconnecting() is
3938  * called during processing of disconnect() call, and soisdisconnected() is
3939  * called when the connection to the peer is totally severed.  The semantics
3940  * of these routines are such that connectionless protocols can call
3941  * soisconnected() and soisdisconnected() only, bypassing the in-progress
3942  * calls when setting up a ``connection'' takes no time.
3943  *
3944  * From the passive side, a socket is created with two queues of sockets:
3945  * so_incomp for connections in progress and so_comp for connections already
3946  * made and awaiting user acceptance.  As a protocol is preparing incoming
3947  * connections, it creates a socket structure queued on so_incomp by calling
3948  * sonewconn().  When the connection is established, soisconnected() is
3949  * called, and transfers the socket structure to so_comp, making it available
3950  * to accept().
3951  *
3952  * If a socket is closed with sockets on either so_incomp or so_comp, these
3953  * sockets are dropped.
3954  *
3955  * If higher-level protocols are implemented in the kernel, the wakeups done
3956  * here will sometimes cause software-interrupt process scheduling.
3957  */
3958 void
3959 soisconnecting(struct socket *so)
3960 {
3961 
3962 	SOCK_LOCK(so);
3963 	so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
3964 	so->so_state |= SS_ISCONNECTING;
3965 	SOCK_UNLOCK(so);
3966 }
3967 
3968 void
3969 soisconnected(struct socket *so)
3970 {
3971 	bool last __diagused;
3972 
3973 	SOCK_LOCK(so);
3974 	so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING);
3975 	so->so_state |= SS_ISCONNECTED;
3976 
3977 	if (so->so_qstate == SQ_INCOMP) {
3978 		struct socket *head = so->so_listen;
3979 		int ret;
3980 
3981 		KASSERT(head, ("%s: so %p on incomp of NULL", __func__, so));
3982 		/*
3983 		 * Promoting a socket from incomplete queue to complete, we
3984 		 * need to go through reverse order of locking.  We first do
3985 		 * trylock, and if that doesn't succeed, we go the hard way
3986 		 * leaving a reference and rechecking consistency after proper
3987 		 * locking.
3988 		 */
3989 		if (__predict_false(SOLISTEN_TRYLOCK(head) == 0)) {
3990 			soref(head);
3991 			SOCK_UNLOCK(so);
3992 			SOLISTEN_LOCK(head);
3993 			SOCK_LOCK(so);
3994 			if (__predict_false(head != so->so_listen)) {
3995 				/*
3996 				 * The socket went off the listen queue,
3997 				 * should be lost race to close(2) of sol.
3998 				 * The socket is about to soabort().
3999 				 */
4000 				SOCK_UNLOCK(so);
4001 				sorele_locked(head);
4002 				return;
4003 			}
4004 			last = refcount_release(&head->so_count);
4005 			KASSERT(!last, ("%s: released last reference for %p",
4006 			    __func__, head));
4007 		}
4008 again:
4009 		if ((so->so_options & SO_ACCEPTFILTER) == 0) {
4010 			TAILQ_REMOVE(&head->sol_incomp, so, so_list);
4011 			head->sol_incqlen--;
4012 			TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list);
4013 			head->sol_qlen++;
4014 			so->so_qstate = SQ_COMP;
4015 			SOCK_UNLOCK(so);
4016 			solisten_wakeup(head);	/* unlocks */
4017 		} else {
4018 			SOCK_RECVBUF_LOCK(so);
4019 			soupcall_set(so, SO_RCV,
4020 			    head->sol_accept_filter->accf_callback,
4021 			    head->sol_accept_filter_arg);
4022 			so->so_options &= ~SO_ACCEPTFILTER;
4023 			ret = head->sol_accept_filter->accf_callback(so,
4024 			    head->sol_accept_filter_arg, M_NOWAIT);
4025 			if (ret == SU_ISCONNECTED) {
4026 				soupcall_clear(so, SO_RCV);
4027 				SOCK_RECVBUF_UNLOCK(so);
4028 				goto again;
4029 			}
4030 			SOCK_RECVBUF_UNLOCK(so);
4031 			SOCK_UNLOCK(so);
4032 			SOLISTEN_UNLOCK(head);
4033 		}
4034 		return;
4035 	}
4036 	SOCK_UNLOCK(so);
4037 	wakeup(&so->so_timeo);
4038 	sorwakeup(so);
4039 	sowwakeup(so);
4040 }
4041 
4042 void
4043 soisdisconnecting(struct socket *so)
4044 {
4045 
4046 	SOCK_LOCK(so);
4047 	so->so_state &= ~SS_ISCONNECTING;
4048 	so->so_state |= SS_ISDISCONNECTING;
4049 
4050 	if (!SOLISTENING(so)) {
4051 		SOCK_RECVBUF_LOCK(so);
4052 		socantrcvmore_locked(so);
4053 		SOCK_SENDBUF_LOCK(so);
4054 		socantsendmore_locked(so);
4055 	}
4056 	SOCK_UNLOCK(so);
4057 	wakeup(&so->so_timeo);
4058 }
4059 
4060 void
4061 soisdisconnected(struct socket *so)
4062 {
4063 
4064 	SOCK_LOCK(so);
4065 
4066 	/*
4067 	 * There is at least one reader of so_state that does not
4068 	 * acquire socket lock, namely soreceive_generic().  Ensure
4069 	 * that it never sees all flags that track connection status
4070 	 * cleared, by ordering the update with a barrier semantic of
4071 	 * our release thread fence.
4072 	 */
4073 	so->so_state |= SS_ISDISCONNECTED;
4074 	atomic_thread_fence_rel();
4075 	so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
4076 
4077 	if (!SOLISTENING(so)) {
4078 		SOCK_UNLOCK(so);
4079 		SOCK_RECVBUF_LOCK(so);
4080 		socantrcvmore_locked(so);
4081 		SOCK_SENDBUF_LOCK(so);
4082 		sbdrop_locked(&so->so_snd, sbused(&so->so_snd));
4083 		socantsendmore_locked(so);
4084 	} else
4085 		SOCK_UNLOCK(so);
4086 	wakeup(&so->so_timeo);
4087 }
4088 
4089 int
4090 soiolock(struct socket *so, struct sx *sx, int flags)
4091 {
4092 	int error;
4093 
4094 	KASSERT((flags & SBL_VALID) == flags,
4095 	    ("soiolock: invalid flags %#x", flags));
4096 
4097 	if ((flags & SBL_WAIT) != 0) {
4098 		if ((flags & SBL_NOINTR) != 0) {
4099 			sx_xlock(sx);
4100 		} else {
4101 			error = sx_xlock_sig(sx);
4102 			if (error != 0)
4103 				return (error);
4104 		}
4105 	} else if (!sx_try_xlock(sx)) {
4106 		return (EWOULDBLOCK);
4107 	}
4108 
4109 	if (__predict_false(SOLISTENING(so))) {
4110 		sx_xunlock(sx);
4111 		return (ENOTCONN);
4112 	}
4113 	return (0);
4114 }
4115 
4116 void
4117 soiounlock(struct sx *sx)
4118 {
4119 	sx_xunlock(sx);
4120 }
4121 
4122 /*
4123  * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
4124  */
4125 struct sockaddr *
4126 sodupsockaddr(const struct sockaddr *sa, int mflags)
4127 {
4128 	struct sockaddr *sa2;
4129 
4130 	sa2 = malloc(sa->sa_len, M_SONAME, mflags);
4131 	if (sa2)
4132 		bcopy(sa, sa2, sa->sa_len);
4133 	return sa2;
4134 }
4135 
4136 /*
4137  * Register per-socket destructor.
4138  */
4139 void
4140 sodtor_set(struct socket *so, so_dtor_t *func)
4141 {
4142 
4143 	SOCK_LOCK_ASSERT(so);
4144 	so->so_dtor = func;
4145 }
4146 
4147 /*
4148  * Register per-socket buffer upcalls.
4149  */
4150 void
4151 soupcall_set(struct socket *so, sb_which which, so_upcall_t func, void *arg)
4152 {
4153 	struct sockbuf *sb;
4154 
4155 	KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so));
4156 
4157 	switch (which) {
4158 	case SO_RCV:
4159 		sb = &so->so_rcv;
4160 		break;
4161 	case SO_SND:
4162 		sb = &so->so_snd;
4163 		break;
4164 	}
4165 	SOCK_BUF_LOCK_ASSERT(so, which);
4166 	sb->sb_upcall = func;
4167 	sb->sb_upcallarg = arg;
4168 	sb->sb_flags |= SB_UPCALL;
4169 }
4170 
4171 void
4172 soupcall_clear(struct socket *so, sb_which which)
4173 {
4174 	struct sockbuf *sb;
4175 
4176 	KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so));
4177 
4178 	switch (which) {
4179 	case SO_RCV:
4180 		sb = &so->so_rcv;
4181 		break;
4182 	case SO_SND:
4183 		sb = &so->so_snd;
4184 		break;
4185 	}
4186 	SOCK_BUF_LOCK_ASSERT(so, which);
4187 	KASSERT(sb->sb_upcall != NULL,
4188 	    ("%s: so %p no upcall to clear", __func__, so));
4189 	sb->sb_upcall = NULL;
4190 	sb->sb_upcallarg = NULL;
4191 	sb->sb_flags &= ~SB_UPCALL;
4192 }
4193 
4194 void
4195 solisten_upcall_set(struct socket *so, so_upcall_t func, void *arg)
4196 {
4197 
4198 	SOLISTEN_LOCK_ASSERT(so);
4199 	so->sol_upcall = func;
4200 	so->sol_upcallarg = arg;
4201 }
4202 
4203 static void
4204 so_rdknl_lock(void *arg)
4205 {
4206 	struct socket *so = arg;
4207 
4208 retry:
4209 	if (SOLISTENING(so)) {
4210 		SOLISTEN_LOCK(so);
4211 	} else {
4212 		SOCK_RECVBUF_LOCK(so);
4213 		if (__predict_false(SOLISTENING(so))) {
4214 			SOCK_RECVBUF_UNLOCK(so);
4215 			goto retry;
4216 		}
4217 	}
4218 }
4219 
4220 static void
4221 so_rdknl_unlock(void *arg)
4222 {
4223 	struct socket *so = arg;
4224 
4225 	if (SOLISTENING(so))
4226 		SOLISTEN_UNLOCK(so);
4227 	else
4228 		SOCK_RECVBUF_UNLOCK(so);
4229 }
4230 
4231 static void
4232 so_rdknl_assert_lock(void *arg, int what)
4233 {
4234 	struct socket *so = arg;
4235 
4236 	if (what == LA_LOCKED) {
4237 		if (SOLISTENING(so))
4238 			SOLISTEN_LOCK_ASSERT(so);
4239 		else
4240 			SOCK_RECVBUF_LOCK_ASSERT(so);
4241 	} else {
4242 		if (SOLISTENING(so))
4243 			SOLISTEN_UNLOCK_ASSERT(so);
4244 		else
4245 			SOCK_RECVBUF_UNLOCK_ASSERT(so);
4246 	}
4247 }
4248 
4249 static void
4250 so_wrknl_lock(void *arg)
4251 {
4252 	struct socket *so = arg;
4253 
4254 retry:
4255 	if (SOLISTENING(so)) {
4256 		SOLISTEN_LOCK(so);
4257 	} else {
4258 		SOCK_SENDBUF_LOCK(so);
4259 		if (__predict_false(SOLISTENING(so))) {
4260 			SOCK_SENDBUF_UNLOCK(so);
4261 			goto retry;
4262 		}
4263 	}
4264 }
4265 
4266 static void
4267 so_wrknl_unlock(void *arg)
4268 {
4269 	struct socket *so = arg;
4270 
4271 	if (SOLISTENING(so))
4272 		SOLISTEN_UNLOCK(so);
4273 	else
4274 		SOCK_SENDBUF_UNLOCK(so);
4275 }
4276 
4277 static void
4278 so_wrknl_assert_lock(void *arg, int what)
4279 {
4280 	struct socket *so = arg;
4281 
4282 	if (what == LA_LOCKED) {
4283 		if (SOLISTENING(so))
4284 			SOLISTEN_LOCK_ASSERT(so);
4285 		else
4286 			SOCK_SENDBUF_LOCK_ASSERT(so);
4287 	} else {
4288 		if (SOLISTENING(so))
4289 			SOLISTEN_UNLOCK_ASSERT(so);
4290 		else
4291 			SOCK_SENDBUF_UNLOCK_ASSERT(so);
4292 	}
4293 }
4294 
4295 /*
4296  * Create an external-format (``xsocket'') structure using the information in
4297  * the kernel-format socket structure pointed to by so.  This is done to
4298  * reduce the spew of irrelevant information over this interface, to isolate
4299  * user code from changes in the kernel structure, and potentially to provide
4300  * information-hiding if we decide that some of this information should be
4301  * hidden from users.
4302  */
4303 void
4304 sotoxsocket(struct socket *so, struct xsocket *xso)
4305 {
4306 
4307 	bzero(xso, sizeof(*xso));
4308 	xso->xso_len = sizeof *xso;
4309 	xso->xso_so = (uintptr_t)so;
4310 	xso->so_type = so->so_type;
4311 	xso->so_options = so->so_options;
4312 	xso->so_linger = so->so_linger;
4313 	xso->so_state = so->so_state;
4314 	xso->so_pcb = (uintptr_t)so->so_pcb;
4315 	xso->xso_protocol = so->so_proto->pr_protocol;
4316 	xso->xso_family = so->so_proto->pr_domain->dom_family;
4317 	xso->so_timeo = so->so_timeo;
4318 	xso->so_error = so->so_error;
4319 	xso->so_uid = so->so_cred->cr_uid;
4320 	xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
4321 	SOCK_LOCK(so);
4322 	if (SOLISTENING(so)) {
4323 		xso->so_qlen = so->sol_qlen;
4324 		xso->so_incqlen = so->sol_incqlen;
4325 		xso->so_qlimit = so->sol_qlimit;
4326 		xso->so_oobmark = 0;
4327 	} else {
4328 		xso->so_state |= so->so_qstate;
4329 		xso->so_qlen = xso->so_incqlen = xso->so_qlimit = 0;
4330 		xso->so_oobmark = so->so_oobmark;
4331 		sbtoxsockbuf(&so->so_snd, &xso->so_snd);
4332 		sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
4333 	}
4334 	SOCK_UNLOCK(so);
4335 }
4336 
4337 int
4338 so_options_get(const struct socket *so)
4339 {
4340 
4341 	return (so->so_options);
4342 }
4343 
4344 void
4345 so_options_set(struct socket *so, int val)
4346 {
4347 
4348 	so->so_options = val;
4349 }
4350 
4351 int
4352 so_error_get(const struct socket *so)
4353 {
4354 
4355 	return (so->so_error);
4356 }
4357 
4358 void
4359 so_error_set(struct socket *so, int val)
4360 {
4361 
4362 	so->so_error = val;
4363 }
4364